git/grep.c

2006 lines
49 KiB
C
Raw Normal View History

#include "git-compat-util.h"
#include "config.h"
#include "gettext.h"
#include "grep.h"
#include "hex.h"
#include "object-store-ll.h"
#include "pretty.h"
#include "userdiff.h"
#include "xdiff-interface.h"
#include "diff.h"
#include "diffcore.h"
#include "commit.h"
#include "quote.h"
#include "help.h"
static int grep_source_load(struct grep_source *gs);
static int grep_source_is_binary(struct grep_source *gs,
struct index_state *istate);
static void std_output(struct grep_opt *opt UNUSED, const void *buf, size_t size)
{
fwrite(buf, size, 1, stdout);
}
static const char *color_grep_slots[] = {
[GREP_COLOR_CONTEXT] = "context",
[GREP_COLOR_FILENAME] = "filename",
[GREP_COLOR_FUNCTION] = "function",
[GREP_COLOR_LINENO] = "lineNumber",
[GREP_COLOR_COLUMNNO] = "column",
[GREP_COLOR_MATCH_CONTEXT] = "matchContext",
[GREP_COLOR_MATCH_SELECTED] = "matchSelected",
[GREP_COLOR_SELECTED] = "selected",
[GREP_COLOR_SEP] = "separator",
};
static int parse_pattern_type_arg(const char *opt, const char *arg)
{
if (!strcmp(arg, "default"))
return GREP_PATTERN_TYPE_UNSPECIFIED;
else if (!strcmp(arg, "basic"))
return GREP_PATTERN_TYPE_BRE;
else if (!strcmp(arg, "extended"))
return GREP_PATTERN_TYPE_ERE;
else if (!strcmp(arg, "fixed"))
return GREP_PATTERN_TYPE_FIXED;
else if (!strcmp(arg, "perl"))
return GREP_PATTERN_TYPE_PCRE;
die("bad %s argument: %s", opt, arg);
}
define_list_config_array_extra(color_grep_slots, {"match"});
/*
* Read the configuration file once and store it in
* the grep_defaults template.
*/
config: add ctx arg to config_fn_t Add a new "const struct config_context *ctx" arg to config_fn_t to hold additional information about the config iteration operation. config_context has a "struct key_value_info kvi" member that holds metadata about the config source being read (e.g. what kind of config source it is, the filename, etc). In this series, we're only interested in .kvi, so we could have just used "struct key_value_info" as an arg, but config_context makes it possible to add/adjust members in the future without changing the config_fn_t signature. We could also consider other ways of organizing the args (e.g. moving the config name and value into config_context or key_value_info), but in my experiments, the incremental benefit doesn't justify the added complexity (e.g. a config_fn_t will sometimes invoke another config_fn_t but with a different config value). In subsequent commits, the .kvi member will replace the global "struct config_reader" in config.c, making config iteration a global-free operation. It requires much more work for the machinery to provide meaningful values of .kvi, so for now, merely change the signature and call sites, pass NULL as a placeholder value, and don't rely on the arg in any meaningful way. Most of the changes are performed by contrib/coccinelle/config_fn_ctx.pending.cocci, which, for every config_fn_t: - Modifies the signature to accept "const struct config_context *ctx" - Passes "ctx" to any inner config_fn_t, if needed - Adds UNUSED attributes to "ctx", if needed Most config_fn_t instances are easily identified by seeing if they are called by the various config functions. Most of the remaining ones are manually named in the .cocci patch. Manual cleanups are still needed, but the majority of it is trivial; it's either adjusting config_fn_t that the .cocci patch didn't catch, or adding forward declarations of "struct config_context ctx" to make the signatures make sense. The non-trivial changes are in cases where we are invoking a config_fn_t outside of config machinery, and we now need to decide what value of "ctx" to pass. These cases are: - trace2/tr2_cfg.c:tr2_cfg_set_fl() This is indirectly called by git_config_set() so that the trace2 machinery can notice the new config values and update its settings using the tr2 config parsing function, i.e. tr2_cfg_cb(). - builtin/checkout.c:checkout_main() This calls git_xmerge_config() as a shorthand for parsing a CLI arg. This might be worth refactoring away in the future, since git_xmerge_config() can call git_default_config(), which can do much more than just parsing. Handle them by creating a KVI_INIT macro that initializes "struct key_value_info" to a reasonable default, and use that to construct the "ctx" arg. Signed-off-by: Glen Choo <chooglen@google.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-06-28 19:26:22 +00:00
int grep_config(const char *var, const char *value,
const struct config_context *ctx, void *cb)
{
grep API: call grep_config() after grep_init() The grep_init() function used the odd pattern of initializing the passed-in "struct grep_opt" with a statically defined "grep_defaults" struct, which would be modified in-place when we invoked grep_config(). So we effectively (b) initialized config, (a) then defaults, (c) followed by user options. Usually those are ordered as "a", "b" and "c" instead. As the comments being removed here show the previous behavior needed to be carefully explained as we'd potentially share the populated configuration among different instances of grep_init(). In practice we didn't do that, but now that it can't be a concern anymore let's remove those comments. This does not change the behavior of any of the configuration variables or options. That would have been the case if we didn't move around the grep_config() call in "builtin/log.c". But now that we call "grep_config" after "git_log_config" and "git_format_config" we'll need to pass in the already initialized "struct grep_opt *". See 6ba9bb76e02 (grep: copy struct in one fell swoop, 2020-11-29) and 7687a0541e0 (grep: move the configuration parsing logic to grep.[ch], 2012-10-09) for the commits that added the comments. The memcpy() pattern here will be optimized away and follows the convention of other *_init() functions. See 5726a6b4012 (*.c *_init(): define in terms of corresponding *_INIT macro, 2021-07-01). Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2022-02-16 00:00:36 +00:00
struct grep_opt *opt = cb;
const char *slot;
if (userdiff_config(var, value) < 0)
return -1;
if (!strcmp(var, "grep.extendedregexp")) {
opt->extended_regexp_option = git_config_bool(var, value);
return 0;
}
if (!strcmp(var, "grep.patterntype")) {
opt->pattern_type_option = parse_pattern_type_arg(var, value);
return 0;
}
if (!strcmp(var, "grep.linenumber")) {
opt->linenum = git_config_bool(var, value);
return 0;
}
if (!strcmp(var, "grep.column")) {
opt->columnnum = git_config_bool(var, value);
return 0;
}
if (!strcmp(var, "grep.fullname")) {
opt->relative = !git_config_bool(var, value);
return 0;
}
if (!strcmp(var, "color.grep"))
opt->color = git_config_colorbool(var, value);
if (!strcmp(var, "color.grep.match")) {
config: add ctx arg to config_fn_t Add a new "const struct config_context *ctx" arg to config_fn_t to hold additional information about the config iteration operation. config_context has a "struct key_value_info kvi" member that holds metadata about the config source being read (e.g. what kind of config source it is, the filename, etc). In this series, we're only interested in .kvi, so we could have just used "struct key_value_info" as an arg, but config_context makes it possible to add/adjust members in the future without changing the config_fn_t signature. We could also consider other ways of organizing the args (e.g. moving the config name and value into config_context or key_value_info), but in my experiments, the incremental benefit doesn't justify the added complexity (e.g. a config_fn_t will sometimes invoke another config_fn_t but with a different config value). In subsequent commits, the .kvi member will replace the global "struct config_reader" in config.c, making config iteration a global-free operation. It requires much more work for the machinery to provide meaningful values of .kvi, so for now, merely change the signature and call sites, pass NULL as a placeholder value, and don't rely on the arg in any meaningful way. Most of the changes are performed by contrib/coccinelle/config_fn_ctx.pending.cocci, which, for every config_fn_t: - Modifies the signature to accept "const struct config_context *ctx" - Passes "ctx" to any inner config_fn_t, if needed - Adds UNUSED attributes to "ctx", if needed Most config_fn_t instances are easily identified by seeing if they are called by the various config functions. Most of the remaining ones are manually named in the .cocci patch. Manual cleanups are still needed, but the majority of it is trivial; it's either adjusting config_fn_t that the .cocci patch didn't catch, or adding forward declarations of "struct config_context ctx" to make the signatures make sense. The non-trivial changes are in cases where we are invoking a config_fn_t outside of config machinery, and we now need to decide what value of "ctx" to pass. These cases are: - trace2/tr2_cfg.c:tr2_cfg_set_fl() This is indirectly called by git_config_set() so that the trace2 machinery can notice the new config values and update its settings using the tr2 config parsing function, i.e. tr2_cfg_cb(). - builtin/checkout.c:checkout_main() This calls git_xmerge_config() as a shorthand for parsing a CLI arg. This might be worth refactoring away in the future, since git_xmerge_config() can call git_default_config(), which can do much more than just parsing. Handle them by creating a KVI_INIT macro that initializes "struct key_value_info" to a reasonable default, and use that to construct the "ctx" arg. Signed-off-by: Glen Choo <chooglen@google.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-06-28 19:26:22 +00:00
if (grep_config("color.grep.matchcontext", value, ctx, cb) < 0)
return -1;
config: add ctx arg to config_fn_t Add a new "const struct config_context *ctx" arg to config_fn_t to hold additional information about the config iteration operation. config_context has a "struct key_value_info kvi" member that holds metadata about the config source being read (e.g. what kind of config source it is, the filename, etc). In this series, we're only interested in .kvi, so we could have just used "struct key_value_info" as an arg, but config_context makes it possible to add/adjust members in the future without changing the config_fn_t signature. We could also consider other ways of organizing the args (e.g. moving the config name and value into config_context or key_value_info), but in my experiments, the incremental benefit doesn't justify the added complexity (e.g. a config_fn_t will sometimes invoke another config_fn_t but with a different config value). In subsequent commits, the .kvi member will replace the global "struct config_reader" in config.c, making config iteration a global-free operation. It requires much more work for the machinery to provide meaningful values of .kvi, so for now, merely change the signature and call sites, pass NULL as a placeholder value, and don't rely on the arg in any meaningful way. Most of the changes are performed by contrib/coccinelle/config_fn_ctx.pending.cocci, which, for every config_fn_t: - Modifies the signature to accept "const struct config_context *ctx" - Passes "ctx" to any inner config_fn_t, if needed - Adds UNUSED attributes to "ctx", if needed Most config_fn_t instances are easily identified by seeing if they are called by the various config functions. Most of the remaining ones are manually named in the .cocci patch. Manual cleanups are still needed, but the majority of it is trivial; it's either adjusting config_fn_t that the .cocci patch didn't catch, or adding forward declarations of "struct config_context ctx" to make the signatures make sense. The non-trivial changes are in cases where we are invoking a config_fn_t outside of config machinery, and we now need to decide what value of "ctx" to pass. These cases are: - trace2/tr2_cfg.c:tr2_cfg_set_fl() This is indirectly called by git_config_set() so that the trace2 machinery can notice the new config values and update its settings using the tr2 config parsing function, i.e. tr2_cfg_cb(). - builtin/checkout.c:checkout_main() This calls git_xmerge_config() as a shorthand for parsing a CLI arg. This might be worth refactoring away in the future, since git_xmerge_config() can call git_default_config(), which can do much more than just parsing. Handle them by creating a KVI_INIT macro that initializes "struct key_value_info" to a reasonable default, and use that to construct the "ctx" arg. Signed-off-by: Glen Choo <chooglen@google.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-06-28 19:26:22 +00:00
if (grep_config("color.grep.matchselected", value, ctx, cb) < 0)
return -1;
} else if (skip_prefix(var, "color.grep.", &slot)) {
int i = LOOKUP_CONFIG(color_grep_slots, slot);
char *color;
if (i < 0)
return -1;
color = opt->colors[i];
if (!value)
return config_error_nonbool(var);
return color_parse(value, color);
}
return 0;
}
built-ins: trust the "prefix" from run_builtin() Change code in "builtin/grep.c" and "builtin/ls-tree.c" to trust the "prefix" passed from "run_builtin()". The "prefix" we get from setup.c is either going to be NULL or a string of length >0, never "". So we can drop the "prefix && *prefix" checks added for "builtin/grep.c" in 0d042fecf2f (git-grep: show pathnames relative to the current directory, 2006-08-11), and for "builtin/ls-tree.c" in a69dd585fca (ls-tree: chomp leading directories when run from a subdirectory, 2005-12-23). As seen in code in revision.c that was added in cd676a51367 (diff --relative: output paths as relative to the current subdirectory, 2008-02-12) we already have existing code that does away with this assertion. This makes it easier to reason about a subsequent change to the "prefix_length" code in grep.c in a subsequent commit, and since we're going to the trouble of doing that let's leave behind an assert() to promise this to any future callers. For "builtin/grep.c" it would be painful to pass the "prefix" down the callchain of: cmd_grep -> grep_tree -> grep_submodule -> grep_cache -> grep_oid -> grep_source_name So for the code that needs it in grep_source_name() let's add a "grep_prefix" variable similar to the existing "ls_tree_prefix". While at it let's move the code in cmd_ls_tree() around so that we assign to the "ls_tree_prefix" right after declaring the variables, and stop assigning to "prefix". We only subsequently used that variable later in the function after clobbering it. Let's just use our own "grep_prefix" instead. Let's also add an assert() in git.c, so that we'll make this promise about the "prefix" to any current and future callers, as well as to any readers of the code. Code history: * The strlen() in "grep.c" hasn't been used since 493b7a08d80 (grep: accept relative paths outside current working directory, 2009-09-05). When that code was added in 0d042fecf2f (git-grep: show pathnames relative to the current directory, 2006-08-11) we used the length. But since 493b7a08d80 we haven't used it for anything except a boolean check that we could have done on the "prefix" member itself. Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2022-02-16 00:00:34 +00:00
void grep_init(struct grep_opt *opt, struct repository *repo)
{
grep API: call grep_config() after grep_init() The grep_init() function used the odd pattern of initializing the passed-in "struct grep_opt" with a statically defined "grep_defaults" struct, which would be modified in-place when we invoked grep_config(). So we effectively (b) initialized config, (a) then defaults, (c) followed by user options. Usually those are ordered as "a", "b" and "c" instead. As the comments being removed here show the previous behavior needed to be carefully explained as we'd potentially share the populated configuration among different instances of grep_init(). In practice we didn't do that, but now that it can't be a concern anymore let's remove those comments. This does not change the behavior of any of the configuration variables or options. That would have been the case if we didn't move around the grep_config() call in "builtin/log.c". But now that we call "grep_config" after "git_log_config" and "git_format_config" we'll need to pass in the already initialized "struct grep_opt *". See 6ba9bb76e02 (grep: copy struct in one fell swoop, 2020-11-29) and 7687a0541e0 (grep: move the configuration parsing logic to grep.[ch], 2012-10-09) for the commits that added the comments. The memcpy() pattern here will be optimized away and follows the convention of other *_init() functions. See 5726a6b4012 (*.c *_init(): define in terms of corresponding *_INIT macro, 2021-07-01). Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2022-02-16 00:00:36 +00:00
struct grep_opt blank = GREP_OPT_INIT;
memcpy(opt, &blank, sizeof(*opt));
opt->repo = repo;
opt->pattern_tail = &opt->pattern_list;
opt->header_tail = &opt->header_list;
}
static struct grep_pat *create_grep_pat(const char *pat, size_t patlen,
const char *origin, int no,
enum grep_pat_token t,
enum grep_header_field field)
log --author/--committer: really match only with name part When we tried to find commits done by AUTHOR, the first implementation tried to pattern match a line with "^author .*AUTHOR", which later was enhanced to strip leading caret and look for "^author AUTHOR" when the search pattern was anchored at the left end (i.e. --author="^AUTHOR"). This had a few problems: * When looking for fixed strings (e.g. "git log -F --author=x --grep=y"), the regexp internally used "^author .*x" would never match anything; * To match at the end (e.g. "git log --author='google.com>$'"), the generated regexp has to also match the trailing timestamp part the commit header lines have. Also, in order to determine if the '$' at the end means "match at the end of the line" or just a literal dollar sign (probably backslash-quoted), we would need to parse the regexp ourselves. An earlier alternative tried to make sure that a line matches "^author " (to limit by field name) and the user supplied pattern at the same time. While it solved the -F problem by introducing a special override for matching the "^author ", it did not solve the trailing timestamp nor tail match problem. It also would have matched every commit if --author=author was asked for, not because the author's email part had this string, but because every commit header line that talks about the author begins with that field name, regardleses of who wrote it. Instead of piling more hacks on top of hacks, this rethinks the grep machinery that is used to look for strings in the commit header, and makes sure that (1) field name matches literally at the beginning of the line, followed by a SP, and (2) the user supplied pattern is matched against the remainder of the line, excluding the trailing timestamp data. Signed-off-by: Junio C Hamano <gitster@pobox.com>
2008-09-05 05:15:02 +00:00
{
struct grep_pat *p = xcalloc(1, sizeof(*p));
p->pattern = xmemdupz(pat, patlen);
p->patternlen = patlen;
p->origin = origin;
p->no = no;
p->token = t;
log --author/--committer: really match only with name part When we tried to find commits done by AUTHOR, the first implementation tried to pattern match a line with "^author .*AUTHOR", which later was enhanced to strip leading caret and look for "^author AUTHOR" when the search pattern was anchored at the left end (i.e. --author="^AUTHOR"). This had a few problems: * When looking for fixed strings (e.g. "git log -F --author=x --grep=y"), the regexp internally used "^author .*x" would never match anything; * To match at the end (e.g. "git log --author='google.com>$'"), the generated regexp has to also match the trailing timestamp part the commit header lines have. Also, in order to determine if the '$' at the end means "match at the end of the line" or just a literal dollar sign (probably backslash-quoted), we would need to parse the regexp ourselves. An earlier alternative tried to make sure that a line matches "^author " (to limit by field name) and the user supplied pattern at the same time. While it solved the -F problem by introducing a special override for matching the "^author ", it did not solve the trailing timestamp nor tail match problem. It also would have matched every commit if --author=author was asked for, not because the author's email part had this string, but because every commit header line that talks about the author begins with that field name, regardleses of who wrote it. Instead of piling more hacks on top of hacks, this rethinks the grep machinery that is used to look for strings in the commit header, and makes sure that (1) field name matches literally at the beginning of the line, followed by a SP, and (2) the user supplied pattern is matched against the remainder of the line, excluding the trailing timestamp data. Signed-off-by: Junio C Hamano <gitster@pobox.com>
2008-09-05 05:15:02 +00:00
p->field = field;
return p;
}
static void do_append_grep_pat(struct grep_pat ***tail, struct grep_pat *p)
{
**tail = p;
*tail = &p->next;
log --author/--committer: really match only with name part When we tried to find commits done by AUTHOR, the first implementation tried to pattern match a line with "^author .*AUTHOR", which later was enhanced to strip leading caret and look for "^author AUTHOR" when the search pattern was anchored at the left end (i.e. --author="^AUTHOR"). This had a few problems: * When looking for fixed strings (e.g. "git log -F --author=x --grep=y"), the regexp internally used "^author .*x" would never match anything; * To match at the end (e.g. "git log --author='google.com>$'"), the generated regexp has to also match the trailing timestamp part the commit header lines have. Also, in order to determine if the '$' at the end means "match at the end of the line" or just a literal dollar sign (probably backslash-quoted), we would need to parse the regexp ourselves. An earlier alternative tried to make sure that a line matches "^author " (to limit by field name) and the user supplied pattern at the same time. While it solved the -F problem by introducing a special override for matching the "^author ", it did not solve the trailing timestamp nor tail match problem. It also would have matched every commit if --author=author was asked for, not because the author's email part had this string, but because every commit header line that talks about the author begins with that field name, regardleses of who wrote it. Instead of piling more hacks on top of hacks, this rethinks the grep machinery that is used to look for strings in the commit header, and makes sure that (1) field name matches literally at the beginning of the line, followed by a SP, and (2) the user supplied pattern is matched against the remainder of the line, excluding the trailing timestamp data. Signed-off-by: Junio C Hamano <gitster@pobox.com>
2008-09-05 05:15:02 +00:00
p->next = NULL;
switch (p->token) {
case GREP_PATTERN: /* atom */
case GREP_PATTERN_HEAD:
case GREP_PATTERN_BODY:
for (;;) {
struct grep_pat *new_pat;
size_t len = 0;
char *cp = p->pattern + p->patternlen, *nl = NULL;
while (++len <= p->patternlen) {
if (*(--cp) == '\n') {
nl = cp;
break;
}
}
if (!nl)
break;
new_pat = create_grep_pat(nl + 1, len - 1, p->origin,
p->no, p->token, p->field);
new_pat->next = p->next;
if (!p->next)
*tail = &new_pat->next;
p->next = new_pat;
*nl = '\0';
p->patternlen -= len;
}
break;
default:
break;
}
}
void append_header_grep_pattern(struct grep_opt *opt,
enum grep_header_field field, const char *pat)
{
struct grep_pat *p = create_grep_pat(pat, strlen(pat), "header", 0,
GREP_PATTERN_HEAD, field);
if (field == GREP_HEADER_REFLOG)
opt->use_reflog_filter = 1;
do_append_grep_pat(&opt->header_tail, p);
log --author/--committer: really match only with name part When we tried to find commits done by AUTHOR, the first implementation tried to pattern match a line with "^author .*AUTHOR", which later was enhanced to strip leading caret and look for "^author AUTHOR" when the search pattern was anchored at the left end (i.e. --author="^AUTHOR"). This had a few problems: * When looking for fixed strings (e.g. "git log -F --author=x --grep=y"), the regexp internally used "^author .*x" would never match anything; * To match at the end (e.g. "git log --author='google.com>$'"), the generated regexp has to also match the trailing timestamp part the commit header lines have. Also, in order to determine if the '$' at the end means "match at the end of the line" or just a literal dollar sign (probably backslash-quoted), we would need to parse the regexp ourselves. An earlier alternative tried to make sure that a line matches "^author " (to limit by field name) and the user supplied pattern at the same time. While it solved the -F problem by introducing a special override for matching the "^author ", it did not solve the trailing timestamp nor tail match problem. It also would have matched every commit if --author=author was asked for, not because the author's email part had this string, but because every commit header line that talks about the author begins with that field name, regardleses of who wrote it. Instead of piling more hacks on top of hacks, this rethinks the grep machinery that is used to look for strings in the commit header, and makes sure that (1) field name matches literally at the beginning of the line, followed by a SP, and (2) the user supplied pattern is matched against the remainder of the line, excluding the trailing timestamp data. Signed-off-by: Junio C Hamano <gitster@pobox.com>
2008-09-05 05:15:02 +00:00
}
void append_grep_pattern(struct grep_opt *opt, const char *pat,
const char *origin, int no, enum grep_pat_token t)
{
append_grep_pat(opt, pat, strlen(pat), origin, no, t);
}
void append_grep_pat(struct grep_opt *opt, const char *pat, size_t patlen,
const char *origin, int no, enum grep_pat_token t)
{
struct grep_pat *p = create_grep_pat(pat, patlen, origin, no, t, 0);
do_append_grep_pat(&opt->pattern_tail, p);
}
struct grep_opt *grep_opt_dup(const struct grep_opt *opt)
{
struct grep_pat *pat;
struct grep_opt *ret = xmalloc(sizeof(struct grep_opt));
*ret = *opt;
ret->pattern_list = NULL;
ret->pattern_tail = &ret->pattern_list;
for(pat = opt->pattern_list; pat != NULL; pat = pat->next)
{
if(pat->token == GREP_PATTERN_HEAD)
append_header_grep_pattern(ret, pat->field,
pat->pattern);
else
append_grep_pat(ret, pat->pattern, pat->patternlen,
pat->origin, pat->no, pat->token);
}
return ret;
}
static NORETURN void compile_regexp_failed(const struct grep_pat *p,
const char *error)
{
char where[1024];
if (p->no)
xsnprintf(where, sizeof(where), "In '%s' at %d, ", p->origin, p->no);
else if (p->origin)
xsnprintf(where, sizeof(where), "%s, ", p->origin);
else
where[0] = 0;
die("%s'%s': %s", where, p->pattern, error);
}
static int is_fixed(const char *s, size_t len)
{
size_t i;
for (i = 0; i < len; i++) {
if (is_regex_special(s[i]))
return 0;
}
return 1;
}
grep: add support for PCRE v2 Add support for v2 of the PCRE API. This is a new major version of PCRE that came out in early 2015[1]. The regular expression syntax is the same, but while the API is similar, pretty much every function is either renamed or takes different arguments. Thus using it via entirely new functions makes sense, as opposed to trying to e.g. have one compile_pcre_pattern() that would call either PCRE v1 or v2 functions. Git can now be compiled with either USE_LIBPCRE1=YesPlease or USE_LIBPCRE2=YesPlease, with USE_LIBPCRE=YesPlease currently being a synonym for the former. Providing both is a compile-time error. With earlier patches to enable JIT for PCRE v1 the performance of the release versions of both libraries is almost exactly the same, with PCRE v2 being around 1% slower. However after I reported this to the pcre-dev mailing list[2] I got a lot of help with the API use from Zoltán Herczeg, he subsequently optimized some of the JIT functionality in v2 of the library. Running the p7820-grep-engines.sh performance test against the latest Subversion trunk of both, with both them and git compiled as -O3, and the test run against linux.git, gives the following results. Just the /perl/ tests shown: $ GIT_PERF_REPEAT_COUNT=30 GIT_PERF_LARGE_REPO=~/g/linux GIT_PERF_MAKE_COMMAND='grep -q LIBPCRE2 Makefile && make -j8 USE_LIBPCRE2=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre2/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre2/inst/lib || make -j8 USE_LIBPCRE=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre/inst/lib' ./run HEAD~5 HEAD~ HEAD p7820-grep-engines.sh [...] Test HEAD~5 HEAD~ HEAD ----------------------------------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.31(1.10+0.48) 0.21(0.35+0.56) -32.3% 0.21(0.34+0.55) -32.3% 7820.7: perl grep '^how to' 0.56(2.70+0.40) 0.24(0.64+0.52) -57.1% 0.20(0.28+0.60) -64.3% 7820.11: perl grep '[how] to' 0.56(2.66+0.38) 0.29(0.95+0.45) -48.2% 0.23(0.45+0.54) -58.9% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 1.02(5.77+0.42) 0.31(1.02+0.54) -69.6% 0.23(0.50+0.54) -77.5% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.38(1.57+0.42) 0.27(0.85+0.46) -28.9% 0.21(0.33+0.57) -44.7% See commit ("perf: add a comparison test of grep regex engines", 2017-04-19) for details on the machine the above test run was executed on. Here HEAD~2 is git with PCRE v1 without JIT, HEAD~ is PCRE v1 with JIT, and HEAD is PCRE v2 (also with JIT). See previous commits of mine mentioning p7820-grep-engines.sh for more details on the test setup. For ease of readability, a different run just of HEAD~ (PCRE v1 with JIT v.s. PCRE v2), again with just the /perl/ tests shown: [...] Test HEAD~ HEAD ---------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.21(0.42+0.52) 0.21(0.31+0.58) +0.0% 7820.7: perl grep '^how to' 0.25(0.65+0.50) 0.20(0.31+0.57) -20.0% 7820.11: perl grep '[how] to' 0.30(0.90+0.50) 0.23(0.46+0.53) -23.3% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 0.30(1.19+0.38) 0.23(0.51+0.51) -23.3% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.27(0.84+0.48) 0.21(0.34+0.57) -22.2% I.e. the two are either neck-to-neck, but PCRE v2 usually pulls ahead, when it does it's around 20% faster. A brief note on thread safety: As noted in pcre2api(3) & pcre2jit(3) the compiled pattern can be shared between threads, but not some of the JIT context, however the grep threading support does all pattern & JIT compilation in separate threads, so this code doesn't need to concern itself with thread safety. See commit 63e7e9d8b6 ("git-grep: Learn PCRE", 2011-05-09) for the initial addition of PCRE v1. This change follows some of the same patterns it did (and which were discussed on list at the time), e.g. mocking up types with typedef instead of ifdef-ing them out when USE_LIBPCRE2 isn't defined. This adds some trivial memory use to the program, but makes the code look nicer. 1. https://lists.exim.org/lurker/message/20150105.162835.0666407a.en.html 2. https://lists.exim.org/lurker/thread/20170419.172322.833ee099.en.html Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-01 18:20:56 +00:00
#ifdef USE_LIBPCRE2
#define GREP_PCRE2_DEBUG_MALLOC 0
static void *pcre2_malloc(PCRE2_SIZE size, MAYBE_UNUSED void *memory_data)
{
void *pointer = malloc(size);
#if GREP_PCRE2_DEBUG_MALLOC
static int count = 1;
fprintf(stderr, "PCRE2:%p -> #%02d: alloc(%lu)\n", pointer, count++, size);
#endif
return pointer;
}
static void pcre2_free(void *pointer, MAYBE_UNUSED void *memory_data)
{
#if GREP_PCRE2_DEBUG_MALLOC
static int count = 1;
if (pointer)
fprintf(stderr, "PCRE2:%p -> #%02d: free()\n", pointer, count++);
#endif
free(pointer);
}
grep: fall back to interpreter if JIT memory allocation fails Under Linux systems with SELinux's 'deny_execmem' or PaX's MPROTECT enabled, the allocation of PCRE2's JIT rwx memory may be prohibited, making pcre2_jit_compile() fail with PCRE2_ERROR_NOMEMORY (-48): [user@fedora git]$ git grep -c PCRE2_JIT grep.c:1 [user@fedora git]$ # Enable SELinux's W^X policy [user@fedora git]$ sudo semanage boolean -m -1 deny_execmem [user@fedora git]$ # JIT memory allocation fails, breaking 'git grep' [user@fedora git]$ git grep -c PCRE2_JIT fatal: Couldn't JIT the PCRE2 pattern 'PCRE2_JIT', got '-48' Instead of failing hard in this case and making 'git grep' unusable on such systems, simply fall back to interpreter mode, leading to a much better user experience. As having a functional PCRE2 JIT compiler is a legitimate use case for performance reasons, we'll only do the fallback if the supposedly available JIT is found to be non-functional by attempting to JIT compile a very simple pattern. If this fails, JIT is deemed to be non-functional and we do the interpreter fallback. For all other cases, i.e. the simple pattern can be compiled but the user provided cannot, we fail hard as we do now as the reason for the failure must be the pattern itself. To aid users in helping themselves change the error message to include a hint about the '(*NO_JIT)' prefix. Also clip the pattern at 64 characters to ensure the hint will be seen by the user and not internally truncated by the die() function. Cc: Carlo Marcelo Arenas Belón <carenas@gmail.com> Signed-off-by: Mathias Krause <minipli@grsecurity.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-01-31 18:56:11 +00:00
static int pcre2_jit_functional(void)
{
static int jit_working = -1;
pcre2_code *code;
size_t off;
int err;
if (jit_working != -1)
return jit_working;
/*
* Try to JIT compile a simple pattern to probe if the JIT is
* working in general. It might fail for systems where creating
* memory mappings for runtime code generation is restricted.
*/
code = pcre2_compile((PCRE2_SPTR)".", 1, 0, &err, &off, NULL);
if (!code)
return 0;
jit_working = pcre2_jit_compile(code, PCRE2_JIT_COMPLETE) == 0;
pcre2_code_free(code);
return jit_working;
}
grep: add support for PCRE v2 Add support for v2 of the PCRE API. This is a new major version of PCRE that came out in early 2015[1]. The regular expression syntax is the same, but while the API is similar, pretty much every function is either renamed or takes different arguments. Thus using it via entirely new functions makes sense, as opposed to trying to e.g. have one compile_pcre_pattern() that would call either PCRE v1 or v2 functions. Git can now be compiled with either USE_LIBPCRE1=YesPlease or USE_LIBPCRE2=YesPlease, with USE_LIBPCRE=YesPlease currently being a synonym for the former. Providing both is a compile-time error. With earlier patches to enable JIT for PCRE v1 the performance of the release versions of both libraries is almost exactly the same, with PCRE v2 being around 1% slower. However after I reported this to the pcre-dev mailing list[2] I got a lot of help with the API use from Zoltán Herczeg, he subsequently optimized some of the JIT functionality in v2 of the library. Running the p7820-grep-engines.sh performance test against the latest Subversion trunk of both, with both them and git compiled as -O3, and the test run against linux.git, gives the following results. Just the /perl/ tests shown: $ GIT_PERF_REPEAT_COUNT=30 GIT_PERF_LARGE_REPO=~/g/linux GIT_PERF_MAKE_COMMAND='grep -q LIBPCRE2 Makefile && make -j8 USE_LIBPCRE2=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre2/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre2/inst/lib || make -j8 USE_LIBPCRE=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre/inst/lib' ./run HEAD~5 HEAD~ HEAD p7820-grep-engines.sh [...] Test HEAD~5 HEAD~ HEAD ----------------------------------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.31(1.10+0.48) 0.21(0.35+0.56) -32.3% 0.21(0.34+0.55) -32.3% 7820.7: perl grep '^how to' 0.56(2.70+0.40) 0.24(0.64+0.52) -57.1% 0.20(0.28+0.60) -64.3% 7820.11: perl grep '[how] to' 0.56(2.66+0.38) 0.29(0.95+0.45) -48.2% 0.23(0.45+0.54) -58.9% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 1.02(5.77+0.42) 0.31(1.02+0.54) -69.6% 0.23(0.50+0.54) -77.5% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.38(1.57+0.42) 0.27(0.85+0.46) -28.9% 0.21(0.33+0.57) -44.7% See commit ("perf: add a comparison test of grep regex engines", 2017-04-19) for details on the machine the above test run was executed on. Here HEAD~2 is git with PCRE v1 without JIT, HEAD~ is PCRE v1 with JIT, and HEAD is PCRE v2 (also with JIT). See previous commits of mine mentioning p7820-grep-engines.sh for more details on the test setup. For ease of readability, a different run just of HEAD~ (PCRE v1 with JIT v.s. PCRE v2), again with just the /perl/ tests shown: [...] Test HEAD~ HEAD ---------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.21(0.42+0.52) 0.21(0.31+0.58) +0.0% 7820.7: perl grep '^how to' 0.25(0.65+0.50) 0.20(0.31+0.57) -20.0% 7820.11: perl grep '[how] to' 0.30(0.90+0.50) 0.23(0.46+0.53) -23.3% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 0.30(1.19+0.38) 0.23(0.51+0.51) -23.3% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.27(0.84+0.48) 0.21(0.34+0.57) -22.2% I.e. the two are either neck-to-neck, but PCRE v2 usually pulls ahead, when it does it's around 20% faster. A brief note on thread safety: As noted in pcre2api(3) & pcre2jit(3) the compiled pattern can be shared between threads, but not some of the JIT context, however the grep threading support does all pattern & JIT compilation in separate threads, so this code doesn't need to concern itself with thread safety. See commit 63e7e9d8b6 ("git-grep: Learn PCRE", 2011-05-09) for the initial addition of PCRE v1. This change follows some of the same patterns it did (and which were discussed on list at the time), e.g. mocking up types with typedef instead of ifdef-ing them out when USE_LIBPCRE2 isn't defined. This adds some trivial memory use to the program, but makes the code look nicer. 1. https://lists.exim.org/lurker/message/20150105.162835.0666407a.en.html 2. https://lists.exim.org/lurker/thread/20170419.172322.833ee099.en.html Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-01 18:20:56 +00:00
static void compile_pcre2_pattern(struct grep_pat *p, const struct grep_opt *opt)
{
int error;
PCRE2_UCHAR errbuf[256];
PCRE2_SIZE erroffset;
int options = PCRE2_MULTILINE;
int jitret;
grep: fix segfault under -P + PCRE2 <=10.30 + (*NO_JIT) Fix a bug in the compilation of PCRE2 patterns under JIT (the most common runtime configuration). Any pattern with a (*NO_JIT) verb would segfault in any currently released PCRE2 version: $ git grep -P '(*NO_JIT)hi.*there' Segmentation fault That this segfaulted was a bug in PCRE2 itself, after reporting it[1] on pcre-dev it's been fixed in a yet-to-be-released version of PCRE (presumably released first as 10.31). Now it'll die with: $ git grep -P '(*NO_JIT)hi.*there' fatal: pcre2_jit_match failed with error code -45: bad JIT option But the cause of the bug is in our own code dating back to my 94da9193a6 ("grep: add support for PCRE v2", 2017-06-01). As explained at more length in the comment being added here, it isn't sufficient to just check pcre2_config() to see whether the JIT should be used, pcre2_pattern_info() also has to be asked. This is something I discovered myself when fiddling around with PCRE2 verbs in patterns passed to git. I don't expect that any user of git has encountered this given the obscurity of passing PCRE2 verbs through to the library, along with the relative obscurity of (*NO_JIT) itself. 1. "How am I supposed to use PCRE2 JIT in the face of (*NO_JIT) ?" (<CACBZZX5mMqDuWuFmi7sRBp3wH6CFyd-ghACukd=v0NN=rBMnJg@mail.gmail.com> & https://lists.exim.org/lurker/thread/20171123.101502.7f0d38ca.en.html) on the pcre-dev mailing list Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-11-23 14:16:58 +00:00
int patinforet;
size_t jitsizearg;
int literal = !opt->ignore_case && (p->fixed || p->is_fixed);
grep: add support for PCRE v2 Add support for v2 of the PCRE API. This is a new major version of PCRE that came out in early 2015[1]. The regular expression syntax is the same, but while the API is similar, pretty much every function is either renamed or takes different arguments. Thus using it via entirely new functions makes sense, as opposed to trying to e.g. have one compile_pcre_pattern() that would call either PCRE v1 or v2 functions. Git can now be compiled with either USE_LIBPCRE1=YesPlease or USE_LIBPCRE2=YesPlease, with USE_LIBPCRE=YesPlease currently being a synonym for the former. Providing both is a compile-time error. With earlier patches to enable JIT for PCRE v1 the performance of the release versions of both libraries is almost exactly the same, with PCRE v2 being around 1% slower. However after I reported this to the pcre-dev mailing list[2] I got a lot of help with the API use from Zoltán Herczeg, he subsequently optimized some of the JIT functionality in v2 of the library. Running the p7820-grep-engines.sh performance test against the latest Subversion trunk of both, with both them and git compiled as -O3, and the test run against linux.git, gives the following results. Just the /perl/ tests shown: $ GIT_PERF_REPEAT_COUNT=30 GIT_PERF_LARGE_REPO=~/g/linux GIT_PERF_MAKE_COMMAND='grep -q LIBPCRE2 Makefile && make -j8 USE_LIBPCRE2=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre2/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre2/inst/lib || make -j8 USE_LIBPCRE=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre/inst/lib' ./run HEAD~5 HEAD~ HEAD p7820-grep-engines.sh [...] Test HEAD~5 HEAD~ HEAD ----------------------------------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.31(1.10+0.48) 0.21(0.35+0.56) -32.3% 0.21(0.34+0.55) -32.3% 7820.7: perl grep '^how to' 0.56(2.70+0.40) 0.24(0.64+0.52) -57.1% 0.20(0.28+0.60) -64.3% 7820.11: perl grep '[how] to' 0.56(2.66+0.38) 0.29(0.95+0.45) -48.2% 0.23(0.45+0.54) -58.9% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 1.02(5.77+0.42) 0.31(1.02+0.54) -69.6% 0.23(0.50+0.54) -77.5% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.38(1.57+0.42) 0.27(0.85+0.46) -28.9% 0.21(0.33+0.57) -44.7% See commit ("perf: add a comparison test of grep regex engines", 2017-04-19) for details on the machine the above test run was executed on. Here HEAD~2 is git with PCRE v1 without JIT, HEAD~ is PCRE v1 with JIT, and HEAD is PCRE v2 (also with JIT). See previous commits of mine mentioning p7820-grep-engines.sh for more details on the test setup. For ease of readability, a different run just of HEAD~ (PCRE v1 with JIT v.s. PCRE v2), again with just the /perl/ tests shown: [...] Test HEAD~ HEAD ---------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.21(0.42+0.52) 0.21(0.31+0.58) +0.0% 7820.7: perl grep '^how to' 0.25(0.65+0.50) 0.20(0.31+0.57) -20.0% 7820.11: perl grep '[how] to' 0.30(0.90+0.50) 0.23(0.46+0.53) -23.3% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 0.30(1.19+0.38) 0.23(0.51+0.51) -23.3% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.27(0.84+0.48) 0.21(0.34+0.57) -22.2% I.e. the two are either neck-to-neck, but PCRE v2 usually pulls ahead, when it does it's around 20% faster. A brief note on thread safety: As noted in pcre2api(3) & pcre2jit(3) the compiled pattern can be shared between threads, but not some of the JIT context, however the grep threading support does all pattern & JIT compilation in separate threads, so this code doesn't need to concern itself with thread safety. See commit 63e7e9d8b6 ("git-grep: Learn PCRE", 2011-05-09) for the initial addition of PCRE v1. This change follows some of the same patterns it did (and which were discussed on list at the time), e.g. mocking up types with typedef instead of ifdef-ing them out when USE_LIBPCRE2 isn't defined. This adds some trivial memory use to the program, but makes the code look nicer. 1. https://lists.exim.org/lurker/message/20150105.162835.0666407a.en.html 2. https://lists.exim.org/lurker/thread/20170419.172322.833ee099.en.html Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-01 18:20:56 +00:00
grep/pcre2: move back to thread-only PCREv2 structures Change the setup of the "pcre2_general_context" to happen per-thread in compile_pcre2_pattern() instead of in grep_init(). This change brings it in line with how the rest of the pcre2_* members in the grep_pat structure are set up. As noted in the preceding commit the approach 513f2b0bbd4 (grep: make PCRE2 aware of custom allocator, 2019-10-16) took to allocate the pcre2_general_context seems to have been initially based on a misunderstanding of how PCREv2 memory allocation works. The approach of creating a global context in grep_init() is just added complexity for almost zero gain. On my system it's 24 bytes saved per-thread. For comparison PCREv2 will then go on to allocate at least a kilobyte for its own thread-local state. As noted in 6d423dd542f (grep: don't redundantly compile throwaway patterns under threading, 2017-05-25) the grep code is intentionally not trying to micro-optimize allocations by e.g. sharing some PCREv2 structures globally, while making others thread-local. So let's remove this special case and make all of them thread-local again for simplicity. With this change we could move the pcre2_{malloc,free} functions around to live closer to their current use. I'm not doing that here to keep this change small, that cleanup will be done in a follow-up commit. See also the discussion in 94da9193a6 (grep: add support for PCRE v2, 2017-06-01) about thread safety, and Johannes's comments[1] to the effect that we should be doing what this patch is doing. 1. https://lore.kernel.org/git/nycvar.QRO.7.76.6.1908052120302.46@tvgsbejvaqbjf.bet/ Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-02-18 00:07:27 +00:00
/*
* Call pcre2_general_context_create() before calling any
* other pcre2_*(). It sets up our malloc()/free() functions
* with which everything else is allocated.
*/
p->pcre2_general_context = pcre2_general_context_create(
pcre2_malloc, pcre2_free, NULL);
if (!p->pcre2_general_context)
die("Couldn't allocate PCRE2 general context");
grep: add support for PCRE v2 Add support for v2 of the PCRE API. This is a new major version of PCRE that came out in early 2015[1]. The regular expression syntax is the same, but while the API is similar, pretty much every function is either renamed or takes different arguments. Thus using it via entirely new functions makes sense, as opposed to trying to e.g. have one compile_pcre_pattern() that would call either PCRE v1 or v2 functions. Git can now be compiled with either USE_LIBPCRE1=YesPlease or USE_LIBPCRE2=YesPlease, with USE_LIBPCRE=YesPlease currently being a synonym for the former. Providing both is a compile-time error. With earlier patches to enable JIT for PCRE v1 the performance of the release versions of both libraries is almost exactly the same, with PCRE v2 being around 1% slower. However after I reported this to the pcre-dev mailing list[2] I got a lot of help with the API use from Zoltán Herczeg, he subsequently optimized some of the JIT functionality in v2 of the library. Running the p7820-grep-engines.sh performance test against the latest Subversion trunk of both, with both them and git compiled as -O3, and the test run against linux.git, gives the following results. Just the /perl/ tests shown: $ GIT_PERF_REPEAT_COUNT=30 GIT_PERF_LARGE_REPO=~/g/linux GIT_PERF_MAKE_COMMAND='grep -q LIBPCRE2 Makefile && make -j8 USE_LIBPCRE2=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre2/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre2/inst/lib || make -j8 USE_LIBPCRE=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre/inst/lib' ./run HEAD~5 HEAD~ HEAD p7820-grep-engines.sh [...] Test HEAD~5 HEAD~ HEAD ----------------------------------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.31(1.10+0.48) 0.21(0.35+0.56) -32.3% 0.21(0.34+0.55) -32.3% 7820.7: perl grep '^how to' 0.56(2.70+0.40) 0.24(0.64+0.52) -57.1% 0.20(0.28+0.60) -64.3% 7820.11: perl grep '[how] to' 0.56(2.66+0.38) 0.29(0.95+0.45) -48.2% 0.23(0.45+0.54) -58.9% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 1.02(5.77+0.42) 0.31(1.02+0.54) -69.6% 0.23(0.50+0.54) -77.5% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.38(1.57+0.42) 0.27(0.85+0.46) -28.9% 0.21(0.33+0.57) -44.7% See commit ("perf: add a comparison test of grep regex engines", 2017-04-19) for details on the machine the above test run was executed on. Here HEAD~2 is git with PCRE v1 without JIT, HEAD~ is PCRE v1 with JIT, and HEAD is PCRE v2 (also with JIT). See previous commits of mine mentioning p7820-grep-engines.sh for more details on the test setup. For ease of readability, a different run just of HEAD~ (PCRE v1 with JIT v.s. PCRE v2), again with just the /perl/ tests shown: [...] Test HEAD~ HEAD ---------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.21(0.42+0.52) 0.21(0.31+0.58) +0.0% 7820.7: perl grep '^how to' 0.25(0.65+0.50) 0.20(0.31+0.57) -20.0% 7820.11: perl grep '[how] to' 0.30(0.90+0.50) 0.23(0.46+0.53) -23.3% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 0.30(1.19+0.38) 0.23(0.51+0.51) -23.3% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.27(0.84+0.48) 0.21(0.34+0.57) -22.2% I.e. the two are either neck-to-neck, but PCRE v2 usually pulls ahead, when it does it's around 20% faster. A brief note on thread safety: As noted in pcre2api(3) & pcre2jit(3) the compiled pattern can be shared between threads, but not some of the JIT context, however the grep threading support does all pattern & JIT compilation in separate threads, so this code doesn't need to concern itself with thread safety. See commit 63e7e9d8b6 ("git-grep: Learn PCRE", 2011-05-09) for the initial addition of PCRE v1. This change follows some of the same patterns it did (and which were discussed on list at the time), e.g. mocking up types with typedef instead of ifdef-ing them out when USE_LIBPCRE2 isn't defined. This adds some trivial memory use to the program, but makes the code look nicer. 1. https://lists.exim.org/lurker/message/20150105.162835.0666407a.en.html 2. https://lists.exim.org/lurker/thread/20170419.172322.833ee099.en.html Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-01 18:20:56 +00:00
if (opt->ignore_case) {
if (!opt->ignore_locale && has_non_ascii(p->pattern)) {
grep/pcre2: move back to thread-only PCREv2 structures Change the setup of the "pcre2_general_context" to happen per-thread in compile_pcre2_pattern() instead of in grep_init(). This change brings it in line with how the rest of the pcre2_* members in the grep_pat structure are set up. As noted in the preceding commit the approach 513f2b0bbd4 (grep: make PCRE2 aware of custom allocator, 2019-10-16) took to allocate the pcre2_general_context seems to have been initially based on a misunderstanding of how PCREv2 memory allocation works. The approach of creating a global context in grep_init() is just added complexity for almost zero gain. On my system it's 24 bytes saved per-thread. For comparison PCREv2 will then go on to allocate at least a kilobyte for its own thread-local state. As noted in 6d423dd542f (grep: don't redundantly compile throwaway patterns under threading, 2017-05-25) the grep code is intentionally not trying to micro-optimize allocations by e.g. sharing some PCREv2 structures globally, while making others thread-local. So let's remove this special case and make all of them thread-local again for simplicity. With this change we could move the pcre2_{malloc,free} functions around to live closer to their current use. I'm not doing that here to keep this change small, that cleanup will be done in a follow-up commit. See also the discussion in 94da9193a6 (grep: add support for PCRE v2, 2017-06-01) about thread safety, and Johannes's comments[1] to the effect that we should be doing what this patch is doing. 1. https://lore.kernel.org/git/nycvar.QRO.7.76.6.1908052120302.46@tvgsbejvaqbjf.bet/ Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-02-18 00:07:27 +00:00
p->pcre2_tables = pcre2_maketables(p->pcre2_general_context);
p->pcre2_compile_context = pcre2_compile_context_create(p->pcre2_general_context);
pcre2_set_character_tables(p->pcre2_compile_context,
p->pcre2_tables);
grep: add support for PCRE v2 Add support for v2 of the PCRE API. This is a new major version of PCRE that came out in early 2015[1]. The regular expression syntax is the same, but while the API is similar, pretty much every function is either renamed or takes different arguments. Thus using it via entirely new functions makes sense, as opposed to trying to e.g. have one compile_pcre_pattern() that would call either PCRE v1 or v2 functions. Git can now be compiled with either USE_LIBPCRE1=YesPlease or USE_LIBPCRE2=YesPlease, with USE_LIBPCRE=YesPlease currently being a synonym for the former. Providing both is a compile-time error. With earlier patches to enable JIT for PCRE v1 the performance of the release versions of both libraries is almost exactly the same, with PCRE v2 being around 1% slower. However after I reported this to the pcre-dev mailing list[2] I got a lot of help with the API use from Zoltán Herczeg, he subsequently optimized some of the JIT functionality in v2 of the library. Running the p7820-grep-engines.sh performance test against the latest Subversion trunk of both, with both them and git compiled as -O3, and the test run against linux.git, gives the following results. Just the /perl/ tests shown: $ GIT_PERF_REPEAT_COUNT=30 GIT_PERF_LARGE_REPO=~/g/linux GIT_PERF_MAKE_COMMAND='grep -q LIBPCRE2 Makefile && make -j8 USE_LIBPCRE2=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre2/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre2/inst/lib || make -j8 USE_LIBPCRE=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre/inst/lib' ./run HEAD~5 HEAD~ HEAD p7820-grep-engines.sh [...] Test HEAD~5 HEAD~ HEAD ----------------------------------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.31(1.10+0.48) 0.21(0.35+0.56) -32.3% 0.21(0.34+0.55) -32.3% 7820.7: perl grep '^how to' 0.56(2.70+0.40) 0.24(0.64+0.52) -57.1% 0.20(0.28+0.60) -64.3% 7820.11: perl grep '[how] to' 0.56(2.66+0.38) 0.29(0.95+0.45) -48.2% 0.23(0.45+0.54) -58.9% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 1.02(5.77+0.42) 0.31(1.02+0.54) -69.6% 0.23(0.50+0.54) -77.5% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.38(1.57+0.42) 0.27(0.85+0.46) -28.9% 0.21(0.33+0.57) -44.7% See commit ("perf: add a comparison test of grep regex engines", 2017-04-19) for details on the machine the above test run was executed on. Here HEAD~2 is git with PCRE v1 without JIT, HEAD~ is PCRE v1 with JIT, and HEAD is PCRE v2 (also with JIT). See previous commits of mine mentioning p7820-grep-engines.sh for more details on the test setup. For ease of readability, a different run just of HEAD~ (PCRE v1 with JIT v.s. PCRE v2), again with just the /perl/ tests shown: [...] Test HEAD~ HEAD ---------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.21(0.42+0.52) 0.21(0.31+0.58) +0.0% 7820.7: perl grep '^how to' 0.25(0.65+0.50) 0.20(0.31+0.57) -20.0% 7820.11: perl grep '[how] to' 0.30(0.90+0.50) 0.23(0.46+0.53) -23.3% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 0.30(1.19+0.38) 0.23(0.51+0.51) -23.3% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.27(0.84+0.48) 0.21(0.34+0.57) -22.2% I.e. the two are either neck-to-neck, but PCRE v2 usually pulls ahead, when it does it's around 20% faster. A brief note on thread safety: As noted in pcre2api(3) & pcre2jit(3) the compiled pattern can be shared between threads, but not some of the JIT context, however the grep threading support does all pattern & JIT compilation in separate threads, so this code doesn't need to concern itself with thread safety. See commit 63e7e9d8b6 ("git-grep: Learn PCRE", 2011-05-09) for the initial addition of PCRE v1. This change follows some of the same patterns it did (and which were discussed on list at the time), e.g. mocking up types with typedef instead of ifdef-ing them out when USE_LIBPCRE2 isn't defined. This adds some trivial memory use to the program, but makes the code look nicer. 1. https://lists.exim.org/lurker/message/20150105.162835.0666407a.en.html 2. https://lists.exim.org/lurker/thread/20170419.172322.833ee099.en.html Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-01 18:20:56 +00:00
}
options |= PCRE2_CASELESS;
}
if (!opt->ignore_locale && is_utf8_locale() && !literal)
options |= (PCRE2_UTF | PCRE2_UCP | PCRE2_MATCH_INVALID_UTF);
#ifndef GIT_PCRE2_VERSION_10_35_OR_HIGHER
/*
* Work around a JIT bug related to invalid Unicode character handling
* fixed in 10.35:
* https://github.com/PCRE2Project/pcre2/commit/c21bd977547d
*/
options &= ~PCRE2_UCP;
#endif
#ifndef GIT_PCRE2_VERSION_10_36_OR_HIGHER
/* Work around https://bugs.exim.org/show_bug.cgi?id=2642 fixed in 10.36 */
if (PCRE2_MATCH_INVALID_UTF && options & (PCRE2_UTF | PCRE2_CASELESS))
options |= PCRE2_NO_START_OPTIMIZE;
#endif
grep: add support for PCRE v2 Add support for v2 of the PCRE API. This is a new major version of PCRE that came out in early 2015[1]. The regular expression syntax is the same, but while the API is similar, pretty much every function is either renamed or takes different arguments. Thus using it via entirely new functions makes sense, as opposed to trying to e.g. have one compile_pcre_pattern() that would call either PCRE v1 or v2 functions. Git can now be compiled with either USE_LIBPCRE1=YesPlease or USE_LIBPCRE2=YesPlease, with USE_LIBPCRE=YesPlease currently being a synonym for the former. Providing both is a compile-time error. With earlier patches to enable JIT for PCRE v1 the performance of the release versions of both libraries is almost exactly the same, with PCRE v2 being around 1% slower. However after I reported this to the pcre-dev mailing list[2] I got a lot of help with the API use from Zoltán Herczeg, he subsequently optimized some of the JIT functionality in v2 of the library. Running the p7820-grep-engines.sh performance test against the latest Subversion trunk of both, with both them and git compiled as -O3, and the test run against linux.git, gives the following results. Just the /perl/ tests shown: $ GIT_PERF_REPEAT_COUNT=30 GIT_PERF_LARGE_REPO=~/g/linux GIT_PERF_MAKE_COMMAND='grep -q LIBPCRE2 Makefile && make -j8 USE_LIBPCRE2=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre2/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre2/inst/lib || make -j8 USE_LIBPCRE=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre/inst/lib' ./run HEAD~5 HEAD~ HEAD p7820-grep-engines.sh [...] Test HEAD~5 HEAD~ HEAD ----------------------------------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.31(1.10+0.48) 0.21(0.35+0.56) -32.3% 0.21(0.34+0.55) -32.3% 7820.7: perl grep '^how to' 0.56(2.70+0.40) 0.24(0.64+0.52) -57.1% 0.20(0.28+0.60) -64.3% 7820.11: perl grep '[how] to' 0.56(2.66+0.38) 0.29(0.95+0.45) -48.2% 0.23(0.45+0.54) -58.9% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 1.02(5.77+0.42) 0.31(1.02+0.54) -69.6% 0.23(0.50+0.54) -77.5% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.38(1.57+0.42) 0.27(0.85+0.46) -28.9% 0.21(0.33+0.57) -44.7% See commit ("perf: add a comparison test of grep regex engines", 2017-04-19) for details on the machine the above test run was executed on. Here HEAD~2 is git with PCRE v1 without JIT, HEAD~ is PCRE v1 with JIT, and HEAD is PCRE v2 (also with JIT). See previous commits of mine mentioning p7820-grep-engines.sh for more details on the test setup. For ease of readability, a different run just of HEAD~ (PCRE v1 with JIT v.s. PCRE v2), again with just the /perl/ tests shown: [...] Test HEAD~ HEAD ---------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.21(0.42+0.52) 0.21(0.31+0.58) +0.0% 7820.7: perl grep '^how to' 0.25(0.65+0.50) 0.20(0.31+0.57) -20.0% 7820.11: perl grep '[how] to' 0.30(0.90+0.50) 0.23(0.46+0.53) -23.3% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 0.30(1.19+0.38) 0.23(0.51+0.51) -23.3% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.27(0.84+0.48) 0.21(0.34+0.57) -22.2% I.e. the two are either neck-to-neck, but PCRE v2 usually pulls ahead, when it does it's around 20% faster. A brief note on thread safety: As noted in pcre2api(3) & pcre2jit(3) the compiled pattern can be shared between threads, but not some of the JIT context, however the grep threading support does all pattern & JIT compilation in separate threads, so this code doesn't need to concern itself with thread safety. See commit 63e7e9d8b6 ("git-grep: Learn PCRE", 2011-05-09) for the initial addition of PCRE v1. This change follows some of the same patterns it did (and which were discussed on list at the time), e.g. mocking up types with typedef instead of ifdef-ing them out when USE_LIBPCRE2 isn't defined. This adds some trivial memory use to the program, but makes the code look nicer. 1. https://lists.exim.org/lurker/message/20150105.162835.0666407a.en.html 2. https://lists.exim.org/lurker/thread/20170419.172322.833ee099.en.html Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-01 18:20:56 +00:00
p->pcre2_pattern = pcre2_compile((PCRE2_SPTR)p->pattern,
p->patternlen, options, &error, &erroffset,
p->pcre2_compile_context);
if (p->pcre2_pattern) {
grep/pcre2: move back to thread-only PCREv2 structures Change the setup of the "pcre2_general_context" to happen per-thread in compile_pcre2_pattern() instead of in grep_init(). This change brings it in line with how the rest of the pcre2_* members in the grep_pat structure are set up. As noted in the preceding commit the approach 513f2b0bbd4 (grep: make PCRE2 aware of custom allocator, 2019-10-16) took to allocate the pcre2_general_context seems to have been initially based on a misunderstanding of how PCREv2 memory allocation works. The approach of creating a global context in grep_init() is just added complexity for almost zero gain. On my system it's 24 bytes saved per-thread. For comparison PCREv2 will then go on to allocate at least a kilobyte for its own thread-local state. As noted in 6d423dd542f (grep: don't redundantly compile throwaway patterns under threading, 2017-05-25) the grep code is intentionally not trying to micro-optimize allocations by e.g. sharing some PCREv2 structures globally, while making others thread-local. So let's remove this special case and make all of them thread-local again for simplicity. With this change we could move the pcre2_{malloc,free} functions around to live closer to their current use. I'm not doing that here to keep this change small, that cleanup will be done in a follow-up commit. See also the discussion in 94da9193a6 (grep: add support for PCRE v2, 2017-06-01) about thread safety, and Johannes's comments[1] to the effect that we should be doing what this patch is doing. 1. https://lore.kernel.org/git/nycvar.QRO.7.76.6.1908052120302.46@tvgsbejvaqbjf.bet/ Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-02-18 00:07:27 +00:00
p->pcre2_match_data = pcre2_match_data_create_from_pattern(p->pcre2_pattern, p->pcre2_general_context);
grep: add support for PCRE v2 Add support for v2 of the PCRE API. This is a new major version of PCRE that came out in early 2015[1]. The regular expression syntax is the same, but while the API is similar, pretty much every function is either renamed or takes different arguments. Thus using it via entirely new functions makes sense, as opposed to trying to e.g. have one compile_pcre_pattern() that would call either PCRE v1 or v2 functions. Git can now be compiled with either USE_LIBPCRE1=YesPlease or USE_LIBPCRE2=YesPlease, with USE_LIBPCRE=YesPlease currently being a synonym for the former. Providing both is a compile-time error. With earlier patches to enable JIT for PCRE v1 the performance of the release versions of both libraries is almost exactly the same, with PCRE v2 being around 1% slower. However after I reported this to the pcre-dev mailing list[2] I got a lot of help with the API use from Zoltán Herczeg, he subsequently optimized some of the JIT functionality in v2 of the library. Running the p7820-grep-engines.sh performance test against the latest Subversion trunk of both, with both them and git compiled as -O3, and the test run against linux.git, gives the following results. Just the /perl/ tests shown: $ GIT_PERF_REPEAT_COUNT=30 GIT_PERF_LARGE_REPO=~/g/linux GIT_PERF_MAKE_COMMAND='grep -q LIBPCRE2 Makefile && make -j8 USE_LIBPCRE2=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre2/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre2/inst/lib || make -j8 USE_LIBPCRE=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre/inst/lib' ./run HEAD~5 HEAD~ HEAD p7820-grep-engines.sh [...] Test HEAD~5 HEAD~ HEAD ----------------------------------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.31(1.10+0.48) 0.21(0.35+0.56) -32.3% 0.21(0.34+0.55) -32.3% 7820.7: perl grep '^how to' 0.56(2.70+0.40) 0.24(0.64+0.52) -57.1% 0.20(0.28+0.60) -64.3% 7820.11: perl grep '[how] to' 0.56(2.66+0.38) 0.29(0.95+0.45) -48.2% 0.23(0.45+0.54) -58.9% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 1.02(5.77+0.42) 0.31(1.02+0.54) -69.6% 0.23(0.50+0.54) -77.5% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.38(1.57+0.42) 0.27(0.85+0.46) -28.9% 0.21(0.33+0.57) -44.7% See commit ("perf: add a comparison test of grep regex engines", 2017-04-19) for details on the machine the above test run was executed on. Here HEAD~2 is git with PCRE v1 without JIT, HEAD~ is PCRE v1 with JIT, and HEAD is PCRE v2 (also with JIT). See previous commits of mine mentioning p7820-grep-engines.sh for more details on the test setup. For ease of readability, a different run just of HEAD~ (PCRE v1 with JIT v.s. PCRE v2), again with just the /perl/ tests shown: [...] Test HEAD~ HEAD ---------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.21(0.42+0.52) 0.21(0.31+0.58) +0.0% 7820.7: perl grep '^how to' 0.25(0.65+0.50) 0.20(0.31+0.57) -20.0% 7820.11: perl grep '[how] to' 0.30(0.90+0.50) 0.23(0.46+0.53) -23.3% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 0.30(1.19+0.38) 0.23(0.51+0.51) -23.3% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.27(0.84+0.48) 0.21(0.34+0.57) -22.2% I.e. the two are either neck-to-neck, but PCRE v2 usually pulls ahead, when it does it's around 20% faster. A brief note on thread safety: As noted in pcre2api(3) & pcre2jit(3) the compiled pattern can be shared between threads, but not some of the JIT context, however the grep threading support does all pattern & JIT compilation in separate threads, so this code doesn't need to concern itself with thread safety. See commit 63e7e9d8b6 ("git-grep: Learn PCRE", 2011-05-09) for the initial addition of PCRE v1. This change follows some of the same patterns it did (and which were discussed on list at the time), e.g. mocking up types with typedef instead of ifdef-ing them out when USE_LIBPCRE2 isn't defined. This adds some trivial memory use to the program, but makes the code look nicer. 1. https://lists.exim.org/lurker/message/20150105.162835.0666407a.en.html 2. https://lists.exim.org/lurker/thread/20170419.172322.833ee099.en.html Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-01 18:20:56 +00:00
if (!p->pcre2_match_data)
die("Couldn't allocate PCRE2 match data");
} else {
pcre2_get_error_message(error, errbuf, sizeof(errbuf));
compile_regexp_failed(p, (const char *)&errbuf);
}
pcre2_config(PCRE2_CONFIG_JIT, &p->pcre2_jit_on);
if (p->pcre2_jit_on) {
grep: add support for PCRE v2 Add support for v2 of the PCRE API. This is a new major version of PCRE that came out in early 2015[1]. The regular expression syntax is the same, but while the API is similar, pretty much every function is either renamed or takes different arguments. Thus using it via entirely new functions makes sense, as opposed to trying to e.g. have one compile_pcre_pattern() that would call either PCRE v1 or v2 functions. Git can now be compiled with either USE_LIBPCRE1=YesPlease or USE_LIBPCRE2=YesPlease, with USE_LIBPCRE=YesPlease currently being a synonym for the former. Providing both is a compile-time error. With earlier patches to enable JIT for PCRE v1 the performance of the release versions of both libraries is almost exactly the same, with PCRE v2 being around 1% slower. However after I reported this to the pcre-dev mailing list[2] I got a lot of help with the API use from Zoltán Herczeg, he subsequently optimized some of the JIT functionality in v2 of the library. Running the p7820-grep-engines.sh performance test against the latest Subversion trunk of both, with both them and git compiled as -O3, and the test run against linux.git, gives the following results. Just the /perl/ tests shown: $ GIT_PERF_REPEAT_COUNT=30 GIT_PERF_LARGE_REPO=~/g/linux GIT_PERF_MAKE_COMMAND='grep -q LIBPCRE2 Makefile && make -j8 USE_LIBPCRE2=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre2/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre2/inst/lib || make -j8 USE_LIBPCRE=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre/inst/lib' ./run HEAD~5 HEAD~ HEAD p7820-grep-engines.sh [...] Test HEAD~5 HEAD~ HEAD ----------------------------------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.31(1.10+0.48) 0.21(0.35+0.56) -32.3% 0.21(0.34+0.55) -32.3% 7820.7: perl grep '^how to' 0.56(2.70+0.40) 0.24(0.64+0.52) -57.1% 0.20(0.28+0.60) -64.3% 7820.11: perl grep '[how] to' 0.56(2.66+0.38) 0.29(0.95+0.45) -48.2% 0.23(0.45+0.54) -58.9% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 1.02(5.77+0.42) 0.31(1.02+0.54) -69.6% 0.23(0.50+0.54) -77.5% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.38(1.57+0.42) 0.27(0.85+0.46) -28.9% 0.21(0.33+0.57) -44.7% See commit ("perf: add a comparison test of grep regex engines", 2017-04-19) for details on the machine the above test run was executed on. Here HEAD~2 is git with PCRE v1 without JIT, HEAD~ is PCRE v1 with JIT, and HEAD is PCRE v2 (also with JIT). See previous commits of mine mentioning p7820-grep-engines.sh for more details on the test setup. For ease of readability, a different run just of HEAD~ (PCRE v1 with JIT v.s. PCRE v2), again with just the /perl/ tests shown: [...] Test HEAD~ HEAD ---------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.21(0.42+0.52) 0.21(0.31+0.58) +0.0% 7820.7: perl grep '^how to' 0.25(0.65+0.50) 0.20(0.31+0.57) -20.0% 7820.11: perl grep '[how] to' 0.30(0.90+0.50) 0.23(0.46+0.53) -23.3% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 0.30(1.19+0.38) 0.23(0.51+0.51) -23.3% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.27(0.84+0.48) 0.21(0.34+0.57) -22.2% I.e. the two are either neck-to-neck, but PCRE v2 usually pulls ahead, when it does it's around 20% faster. A brief note on thread safety: As noted in pcre2api(3) & pcre2jit(3) the compiled pattern can be shared between threads, but not some of the JIT context, however the grep threading support does all pattern & JIT compilation in separate threads, so this code doesn't need to concern itself with thread safety. See commit 63e7e9d8b6 ("git-grep: Learn PCRE", 2011-05-09) for the initial addition of PCRE v1. This change follows some of the same patterns it did (and which were discussed on list at the time), e.g. mocking up types with typedef instead of ifdef-ing them out when USE_LIBPCRE2 isn't defined. This adds some trivial memory use to the program, but makes the code look nicer. 1. https://lists.exim.org/lurker/message/20150105.162835.0666407a.en.html 2. https://lists.exim.org/lurker/thread/20170419.172322.833ee099.en.html Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-01 18:20:56 +00:00
jitret = pcre2_jit_compile(p->pcre2_pattern, PCRE2_JIT_COMPLETE);
grep: fall back to interpreter if JIT memory allocation fails Under Linux systems with SELinux's 'deny_execmem' or PaX's MPROTECT enabled, the allocation of PCRE2's JIT rwx memory may be prohibited, making pcre2_jit_compile() fail with PCRE2_ERROR_NOMEMORY (-48): [user@fedora git]$ git grep -c PCRE2_JIT grep.c:1 [user@fedora git]$ # Enable SELinux's W^X policy [user@fedora git]$ sudo semanage boolean -m -1 deny_execmem [user@fedora git]$ # JIT memory allocation fails, breaking 'git grep' [user@fedora git]$ git grep -c PCRE2_JIT fatal: Couldn't JIT the PCRE2 pattern 'PCRE2_JIT', got '-48' Instead of failing hard in this case and making 'git grep' unusable on such systems, simply fall back to interpreter mode, leading to a much better user experience. As having a functional PCRE2 JIT compiler is a legitimate use case for performance reasons, we'll only do the fallback if the supposedly available JIT is found to be non-functional by attempting to JIT compile a very simple pattern. If this fails, JIT is deemed to be non-functional and we do the interpreter fallback. For all other cases, i.e. the simple pattern can be compiled but the user provided cannot, we fail hard as we do now as the reason for the failure must be the pattern itself. To aid users in helping themselves change the error message to include a hint about the '(*NO_JIT)' prefix. Also clip the pattern at 64 characters to ensure the hint will be seen by the user and not internally truncated by the die() function. Cc: Carlo Marcelo Arenas Belón <carenas@gmail.com> Signed-off-by: Mathias Krause <minipli@grsecurity.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-01-31 18:56:11 +00:00
if (jitret == PCRE2_ERROR_NOMEMORY && !pcre2_jit_functional()) {
/*
* Even though pcre2_config(PCRE2_CONFIG_JIT, ...)
* indicated JIT support, the library might still
* fail to generate JIT code for various reasons,
* e.g. when SELinux's 'deny_execmem' or PaX's
* MPROTECT prevent creating W|X memory mappings.
*
* Instead of faling hard, fall back to interpreter
* mode, just as if the pattern was prefixed with
* '(*NO_JIT)'.
*/
p->pcre2_jit_on = 0;
return;
} else if (jitret) {
int need_clip = p->patternlen > 64;
int clip_len = need_clip ? 64 : p->patternlen;
die("Couldn't JIT the PCRE2 pattern '%.*s'%s, got '%d'%s",
clip_len, p->pattern, need_clip ? "..." : "", jitret,
pcre2_jit_functional()
? "\nPerhaps prefix (*NO_JIT) to your pattern?"
: "");
}
grep: fix segfault under -P + PCRE2 <=10.30 + (*NO_JIT) Fix a bug in the compilation of PCRE2 patterns under JIT (the most common runtime configuration). Any pattern with a (*NO_JIT) verb would segfault in any currently released PCRE2 version: $ git grep -P '(*NO_JIT)hi.*there' Segmentation fault That this segfaulted was a bug in PCRE2 itself, after reporting it[1] on pcre-dev it's been fixed in a yet-to-be-released version of PCRE (presumably released first as 10.31). Now it'll die with: $ git grep -P '(*NO_JIT)hi.*there' fatal: pcre2_jit_match failed with error code -45: bad JIT option But the cause of the bug is in our own code dating back to my 94da9193a6 ("grep: add support for PCRE v2", 2017-06-01). As explained at more length in the comment being added here, it isn't sufficient to just check pcre2_config() to see whether the JIT should be used, pcre2_pattern_info() also has to be asked. This is something I discovered myself when fiddling around with PCRE2 verbs in patterns passed to git. I don't expect that any user of git has encountered this given the obscurity of passing PCRE2 verbs through to the library, along with the relative obscurity of (*NO_JIT) itself. 1. "How am I supposed to use PCRE2 JIT in the face of (*NO_JIT) ?" (<CACBZZX5mMqDuWuFmi7sRBp3wH6CFyd-ghACukd=v0NN=rBMnJg@mail.gmail.com> & https://lists.exim.org/lurker/thread/20171123.101502.7f0d38ca.en.html) on the pcre-dev mailing list Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-11-23 14:16:58 +00:00
/*
* The pcre2_config(PCRE2_CONFIG_JIT, ...) call just
* tells us whether the library itself supports JIT,
* but to see whether we're going to be actually using
* JIT we need to extract PCRE2_INFO_JITSIZE from the
* pattern *after* we do pcre2_jit_compile() above.
*
* This is because if the pattern contains the
* (*NO_JIT) verb (see pcre2syntax(3))
* pcre2_jit_compile() will exit early with 0. If we
* then proceed to call pcre2_jit_match() further down
* the line instead of pcre2_match() we'll either
* segfault (pre PCRE 10.31) or run into a fatal error
* (post PCRE2 10.31)
*/
patinforet = pcre2_pattern_info(p->pcre2_pattern, PCRE2_INFO_JITSIZE, &jitsizearg);
if (patinforet)
BUG("pcre2_pattern_info() failed: %d", patinforet);
if (jitsizearg == 0) {
p->pcre2_jit_on = 0;
return;
}
grep: add support for PCRE v2 Add support for v2 of the PCRE API. This is a new major version of PCRE that came out in early 2015[1]. The regular expression syntax is the same, but while the API is similar, pretty much every function is either renamed or takes different arguments. Thus using it via entirely new functions makes sense, as opposed to trying to e.g. have one compile_pcre_pattern() that would call either PCRE v1 or v2 functions. Git can now be compiled with either USE_LIBPCRE1=YesPlease or USE_LIBPCRE2=YesPlease, with USE_LIBPCRE=YesPlease currently being a synonym for the former. Providing both is a compile-time error. With earlier patches to enable JIT for PCRE v1 the performance of the release versions of both libraries is almost exactly the same, with PCRE v2 being around 1% slower. However after I reported this to the pcre-dev mailing list[2] I got a lot of help with the API use from Zoltán Herczeg, he subsequently optimized some of the JIT functionality in v2 of the library. Running the p7820-grep-engines.sh performance test against the latest Subversion trunk of both, with both them and git compiled as -O3, and the test run against linux.git, gives the following results. Just the /perl/ tests shown: $ GIT_PERF_REPEAT_COUNT=30 GIT_PERF_LARGE_REPO=~/g/linux GIT_PERF_MAKE_COMMAND='grep -q LIBPCRE2 Makefile && make -j8 USE_LIBPCRE2=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre2/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre2/inst/lib || make -j8 USE_LIBPCRE=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre/inst/lib' ./run HEAD~5 HEAD~ HEAD p7820-grep-engines.sh [...] Test HEAD~5 HEAD~ HEAD ----------------------------------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.31(1.10+0.48) 0.21(0.35+0.56) -32.3% 0.21(0.34+0.55) -32.3% 7820.7: perl grep '^how to' 0.56(2.70+0.40) 0.24(0.64+0.52) -57.1% 0.20(0.28+0.60) -64.3% 7820.11: perl grep '[how] to' 0.56(2.66+0.38) 0.29(0.95+0.45) -48.2% 0.23(0.45+0.54) -58.9% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 1.02(5.77+0.42) 0.31(1.02+0.54) -69.6% 0.23(0.50+0.54) -77.5% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.38(1.57+0.42) 0.27(0.85+0.46) -28.9% 0.21(0.33+0.57) -44.7% See commit ("perf: add a comparison test of grep regex engines", 2017-04-19) for details on the machine the above test run was executed on. Here HEAD~2 is git with PCRE v1 without JIT, HEAD~ is PCRE v1 with JIT, and HEAD is PCRE v2 (also with JIT). See previous commits of mine mentioning p7820-grep-engines.sh for more details on the test setup. For ease of readability, a different run just of HEAD~ (PCRE v1 with JIT v.s. PCRE v2), again with just the /perl/ tests shown: [...] Test HEAD~ HEAD ---------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.21(0.42+0.52) 0.21(0.31+0.58) +0.0% 7820.7: perl grep '^how to' 0.25(0.65+0.50) 0.20(0.31+0.57) -20.0% 7820.11: perl grep '[how] to' 0.30(0.90+0.50) 0.23(0.46+0.53) -23.3% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 0.30(1.19+0.38) 0.23(0.51+0.51) -23.3% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.27(0.84+0.48) 0.21(0.34+0.57) -22.2% I.e. the two are either neck-to-neck, but PCRE v2 usually pulls ahead, when it does it's around 20% faster. A brief note on thread safety: As noted in pcre2api(3) & pcre2jit(3) the compiled pattern can be shared between threads, but not some of the JIT context, however the grep threading support does all pattern & JIT compilation in separate threads, so this code doesn't need to concern itself with thread safety. See commit 63e7e9d8b6 ("git-grep: Learn PCRE", 2011-05-09) for the initial addition of PCRE v1. This change follows some of the same patterns it did (and which were discussed on list at the time), e.g. mocking up types with typedef instead of ifdef-ing them out when USE_LIBPCRE2 isn't defined. This adds some trivial memory use to the program, but makes the code look nicer. 1. https://lists.exim.org/lurker/message/20150105.162835.0666407a.en.html 2. https://lists.exim.org/lurker/thread/20170419.172322.833ee099.en.html Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-01 18:20:56 +00:00
}
}
static int pcre2match(struct grep_pat *p, const char *line, const char *eol,
regmatch_t *match, int eflags)
{
int ret, flags = 0;
PCRE2_SIZE *ovector;
PCRE2_UCHAR errbuf[256];
if (eflags & REG_NOTBOL)
flags |= PCRE2_NOTBOL;
if (p->pcre2_jit_on)
ret = pcre2_jit_match(p->pcre2_pattern, (unsigned char *)line,
eol - line, 0, flags, p->pcre2_match_data,
NULL);
else
ret = pcre2_match(p->pcre2_pattern, (unsigned char *)line,
eol - line, 0, flags, p->pcre2_match_data,
NULL);
if (ret < 0 && ret != PCRE2_ERROR_NOMATCH) {
pcre2_get_error_message(ret, errbuf, sizeof(errbuf));
die("%s failed with error code %d: %s",
(p->pcre2_jit_on ? "pcre2_jit_match" : "pcre2_match"), ret,
errbuf);
}
if (ret > 0) {
ovector = pcre2_get_ovector_pointer(p->pcre2_match_data);
ret = 0;
match->rm_so = (int)ovector[0];
match->rm_eo = (int)ovector[1];
}
return ret;
}
static void free_pcre2_pattern(struct grep_pat *p)
{
pcre2_compile_context_free(p->pcre2_compile_context);
pcre2_code_free(p->pcre2_pattern);
pcre2_match_data_free(p->pcre2_match_data);
#ifdef GIT_PCRE2_VERSION_10_34_OR_HIGHER
grep/pcre2: move back to thread-only PCREv2 structures Change the setup of the "pcre2_general_context" to happen per-thread in compile_pcre2_pattern() instead of in grep_init(). This change brings it in line with how the rest of the pcre2_* members in the grep_pat structure are set up. As noted in the preceding commit the approach 513f2b0bbd4 (grep: make PCRE2 aware of custom allocator, 2019-10-16) took to allocate the pcre2_general_context seems to have been initially based on a misunderstanding of how PCREv2 memory allocation works. The approach of creating a global context in grep_init() is just added complexity for almost zero gain. On my system it's 24 bytes saved per-thread. For comparison PCREv2 will then go on to allocate at least a kilobyte for its own thread-local state. As noted in 6d423dd542f (grep: don't redundantly compile throwaway patterns under threading, 2017-05-25) the grep code is intentionally not trying to micro-optimize allocations by e.g. sharing some PCREv2 structures globally, while making others thread-local. So let's remove this special case and make all of them thread-local again for simplicity. With this change we could move the pcre2_{malloc,free} functions around to live closer to their current use. I'm not doing that here to keep this change small, that cleanup will be done in a follow-up commit. See also the discussion in 94da9193a6 (grep: add support for PCRE v2, 2017-06-01) about thread safety, and Johannes's comments[1] to the effect that we should be doing what this patch is doing. 1. https://lore.kernel.org/git/nycvar.QRO.7.76.6.1908052120302.46@tvgsbejvaqbjf.bet/ Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-02-18 00:07:27 +00:00
pcre2_maketables_free(p->pcre2_general_context, p->pcre2_tables);
#else
free((void *)p->pcre2_tables);
#endif
grep/pcre2: move back to thread-only PCREv2 structures Change the setup of the "pcre2_general_context" to happen per-thread in compile_pcre2_pattern() instead of in grep_init(). This change brings it in line with how the rest of the pcre2_* members in the grep_pat structure are set up. As noted in the preceding commit the approach 513f2b0bbd4 (grep: make PCRE2 aware of custom allocator, 2019-10-16) took to allocate the pcre2_general_context seems to have been initially based on a misunderstanding of how PCREv2 memory allocation works. The approach of creating a global context in grep_init() is just added complexity for almost zero gain. On my system it's 24 bytes saved per-thread. For comparison PCREv2 will then go on to allocate at least a kilobyte for its own thread-local state. As noted in 6d423dd542f (grep: don't redundantly compile throwaway patterns under threading, 2017-05-25) the grep code is intentionally not trying to micro-optimize allocations by e.g. sharing some PCREv2 structures globally, while making others thread-local. So let's remove this special case and make all of them thread-local again for simplicity. With this change we could move the pcre2_{malloc,free} functions around to live closer to their current use. I'm not doing that here to keep this change small, that cleanup will be done in a follow-up commit. See also the discussion in 94da9193a6 (grep: add support for PCRE v2, 2017-06-01) about thread safety, and Johannes's comments[1] to the effect that we should be doing what this patch is doing. 1. https://lore.kernel.org/git/nycvar.QRO.7.76.6.1908052120302.46@tvgsbejvaqbjf.bet/ Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-02-18 00:07:27 +00:00
pcre2_general_context_free(p->pcre2_general_context);
grep: add support for PCRE v2 Add support for v2 of the PCRE API. This is a new major version of PCRE that came out in early 2015[1]. The regular expression syntax is the same, but while the API is similar, pretty much every function is either renamed or takes different arguments. Thus using it via entirely new functions makes sense, as opposed to trying to e.g. have one compile_pcre_pattern() that would call either PCRE v1 or v2 functions. Git can now be compiled with either USE_LIBPCRE1=YesPlease or USE_LIBPCRE2=YesPlease, with USE_LIBPCRE=YesPlease currently being a synonym for the former. Providing both is a compile-time error. With earlier patches to enable JIT for PCRE v1 the performance of the release versions of both libraries is almost exactly the same, with PCRE v2 being around 1% slower. However after I reported this to the pcre-dev mailing list[2] I got a lot of help with the API use from Zoltán Herczeg, he subsequently optimized some of the JIT functionality in v2 of the library. Running the p7820-grep-engines.sh performance test against the latest Subversion trunk of both, with both them and git compiled as -O3, and the test run against linux.git, gives the following results. Just the /perl/ tests shown: $ GIT_PERF_REPEAT_COUNT=30 GIT_PERF_LARGE_REPO=~/g/linux GIT_PERF_MAKE_COMMAND='grep -q LIBPCRE2 Makefile && make -j8 USE_LIBPCRE2=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre2/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre2/inst/lib || make -j8 USE_LIBPCRE=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre/inst/lib' ./run HEAD~5 HEAD~ HEAD p7820-grep-engines.sh [...] Test HEAD~5 HEAD~ HEAD ----------------------------------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.31(1.10+0.48) 0.21(0.35+0.56) -32.3% 0.21(0.34+0.55) -32.3% 7820.7: perl grep '^how to' 0.56(2.70+0.40) 0.24(0.64+0.52) -57.1% 0.20(0.28+0.60) -64.3% 7820.11: perl grep '[how] to' 0.56(2.66+0.38) 0.29(0.95+0.45) -48.2% 0.23(0.45+0.54) -58.9% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 1.02(5.77+0.42) 0.31(1.02+0.54) -69.6% 0.23(0.50+0.54) -77.5% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.38(1.57+0.42) 0.27(0.85+0.46) -28.9% 0.21(0.33+0.57) -44.7% See commit ("perf: add a comparison test of grep regex engines", 2017-04-19) for details on the machine the above test run was executed on. Here HEAD~2 is git with PCRE v1 without JIT, HEAD~ is PCRE v1 with JIT, and HEAD is PCRE v2 (also with JIT). See previous commits of mine mentioning p7820-grep-engines.sh for more details on the test setup. For ease of readability, a different run just of HEAD~ (PCRE v1 with JIT v.s. PCRE v2), again with just the /perl/ tests shown: [...] Test HEAD~ HEAD ---------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.21(0.42+0.52) 0.21(0.31+0.58) +0.0% 7820.7: perl grep '^how to' 0.25(0.65+0.50) 0.20(0.31+0.57) -20.0% 7820.11: perl grep '[how] to' 0.30(0.90+0.50) 0.23(0.46+0.53) -23.3% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 0.30(1.19+0.38) 0.23(0.51+0.51) -23.3% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.27(0.84+0.48) 0.21(0.34+0.57) -22.2% I.e. the two are either neck-to-neck, but PCRE v2 usually pulls ahead, when it does it's around 20% faster. A brief note on thread safety: As noted in pcre2api(3) & pcre2jit(3) the compiled pattern can be shared between threads, but not some of the JIT context, however the grep threading support does all pattern & JIT compilation in separate threads, so this code doesn't need to concern itself with thread safety. See commit 63e7e9d8b6 ("git-grep: Learn PCRE", 2011-05-09) for the initial addition of PCRE v1. This change follows some of the same patterns it did (and which were discussed on list at the time), e.g. mocking up types with typedef instead of ifdef-ing them out when USE_LIBPCRE2 isn't defined. This adds some trivial memory use to the program, but makes the code look nicer. 1. https://lists.exim.org/lurker/message/20150105.162835.0666407a.en.html 2. https://lists.exim.org/lurker/thread/20170419.172322.833ee099.en.html Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-01 18:20:56 +00:00
}
#else /* !USE_LIBPCRE2 */
static void compile_pcre2_pattern(struct grep_pat *p UNUSED,
const struct grep_opt *opt UNUSED)
grep: add support for PCRE v2 Add support for v2 of the PCRE API. This is a new major version of PCRE that came out in early 2015[1]. The regular expression syntax is the same, but while the API is similar, pretty much every function is either renamed or takes different arguments. Thus using it via entirely new functions makes sense, as opposed to trying to e.g. have one compile_pcre_pattern() that would call either PCRE v1 or v2 functions. Git can now be compiled with either USE_LIBPCRE1=YesPlease or USE_LIBPCRE2=YesPlease, with USE_LIBPCRE=YesPlease currently being a synonym for the former. Providing both is a compile-time error. With earlier patches to enable JIT for PCRE v1 the performance of the release versions of both libraries is almost exactly the same, with PCRE v2 being around 1% slower. However after I reported this to the pcre-dev mailing list[2] I got a lot of help with the API use from Zoltán Herczeg, he subsequently optimized some of the JIT functionality in v2 of the library. Running the p7820-grep-engines.sh performance test against the latest Subversion trunk of both, with both them and git compiled as -O3, and the test run against linux.git, gives the following results. Just the /perl/ tests shown: $ GIT_PERF_REPEAT_COUNT=30 GIT_PERF_LARGE_REPO=~/g/linux GIT_PERF_MAKE_COMMAND='grep -q LIBPCRE2 Makefile && make -j8 USE_LIBPCRE2=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre2/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre2/inst/lib || make -j8 USE_LIBPCRE=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre/inst/lib' ./run HEAD~5 HEAD~ HEAD p7820-grep-engines.sh [...] Test HEAD~5 HEAD~ HEAD ----------------------------------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.31(1.10+0.48) 0.21(0.35+0.56) -32.3% 0.21(0.34+0.55) -32.3% 7820.7: perl grep '^how to' 0.56(2.70+0.40) 0.24(0.64+0.52) -57.1% 0.20(0.28+0.60) -64.3% 7820.11: perl grep '[how] to' 0.56(2.66+0.38) 0.29(0.95+0.45) -48.2% 0.23(0.45+0.54) -58.9% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 1.02(5.77+0.42) 0.31(1.02+0.54) -69.6% 0.23(0.50+0.54) -77.5% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.38(1.57+0.42) 0.27(0.85+0.46) -28.9% 0.21(0.33+0.57) -44.7% See commit ("perf: add a comparison test of grep regex engines", 2017-04-19) for details on the machine the above test run was executed on. Here HEAD~2 is git with PCRE v1 without JIT, HEAD~ is PCRE v1 with JIT, and HEAD is PCRE v2 (also with JIT). See previous commits of mine mentioning p7820-grep-engines.sh for more details on the test setup. For ease of readability, a different run just of HEAD~ (PCRE v1 with JIT v.s. PCRE v2), again with just the /perl/ tests shown: [...] Test HEAD~ HEAD ---------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.21(0.42+0.52) 0.21(0.31+0.58) +0.0% 7820.7: perl grep '^how to' 0.25(0.65+0.50) 0.20(0.31+0.57) -20.0% 7820.11: perl grep '[how] to' 0.30(0.90+0.50) 0.23(0.46+0.53) -23.3% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 0.30(1.19+0.38) 0.23(0.51+0.51) -23.3% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.27(0.84+0.48) 0.21(0.34+0.57) -22.2% I.e. the two are either neck-to-neck, but PCRE v2 usually pulls ahead, when it does it's around 20% faster. A brief note on thread safety: As noted in pcre2api(3) & pcre2jit(3) the compiled pattern can be shared between threads, but not some of the JIT context, however the grep threading support does all pattern & JIT compilation in separate threads, so this code doesn't need to concern itself with thread safety. See commit 63e7e9d8b6 ("git-grep: Learn PCRE", 2011-05-09) for the initial addition of PCRE v1. This change follows some of the same patterns it did (and which were discussed on list at the time), e.g. mocking up types with typedef instead of ifdef-ing them out when USE_LIBPCRE2 isn't defined. This adds some trivial memory use to the program, but makes the code look nicer. 1. https://lists.exim.org/lurker/message/20150105.162835.0666407a.en.html 2. https://lists.exim.org/lurker/thread/20170419.172322.833ee099.en.html Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-01 18:20:56 +00:00
{
die("cannot use Perl-compatible regexes when not compiled with USE_LIBPCRE");
}
static int pcre2match(struct grep_pat *p UNUSED, const char *line UNUSED,
const char *eol UNUSED, regmatch_t *match UNUSED,
int eflags UNUSED)
grep: add support for PCRE v2 Add support for v2 of the PCRE API. This is a new major version of PCRE that came out in early 2015[1]. The regular expression syntax is the same, but while the API is similar, pretty much every function is either renamed or takes different arguments. Thus using it via entirely new functions makes sense, as opposed to trying to e.g. have one compile_pcre_pattern() that would call either PCRE v1 or v2 functions. Git can now be compiled with either USE_LIBPCRE1=YesPlease or USE_LIBPCRE2=YesPlease, with USE_LIBPCRE=YesPlease currently being a synonym for the former. Providing both is a compile-time error. With earlier patches to enable JIT for PCRE v1 the performance of the release versions of both libraries is almost exactly the same, with PCRE v2 being around 1% slower. However after I reported this to the pcre-dev mailing list[2] I got a lot of help with the API use from Zoltán Herczeg, he subsequently optimized some of the JIT functionality in v2 of the library. Running the p7820-grep-engines.sh performance test against the latest Subversion trunk of both, with both them and git compiled as -O3, and the test run against linux.git, gives the following results. Just the /perl/ tests shown: $ GIT_PERF_REPEAT_COUNT=30 GIT_PERF_LARGE_REPO=~/g/linux GIT_PERF_MAKE_COMMAND='grep -q LIBPCRE2 Makefile && make -j8 USE_LIBPCRE2=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre2/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre2/inst/lib || make -j8 USE_LIBPCRE=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre/inst/lib' ./run HEAD~5 HEAD~ HEAD p7820-grep-engines.sh [...] Test HEAD~5 HEAD~ HEAD ----------------------------------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.31(1.10+0.48) 0.21(0.35+0.56) -32.3% 0.21(0.34+0.55) -32.3% 7820.7: perl grep '^how to' 0.56(2.70+0.40) 0.24(0.64+0.52) -57.1% 0.20(0.28+0.60) -64.3% 7820.11: perl grep '[how] to' 0.56(2.66+0.38) 0.29(0.95+0.45) -48.2% 0.23(0.45+0.54) -58.9% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 1.02(5.77+0.42) 0.31(1.02+0.54) -69.6% 0.23(0.50+0.54) -77.5% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.38(1.57+0.42) 0.27(0.85+0.46) -28.9% 0.21(0.33+0.57) -44.7% See commit ("perf: add a comparison test of grep regex engines", 2017-04-19) for details on the machine the above test run was executed on. Here HEAD~2 is git with PCRE v1 without JIT, HEAD~ is PCRE v1 with JIT, and HEAD is PCRE v2 (also with JIT). See previous commits of mine mentioning p7820-grep-engines.sh for more details on the test setup. For ease of readability, a different run just of HEAD~ (PCRE v1 with JIT v.s. PCRE v2), again with just the /perl/ tests shown: [...] Test HEAD~ HEAD ---------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.21(0.42+0.52) 0.21(0.31+0.58) +0.0% 7820.7: perl grep '^how to' 0.25(0.65+0.50) 0.20(0.31+0.57) -20.0% 7820.11: perl grep '[how] to' 0.30(0.90+0.50) 0.23(0.46+0.53) -23.3% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 0.30(1.19+0.38) 0.23(0.51+0.51) -23.3% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.27(0.84+0.48) 0.21(0.34+0.57) -22.2% I.e. the two are either neck-to-neck, but PCRE v2 usually pulls ahead, when it does it's around 20% faster. A brief note on thread safety: As noted in pcre2api(3) & pcre2jit(3) the compiled pattern can be shared between threads, but not some of the JIT context, however the grep threading support does all pattern & JIT compilation in separate threads, so this code doesn't need to concern itself with thread safety. See commit 63e7e9d8b6 ("git-grep: Learn PCRE", 2011-05-09) for the initial addition of PCRE v1. This change follows some of the same patterns it did (and which were discussed on list at the time), e.g. mocking up types with typedef instead of ifdef-ing them out when USE_LIBPCRE2 isn't defined. This adds some trivial memory use to the program, but makes the code look nicer. 1. https://lists.exim.org/lurker/message/20150105.162835.0666407a.en.html 2. https://lists.exim.org/lurker/thread/20170419.172322.833ee099.en.html Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-01 18:20:56 +00:00
{
return 1;
}
static void free_pcre2_pattern(struct grep_pat *p UNUSED)
grep: add support for PCRE v2 Add support for v2 of the PCRE API. This is a new major version of PCRE that came out in early 2015[1]. The regular expression syntax is the same, but while the API is similar, pretty much every function is either renamed or takes different arguments. Thus using it via entirely new functions makes sense, as opposed to trying to e.g. have one compile_pcre_pattern() that would call either PCRE v1 or v2 functions. Git can now be compiled with either USE_LIBPCRE1=YesPlease or USE_LIBPCRE2=YesPlease, with USE_LIBPCRE=YesPlease currently being a synonym for the former. Providing both is a compile-time error. With earlier patches to enable JIT for PCRE v1 the performance of the release versions of both libraries is almost exactly the same, with PCRE v2 being around 1% slower. However after I reported this to the pcre-dev mailing list[2] I got a lot of help with the API use from Zoltán Herczeg, he subsequently optimized some of the JIT functionality in v2 of the library. Running the p7820-grep-engines.sh performance test against the latest Subversion trunk of both, with both them and git compiled as -O3, and the test run against linux.git, gives the following results. Just the /perl/ tests shown: $ GIT_PERF_REPEAT_COUNT=30 GIT_PERF_LARGE_REPO=~/g/linux GIT_PERF_MAKE_COMMAND='grep -q LIBPCRE2 Makefile && make -j8 USE_LIBPCRE2=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre2/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre2/inst/lib || make -j8 USE_LIBPCRE=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre/inst/lib' ./run HEAD~5 HEAD~ HEAD p7820-grep-engines.sh [...] Test HEAD~5 HEAD~ HEAD ----------------------------------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.31(1.10+0.48) 0.21(0.35+0.56) -32.3% 0.21(0.34+0.55) -32.3% 7820.7: perl grep '^how to' 0.56(2.70+0.40) 0.24(0.64+0.52) -57.1% 0.20(0.28+0.60) -64.3% 7820.11: perl grep '[how] to' 0.56(2.66+0.38) 0.29(0.95+0.45) -48.2% 0.23(0.45+0.54) -58.9% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 1.02(5.77+0.42) 0.31(1.02+0.54) -69.6% 0.23(0.50+0.54) -77.5% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.38(1.57+0.42) 0.27(0.85+0.46) -28.9% 0.21(0.33+0.57) -44.7% See commit ("perf: add a comparison test of grep regex engines", 2017-04-19) for details on the machine the above test run was executed on. Here HEAD~2 is git with PCRE v1 without JIT, HEAD~ is PCRE v1 with JIT, and HEAD is PCRE v2 (also with JIT). See previous commits of mine mentioning p7820-grep-engines.sh for more details on the test setup. For ease of readability, a different run just of HEAD~ (PCRE v1 with JIT v.s. PCRE v2), again with just the /perl/ tests shown: [...] Test HEAD~ HEAD ---------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.21(0.42+0.52) 0.21(0.31+0.58) +0.0% 7820.7: perl grep '^how to' 0.25(0.65+0.50) 0.20(0.31+0.57) -20.0% 7820.11: perl grep '[how] to' 0.30(0.90+0.50) 0.23(0.46+0.53) -23.3% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 0.30(1.19+0.38) 0.23(0.51+0.51) -23.3% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.27(0.84+0.48) 0.21(0.34+0.57) -22.2% I.e. the two are either neck-to-neck, but PCRE v2 usually pulls ahead, when it does it's around 20% faster. A brief note on thread safety: As noted in pcre2api(3) & pcre2jit(3) the compiled pattern can be shared between threads, but not some of the JIT context, however the grep threading support does all pattern & JIT compilation in separate threads, so this code doesn't need to concern itself with thread safety. See commit 63e7e9d8b6 ("git-grep: Learn PCRE", 2011-05-09) for the initial addition of PCRE v1. This change follows some of the same patterns it did (and which were discussed on list at the time), e.g. mocking up types with typedef instead of ifdef-ing them out when USE_LIBPCRE2 isn't defined. This adds some trivial memory use to the program, but makes the code look nicer. 1. https://lists.exim.org/lurker/message/20150105.162835.0666407a.en.html 2. https://lists.exim.org/lurker/thread/20170419.172322.833ee099.en.html Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-01 18:20:56 +00:00
{
}
static void compile_fixed_regexp(struct grep_pat *p, struct grep_opt *opt)
{
struct strbuf sb = STRBUF_INIT;
int err;
int regflags = 0;
basic_regex_quote_buf(&sb, p->pattern);
if (opt->ignore_case)
regflags |= REG_ICASE;
err = regcomp(&p->regexp, sb.buf, regflags);
strbuf_release(&sb);
if (err) {
char errbuf[1024];
regerror(err, &p->regexp, errbuf, sizeof(errbuf));
compile_regexp_failed(p, errbuf);
}
}
grep: use PCRE v2 for optimized fixed-string search Bring back optimized fixed-string search for "grep", this time with PCRE v2 as an optional backend. As noted in [1] with kwset we were slower than PCRE v1 and v2 JIT with the kwset backend, so that optimization was counterproductive. This brings back the optimization for "--fixed-strings", without changing the semantics of having a NUL-byte in patterns. As seen in previous commits in this series we could support it now, but I'd rather just leave that edge-case aside so we don't have one behavior or the other depending what "--fixed-strings" backend we're using. It makes the behavior harder to understand and document, and makes tests for the different backends more painful. This does change the behavior under non-C locales when "log"'s "--encoding" option is used and the heystack/needle in the content/command-line doesn't have a matching encoding. See the recent change in "t4210: skip more command-line encoding tests on MinGW" in this series. I think that's OK. We did nothing sensible before then (just compared raw bytes that had no hope of matching). At least now the user will get some idea why their grep/log never matches in that edge case. I could also support the PCRE v1 backend here, but that would make the code more complex. I'd rather aim for simplicity here and in future changes to the diffcore. We're not going to have someone who absolutely must have faster search, but for whom building PCRE v2 isn't acceptable. The difference between this series of commits and the current "master" is, using the same t/perf commands shown in the last commit: plain grep: Test origin/master HEAD ------------------------------------------------------------------------- 7821.1: fixed grep int 0.55(1.67+0.56) 0.41(0.98+0.60) -25.5% 7821.2: basic grep int 0.58(1.65+0.52) 0.41(0.96+0.57) -29.3% 7821.3: extended grep int 0.57(1.66+0.49) 0.42(0.93+0.60) -26.3% 7821.4: perl grep int 0.54(1.67+0.50) 0.43(0.88+0.65) -20.4% 7821.6: fixed grep uncommon 0.21(0.52+0.42) 0.16(0.24+0.51) -23.8% 7821.7: basic grep uncommon 0.20(0.49+0.45) 0.17(0.28+0.47) -15.0% 7821.8: extended grep uncommon 0.20(0.54+0.39) 0.16(0.25+0.50) -20.0% 7821.9: perl grep uncommon 0.20(0.58+0.36) 0.16(0.23+0.50) -20.0% 7821.11: fixed grep æ 0.35(1.24+0.43) 0.16(0.23+0.50) -54.3% 7821.12: basic grep æ 0.36(1.29+0.38) 0.16(0.20+0.54) -55.6% 7821.13: extended grep æ 0.35(1.23+0.44) 0.16(0.24+0.50) -54.3% 7821.14: perl grep æ 0.35(1.33+0.34) 0.16(0.28+0.46) -54.3% grep with -i: Test origin/master HEAD ---------------------------------------------------------------------------- 7821.1: fixed grep -i int 0.62(1.81+0.70) 0.47(1.11+0.64) -24.2% 7821.2: basic grep -i int 0.67(1.90+0.53) 0.46(1.07+0.62) -31.3% 7821.3: extended grep -i int 0.62(1.92+0.53) 0.53(1.12+0.58) -14.5% 7821.4: perl grep -i int 0.66(1.85+0.58) 0.45(1.10+0.59) -31.8% 7821.6: fixed grep -i uncommon 0.21(0.54+0.43) 0.17(0.20+0.55) -19.0% 7821.7: basic grep -i uncommon 0.20(0.52+0.45) 0.17(0.29+0.48) -15.0% 7821.8: extended grep -i uncommon 0.21(0.52+0.44) 0.17(0.26+0.50) -19.0% 7821.9: perl grep -i uncommon 0.21(0.53+0.44) 0.17(0.20+0.56) -19.0% 7821.11: fixed grep -i æ 0.26(0.79+0.44) 0.16(0.29+0.46) -38.5% 7821.12: basic grep -i æ 0.26(0.79+0.42) 0.16(0.20+0.54) -38.5% 7821.13: extended grep -i æ 0.26(0.84+0.39) 0.16(0.24+0.50) -38.5% 7821.14: perl grep -i æ 0.16(0.24+0.49) 0.17(0.25+0.51) +6.3% plain log: Test origin/master HEAD -------------------------------------------------------------------------------- 4221.1: fixed log --grep='int' 7.24(6.95+0.28) 7.20(6.95+0.18) -0.6% 4221.2: basic log --grep='int' 7.31(6.97+0.22) 7.20(6.93+0.21) -1.5% 4221.3: extended log --grep='int' 7.37(7.04+0.24) 7.22(6.91+0.25) -2.0% 4221.4: perl log --grep='int' 7.31(7.04+0.21) 7.19(6.89+0.21) -1.6% 4221.6: fixed log --grep='uncommon' 6.93(6.59+0.32) 7.04(6.66+0.37) +1.6% 4221.7: basic log --grep='uncommon' 6.92(6.58+0.29) 7.08(6.75+0.29) +2.3% 4221.8: extended log --grep='uncommon' 6.92(6.55+0.31) 7.00(6.68+0.31) +1.2% 4221.9: perl log --grep='uncommon' 7.03(6.59+0.33) 7.12(6.73+0.34) +1.3% 4221.11: fixed log --grep='æ' 7.41(7.08+0.28) 7.05(6.76+0.29) -4.9% 4221.12: basic log --grep='æ' 7.39(6.99+0.33) 7.00(6.68+0.25) -5.3% 4221.13: extended log --grep='æ' 7.34(7.00+0.25) 7.15(6.81+0.31) -2.6% 4221.14: perl log --grep='æ' 7.43(7.13+0.26) 7.01(6.60+0.36) -5.7% log with -i: Test origin/master HEAD ------------------------------------------------------------------------------------ 4221.1: fixed log -i --grep='int' 7.31(7.07+0.24) 7.23(7.00+0.22) -1.1% 4221.2: basic log -i --grep='int' 7.40(7.08+0.28) 7.19(6.92+0.20) -2.8% 4221.3: extended log -i --grep='int' 7.43(7.13+0.25) 7.27(6.99+0.21) -2.2% 4221.4: perl log -i --grep='int' 7.34(7.10+0.24) 7.10(6.90+0.19) -3.3% 4221.6: fixed log -i --grep='uncommon' 7.07(6.71+0.32) 7.11(6.77+0.28) +0.6% 4221.7: basic log -i --grep='uncommon' 6.99(6.64+0.28) 7.12(6.69+0.38) +1.9% 4221.8: extended log -i --grep='uncommon' 7.11(6.74+0.32) 7.10(6.77+0.27) -0.1% 4221.9: perl log -i --grep='uncommon' 6.98(6.60+0.29) 7.05(6.64+0.34) +1.0% 4221.11: fixed log -i --grep='æ' 7.85(7.45+0.34) 7.03(6.68+0.32) -10.4% 4221.12: basic log -i --grep='æ' 7.87(7.49+0.29) 7.06(6.69+0.31) -10.3% 4221.13: extended log -i --grep='æ' 7.87(7.54+0.31) 7.09(6.69+0.31) -9.9% 4221.14: perl log -i --grep='æ' 7.06(6.77+0.28) 6.91(6.57+0.31) -2.1% So as with e05b027627 ("grep: use PCRE v2 for optimized fixed-string search", 2019-06-26) there's a huge improvement in performance for "grep", but in "log" most of our time is spent elsewhere, so we don't notice it that much. Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-07-01 21:21:00 +00:00
#endif /* !USE_LIBPCRE2 */
static void compile_regexp(struct grep_pat *p, struct grep_opt *opt)
{
int err;
grep: remove regflags from the public grep_opt API Refactor calls to the grep machinery to always pass opt.ignore_case & opt.extended_regexp_option instead of setting the equivalent regflags bits. The bug fixed when making -i work with -P in commit 9e3cbc59d5 ("log: make --regexp-ignore-case work with --perl-regexp", 2017-05-20) was really just plastering over the code smell which this change fixes. The reason for adding the extensive commentary here is that I discovered some subtle complexity in implementing this that really should be called out explicitly to future readers. Before this change we'd rely on the difference between `extended_regexp_option` and `regflags` to serve as a membrane between our preliminary parsing of grep.extendedRegexp and grep.patternType, and what we decided to do internally. Now that those two are the same thing, it's necessary to unset `extended_regexp_option` just before we commit in cases where both of those config variables are set. See 84befcd0a4 ("grep: add a grep.patternType configuration setting", 2012-08-03) for the code and documentation related to that. The explanation of why the if/else branches in grep_commit_pattern_type() are ordered the way they are exists in that commit message, but I think it's worth calling this subtlety out explicitly with a comment for future readers. Even though grep_commit_pattern_type() is the only caller of grep_set_pattern_type_option() it's simpler to reset the extended_regexp_option flag in the latter, since 2/3 branches in the former would otherwise need to reset it, this way we can do it in one place. Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-29 22:22:21 +00:00
int regflags = REG_NEWLINE;
grep: simplify config parsing and option parsing Simplify the parsing of "grep.patternType" and "grep.extendedRegexp". This changes no behavior, but gets rid of complex parsing logic that isn't needed anymore. When "grep.patternType" was introduced in 84befcd0a4a (grep: add a grep.patternType configuration setting, 2012-08-03) we promised that: 1. You can set "grep.patternType", and "[setting it to] 'default' will return to the default matching behavior". In that context "the default" meant whatever the configuration system specified before that change, i.e. via grep.extendedRegexp. 2. We'd support the existing "grep.extendedRegexp" option, but ignore it when the new "grep.patternType" option is set. We said we'd only ignore the older "grep.extendedRegexp" option "when the `grep.patternType` option is set to a value other than 'default'". In a preceding commit we changed grep_config() to be called after grep_init(), which means that much of the complexity here can go away. As before both "grep.patternType" and "grep.extendedRegexp" are last-one-wins variable, with "grep.extendedRegexp" yielding to "grep.patternType", except when "grep.patternType=default". Note that as the previously added tests indicate this cannot be done on-the-fly as we see the config variables, without introducing more state keeping. I.e. if we see: -c grep.extendedRegexp=false -c grep.patternType=default -c extendedRegexp=true We need to select ERE, since grep.patternType=default unselects that variable, which normally has higher precedence, but we also need to select BRE in cases of: -c grep.extendedRegexp=true \ -c grep.extendedRegexp=false Which would not be the case for this, which select ERE: -c grep.patternType=extended \ -c grep.extendedRegexp=false Therefore we cannot do this on-the-fly in grep_config without also introducing tracking variables for not only the pattern type, but what the source of that pattern type was. So we need to decide on the pattern after our config was fully parsed. Let's do that by deferring the decision on the pattern type until it's time to compile it in compile_regexp(). By that time we've not only parsed the config, but also handled the command-line options. Those will set "opt.pattern_type_option" (*not* "opt.extended_regexp_option"!). At that point all we need to do is see if "grep.patternType" was UNSPECIFIED in the end (including an explicit "=default"), if so we'll use the "grep.extendedRegexp" configuration, if any. See my 07a3d411739 (grep: remove regflags from the public grep_opt API, 2017-06-29) for addition of the two comments being removed here, i.e. the complexity noted in that commit is now going away. 1. https://lore.kernel.org/git/patch-v8-09.10-c211bb0c69d-20220118T155211Z-avarab@gmail.com/ Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2022-02-16 00:00:39 +00:00
if (opt->pattern_type_option == GREP_PATTERN_TYPE_UNSPECIFIED)
opt->pattern_type_option = (opt->extended_regexp_option
? GREP_PATTERN_TYPE_ERE
: GREP_PATTERN_TYPE_BRE);
p->word_regexp = opt->word_regexp;
p->ignore_case = opt->ignore_case;
grep: simplify config parsing and option parsing Simplify the parsing of "grep.patternType" and "grep.extendedRegexp". This changes no behavior, but gets rid of complex parsing logic that isn't needed anymore. When "grep.patternType" was introduced in 84befcd0a4a (grep: add a grep.patternType configuration setting, 2012-08-03) we promised that: 1. You can set "grep.patternType", and "[setting it to] 'default' will return to the default matching behavior". In that context "the default" meant whatever the configuration system specified before that change, i.e. via grep.extendedRegexp. 2. We'd support the existing "grep.extendedRegexp" option, but ignore it when the new "grep.patternType" option is set. We said we'd only ignore the older "grep.extendedRegexp" option "when the `grep.patternType` option is set to a value other than 'default'". In a preceding commit we changed grep_config() to be called after grep_init(), which means that much of the complexity here can go away. As before both "grep.patternType" and "grep.extendedRegexp" are last-one-wins variable, with "grep.extendedRegexp" yielding to "grep.patternType", except when "grep.patternType=default". Note that as the previously added tests indicate this cannot be done on-the-fly as we see the config variables, without introducing more state keeping. I.e. if we see: -c grep.extendedRegexp=false -c grep.patternType=default -c extendedRegexp=true We need to select ERE, since grep.patternType=default unselects that variable, which normally has higher precedence, but we also need to select BRE in cases of: -c grep.extendedRegexp=true \ -c grep.extendedRegexp=false Which would not be the case for this, which select ERE: -c grep.patternType=extended \ -c grep.extendedRegexp=false Therefore we cannot do this on-the-fly in grep_config without also introducing tracking variables for not only the pattern type, but what the source of that pattern type was. So we need to decide on the pattern after our config was fully parsed. Let's do that by deferring the decision on the pattern type until it's time to compile it in compile_regexp(). By that time we've not only parsed the config, but also handled the command-line options. Those will set "opt.pattern_type_option" (*not* "opt.extended_regexp_option"!). At that point all we need to do is see if "grep.patternType" was UNSPECIFIED in the end (including an explicit "=default"), if so we'll use the "grep.extendedRegexp" configuration, if any. See my 07a3d411739 (grep: remove regflags from the public grep_opt API, 2017-06-29) for addition of the two comments being removed here, i.e. the complexity noted in that commit is now going away. 1. https://lore.kernel.org/git/patch-v8-09.10-c211bb0c69d-20220118T155211Z-avarab@gmail.com/ Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2022-02-16 00:00:39 +00:00
p->fixed = opt->pattern_type_option == GREP_PATTERN_TYPE_FIXED;
grep: simplify config parsing and option parsing Simplify the parsing of "grep.patternType" and "grep.extendedRegexp". This changes no behavior, but gets rid of complex parsing logic that isn't needed anymore. When "grep.patternType" was introduced in 84befcd0a4a (grep: add a grep.patternType configuration setting, 2012-08-03) we promised that: 1. You can set "grep.patternType", and "[setting it to] 'default' will return to the default matching behavior". In that context "the default" meant whatever the configuration system specified before that change, i.e. via grep.extendedRegexp. 2. We'd support the existing "grep.extendedRegexp" option, but ignore it when the new "grep.patternType" option is set. We said we'd only ignore the older "grep.extendedRegexp" option "when the `grep.patternType` option is set to a value other than 'default'". In a preceding commit we changed grep_config() to be called after grep_init(), which means that much of the complexity here can go away. As before both "grep.patternType" and "grep.extendedRegexp" are last-one-wins variable, with "grep.extendedRegexp" yielding to "grep.patternType", except when "grep.patternType=default". Note that as the previously added tests indicate this cannot be done on-the-fly as we see the config variables, without introducing more state keeping. I.e. if we see: -c grep.extendedRegexp=false -c grep.patternType=default -c extendedRegexp=true We need to select ERE, since grep.patternType=default unselects that variable, which normally has higher precedence, but we also need to select BRE in cases of: -c grep.extendedRegexp=true \ -c grep.extendedRegexp=false Which would not be the case for this, which select ERE: -c grep.patternType=extended \ -c grep.extendedRegexp=false Therefore we cannot do this on-the-fly in grep_config without also introducing tracking variables for not only the pattern type, but what the source of that pattern type was. So we need to decide on the pattern after our config was fully parsed. Let's do that by deferring the decision on the pattern type until it's time to compile it in compile_regexp(). By that time we've not only parsed the config, but also handled the command-line options. Those will set "opt.pattern_type_option" (*not* "opt.extended_regexp_option"!). At that point all we need to do is see if "grep.patternType" was UNSPECIFIED in the end (including an explicit "=default"), if so we'll use the "grep.extendedRegexp" configuration, if any. See my 07a3d411739 (grep: remove regflags from the public grep_opt API, 2017-06-29) for addition of the two comments being removed here, i.e. the complexity noted in that commit is now going away. 1. https://lore.kernel.org/git/patch-v8-09.10-c211bb0c69d-20220118T155211Z-avarab@gmail.com/ Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2022-02-16 00:00:39 +00:00
if (opt->pattern_type_option != GREP_PATTERN_TYPE_PCRE &&
memchr(p->pattern, 0, p->patternlen))
die(_("given pattern contains NULL byte (via -f <file>). This is only supported with -P under PCRE v2"));
p->is_fixed = is_fixed(p->pattern, p->patternlen);
grep: stess test PCRE v2 on invalid UTF-8 data Since my b65abcafc7 ("grep: use PCRE v2 for optimized fixed-string search", 2019-07-01) we've been dying on invalid UTF-8 data when grepping for fixed strings if the following are all true: * The subject string is non-ASCII (e.g. "ævar") * We're under a is_utf8_locale(), e.g. "en_US.UTF-8", not "C" * We compiled with PCRE v2 * That PCRE v2 did not have JIT support The last of those is why this wasn't caught earlier, per pcre2jit(3): "unless PCRE2_NO_UTF_CHECK is set, a UTF subject string is tested for validity. In the interests of speed, these checks do not happen on the JIT fast path, and if invalid data is passed, the result is undefined." I.e. the subject being matched against our pattern was invalid, but we were lucky and getting away with it on the JIT path, but the non-JIT one is stricter. This patch does nothing to fix that, instead we sneak in support for fixed patterns starting with "(*NO_JIT)", this disables the PCRE v2 jit with implicit fixed-string matching for testing, see pcre2syntax(3) the syntax. This is technically a change in behavior, but it's so obscure that I figured it was OK. We'd previously consider this an invalid regular expression as regcomp() would die on it, now we feed it to the PCRE v2 fixed-string path. I thought this was better than introducing yet another GIT_TEST_* environment variable. We're also relying on a behavior of PCRE v2 that technically could change, but I think the test coverage is worth dipping our toe into some somewhat undefined behavior. Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-07-26 15:08:16 +00:00
#ifdef USE_LIBPCRE2
if (!p->fixed && !p->is_fixed) {
const char *no_jit = "(*NO_JIT)";
const int no_jit_len = strlen(no_jit);
if (starts_with(p->pattern, no_jit) &&
is_fixed(p->pattern + no_jit_len,
p->patternlen - no_jit_len))
p->is_fixed = 1;
}
#endif
if (p->fixed || p->is_fixed) {
grep: use PCRE v2 for optimized fixed-string search Bring back optimized fixed-string search for "grep", this time with PCRE v2 as an optional backend. As noted in [1] with kwset we were slower than PCRE v1 and v2 JIT with the kwset backend, so that optimization was counterproductive. This brings back the optimization for "--fixed-strings", without changing the semantics of having a NUL-byte in patterns. As seen in previous commits in this series we could support it now, but I'd rather just leave that edge-case aside so we don't have one behavior or the other depending what "--fixed-strings" backend we're using. It makes the behavior harder to understand and document, and makes tests for the different backends more painful. This does change the behavior under non-C locales when "log"'s "--encoding" option is used and the heystack/needle in the content/command-line doesn't have a matching encoding. See the recent change in "t4210: skip more command-line encoding tests on MinGW" in this series. I think that's OK. We did nothing sensible before then (just compared raw bytes that had no hope of matching). At least now the user will get some idea why their grep/log never matches in that edge case. I could also support the PCRE v1 backend here, but that would make the code more complex. I'd rather aim for simplicity here and in future changes to the diffcore. We're not going to have someone who absolutely must have faster search, but for whom building PCRE v2 isn't acceptable. The difference between this series of commits and the current "master" is, using the same t/perf commands shown in the last commit: plain grep: Test origin/master HEAD ------------------------------------------------------------------------- 7821.1: fixed grep int 0.55(1.67+0.56) 0.41(0.98+0.60) -25.5% 7821.2: basic grep int 0.58(1.65+0.52) 0.41(0.96+0.57) -29.3% 7821.3: extended grep int 0.57(1.66+0.49) 0.42(0.93+0.60) -26.3% 7821.4: perl grep int 0.54(1.67+0.50) 0.43(0.88+0.65) -20.4% 7821.6: fixed grep uncommon 0.21(0.52+0.42) 0.16(0.24+0.51) -23.8% 7821.7: basic grep uncommon 0.20(0.49+0.45) 0.17(0.28+0.47) -15.0% 7821.8: extended grep uncommon 0.20(0.54+0.39) 0.16(0.25+0.50) -20.0% 7821.9: perl grep uncommon 0.20(0.58+0.36) 0.16(0.23+0.50) -20.0% 7821.11: fixed grep æ 0.35(1.24+0.43) 0.16(0.23+0.50) -54.3% 7821.12: basic grep æ 0.36(1.29+0.38) 0.16(0.20+0.54) -55.6% 7821.13: extended grep æ 0.35(1.23+0.44) 0.16(0.24+0.50) -54.3% 7821.14: perl grep æ 0.35(1.33+0.34) 0.16(0.28+0.46) -54.3% grep with -i: Test origin/master HEAD ---------------------------------------------------------------------------- 7821.1: fixed grep -i int 0.62(1.81+0.70) 0.47(1.11+0.64) -24.2% 7821.2: basic grep -i int 0.67(1.90+0.53) 0.46(1.07+0.62) -31.3% 7821.3: extended grep -i int 0.62(1.92+0.53) 0.53(1.12+0.58) -14.5% 7821.4: perl grep -i int 0.66(1.85+0.58) 0.45(1.10+0.59) -31.8% 7821.6: fixed grep -i uncommon 0.21(0.54+0.43) 0.17(0.20+0.55) -19.0% 7821.7: basic grep -i uncommon 0.20(0.52+0.45) 0.17(0.29+0.48) -15.0% 7821.8: extended grep -i uncommon 0.21(0.52+0.44) 0.17(0.26+0.50) -19.0% 7821.9: perl grep -i uncommon 0.21(0.53+0.44) 0.17(0.20+0.56) -19.0% 7821.11: fixed grep -i æ 0.26(0.79+0.44) 0.16(0.29+0.46) -38.5% 7821.12: basic grep -i æ 0.26(0.79+0.42) 0.16(0.20+0.54) -38.5% 7821.13: extended grep -i æ 0.26(0.84+0.39) 0.16(0.24+0.50) -38.5% 7821.14: perl grep -i æ 0.16(0.24+0.49) 0.17(0.25+0.51) +6.3% plain log: Test origin/master HEAD -------------------------------------------------------------------------------- 4221.1: fixed log --grep='int' 7.24(6.95+0.28) 7.20(6.95+0.18) -0.6% 4221.2: basic log --grep='int' 7.31(6.97+0.22) 7.20(6.93+0.21) -1.5% 4221.3: extended log --grep='int' 7.37(7.04+0.24) 7.22(6.91+0.25) -2.0% 4221.4: perl log --grep='int' 7.31(7.04+0.21) 7.19(6.89+0.21) -1.6% 4221.6: fixed log --grep='uncommon' 6.93(6.59+0.32) 7.04(6.66+0.37) +1.6% 4221.7: basic log --grep='uncommon' 6.92(6.58+0.29) 7.08(6.75+0.29) +2.3% 4221.8: extended log --grep='uncommon' 6.92(6.55+0.31) 7.00(6.68+0.31) +1.2% 4221.9: perl log --grep='uncommon' 7.03(6.59+0.33) 7.12(6.73+0.34) +1.3% 4221.11: fixed log --grep='æ' 7.41(7.08+0.28) 7.05(6.76+0.29) -4.9% 4221.12: basic log --grep='æ' 7.39(6.99+0.33) 7.00(6.68+0.25) -5.3% 4221.13: extended log --grep='æ' 7.34(7.00+0.25) 7.15(6.81+0.31) -2.6% 4221.14: perl log --grep='æ' 7.43(7.13+0.26) 7.01(6.60+0.36) -5.7% log with -i: Test origin/master HEAD ------------------------------------------------------------------------------------ 4221.1: fixed log -i --grep='int' 7.31(7.07+0.24) 7.23(7.00+0.22) -1.1% 4221.2: basic log -i --grep='int' 7.40(7.08+0.28) 7.19(6.92+0.20) -2.8% 4221.3: extended log -i --grep='int' 7.43(7.13+0.25) 7.27(6.99+0.21) -2.2% 4221.4: perl log -i --grep='int' 7.34(7.10+0.24) 7.10(6.90+0.19) -3.3% 4221.6: fixed log -i --grep='uncommon' 7.07(6.71+0.32) 7.11(6.77+0.28) +0.6% 4221.7: basic log -i --grep='uncommon' 6.99(6.64+0.28) 7.12(6.69+0.38) +1.9% 4221.8: extended log -i --grep='uncommon' 7.11(6.74+0.32) 7.10(6.77+0.27) -0.1% 4221.9: perl log -i --grep='uncommon' 6.98(6.60+0.29) 7.05(6.64+0.34) +1.0% 4221.11: fixed log -i --grep='æ' 7.85(7.45+0.34) 7.03(6.68+0.32) -10.4% 4221.12: basic log -i --grep='æ' 7.87(7.49+0.29) 7.06(6.69+0.31) -10.3% 4221.13: extended log -i --grep='æ' 7.87(7.54+0.31) 7.09(6.69+0.31) -9.9% 4221.14: perl log -i --grep='æ' 7.06(6.77+0.28) 6.91(6.57+0.31) -2.1% So as with e05b027627 ("grep: use PCRE v2 for optimized fixed-string search", 2019-06-26) there's a huge improvement in performance for "grep", but in "log" most of our time is spent elsewhere, so we don't notice it that much. Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-07-01 21:21:00 +00:00
#ifdef USE_LIBPCRE2
if (p->is_fixed) {
grep: use PCRE v2 for optimized fixed-string search Bring back optimized fixed-string search for "grep", this time with PCRE v2 as an optional backend. As noted in [1] with kwset we were slower than PCRE v1 and v2 JIT with the kwset backend, so that optimization was counterproductive. This brings back the optimization for "--fixed-strings", without changing the semantics of having a NUL-byte in patterns. As seen in previous commits in this series we could support it now, but I'd rather just leave that edge-case aside so we don't have one behavior or the other depending what "--fixed-strings" backend we're using. It makes the behavior harder to understand and document, and makes tests for the different backends more painful. This does change the behavior under non-C locales when "log"'s "--encoding" option is used and the heystack/needle in the content/command-line doesn't have a matching encoding. See the recent change in "t4210: skip more command-line encoding tests on MinGW" in this series. I think that's OK. We did nothing sensible before then (just compared raw bytes that had no hope of matching). At least now the user will get some idea why their grep/log never matches in that edge case. I could also support the PCRE v1 backend here, but that would make the code more complex. I'd rather aim for simplicity here and in future changes to the diffcore. We're not going to have someone who absolutely must have faster search, but for whom building PCRE v2 isn't acceptable. The difference between this series of commits and the current "master" is, using the same t/perf commands shown in the last commit: plain grep: Test origin/master HEAD ------------------------------------------------------------------------- 7821.1: fixed grep int 0.55(1.67+0.56) 0.41(0.98+0.60) -25.5% 7821.2: basic grep int 0.58(1.65+0.52) 0.41(0.96+0.57) -29.3% 7821.3: extended grep int 0.57(1.66+0.49) 0.42(0.93+0.60) -26.3% 7821.4: perl grep int 0.54(1.67+0.50) 0.43(0.88+0.65) -20.4% 7821.6: fixed grep uncommon 0.21(0.52+0.42) 0.16(0.24+0.51) -23.8% 7821.7: basic grep uncommon 0.20(0.49+0.45) 0.17(0.28+0.47) -15.0% 7821.8: extended grep uncommon 0.20(0.54+0.39) 0.16(0.25+0.50) -20.0% 7821.9: perl grep uncommon 0.20(0.58+0.36) 0.16(0.23+0.50) -20.0% 7821.11: fixed grep æ 0.35(1.24+0.43) 0.16(0.23+0.50) -54.3% 7821.12: basic grep æ 0.36(1.29+0.38) 0.16(0.20+0.54) -55.6% 7821.13: extended grep æ 0.35(1.23+0.44) 0.16(0.24+0.50) -54.3% 7821.14: perl grep æ 0.35(1.33+0.34) 0.16(0.28+0.46) -54.3% grep with -i: Test origin/master HEAD ---------------------------------------------------------------------------- 7821.1: fixed grep -i int 0.62(1.81+0.70) 0.47(1.11+0.64) -24.2% 7821.2: basic grep -i int 0.67(1.90+0.53) 0.46(1.07+0.62) -31.3% 7821.3: extended grep -i int 0.62(1.92+0.53) 0.53(1.12+0.58) -14.5% 7821.4: perl grep -i int 0.66(1.85+0.58) 0.45(1.10+0.59) -31.8% 7821.6: fixed grep -i uncommon 0.21(0.54+0.43) 0.17(0.20+0.55) -19.0% 7821.7: basic grep -i uncommon 0.20(0.52+0.45) 0.17(0.29+0.48) -15.0% 7821.8: extended grep -i uncommon 0.21(0.52+0.44) 0.17(0.26+0.50) -19.0% 7821.9: perl grep -i uncommon 0.21(0.53+0.44) 0.17(0.20+0.56) -19.0% 7821.11: fixed grep -i æ 0.26(0.79+0.44) 0.16(0.29+0.46) -38.5% 7821.12: basic grep -i æ 0.26(0.79+0.42) 0.16(0.20+0.54) -38.5% 7821.13: extended grep -i æ 0.26(0.84+0.39) 0.16(0.24+0.50) -38.5% 7821.14: perl grep -i æ 0.16(0.24+0.49) 0.17(0.25+0.51) +6.3% plain log: Test origin/master HEAD -------------------------------------------------------------------------------- 4221.1: fixed log --grep='int' 7.24(6.95+0.28) 7.20(6.95+0.18) -0.6% 4221.2: basic log --grep='int' 7.31(6.97+0.22) 7.20(6.93+0.21) -1.5% 4221.3: extended log --grep='int' 7.37(7.04+0.24) 7.22(6.91+0.25) -2.0% 4221.4: perl log --grep='int' 7.31(7.04+0.21) 7.19(6.89+0.21) -1.6% 4221.6: fixed log --grep='uncommon' 6.93(6.59+0.32) 7.04(6.66+0.37) +1.6% 4221.7: basic log --grep='uncommon' 6.92(6.58+0.29) 7.08(6.75+0.29) +2.3% 4221.8: extended log --grep='uncommon' 6.92(6.55+0.31) 7.00(6.68+0.31) +1.2% 4221.9: perl log --grep='uncommon' 7.03(6.59+0.33) 7.12(6.73+0.34) +1.3% 4221.11: fixed log --grep='æ' 7.41(7.08+0.28) 7.05(6.76+0.29) -4.9% 4221.12: basic log --grep='æ' 7.39(6.99+0.33) 7.00(6.68+0.25) -5.3% 4221.13: extended log --grep='æ' 7.34(7.00+0.25) 7.15(6.81+0.31) -2.6% 4221.14: perl log --grep='æ' 7.43(7.13+0.26) 7.01(6.60+0.36) -5.7% log with -i: Test origin/master HEAD ------------------------------------------------------------------------------------ 4221.1: fixed log -i --grep='int' 7.31(7.07+0.24) 7.23(7.00+0.22) -1.1% 4221.2: basic log -i --grep='int' 7.40(7.08+0.28) 7.19(6.92+0.20) -2.8% 4221.3: extended log -i --grep='int' 7.43(7.13+0.25) 7.27(6.99+0.21) -2.2% 4221.4: perl log -i --grep='int' 7.34(7.10+0.24) 7.10(6.90+0.19) -3.3% 4221.6: fixed log -i --grep='uncommon' 7.07(6.71+0.32) 7.11(6.77+0.28) +0.6% 4221.7: basic log -i --grep='uncommon' 6.99(6.64+0.28) 7.12(6.69+0.38) +1.9% 4221.8: extended log -i --grep='uncommon' 7.11(6.74+0.32) 7.10(6.77+0.27) -0.1% 4221.9: perl log -i --grep='uncommon' 6.98(6.60+0.29) 7.05(6.64+0.34) +1.0% 4221.11: fixed log -i --grep='æ' 7.85(7.45+0.34) 7.03(6.68+0.32) -10.4% 4221.12: basic log -i --grep='æ' 7.87(7.49+0.29) 7.06(6.69+0.31) -10.3% 4221.13: extended log -i --grep='æ' 7.87(7.54+0.31) 7.09(6.69+0.31) -9.9% 4221.14: perl log -i --grep='æ' 7.06(6.77+0.28) 6.91(6.57+0.31) -2.1% So as with e05b027627 ("grep: use PCRE v2 for optimized fixed-string search", 2019-06-26) there's a huge improvement in performance for "grep", but in "log" most of our time is spent elsewhere, so we don't notice it that much. Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-07-01 21:21:00 +00:00
compile_pcre2_pattern(p, opt);
} else {
/*
* E.g. t7811-grep-open.sh relies on the
* pattern being restored.
*/
char *old_pattern = p->pattern;
size_t old_patternlen = p->patternlen;
struct strbuf sb = STRBUF_INIT;
/*
* There is the PCRE2_LITERAL flag, but it's
* only in PCRE v2 10.30 and later. Needing to
* ifdef our way around that and dealing with
* it + PCRE2_MULTILINE being an error is more
* complex than just quoting this ourselves.
*/
strbuf_add(&sb, "\\Q", 2);
strbuf_add(&sb, p->pattern, p->patternlen);
strbuf_add(&sb, "\\E", 2);
p->pattern = sb.buf;
p->patternlen = sb.len;
compile_pcre2_pattern(p, opt);
p->pattern = old_pattern;
p->patternlen = old_patternlen;
strbuf_release(&sb);
}
#else /* !USE_LIBPCRE2 */
compile_fixed_regexp(p, opt);
grep: use PCRE v2 for optimized fixed-string search Bring back optimized fixed-string search for "grep", this time with PCRE v2 as an optional backend. As noted in [1] with kwset we were slower than PCRE v1 and v2 JIT with the kwset backend, so that optimization was counterproductive. This brings back the optimization for "--fixed-strings", without changing the semantics of having a NUL-byte in patterns. As seen in previous commits in this series we could support it now, but I'd rather just leave that edge-case aside so we don't have one behavior or the other depending what "--fixed-strings" backend we're using. It makes the behavior harder to understand and document, and makes tests for the different backends more painful. This does change the behavior under non-C locales when "log"'s "--encoding" option is used and the heystack/needle in the content/command-line doesn't have a matching encoding. See the recent change in "t4210: skip more command-line encoding tests on MinGW" in this series. I think that's OK. We did nothing sensible before then (just compared raw bytes that had no hope of matching). At least now the user will get some idea why their grep/log never matches in that edge case. I could also support the PCRE v1 backend here, but that would make the code more complex. I'd rather aim for simplicity here and in future changes to the diffcore. We're not going to have someone who absolutely must have faster search, but for whom building PCRE v2 isn't acceptable. The difference between this series of commits and the current "master" is, using the same t/perf commands shown in the last commit: plain grep: Test origin/master HEAD ------------------------------------------------------------------------- 7821.1: fixed grep int 0.55(1.67+0.56) 0.41(0.98+0.60) -25.5% 7821.2: basic grep int 0.58(1.65+0.52) 0.41(0.96+0.57) -29.3% 7821.3: extended grep int 0.57(1.66+0.49) 0.42(0.93+0.60) -26.3% 7821.4: perl grep int 0.54(1.67+0.50) 0.43(0.88+0.65) -20.4% 7821.6: fixed grep uncommon 0.21(0.52+0.42) 0.16(0.24+0.51) -23.8% 7821.7: basic grep uncommon 0.20(0.49+0.45) 0.17(0.28+0.47) -15.0% 7821.8: extended grep uncommon 0.20(0.54+0.39) 0.16(0.25+0.50) -20.0% 7821.9: perl grep uncommon 0.20(0.58+0.36) 0.16(0.23+0.50) -20.0% 7821.11: fixed grep æ 0.35(1.24+0.43) 0.16(0.23+0.50) -54.3% 7821.12: basic grep æ 0.36(1.29+0.38) 0.16(0.20+0.54) -55.6% 7821.13: extended grep æ 0.35(1.23+0.44) 0.16(0.24+0.50) -54.3% 7821.14: perl grep æ 0.35(1.33+0.34) 0.16(0.28+0.46) -54.3% grep with -i: Test origin/master HEAD ---------------------------------------------------------------------------- 7821.1: fixed grep -i int 0.62(1.81+0.70) 0.47(1.11+0.64) -24.2% 7821.2: basic grep -i int 0.67(1.90+0.53) 0.46(1.07+0.62) -31.3% 7821.3: extended grep -i int 0.62(1.92+0.53) 0.53(1.12+0.58) -14.5% 7821.4: perl grep -i int 0.66(1.85+0.58) 0.45(1.10+0.59) -31.8% 7821.6: fixed grep -i uncommon 0.21(0.54+0.43) 0.17(0.20+0.55) -19.0% 7821.7: basic grep -i uncommon 0.20(0.52+0.45) 0.17(0.29+0.48) -15.0% 7821.8: extended grep -i uncommon 0.21(0.52+0.44) 0.17(0.26+0.50) -19.0% 7821.9: perl grep -i uncommon 0.21(0.53+0.44) 0.17(0.20+0.56) -19.0% 7821.11: fixed grep -i æ 0.26(0.79+0.44) 0.16(0.29+0.46) -38.5% 7821.12: basic grep -i æ 0.26(0.79+0.42) 0.16(0.20+0.54) -38.5% 7821.13: extended grep -i æ 0.26(0.84+0.39) 0.16(0.24+0.50) -38.5% 7821.14: perl grep -i æ 0.16(0.24+0.49) 0.17(0.25+0.51) +6.3% plain log: Test origin/master HEAD -------------------------------------------------------------------------------- 4221.1: fixed log --grep='int' 7.24(6.95+0.28) 7.20(6.95+0.18) -0.6% 4221.2: basic log --grep='int' 7.31(6.97+0.22) 7.20(6.93+0.21) -1.5% 4221.3: extended log --grep='int' 7.37(7.04+0.24) 7.22(6.91+0.25) -2.0% 4221.4: perl log --grep='int' 7.31(7.04+0.21) 7.19(6.89+0.21) -1.6% 4221.6: fixed log --grep='uncommon' 6.93(6.59+0.32) 7.04(6.66+0.37) +1.6% 4221.7: basic log --grep='uncommon' 6.92(6.58+0.29) 7.08(6.75+0.29) +2.3% 4221.8: extended log --grep='uncommon' 6.92(6.55+0.31) 7.00(6.68+0.31) +1.2% 4221.9: perl log --grep='uncommon' 7.03(6.59+0.33) 7.12(6.73+0.34) +1.3% 4221.11: fixed log --grep='æ' 7.41(7.08+0.28) 7.05(6.76+0.29) -4.9% 4221.12: basic log --grep='æ' 7.39(6.99+0.33) 7.00(6.68+0.25) -5.3% 4221.13: extended log --grep='æ' 7.34(7.00+0.25) 7.15(6.81+0.31) -2.6% 4221.14: perl log --grep='æ' 7.43(7.13+0.26) 7.01(6.60+0.36) -5.7% log with -i: Test origin/master HEAD ------------------------------------------------------------------------------------ 4221.1: fixed log -i --grep='int' 7.31(7.07+0.24) 7.23(7.00+0.22) -1.1% 4221.2: basic log -i --grep='int' 7.40(7.08+0.28) 7.19(6.92+0.20) -2.8% 4221.3: extended log -i --grep='int' 7.43(7.13+0.25) 7.27(6.99+0.21) -2.2% 4221.4: perl log -i --grep='int' 7.34(7.10+0.24) 7.10(6.90+0.19) -3.3% 4221.6: fixed log -i --grep='uncommon' 7.07(6.71+0.32) 7.11(6.77+0.28) +0.6% 4221.7: basic log -i --grep='uncommon' 6.99(6.64+0.28) 7.12(6.69+0.38) +1.9% 4221.8: extended log -i --grep='uncommon' 7.11(6.74+0.32) 7.10(6.77+0.27) -0.1% 4221.9: perl log -i --grep='uncommon' 6.98(6.60+0.29) 7.05(6.64+0.34) +1.0% 4221.11: fixed log -i --grep='æ' 7.85(7.45+0.34) 7.03(6.68+0.32) -10.4% 4221.12: basic log -i --grep='æ' 7.87(7.49+0.29) 7.06(6.69+0.31) -10.3% 4221.13: extended log -i --grep='æ' 7.87(7.54+0.31) 7.09(6.69+0.31) -9.9% 4221.14: perl log -i --grep='æ' 7.06(6.77+0.28) 6.91(6.57+0.31) -2.1% So as with e05b027627 ("grep: use PCRE v2 for optimized fixed-string search", 2019-06-26) there's a huge improvement in performance for "grep", but in "log" most of our time is spent elsewhere, so we don't notice it that much. Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-07-01 21:21:00 +00:00
#endif /* !USE_LIBPCRE2 */
return;
Use kwset in grep Benchmarks for the hot cache case: before: $ perf stat --repeat=5 git grep qwerty > /dev/null Performance counter stats for 'git grep qwerty' (5 runs): 3,478,085 cache-misses # 2.322 M/sec ( +- 2.690% ) 11,356,177 cache-references # 7.582 M/sec ( +- 2.598% ) 3,872,184 branch-misses # 0.363 % ( +- 0.258% ) 1,067,367,848 branches # 712.673 M/sec ( +- 2.622% ) 3,828,370,782 instructions # 0.947 IPC ( +- 0.033% ) 4,043,832,831 cycles # 2700.037 M/sec ( +- 0.167% ) 8,518 page-faults # 0.006 M/sec ( +- 3.648% ) 847 CPU-migrations # 0.001 M/sec ( +- 3.262% ) 6,546 context-switches # 0.004 M/sec ( +- 2.292% ) 1497.695495 task-clock-msecs # 3.303 CPUs ( +- 2.550% ) 0.453394396 seconds time elapsed ( +- 0.912% ) after: $ perf stat --repeat=5 git grep qwerty > /dev/null Performance counter stats for 'git grep qwerty' (5 runs): 2,989,918 cache-misses # 3.166 M/sec ( +- 5.013% ) 10,986,041 cache-references # 11.633 M/sec ( +- 4.899% ) (scaled from 95.06%) 3,511,993 branch-misses # 1.422 % ( +- 0.785% ) 246,893,561 branches # 261.433 M/sec ( +- 3.967% ) 1,392,727,757 instructions # 0.564 IPC ( +- 0.040% ) 2,468,142,397 cycles # 2613.494 M/sec ( +- 0.110% ) 7,747 page-faults # 0.008 M/sec ( +- 3.995% ) 897 CPU-migrations # 0.001 M/sec ( +- 2.383% ) 6,535 context-switches # 0.007 M/sec ( +- 1.993% ) 944.384228 task-clock-msecs # 3.177 CPUs ( +- 0.268% ) 0.297257643 seconds time elapsed ( +- 0.450% ) So we gain about 35% by using the kwset code. As a side effect of using kwset two grep tests are fixed by this patch. The first is fixed because kwset can deal with case-insensitive search containing NULs, something strcasestr cannot do. The second one is fixed because we consider patterns containing NULs as fixed strings (regcomp cannot accept patterns with NULs). Signed-off-by: Fredrik Kuivinen <frekui@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2011-08-20 22:42:18 +00:00
}
grep: simplify config parsing and option parsing Simplify the parsing of "grep.patternType" and "grep.extendedRegexp". This changes no behavior, but gets rid of complex parsing logic that isn't needed anymore. When "grep.patternType" was introduced in 84befcd0a4a (grep: add a grep.patternType configuration setting, 2012-08-03) we promised that: 1. You can set "grep.patternType", and "[setting it to] 'default' will return to the default matching behavior". In that context "the default" meant whatever the configuration system specified before that change, i.e. via grep.extendedRegexp. 2. We'd support the existing "grep.extendedRegexp" option, but ignore it when the new "grep.patternType" option is set. We said we'd only ignore the older "grep.extendedRegexp" option "when the `grep.patternType` option is set to a value other than 'default'". In a preceding commit we changed grep_config() to be called after grep_init(), which means that much of the complexity here can go away. As before both "grep.patternType" and "grep.extendedRegexp" are last-one-wins variable, with "grep.extendedRegexp" yielding to "grep.patternType", except when "grep.patternType=default". Note that as the previously added tests indicate this cannot be done on-the-fly as we see the config variables, without introducing more state keeping. I.e. if we see: -c grep.extendedRegexp=false -c grep.patternType=default -c extendedRegexp=true We need to select ERE, since grep.patternType=default unselects that variable, which normally has higher precedence, but we also need to select BRE in cases of: -c grep.extendedRegexp=true \ -c grep.extendedRegexp=false Which would not be the case for this, which select ERE: -c grep.patternType=extended \ -c grep.extendedRegexp=false Therefore we cannot do this on-the-fly in grep_config without also introducing tracking variables for not only the pattern type, but what the source of that pattern type was. So we need to decide on the pattern after our config was fully parsed. Let's do that by deferring the decision on the pattern type until it's time to compile it in compile_regexp(). By that time we've not only parsed the config, but also handled the command-line options. Those will set "opt.pattern_type_option" (*not* "opt.extended_regexp_option"!). At that point all we need to do is see if "grep.patternType" was UNSPECIFIED in the end (including an explicit "=default"), if so we'll use the "grep.extendedRegexp" configuration, if any. See my 07a3d411739 (grep: remove regflags from the public grep_opt API, 2017-06-29) for addition of the two comments being removed here, i.e. the complexity noted in that commit is now going away. 1. https://lore.kernel.org/git/patch-v8-09.10-c211bb0c69d-20220118T155211Z-avarab@gmail.com/ Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2022-02-16 00:00:39 +00:00
if (opt->pattern_type_option == GREP_PATTERN_TYPE_PCRE) {
grep: add support for PCRE v2 Add support for v2 of the PCRE API. This is a new major version of PCRE that came out in early 2015[1]. The regular expression syntax is the same, but while the API is similar, pretty much every function is either renamed or takes different arguments. Thus using it via entirely new functions makes sense, as opposed to trying to e.g. have one compile_pcre_pattern() that would call either PCRE v1 or v2 functions. Git can now be compiled with either USE_LIBPCRE1=YesPlease or USE_LIBPCRE2=YesPlease, with USE_LIBPCRE=YesPlease currently being a synonym for the former. Providing both is a compile-time error. With earlier patches to enable JIT for PCRE v1 the performance of the release versions of both libraries is almost exactly the same, with PCRE v2 being around 1% slower. However after I reported this to the pcre-dev mailing list[2] I got a lot of help with the API use from Zoltán Herczeg, he subsequently optimized some of the JIT functionality in v2 of the library. Running the p7820-grep-engines.sh performance test against the latest Subversion trunk of both, with both them and git compiled as -O3, and the test run against linux.git, gives the following results. Just the /perl/ tests shown: $ GIT_PERF_REPEAT_COUNT=30 GIT_PERF_LARGE_REPO=~/g/linux GIT_PERF_MAKE_COMMAND='grep -q LIBPCRE2 Makefile && make -j8 USE_LIBPCRE2=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre2/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre2/inst/lib || make -j8 USE_LIBPCRE=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre/inst/lib' ./run HEAD~5 HEAD~ HEAD p7820-grep-engines.sh [...] Test HEAD~5 HEAD~ HEAD ----------------------------------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.31(1.10+0.48) 0.21(0.35+0.56) -32.3% 0.21(0.34+0.55) -32.3% 7820.7: perl grep '^how to' 0.56(2.70+0.40) 0.24(0.64+0.52) -57.1% 0.20(0.28+0.60) -64.3% 7820.11: perl grep '[how] to' 0.56(2.66+0.38) 0.29(0.95+0.45) -48.2% 0.23(0.45+0.54) -58.9% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 1.02(5.77+0.42) 0.31(1.02+0.54) -69.6% 0.23(0.50+0.54) -77.5% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.38(1.57+0.42) 0.27(0.85+0.46) -28.9% 0.21(0.33+0.57) -44.7% See commit ("perf: add a comparison test of grep regex engines", 2017-04-19) for details on the machine the above test run was executed on. Here HEAD~2 is git with PCRE v1 without JIT, HEAD~ is PCRE v1 with JIT, and HEAD is PCRE v2 (also with JIT). See previous commits of mine mentioning p7820-grep-engines.sh for more details on the test setup. For ease of readability, a different run just of HEAD~ (PCRE v1 with JIT v.s. PCRE v2), again with just the /perl/ tests shown: [...] Test HEAD~ HEAD ---------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.21(0.42+0.52) 0.21(0.31+0.58) +0.0% 7820.7: perl grep '^how to' 0.25(0.65+0.50) 0.20(0.31+0.57) -20.0% 7820.11: perl grep '[how] to' 0.30(0.90+0.50) 0.23(0.46+0.53) -23.3% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 0.30(1.19+0.38) 0.23(0.51+0.51) -23.3% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.27(0.84+0.48) 0.21(0.34+0.57) -22.2% I.e. the two are either neck-to-neck, but PCRE v2 usually pulls ahead, when it does it's around 20% faster. A brief note on thread safety: As noted in pcre2api(3) & pcre2jit(3) the compiled pattern can be shared between threads, but not some of the JIT context, however the grep threading support does all pattern & JIT compilation in separate threads, so this code doesn't need to concern itself with thread safety. See commit 63e7e9d8b6 ("git-grep: Learn PCRE", 2011-05-09) for the initial addition of PCRE v1. This change follows some of the same patterns it did (and which were discussed on list at the time), e.g. mocking up types with typedef instead of ifdef-ing them out when USE_LIBPCRE2 isn't defined. This adds some trivial memory use to the program, but makes the code look nicer. 1. https://lists.exim.org/lurker/message/20150105.162835.0666407a.en.html 2. https://lists.exim.org/lurker/thread/20170419.172322.833ee099.en.html Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-01 18:20:56 +00:00
compile_pcre2_pattern(p, opt);
return;
}
grep: remove regflags from the public grep_opt API Refactor calls to the grep machinery to always pass opt.ignore_case & opt.extended_regexp_option instead of setting the equivalent regflags bits. The bug fixed when making -i work with -P in commit 9e3cbc59d5 ("log: make --regexp-ignore-case work with --perl-regexp", 2017-05-20) was really just plastering over the code smell which this change fixes. The reason for adding the extensive commentary here is that I discovered some subtle complexity in implementing this that really should be called out explicitly to future readers. Before this change we'd rely on the difference between `extended_regexp_option` and `regflags` to serve as a membrane between our preliminary parsing of grep.extendedRegexp and grep.patternType, and what we decided to do internally. Now that those two are the same thing, it's necessary to unset `extended_regexp_option` just before we commit in cases where both of those config variables are set. See 84befcd0a4 ("grep: add a grep.patternType configuration setting", 2012-08-03) for the code and documentation related to that. The explanation of why the if/else branches in grep_commit_pattern_type() are ordered the way they are exists in that commit message, but I think it's worth calling this subtlety out explicitly with a comment for future readers. Even though grep_commit_pattern_type() is the only caller of grep_set_pattern_type_option() it's simpler to reset the extended_regexp_option flag in the latter, since 2/3 branches in the former would otherwise need to reset it, this way we can do it in one place. Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-29 22:22:21 +00:00
if (p->ignore_case)
regflags |= REG_ICASE;
grep: simplify config parsing and option parsing Simplify the parsing of "grep.patternType" and "grep.extendedRegexp". This changes no behavior, but gets rid of complex parsing logic that isn't needed anymore. When "grep.patternType" was introduced in 84befcd0a4a (grep: add a grep.patternType configuration setting, 2012-08-03) we promised that: 1. You can set "grep.patternType", and "[setting it to] 'default' will return to the default matching behavior". In that context "the default" meant whatever the configuration system specified before that change, i.e. via grep.extendedRegexp. 2. We'd support the existing "grep.extendedRegexp" option, but ignore it when the new "grep.patternType" option is set. We said we'd only ignore the older "grep.extendedRegexp" option "when the `grep.patternType` option is set to a value other than 'default'". In a preceding commit we changed grep_config() to be called after grep_init(), which means that much of the complexity here can go away. As before both "grep.patternType" and "grep.extendedRegexp" are last-one-wins variable, with "grep.extendedRegexp" yielding to "grep.patternType", except when "grep.patternType=default". Note that as the previously added tests indicate this cannot be done on-the-fly as we see the config variables, without introducing more state keeping. I.e. if we see: -c grep.extendedRegexp=false -c grep.patternType=default -c extendedRegexp=true We need to select ERE, since grep.patternType=default unselects that variable, which normally has higher precedence, but we also need to select BRE in cases of: -c grep.extendedRegexp=true \ -c grep.extendedRegexp=false Which would not be the case for this, which select ERE: -c grep.patternType=extended \ -c grep.extendedRegexp=false Therefore we cannot do this on-the-fly in grep_config without also introducing tracking variables for not only the pattern type, but what the source of that pattern type was. So we need to decide on the pattern after our config was fully parsed. Let's do that by deferring the decision on the pattern type until it's time to compile it in compile_regexp(). By that time we've not only parsed the config, but also handled the command-line options. Those will set "opt.pattern_type_option" (*not* "opt.extended_regexp_option"!). At that point all we need to do is see if "grep.patternType" was UNSPECIFIED in the end (including an explicit "=default"), if so we'll use the "grep.extendedRegexp" configuration, if any. See my 07a3d411739 (grep: remove regflags from the public grep_opt API, 2017-06-29) for addition of the two comments being removed here, i.e. the complexity noted in that commit is now going away. 1. https://lore.kernel.org/git/patch-v8-09.10-c211bb0c69d-20220118T155211Z-avarab@gmail.com/ Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2022-02-16 00:00:39 +00:00
if (opt->pattern_type_option == GREP_PATTERN_TYPE_ERE)
grep: remove regflags from the public grep_opt API Refactor calls to the grep machinery to always pass opt.ignore_case & opt.extended_regexp_option instead of setting the equivalent regflags bits. The bug fixed when making -i work with -P in commit 9e3cbc59d5 ("log: make --regexp-ignore-case work with --perl-regexp", 2017-05-20) was really just plastering over the code smell which this change fixes. The reason for adding the extensive commentary here is that I discovered some subtle complexity in implementing this that really should be called out explicitly to future readers. Before this change we'd rely on the difference between `extended_regexp_option` and `regflags` to serve as a membrane between our preliminary parsing of grep.extendedRegexp and grep.patternType, and what we decided to do internally. Now that those two are the same thing, it's necessary to unset `extended_regexp_option` just before we commit in cases where both of those config variables are set. See 84befcd0a4 ("grep: add a grep.patternType configuration setting", 2012-08-03) for the code and documentation related to that. The explanation of why the if/else branches in grep_commit_pattern_type() are ordered the way they are exists in that commit message, but I think it's worth calling this subtlety out explicitly with a comment for future readers. Even though grep_commit_pattern_type() is the only caller of grep_set_pattern_type_option() it's simpler to reset the extended_regexp_option flag in the latter, since 2/3 branches in the former would otherwise need to reset it, this way we can do it in one place. Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-29 22:22:21 +00:00
regflags |= REG_EXTENDED;
err = regcomp(&p->regexp, p->pattern, regflags);
if (err) {
char errbuf[1024];
regerror(err, &p->regexp, errbuf, 1024);
compile_regexp_failed(p, errbuf);
}
}
static struct grep_expr *grep_not_expr(struct grep_expr *expr)
{
struct grep_expr *z = xcalloc(1, sizeof(*z));
z->node = GREP_NODE_NOT;
z->u.unary = expr;
return z;
}
static struct grep_expr *grep_binexp(enum grep_expr_node kind,
struct grep_expr *left,
struct grep_expr *right)
{
struct grep_expr *z = xcalloc(1, sizeof(*z));
z->node = kind;
z->u.binary.left = left;
z->u.binary.right = right;
return z;
}
static struct grep_expr *grep_or_expr(struct grep_expr *left, struct grep_expr *right)
{
return grep_binexp(GREP_NODE_OR, left, right);
}
static struct grep_expr *grep_and_expr(struct grep_expr *left, struct grep_expr *right)
{
return grep_binexp(GREP_NODE_AND, left, right);
}
static struct grep_expr *compile_pattern_or(struct grep_pat **);
static struct grep_expr *compile_pattern_atom(struct grep_pat **list)
{
struct grep_pat *p;
struct grep_expr *x;
p = *list;
if (!p)
return NULL;
switch (p->token) {
case GREP_PATTERN: /* atom */
case GREP_PATTERN_HEAD:
case GREP_PATTERN_BODY:
CALLOC_ARRAY(x, 1);
x->node = GREP_NODE_ATOM;
x->u.atom = p;
*list = p->next;
return x;
case GREP_OPEN_PAREN:
*list = p->next;
x = compile_pattern_or(list);
if (!*list || (*list)->token != GREP_CLOSE_PAREN)
die("unmatched parenthesis");
*list = (*list)->next;
return x;
default:
return NULL;
}
}
static struct grep_expr *compile_pattern_not(struct grep_pat **list)
{
struct grep_pat *p;
struct grep_expr *x;
p = *list;
if (!p)
return NULL;
switch (p->token) {
case GREP_NOT:
if (!p->next)
die("--not not followed by pattern expression");
*list = p->next;
x = compile_pattern_not(list);
if (!x)
die("--not followed by non pattern expression");
return grep_not_expr(x);
default:
return compile_pattern_atom(list);
}
}
static struct grep_expr *compile_pattern_and(struct grep_pat **list)
{
struct grep_pat *p;
struct grep_expr *x, *y;
x = compile_pattern_not(list);
p = *list;
if (p && p->token == GREP_AND) {
if (!x)
die("--and not preceded by pattern expression");
if (!p->next)
die("--and not followed by pattern expression");
*list = p->next;
y = compile_pattern_and(list);
if (!y)
die("--and not followed by pattern expression");
return grep_and_expr(x, y);
}
return x;
}
static struct grep_expr *compile_pattern_or(struct grep_pat **list)
{
struct grep_pat *p;
struct grep_expr *x, *y;
x = compile_pattern_and(list);
p = *list;
if (x && p && p->token != GREP_CLOSE_PAREN) {
y = compile_pattern_or(list);
if (!y)
die("not a pattern expression %s", p->pattern);
return grep_or_expr(x, y);
}
return x;
}
static struct grep_expr *compile_pattern_expr(struct grep_pat **list)
{
return compile_pattern_or(list);
}
static struct grep_expr *grep_true_expr(void)
{
struct grep_expr *z = xcalloc(1, sizeof(*z));
z->node = GREP_NODE_TRUE;
return z;
}
static struct grep_expr *prep_header_patterns(struct grep_opt *opt)
{
struct grep_pat *p;
struct grep_expr *header_expr;
struct grep_expr *(header_group[GREP_HEADER_FIELD_MAX]);
enum grep_header_field fld;
if (!opt->header_list)
return NULL;
for (p = opt->header_list; p; p = p->next) {
if (p->token != GREP_PATTERN_HEAD)
BUG("a non-header pattern in grep header list.");
if (p->field < GREP_HEADER_FIELD_MIN ||
GREP_HEADER_FIELD_MAX <= p->field)
BUG("unknown header field %d", p->field);
compile_regexp(p, opt);
}
for (fld = 0; fld < GREP_HEADER_FIELD_MAX; fld++)
header_group[fld] = NULL;
for (p = opt->header_list; p; p = p->next) {
struct grep_expr *h;
struct grep_pat *pp = p;
h = compile_pattern_atom(&pp);
if (!h || pp != p->next)
BUG("malformed header expr");
if (!header_group[p->field]) {
header_group[p->field] = h;
continue;
}
header_group[p->field] = grep_or_expr(h, header_group[p->field]);
}
header_expr = NULL;
for (fld = 0; fld < GREP_HEADER_FIELD_MAX; fld++) {
if (!header_group[fld])
continue;
if (!header_expr)
header_expr = grep_true_expr();
header_expr = grep_or_expr(header_group[fld], header_expr);
}
return header_expr;
}
log --grep/--author: honor --all-match honored for multiple --grep patterns When we have both header expression (which has to be an OR node by construction) and a pattern expression (which could be anything), we create a new top-level OR node to bind them together, and the resulting expression structure looks like this: OR / \ / \ pattern OR / \ / \ ..... committer OR / \ author TRUE The three elements on the top-level backbone that are inspected by the "all-match" logic are "pattern", "committer" and "author". When there are more than one elements in the "pattern", the top-level node of the "pattern" part of the subtree is an OR, and that node is inspected by "all-match". The result ends up ignoring the "--all-match" given from the command line. A match on either side of the pattern is considered a match, hence: git log --grep=A --grep=B --author=C --all-match shows the same "authored by C and has either A or B" that is correct only when run without "--all-match". Fix this by turning the resulting expression around when "--all-match" is in effect, like this: OR / \ / \ / OR committer / \ author \ pattern The set of nodes on the top-level backbone in the resulting expression becomes "committer", "author", and the nodes that are on the top-level backbone of the "pattern" subexpression. This makes the "all-match" logic inspect the same nodes in "pattern" as the case without the author and/or the committer restriction, and makes the earlier "log" example to show "authored by C and has A and has B", which is what the command line expects. Signed-off-by: Junio C Hamano <gitster@pobox.com>
2012-09-13 23:26:57 +00:00
static struct grep_expr *grep_splice_or(struct grep_expr *x, struct grep_expr *y)
{
struct grep_expr *z = x;
while (x) {
assert(x->node == GREP_NODE_OR);
if (x->u.binary.right &&
x->u.binary.right->node == GREP_NODE_TRUE) {
x->u.binary.right = y;
break;
}
x = x->u.binary.right;
}
return z;
}
grep/log: remove hidden --debug and --grep-debug options Remove the hidden "grep --debug" and "log --grep-debug" options added in 17bf35a3c7b (grep: teach --debug option to dump the parse tree, 2012-09-13). At the time these options seem to have been intended to go along with a documentation discussion and to help the author of relevant tests to perform ad-hoc debugging on them[1]. Reasons to want this gone: 1. They were never documented, and the only (rather trivial) use of them in our own codebase for testing is something I removed back in e01b4dab01e (grep: change non-ASCII -i test to stop using --debug, 2017-05-20). 2. Googling around doesn't show any in-the-wild uses I could dig up, and on the Git ML the only mentions after the original discussion seem to have been when they came up in unrelated diff contexts, or that test commit of mine. 3. An exception to that is c581e4a7499 (grep: under --debug, show whether PCRE JIT is enabled, 2019-08-18) where we added the ability to dump out when PCREv2 has the JIT in effect. The combination of that and my earlier b65abcafc7a (grep: use PCRE v2 for optimized fixed-string search, 2019-07-01) means Git prints this out in its most common in-the-wild configuration: $ git log --grep-debug --grep=foo --grep=bar --grep=baz --all-match pcre2_jit_on=1 pcre2_jit_on=1 pcre2_jit_on=1 [all-match] (or pattern_body<body>foo (or pattern_body<body>bar pattern_body<body>baz ) ) $ git grep --debug \( -e foo --and -e bar \) --or -e baz pcre2_jit_on=1 pcre2_jit_on=1 pcre2_jit_on=1 (or (and patternfoo patternbar ) patternbaz ) I.e. for each pattern we're considering for the and/or/--all-match etc. debugging we'll now diligently spew out another identical line saying whether the PCREv2 JIT is on or not. I think that nobody's complained about that rather glaringly obviously bad output says something about how much this is used, i.e. it's not. The need for this debugging aid for the composed grep/log patterns seems to have passed, and the desire to dump the JIT config seems to have been another one-off around the time we had JIT-related issues on the PCREv2 codepath. That the original author of this debugging facility seemingly hasn't noticed the bad output since then[2] is probably some indicator. 1. https://lore.kernel.org/git/cover.1347615361.git.git@drmicha.warpmail.net/ 2. https://lore.kernel.org/git/xmqqk1b8x0ac.fsf@gitster-ct.c.googlers.com/ Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-01-25 23:36:51 +00:00
void compile_grep_patterns(struct grep_opt *opt)
{
struct grep_pat *p;
struct grep_expr *header_expr = prep_header_patterns(opt);
grep.c: remove "extended" in favor of "pattern_expression", fix segfault Since 79d3696cfb4 (git-grep: boolean expression on pattern matching., 2006-06-30) the "pattern_expression" member has been used for complex queries (AND/OR...), with "pattern_list" being used for the simple OR queries. Since then we've used both "pattern_expression" and its associated boolean "extended" member to see if we have a complex expression. Since f41fb662f57 (revisions API: have release_revisions() release "grep_filter", 2022-04-13) we've had a subtle bug relating to that: If we supplied options that were only used for "complex queries", but didn't supply the query itself we'd set "opt->extended", but would have a NULL "pattern_expression". As a result these would segfault as we tried to call "free_grep_patterns()" from "release_revisions()": git -P log -1 --invert-grep git -P log -1 --all-match The root cause of this is that we were conflating the state management we needed in "compile_grep_patterns()" itself with whether or not we had an "opt->pattern_expression" later on. In this cases as we're going through "compile_grep_patterns()" we have no "opt->pattern_list" but have "opt->no_body_match" or "opt->all_match". So we'd set "opt->extended = 1", but not "return" on "opt->extended" as that's an "else if" in the same "if" statement. That behavior is intentional and required, as the common case is that we have an "opt->pattern_list" that we're about to parse into the "opt->pattern_expression". But we don't need to keep track of this "extended" flag beyond the state management in compile_grep_patterns() itself. It needs it, but once we're out of that function we can rely on "opt->pattern_expression" being non-NULL instead for using these extended patterns. As 79d3696cfb4 itself shows we've assumed that there's a one-to-one mapping between the two since the very beginning. I.e. "match_line()" would check "opt->extended" to see if it should call "match_expr()", and the first thing we do in that function is assume that we have a "opt->pattern_expression". We'd then call "match_expr_eval()", which would have died if that "opt->pattern_expression" was NULL. The "die" was added in c922b01f54c (grep: fix segfault when "git grep '('" is given, 2009-04-27), and can now be removed as it's now clearly unreachable. We still do the right thing in the case that prompted that fix: git grep '(' fatal: unmatched parenthesis Arguably neither the "--invert-grep" option added in [1] nor the earlier "--all-match" option added in [2] were intended to be used stand-alone, and another approach[3] would be to error out in those cases. But since we've been treating them as a NOOP when given without --grep for a long time let's keep doing that. We could also return in "free_pattern_expr()" if the argument is non-NULL, as an alternative fix for this segfault does [4]. That would be more elegant in making the "free_*()" function behave like "free()", but it would also remove a sanity check: The "free_pattern_expr()" function calls itself recursively, and only the top-level is allowed to be NULL, let's not conflate those two conditions. 1. 22dfa8a23de (log: teach --invert-grep option, 2015-01-12) 2. 0ab7befa31d (grep --all-match, 2006-09-27) 3. https://lore.kernel.org/git/patch-1.1-f4b90799fce-20221010T165711Z-avarab@gmail.com/ 4. http://lore.kernel.org/git/7e094882c2a71894416089f894557a9eae07e8f8.1665423686.git.me@ttaylorr.com Reported-by: orygaw <orygaw@protonmail.com> Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2022-10-11 09:48:45 +00:00
int extended = 0;
for (p = opt->pattern_list; p; p = p->next) {
switch (p->token) {
case GREP_PATTERN: /* atom */
case GREP_PATTERN_HEAD:
case GREP_PATTERN_BODY:
compile_regexp(p, opt);
break;
default:
grep.c: remove "extended" in favor of "pattern_expression", fix segfault Since 79d3696cfb4 (git-grep: boolean expression on pattern matching., 2006-06-30) the "pattern_expression" member has been used for complex queries (AND/OR...), with "pattern_list" being used for the simple OR queries. Since then we've used both "pattern_expression" and its associated boolean "extended" member to see if we have a complex expression. Since f41fb662f57 (revisions API: have release_revisions() release "grep_filter", 2022-04-13) we've had a subtle bug relating to that: If we supplied options that were only used for "complex queries", but didn't supply the query itself we'd set "opt->extended", but would have a NULL "pattern_expression". As a result these would segfault as we tried to call "free_grep_patterns()" from "release_revisions()": git -P log -1 --invert-grep git -P log -1 --all-match The root cause of this is that we were conflating the state management we needed in "compile_grep_patterns()" itself with whether or not we had an "opt->pattern_expression" later on. In this cases as we're going through "compile_grep_patterns()" we have no "opt->pattern_list" but have "opt->no_body_match" or "opt->all_match". So we'd set "opt->extended = 1", but not "return" on "opt->extended" as that's an "else if" in the same "if" statement. That behavior is intentional and required, as the common case is that we have an "opt->pattern_list" that we're about to parse into the "opt->pattern_expression". But we don't need to keep track of this "extended" flag beyond the state management in compile_grep_patterns() itself. It needs it, but once we're out of that function we can rely on "opt->pattern_expression" being non-NULL instead for using these extended patterns. As 79d3696cfb4 itself shows we've assumed that there's a one-to-one mapping between the two since the very beginning. I.e. "match_line()" would check "opt->extended" to see if it should call "match_expr()", and the first thing we do in that function is assume that we have a "opt->pattern_expression". We'd then call "match_expr_eval()", which would have died if that "opt->pattern_expression" was NULL. The "die" was added in c922b01f54c (grep: fix segfault when "git grep '('" is given, 2009-04-27), and can now be removed as it's now clearly unreachable. We still do the right thing in the case that prompted that fix: git grep '(' fatal: unmatched parenthesis Arguably neither the "--invert-grep" option added in [1] nor the earlier "--all-match" option added in [2] were intended to be used stand-alone, and another approach[3] would be to error out in those cases. But since we've been treating them as a NOOP when given without --grep for a long time let's keep doing that. We could also return in "free_pattern_expr()" if the argument is non-NULL, as an alternative fix for this segfault does [4]. That would be more elegant in making the "free_*()" function behave like "free()", but it would also remove a sanity check: The "free_pattern_expr()" function calls itself recursively, and only the top-level is allowed to be NULL, let's not conflate those two conditions. 1. 22dfa8a23de (log: teach --invert-grep option, 2015-01-12) 2. 0ab7befa31d (grep --all-match, 2006-09-27) 3. https://lore.kernel.org/git/patch-1.1-f4b90799fce-20221010T165711Z-avarab@gmail.com/ 4. http://lore.kernel.org/git/7e094882c2a71894416089f894557a9eae07e8f8.1665423686.git.me@ttaylorr.com Reported-by: orygaw <orygaw@protonmail.com> Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2022-10-11 09:48:45 +00:00
extended = 1;
break;
}
}
if (opt->all_match || opt->no_body_match || header_expr)
grep.c: remove "extended" in favor of "pattern_expression", fix segfault Since 79d3696cfb4 (git-grep: boolean expression on pattern matching., 2006-06-30) the "pattern_expression" member has been used for complex queries (AND/OR...), with "pattern_list" being used for the simple OR queries. Since then we've used both "pattern_expression" and its associated boolean "extended" member to see if we have a complex expression. Since f41fb662f57 (revisions API: have release_revisions() release "grep_filter", 2022-04-13) we've had a subtle bug relating to that: If we supplied options that were only used for "complex queries", but didn't supply the query itself we'd set "opt->extended", but would have a NULL "pattern_expression". As a result these would segfault as we tried to call "free_grep_patterns()" from "release_revisions()": git -P log -1 --invert-grep git -P log -1 --all-match The root cause of this is that we were conflating the state management we needed in "compile_grep_patterns()" itself with whether or not we had an "opt->pattern_expression" later on. In this cases as we're going through "compile_grep_patterns()" we have no "opt->pattern_list" but have "opt->no_body_match" or "opt->all_match". So we'd set "opt->extended = 1", but not "return" on "opt->extended" as that's an "else if" in the same "if" statement. That behavior is intentional and required, as the common case is that we have an "opt->pattern_list" that we're about to parse into the "opt->pattern_expression". But we don't need to keep track of this "extended" flag beyond the state management in compile_grep_patterns() itself. It needs it, but once we're out of that function we can rely on "opt->pattern_expression" being non-NULL instead for using these extended patterns. As 79d3696cfb4 itself shows we've assumed that there's a one-to-one mapping between the two since the very beginning. I.e. "match_line()" would check "opt->extended" to see if it should call "match_expr()", and the first thing we do in that function is assume that we have a "opt->pattern_expression". We'd then call "match_expr_eval()", which would have died if that "opt->pattern_expression" was NULL. The "die" was added in c922b01f54c (grep: fix segfault when "git grep '('" is given, 2009-04-27), and can now be removed as it's now clearly unreachable. We still do the right thing in the case that prompted that fix: git grep '(' fatal: unmatched parenthesis Arguably neither the "--invert-grep" option added in [1] nor the earlier "--all-match" option added in [2] were intended to be used stand-alone, and another approach[3] would be to error out in those cases. But since we've been treating them as a NOOP when given without --grep for a long time let's keep doing that. We could also return in "free_pattern_expr()" if the argument is non-NULL, as an alternative fix for this segfault does [4]. That would be more elegant in making the "free_*()" function behave like "free()", but it would also remove a sanity check: The "free_pattern_expr()" function calls itself recursively, and only the top-level is allowed to be NULL, let's not conflate those two conditions. 1. 22dfa8a23de (log: teach --invert-grep option, 2015-01-12) 2. 0ab7befa31d (grep --all-match, 2006-09-27) 3. https://lore.kernel.org/git/patch-1.1-f4b90799fce-20221010T165711Z-avarab@gmail.com/ 4. http://lore.kernel.org/git/7e094882c2a71894416089f894557a9eae07e8f8.1665423686.git.me@ttaylorr.com Reported-by: orygaw <orygaw@protonmail.com> Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2022-10-11 09:48:45 +00:00
extended = 1;
else if (!extended)
return;
p = opt->pattern_list;
if (p)
opt->pattern_expression = compile_pattern_expr(&p);
if (p)
die("incomplete pattern expression: %s", p->pattern);
if (opt->no_body_match && opt->pattern_expression)
opt->pattern_expression = grep_not_expr(opt->pattern_expression);
if (!header_expr)
return;
if (!opt->pattern_expression)
opt->pattern_expression = header_expr;
log --grep/--author: honor --all-match honored for multiple --grep patterns When we have both header expression (which has to be an OR node by construction) and a pattern expression (which could be anything), we create a new top-level OR node to bind them together, and the resulting expression structure looks like this: OR / \ / \ pattern OR / \ / \ ..... committer OR / \ author TRUE The three elements on the top-level backbone that are inspected by the "all-match" logic are "pattern", "committer" and "author". When there are more than one elements in the "pattern", the top-level node of the "pattern" part of the subtree is an OR, and that node is inspected by "all-match". The result ends up ignoring the "--all-match" given from the command line. A match on either side of the pattern is considered a match, hence: git log --grep=A --grep=B --author=C --all-match shows the same "authored by C and has either A or B" that is correct only when run without "--all-match". Fix this by turning the resulting expression around when "--all-match" is in effect, like this: OR / \ / \ / OR committer / \ author \ pattern The set of nodes on the top-level backbone in the resulting expression becomes "committer", "author", and the nodes that are on the top-level backbone of the "pattern" subexpression. This makes the "all-match" logic inspect the same nodes in "pattern" as the case without the author and/or the committer restriction, and makes the earlier "log" example to show "authored by C and has A and has B", which is what the command line expects. Signed-off-by: Junio C Hamano <gitster@pobox.com>
2012-09-13 23:26:57 +00:00
else if (opt->all_match)
opt->pattern_expression = grep_splice_or(header_expr,
opt->pattern_expression);
else
opt->pattern_expression = grep_or_expr(opt->pattern_expression,
header_expr);
opt->all_match = 1;
}
static void free_pattern_expr(struct grep_expr *x)
{
switch (x->node) {
case GREP_NODE_TRUE:
case GREP_NODE_ATOM:
break;
case GREP_NODE_NOT:
free_pattern_expr(x->u.unary);
break;
case GREP_NODE_AND:
case GREP_NODE_OR:
free_pattern_expr(x->u.binary.left);
free_pattern_expr(x->u.binary.right);
break;
}
free(x);
}
static void free_grep_pat(struct grep_pat *pattern)
{
struct grep_pat *p, *n;
for (p = pattern; p; p = n) {
n = p->next;
switch (p->token) {
case GREP_PATTERN: /* atom */
case GREP_PATTERN_HEAD:
case GREP_PATTERN_BODY:
if (p->pcre2_pattern)
grep: add support for PCRE v2 Add support for v2 of the PCRE API. This is a new major version of PCRE that came out in early 2015[1]. The regular expression syntax is the same, but while the API is similar, pretty much every function is either renamed or takes different arguments. Thus using it via entirely new functions makes sense, as opposed to trying to e.g. have one compile_pcre_pattern() that would call either PCRE v1 or v2 functions. Git can now be compiled with either USE_LIBPCRE1=YesPlease or USE_LIBPCRE2=YesPlease, with USE_LIBPCRE=YesPlease currently being a synonym for the former. Providing both is a compile-time error. With earlier patches to enable JIT for PCRE v1 the performance of the release versions of both libraries is almost exactly the same, with PCRE v2 being around 1% slower. However after I reported this to the pcre-dev mailing list[2] I got a lot of help with the API use from Zoltán Herczeg, he subsequently optimized some of the JIT functionality in v2 of the library. Running the p7820-grep-engines.sh performance test against the latest Subversion trunk of both, with both them and git compiled as -O3, and the test run against linux.git, gives the following results. Just the /perl/ tests shown: $ GIT_PERF_REPEAT_COUNT=30 GIT_PERF_LARGE_REPO=~/g/linux GIT_PERF_MAKE_COMMAND='grep -q LIBPCRE2 Makefile && make -j8 USE_LIBPCRE2=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre2/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre2/inst/lib || make -j8 USE_LIBPCRE=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre/inst/lib' ./run HEAD~5 HEAD~ HEAD p7820-grep-engines.sh [...] Test HEAD~5 HEAD~ HEAD ----------------------------------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.31(1.10+0.48) 0.21(0.35+0.56) -32.3% 0.21(0.34+0.55) -32.3% 7820.7: perl grep '^how to' 0.56(2.70+0.40) 0.24(0.64+0.52) -57.1% 0.20(0.28+0.60) -64.3% 7820.11: perl grep '[how] to' 0.56(2.66+0.38) 0.29(0.95+0.45) -48.2% 0.23(0.45+0.54) -58.9% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 1.02(5.77+0.42) 0.31(1.02+0.54) -69.6% 0.23(0.50+0.54) -77.5% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.38(1.57+0.42) 0.27(0.85+0.46) -28.9% 0.21(0.33+0.57) -44.7% See commit ("perf: add a comparison test of grep regex engines", 2017-04-19) for details on the machine the above test run was executed on. Here HEAD~2 is git with PCRE v1 without JIT, HEAD~ is PCRE v1 with JIT, and HEAD is PCRE v2 (also with JIT). See previous commits of mine mentioning p7820-grep-engines.sh for more details on the test setup. For ease of readability, a different run just of HEAD~ (PCRE v1 with JIT v.s. PCRE v2), again with just the /perl/ tests shown: [...] Test HEAD~ HEAD ---------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.21(0.42+0.52) 0.21(0.31+0.58) +0.0% 7820.7: perl grep '^how to' 0.25(0.65+0.50) 0.20(0.31+0.57) -20.0% 7820.11: perl grep '[how] to' 0.30(0.90+0.50) 0.23(0.46+0.53) -23.3% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 0.30(1.19+0.38) 0.23(0.51+0.51) -23.3% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.27(0.84+0.48) 0.21(0.34+0.57) -22.2% I.e. the two are either neck-to-neck, but PCRE v2 usually pulls ahead, when it does it's around 20% faster. A brief note on thread safety: As noted in pcre2api(3) & pcre2jit(3) the compiled pattern can be shared between threads, but not some of the JIT context, however the grep threading support does all pattern & JIT compilation in separate threads, so this code doesn't need to concern itself with thread safety. See commit 63e7e9d8b6 ("git-grep: Learn PCRE", 2011-05-09) for the initial addition of PCRE v1. This change follows some of the same patterns it did (and which were discussed on list at the time), e.g. mocking up types with typedef instead of ifdef-ing them out when USE_LIBPCRE2 isn't defined. This adds some trivial memory use to the program, but makes the code look nicer. 1. https://lists.exim.org/lurker/message/20150105.162835.0666407a.en.html 2. https://lists.exim.org/lurker/thread/20170419.172322.833ee099.en.html Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-01 18:20:56 +00:00
free_pcre2_pattern(p);
else
regfree(&p->regexp);
free(p->pattern);
break;
default:
break;
}
free(p);
}
}
void free_grep_patterns(struct grep_opt *opt)
{
free_grep_pat(opt->pattern_list);
free_grep_pat(opt->header_list);
if (opt->pattern_expression)
free_pattern_expr(opt->pattern_expression);
}
static const char *end_of_line(const char *cp, unsigned long *left)
{
unsigned long l = *left;
while (l && *cp != '\n') {
l--;
cp++;
}
*left = l;
return cp;
}
static int word_char(char ch)
{
return isalnum(ch) || ch == '_';
}
static void output_color(struct grep_opt *opt, const void *data, size_t size,
const char *color)
{
color: delay auto-color decision until point of use When we read a color value either from a config file or from the command line, we use git_config_colorbool to convert it from the tristate always/never/auto into a single yes/no boolean value. This has some timing implications with respect to starting a pager. If we start (or decide not to start) the pager before checking the colorbool, everything is fine. Either isatty(1) will give us the right information, or we will properly check for pager_in_use(). However, if we decide to start a pager after we have checked the colorbool, things are not so simple. If stdout is a tty, then we will have already decided to use color. However, the user may also have configured color.pager not to use color with the pager. In this case, we need to actually turn off color. Unfortunately, the pager code has no idea which color variables were turned on (and there are many of them throughout the code, and they may even have been manipulated after the colorbool selection by something like "--color" on the command line). This bug can be seen any time a pager is started after config and command line options are checked. This has affected "git diff" since 89d07f7 (diff: don't run pager if user asked for a diff style exit code, 2007-08-12). It has also affect the log family since 1fda91b (Fix 'git log' early pager startup error case, 2010-08-24). This patch splits the notion of parsing a colorbool and actually checking the configuration. The "use_color" variables now have an additional possible value, GIT_COLOR_AUTO. Users of the variable should use the new "want_color()" wrapper, which will lazily determine and cache the auto-color decision. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2011-08-18 05:04:23 +00:00
if (want_color(opt->color) && color && color[0]) {
opt->output(opt, color, strlen(color));
opt->output(opt, data, size);
opt->output(opt, GIT_COLOR_RESET, strlen(GIT_COLOR_RESET));
} else
opt->output(opt, data, size);
}
static void output_sep(struct grep_opt *opt, char sign)
{
if (opt->null_following_name)
opt->output(opt, "\0", 1);
else
output_color(opt, &sign, 1, opt->colors[GREP_COLOR_SEP]);
}
static void show_name(struct grep_opt *opt, const char *name)
{
output_color(opt, name, strlen(name), opt->colors[GREP_COLOR_FILENAME]);
opt->output(opt, opt->null_following_name ? "\0" : "\n", 1);
}
static int patmatch(struct grep_pat *p,
const char *line, const char *eol,
regmatch_t *match, int eflags)
{
int hit;
if (p->pcre2_pattern)
grep: add support for PCRE v2 Add support for v2 of the PCRE API. This is a new major version of PCRE that came out in early 2015[1]. The regular expression syntax is the same, but while the API is similar, pretty much every function is either renamed or takes different arguments. Thus using it via entirely new functions makes sense, as opposed to trying to e.g. have one compile_pcre_pattern() that would call either PCRE v1 or v2 functions. Git can now be compiled with either USE_LIBPCRE1=YesPlease or USE_LIBPCRE2=YesPlease, with USE_LIBPCRE=YesPlease currently being a synonym for the former. Providing both is a compile-time error. With earlier patches to enable JIT for PCRE v1 the performance of the release versions of both libraries is almost exactly the same, with PCRE v2 being around 1% slower. However after I reported this to the pcre-dev mailing list[2] I got a lot of help with the API use from Zoltán Herczeg, he subsequently optimized some of the JIT functionality in v2 of the library. Running the p7820-grep-engines.sh performance test against the latest Subversion trunk of both, with both them and git compiled as -O3, and the test run against linux.git, gives the following results. Just the /perl/ tests shown: $ GIT_PERF_REPEAT_COUNT=30 GIT_PERF_LARGE_REPO=~/g/linux GIT_PERF_MAKE_COMMAND='grep -q LIBPCRE2 Makefile && make -j8 USE_LIBPCRE2=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre2/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre2/inst/lib || make -j8 USE_LIBPCRE=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre/inst/lib' ./run HEAD~5 HEAD~ HEAD p7820-grep-engines.sh [...] Test HEAD~5 HEAD~ HEAD ----------------------------------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.31(1.10+0.48) 0.21(0.35+0.56) -32.3% 0.21(0.34+0.55) -32.3% 7820.7: perl grep '^how to' 0.56(2.70+0.40) 0.24(0.64+0.52) -57.1% 0.20(0.28+0.60) -64.3% 7820.11: perl grep '[how] to' 0.56(2.66+0.38) 0.29(0.95+0.45) -48.2% 0.23(0.45+0.54) -58.9% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 1.02(5.77+0.42) 0.31(1.02+0.54) -69.6% 0.23(0.50+0.54) -77.5% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.38(1.57+0.42) 0.27(0.85+0.46) -28.9% 0.21(0.33+0.57) -44.7% See commit ("perf: add a comparison test of grep regex engines", 2017-04-19) for details on the machine the above test run was executed on. Here HEAD~2 is git with PCRE v1 without JIT, HEAD~ is PCRE v1 with JIT, and HEAD is PCRE v2 (also with JIT). See previous commits of mine mentioning p7820-grep-engines.sh for more details on the test setup. For ease of readability, a different run just of HEAD~ (PCRE v1 with JIT v.s. PCRE v2), again with just the /perl/ tests shown: [...] Test HEAD~ HEAD ---------------------------------------------------------------------------------------- 7820.3: perl grep 'how.to' 0.21(0.42+0.52) 0.21(0.31+0.58) +0.0% 7820.7: perl grep '^how to' 0.25(0.65+0.50) 0.20(0.31+0.57) -20.0% 7820.11: perl grep '[how] to' 0.30(0.90+0.50) 0.23(0.46+0.53) -23.3% 7820.15: perl grep '(e.t[^ ]*|v.ry) rare' 0.30(1.19+0.38) 0.23(0.51+0.51) -23.3% 7820.19: perl grep 'm(ú|u)lt.b(æ|y)te' 0.27(0.84+0.48) 0.21(0.34+0.57) -22.2% I.e. the two are either neck-to-neck, but PCRE v2 usually pulls ahead, when it does it's around 20% faster. A brief note on thread safety: As noted in pcre2api(3) & pcre2jit(3) the compiled pattern can be shared between threads, but not some of the JIT context, however the grep threading support does all pattern & JIT compilation in separate threads, so this code doesn't need to concern itself with thread safety. See commit 63e7e9d8b6 ("git-grep: Learn PCRE", 2011-05-09) for the initial addition of PCRE v1. This change follows some of the same patterns it did (and which were discussed on list at the time), e.g. mocking up types with typedef instead of ifdef-ing them out when USE_LIBPCRE2 isn't defined. This adds some trivial memory use to the program, but makes the code look nicer. 1. https://lists.exim.org/lurker/message/20150105.162835.0666407a.en.html 2. https://lists.exim.org/lurker/thread/20170419.172322.833ee099.en.html Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-06-01 18:20:56 +00:00
hit = !pcre2match(p, line, eol, match, eflags);
else
hit = !regexec_buf(&p->regexp, line, eol - line, 1, match,
eflags);
return hit;
}
static void strip_timestamp(const char *bol, const char **eol_p)
log --author/--committer: really match only with name part When we tried to find commits done by AUTHOR, the first implementation tried to pattern match a line with "^author .*AUTHOR", which later was enhanced to strip leading caret and look for "^author AUTHOR" when the search pattern was anchored at the left end (i.e. --author="^AUTHOR"). This had a few problems: * When looking for fixed strings (e.g. "git log -F --author=x --grep=y"), the regexp internally used "^author .*x" would never match anything; * To match at the end (e.g. "git log --author='google.com>$'"), the generated regexp has to also match the trailing timestamp part the commit header lines have. Also, in order to determine if the '$' at the end means "match at the end of the line" or just a literal dollar sign (probably backslash-quoted), we would need to parse the regexp ourselves. An earlier alternative tried to make sure that a line matches "^author " (to limit by field name) and the user supplied pattern at the same time. While it solved the -F problem by introducing a special override for matching the "^author ", it did not solve the trailing timestamp nor tail match problem. It also would have matched every commit if --author=author was asked for, not because the author's email part had this string, but because every commit header line that talks about the author begins with that field name, regardleses of who wrote it. Instead of piling more hacks on top of hacks, this rethinks the grep machinery that is used to look for strings in the commit header, and makes sure that (1) field name matches literally at the beginning of the line, followed by a SP, and (2) the user supplied pattern is matched against the remainder of the line, excluding the trailing timestamp data. Signed-off-by: Junio C Hamano <gitster@pobox.com>
2008-09-05 05:15:02 +00:00
{
const char *eol = *eol_p;
log --author/--committer: really match only with name part When we tried to find commits done by AUTHOR, the first implementation tried to pattern match a line with "^author .*AUTHOR", which later was enhanced to strip leading caret and look for "^author AUTHOR" when the search pattern was anchored at the left end (i.e. --author="^AUTHOR"). This had a few problems: * When looking for fixed strings (e.g. "git log -F --author=x --grep=y"), the regexp internally used "^author .*x" would never match anything; * To match at the end (e.g. "git log --author='google.com>$'"), the generated regexp has to also match the trailing timestamp part the commit header lines have. Also, in order to determine if the '$' at the end means "match at the end of the line" or just a literal dollar sign (probably backslash-quoted), we would need to parse the regexp ourselves. An earlier alternative tried to make sure that a line matches "^author " (to limit by field name) and the user supplied pattern at the same time. While it solved the -F problem by introducing a special override for matching the "^author ", it did not solve the trailing timestamp nor tail match problem. It also would have matched every commit if --author=author was asked for, not because the author's email part had this string, but because every commit header line that talks about the author begins with that field name, regardleses of who wrote it. Instead of piling more hacks on top of hacks, this rethinks the grep machinery that is used to look for strings in the commit header, and makes sure that (1) field name matches literally at the beginning of the line, followed by a SP, and (2) the user supplied pattern is matched against the remainder of the line, excluding the trailing timestamp data. Signed-off-by: Junio C Hamano <gitster@pobox.com>
2008-09-05 05:15:02 +00:00
while (bol < --eol) {
if (*eol != '>')
continue;
*eol_p = ++eol;
break;
log --author/--committer: really match only with name part When we tried to find commits done by AUTHOR, the first implementation tried to pattern match a line with "^author .*AUTHOR", which later was enhanced to strip leading caret and look for "^author AUTHOR" when the search pattern was anchored at the left end (i.e. --author="^AUTHOR"). This had a few problems: * When looking for fixed strings (e.g. "git log -F --author=x --grep=y"), the regexp internally used "^author .*x" would never match anything; * To match at the end (e.g. "git log --author='google.com>$'"), the generated regexp has to also match the trailing timestamp part the commit header lines have. Also, in order to determine if the '$' at the end means "match at the end of the line" or just a literal dollar sign (probably backslash-quoted), we would need to parse the regexp ourselves. An earlier alternative tried to make sure that a line matches "^author " (to limit by field name) and the user supplied pattern at the same time. While it solved the -F problem by introducing a special override for matching the "^author ", it did not solve the trailing timestamp nor tail match problem. It also would have matched every commit if --author=author was asked for, not because the author's email part had this string, but because every commit header line that talks about the author begins with that field name, regardleses of who wrote it. Instead of piling more hacks on top of hacks, this rethinks the grep machinery that is used to look for strings in the commit header, and makes sure that (1) field name matches literally at the beginning of the line, followed by a SP, and (2) the user supplied pattern is matched against the remainder of the line, excluding the trailing timestamp data. Signed-off-by: Junio C Hamano <gitster@pobox.com>
2008-09-05 05:15:02 +00:00
}
}
static struct {
const char *field;
size_t len;
} header_field[] = {
{ "author ", 7 },
{ "committer ", 10 },
{ "reflog ", 7 },
log --author/--committer: really match only with name part When we tried to find commits done by AUTHOR, the first implementation tried to pattern match a line with "^author .*AUTHOR", which later was enhanced to strip leading caret and look for "^author AUTHOR" when the search pattern was anchored at the left end (i.e. --author="^AUTHOR"). This had a few problems: * When looking for fixed strings (e.g. "git log -F --author=x --grep=y"), the regexp internally used "^author .*x" would never match anything; * To match at the end (e.g. "git log --author='google.com>$'"), the generated regexp has to also match the trailing timestamp part the commit header lines have. Also, in order to determine if the '$' at the end means "match at the end of the line" or just a literal dollar sign (probably backslash-quoted), we would need to parse the regexp ourselves. An earlier alternative tried to make sure that a line matches "^author " (to limit by field name) and the user supplied pattern at the same time. While it solved the -F problem by introducing a special override for matching the "^author ", it did not solve the trailing timestamp nor tail match problem. It also would have matched every commit if --author=author was asked for, not because the author's email part had this string, but because every commit header line that talks about the author begins with that field name, regardleses of who wrote it. Instead of piling more hacks on top of hacks, this rethinks the grep machinery that is used to look for strings in the commit header, and makes sure that (1) field name matches literally at the beginning of the line, followed by a SP, and (2) the user supplied pattern is matched against the remainder of the line, excluding the trailing timestamp data. Signed-off-by: Junio C Hamano <gitster@pobox.com>
2008-09-05 05:15:02 +00:00
};
static int headerless_match_one_pattern(struct grep_pat *p,
const char *bol, const char *eol,
enum grep_context ctx,
regmatch_t *pmatch, int eflags)
{
int hit = 0;
const char *start = bol;
if ((p->token != GREP_PATTERN) &&
((p->token == GREP_PATTERN_HEAD) != (ctx == GREP_CONTEXT_HEAD)))
return 0;
again:
hit = patmatch(p, bol, eol, pmatch, eflags);
if (hit && p->word_regexp) {
if ((pmatch[0].rm_so < 0) ||
(eol - bol) < pmatch[0].rm_so ||
(pmatch[0].rm_eo < 0) ||
(eol - bol) < pmatch[0].rm_eo)
die("regexp returned nonsense");
/* Match beginning must be either beginning of the
* line, or at word boundary (i.e. the last char must
* not be a word char). Similarly, match end must be
* either end of the line, or at word boundary
* (i.e. the next char must not be a word char).
*/
if ( ((pmatch[0].rm_so == 0) ||
!word_char(bol[pmatch[0].rm_so-1])) &&
((pmatch[0].rm_eo == (eol-bol)) ||
!word_char(bol[pmatch[0].rm_eo])) )
;
else
hit = 0;
/* Words consist of at least one character. */
if (pmatch->rm_so == pmatch->rm_eo)
hit = 0;
if (!hit && pmatch[0].rm_so + bol + 1 < eol) {
/* There could be more than one match on the
* line, and the first match might not be
* strict word match. But later ones could be!
* Forward to the next possible start, i.e. the
* next position following a non-word char.
*/
bol = pmatch[0].rm_so + bol + 1;
while (word_char(bol[-1]) && bol < eol)
bol++;
eflags |= REG_NOTBOL;
if (bol < eol)
goto again;
}
}
if (hit) {
pmatch[0].rm_so += bol - start;
pmatch[0].rm_eo += bol - start;
}
return hit;
}
static int match_one_pattern(struct grep_pat *p,
const char *bol, const char *eol,
enum grep_context ctx, regmatch_t *pmatch,
int eflags)
{
const char *field;
size_t len;
if (p->token == GREP_PATTERN_HEAD) {
assert(p->field < ARRAY_SIZE(header_field));
field = header_field[p->field].field;
len = header_field[p->field].len;
if (strncmp(bol, field, len))
return 0;
bol += len;
switch (p->field) {
case GREP_HEADER_AUTHOR:
case GREP_HEADER_COMMITTER:
strip_timestamp(bol, &eol);
break;
default:
break;
}
}
return headerless_match_one_pattern(p, bol, eol, ctx, pmatch, eflags);
}
static int match_expr_eval(struct grep_opt *opt, struct grep_expr *x,
const char *bol, const char *eol,
enum grep_context ctx, ssize_t *col,
grep.c: expose {,inverted} match column in match_line() When calling match_line(), callers presently cannot determine the relative offset of the match because match_line() discards the 'regmatch_t' that contains this information. Instead, teach match_line() to take in two 'ssize_t's. Fill the first with the offset of the match produced by the given expression. If extended, fill the later with the offset of the match produced as if --invert were given. For instance, matching "--not -e x" on this line produces a columnar offset of 0, (i.e., the whole line does not contain an x), but "--invert --not -e -x" will fill the later ssize_t of the column containing an "x", because this expression is semantically equivalent to "-e x". To determine the column for the inverted and non-inverted case, do the following: - If matching an atom, the non-inverted column is as given from match_one_pattern(), and the inverted column is unset. - If matching a --not, the inverted column and non-inverted column swap. - If matching an --and, or --or, the non-inverted column is the minimum of the two children. Presently, the existing short-circuiting logic for AND and OR applies as before. This will change in the following commit when we add options to configure the --column flag. Taken together, this and the forthcoming change will always yield the earlier column on a given line. This patch will become useful when we later pick between the two new results in order to display the column number of the first match on a line with --column. Co-authored-by: Jeff King <peff@peff.net> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-06-22 15:49:35 +00:00
ssize_t *icol, int collect_hits)
{
int h = 0;
switch (x->node) {
case GREP_NODE_TRUE:
h = 1;
break;
case GREP_NODE_ATOM:
grep.c: expose {,inverted} match column in match_line() When calling match_line(), callers presently cannot determine the relative offset of the match because match_line() discards the 'regmatch_t' that contains this information. Instead, teach match_line() to take in two 'ssize_t's. Fill the first with the offset of the match produced by the given expression. If extended, fill the later with the offset of the match produced as if --invert were given. For instance, matching "--not -e x" on this line produces a columnar offset of 0, (i.e., the whole line does not contain an x), but "--invert --not -e -x" will fill the later ssize_t of the column containing an "x", because this expression is semantically equivalent to "-e x". To determine the column for the inverted and non-inverted case, do the following: - If matching an atom, the non-inverted column is as given from match_one_pattern(), and the inverted column is unset. - If matching a --not, the inverted column and non-inverted column swap. - If matching an --and, or --or, the non-inverted column is the minimum of the two children. Presently, the existing short-circuiting logic for AND and OR applies as before. This will change in the following commit when we add options to configure the --column flag. Taken together, this and the forthcoming change will always yield the earlier column on a given line. This patch will become useful when we later pick between the two new results in order to display the column number of the first match on a line with --column. Co-authored-by: Jeff King <peff@peff.net> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-06-22 15:49:35 +00:00
{
regmatch_t tmp;
h = match_one_pattern(x->u.atom, bol, eol, ctx,
&tmp, 0);
if (h && (*col < 0 || tmp.rm_so < *col))
*col = tmp.rm_so;
}
if (x->u.atom->token == GREP_PATTERN_BODY)
opt->body_hit |= h;
break;
case GREP_NODE_NOT:
grep.c: expose {,inverted} match column in match_line() When calling match_line(), callers presently cannot determine the relative offset of the match because match_line() discards the 'regmatch_t' that contains this information. Instead, teach match_line() to take in two 'ssize_t's. Fill the first with the offset of the match produced by the given expression. If extended, fill the later with the offset of the match produced as if --invert were given. For instance, matching "--not -e x" on this line produces a columnar offset of 0, (i.e., the whole line does not contain an x), but "--invert --not -e -x" will fill the later ssize_t of the column containing an "x", because this expression is semantically equivalent to "-e x". To determine the column for the inverted and non-inverted case, do the following: - If matching an atom, the non-inverted column is as given from match_one_pattern(), and the inverted column is unset. - If matching a --not, the inverted column and non-inverted column swap. - If matching an --and, or --or, the non-inverted column is the minimum of the two children. Presently, the existing short-circuiting logic for AND and OR applies as before. This will change in the following commit when we add options to configure the --column flag. Taken together, this and the forthcoming change will always yield the earlier column on a given line. This patch will become useful when we later pick between the two new results in order to display the column number of the first match on a line with --column. Co-authored-by: Jeff King <peff@peff.net> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-06-22 15:49:35 +00:00
/*
* Upon visiting a GREP_NODE_NOT, col and icol become swapped.
*/
h = !match_expr_eval(opt, x->u.unary, bol, eol, ctx, icol, col,
0);
break;
case GREP_NODE_AND:
h = match_expr_eval(opt, x->u.binary.left, bol, eol, ctx, col,
grep.c: expose {,inverted} match column in match_line() When calling match_line(), callers presently cannot determine the relative offset of the match because match_line() discards the 'regmatch_t' that contains this information. Instead, teach match_line() to take in two 'ssize_t's. Fill the first with the offset of the match produced by the given expression. If extended, fill the later with the offset of the match produced as if --invert were given. For instance, matching "--not -e x" on this line produces a columnar offset of 0, (i.e., the whole line does not contain an x), but "--invert --not -e -x" will fill the later ssize_t of the column containing an "x", because this expression is semantically equivalent to "-e x". To determine the column for the inverted and non-inverted case, do the following: - If matching an atom, the non-inverted column is as given from match_one_pattern(), and the inverted column is unset. - If matching a --not, the inverted column and non-inverted column swap. - If matching an --and, or --or, the non-inverted column is the minimum of the two children. Presently, the existing short-circuiting logic for AND and OR applies as before. This will change in the following commit when we add options to configure the --column flag. Taken together, this and the forthcoming change will always yield the earlier column on a given line. This patch will become useful when we later pick between the two new results in order to display the column number of the first match on a line with --column. Co-authored-by: Jeff King <peff@peff.net> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-06-22 15:49:35 +00:00
icol, 0);
if (h || opt->columnnum) {
/*
* Don't short-circuit AND when given --column, since a
* NOT earlier in the tree may turn this into an OR. In
* this case, see the below comment.
*/
h &= match_expr_eval(opt, x->u.binary.right, bol, eol,
ctx, col, icol, 0);
}
break;
case GREP_NODE_OR:
if (!(collect_hits || opt->columnnum)) {
/*
* Don't short-circuit OR when given --column (or
* collecting hits) to ensure we don't skip a later
* child that would produce an earlier match.
*/
grep.c: expose {,inverted} match column in match_line() When calling match_line(), callers presently cannot determine the relative offset of the match because match_line() discards the 'regmatch_t' that contains this information. Instead, teach match_line() to take in two 'ssize_t's. Fill the first with the offset of the match produced by the given expression. If extended, fill the later with the offset of the match produced as if --invert were given. For instance, matching "--not -e x" on this line produces a columnar offset of 0, (i.e., the whole line does not contain an x), but "--invert --not -e -x" will fill the later ssize_t of the column containing an "x", because this expression is semantically equivalent to "-e x". To determine the column for the inverted and non-inverted case, do the following: - If matching an atom, the non-inverted column is as given from match_one_pattern(), and the inverted column is unset. - If matching a --not, the inverted column and non-inverted column swap. - If matching an --and, or --or, the non-inverted column is the minimum of the two children. Presently, the existing short-circuiting logic for AND and OR applies as before. This will change in the following commit when we add options to configure the --column flag. Taken together, this and the forthcoming change will always yield the earlier column on a given line. This patch will become useful when we later pick between the two new results in order to display the column number of the first match on a line with --column. Co-authored-by: Jeff King <peff@peff.net> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-06-22 15:49:35 +00:00
return (match_expr_eval(opt, x->u.binary.left, bol, eol,
ctx, col, icol, 0) ||
match_expr_eval(opt, x->u.binary.right, bol,
eol, ctx, col, icol, 0));
}
grep.c: expose {,inverted} match column in match_line() When calling match_line(), callers presently cannot determine the relative offset of the match because match_line() discards the 'regmatch_t' that contains this information. Instead, teach match_line() to take in two 'ssize_t's. Fill the first with the offset of the match produced by the given expression. If extended, fill the later with the offset of the match produced as if --invert were given. For instance, matching "--not -e x" on this line produces a columnar offset of 0, (i.e., the whole line does not contain an x), but "--invert --not -e -x" will fill the later ssize_t of the column containing an "x", because this expression is semantically equivalent to "-e x". To determine the column for the inverted and non-inverted case, do the following: - If matching an atom, the non-inverted column is as given from match_one_pattern(), and the inverted column is unset. - If matching a --not, the inverted column and non-inverted column swap. - If matching an --and, or --or, the non-inverted column is the minimum of the two children. Presently, the existing short-circuiting logic for AND and OR applies as before. This will change in the following commit when we add options to configure the --column flag. Taken together, this and the forthcoming change will always yield the earlier column on a given line. This patch will become useful when we later pick between the two new results in order to display the column number of the first match on a line with --column. Co-authored-by: Jeff King <peff@peff.net> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-06-22 15:49:35 +00:00
h = match_expr_eval(opt, x->u.binary.left, bol, eol, ctx, col,
icol, 0);
if (collect_hits)
x->u.binary.left->hit |= h;
grep.c: expose {,inverted} match column in match_line() When calling match_line(), callers presently cannot determine the relative offset of the match because match_line() discards the 'regmatch_t' that contains this information. Instead, teach match_line() to take in two 'ssize_t's. Fill the first with the offset of the match produced by the given expression. If extended, fill the later with the offset of the match produced as if --invert were given. For instance, matching "--not -e x" on this line produces a columnar offset of 0, (i.e., the whole line does not contain an x), but "--invert --not -e -x" will fill the later ssize_t of the column containing an "x", because this expression is semantically equivalent to "-e x". To determine the column for the inverted and non-inverted case, do the following: - If matching an atom, the non-inverted column is as given from match_one_pattern(), and the inverted column is unset. - If matching a --not, the inverted column and non-inverted column swap. - If matching an --and, or --or, the non-inverted column is the minimum of the two children. Presently, the existing short-circuiting logic for AND and OR applies as before. This will change in the following commit when we add options to configure the --column flag. Taken together, this and the forthcoming change will always yield the earlier column on a given line. This patch will become useful when we later pick between the two new results in order to display the column number of the first match on a line with --column. Co-authored-by: Jeff King <peff@peff.net> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-06-22 15:49:35 +00:00
h |= match_expr_eval(opt, x->u.binary.right, bol, eol, ctx, col,
icol, collect_hits);
break;
default:
die("Unexpected node type (internal error) %d", x->node);
}
if (collect_hits)
x->hit |= h;
return h;
}
static int match_expr(struct grep_opt *opt,
const char *bol, const char *eol,
grep.c: expose {,inverted} match column in match_line() When calling match_line(), callers presently cannot determine the relative offset of the match because match_line() discards the 'regmatch_t' that contains this information. Instead, teach match_line() to take in two 'ssize_t's. Fill the first with the offset of the match produced by the given expression. If extended, fill the later with the offset of the match produced as if --invert were given. For instance, matching "--not -e x" on this line produces a columnar offset of 0, (i.e., the whole line does not contain an x), but "--invert --not -e -x" will fill the later ssize_t of the column containing an "x", because this expression is semantically equivalent to "-e x". To determine the column for the inverted and non-inverted case, do the following: - If matching an atom, the non-inverted column is as given from match_one_pattern(), and the inverted column is unset. - If matching a --not, the inverted column and non-inverted column swap. - If matching an --and, or --or, the non-inverted column is the minimum of the two children. Presently, the existing short-circuiting logic for AND and OR applies as before. This will change in the following commit when we add options to configure the --column flag. Taken together, this and the forthcoming change will always yield the earlier column on a given line. This patch will become useful when we later pick between the two new results in order to display the column number of the first match on a line with --column. Co-authored-by: Jeff King <peff@peff.net> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-06-22 15:49:35 +00:00
enum grep_context ctx, ssize_t *col,
ssize_t *icol, int collect_hits)
{
struct grep_expr *x = opt->pattern_expression;
grep.c: expose {,inverted} match column in match_line() When calling match_line(), callers presently cannot determine the relative offset of the match because match_line() discards the 'regmatch_t' that contains this information. Instead, teach match_line() to take in two 'ssize_t's. Fill the first with the offset of the match produced by the given expression. If extended, fill the later with the offset of the match produced as if --invert were given. For instance, matching "--not -e x" on this line produces a columnar offset of 0, (i.e., the whole line does not contain an x), but "--invert --not -e -x" will fill the later ssize_t of the column containing an "x", because this expression is semantically equivalent to "-e x". To determine the column for the inverted and non-inverted case, do the following: - If matching an atom, the non-inverted column is as given from match_one_pattern(), and the inverted column is unset. - If matching a --not, the inverted column and non-inverted column swap. - If matching an --and, or --or, the non-inverted column is the minimum of the two children. Presently, the existing short-circuiting logic for AND and OR applies as before. This will change in the following commit when we add options to configure the --column flag. Taken together, this and the forthcoming change will always yield the earlier column on a given line. This patch will become useful when we later pick between the two new results in order to display the column number of the first match on a line with --column. Co-authored-by: Jeff King <peff@peff.net> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-06-22 15:49:35 +00:00
return match_expr_eval(opt, x, bol, eol, ctx, col, icol, collect_hits);
}
static int match_line(struct grep_opt *opt,
const char *bol, const char *eol,
grep.c: expose {,inverted} match column in match_line() When calling match_line(), callers presently cannot determine the relative offset of the match because match_line() discards the 'regmatch_t' that contains this information. Instead, teach match_line() to take in two 'ssize_t's. Fill the first with the offset of the match produced by the given expression. If extended, fill the later with the offset of the match produced as if --invert were given. For instance, matching "--not -e x" on this line produces a columnar offset of 0, (i.e., the whole line does not contain an x), but "--invert --not -e -x" will fill the later ssize_t of the column containing an "x", because this expression is semantically equivalent to "-e x". To determine the column for the inverted and non-inverted case, do the following: - If matching an atom, the non-inverted column is as given from match_one_pattern(), and the inverted column is unset. - If matching a --not, the inverted column and non-inverted column swap. - If matching an --and, or --or, the non-inverted column is the minimum of the two children. Presently, the existing short-circuiting logic for AND and OR applies as before. This will change in the following commit when we add options to configure the --column flag. Taken together, this and the forthcoming change will always yield the earlier column on a given line. This patch will become useful when we later pick between the two new results in order to display the column number of the first match on a line with --column. Co-authored-by: Jeff King <peff@peff.net> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-06-22 15:49:35 +00:00
ssize_t *col, ssize_t *icol,
enum grep_context ctx, int collect_hits)
{
struct grep_pat *p;
int hit = 0;
grep.c: remove "extended" in favor of "pattern_expression", fix segfault Since 79d3696cfb4 (git-grep: boolean expression on pattern matching., 2006-06-30) the "pattern_expression" member has been used for complex queries (AND/OR...), with "pattern_list" being used for the simple OR queries. Since then we've used both "pattern_expression" and its associated boolean "extended" member to see if we have a complex expression. Since f41fb662f57 (revisions API: have release_revisions() release "grep_filter", 2022-04-13) we've had a subtle bug relating to that: If we supplied options that were only used for "complex queries", but didn't supply the query itself we'd set "opt->extended", but would have a NULL "pattern_expression". As a result these would segfault as we tried to call "free_grep_patterns()" from "release_revisions()": git -P log -1 --invert-grep git -P log -1 --all-match The root cause of this is that we were conflating the state management we needed in "compile_grep_patterns()" itself with whether or not we had an "opt->pattern_expression" later on. In this cases as we're going through "compile_grep_patterns()" we have no "opt->pattern_list" but have "opt->no_body_match" or "opt->all_match". So we'd set "opt->extended = 1", but not "return" on "opt->extended" as that's an "else if" in the same "if" statement. That behavior is intentional and required, as the common case is that we have an "opt->pattern_list" that we're about to parse into the "opt->pattern_expression". But we don't need to keep track of this "extended" flag beyond the state management in compile_grep_patterns() itself. It needs it, but once we're out of that function we can rely on "opt->pattern_expression" being non-NULL instead for using these extended patterns. As 79d3696cfb4 itself shows we've assumed that there's a one-to-one mapping between the two since the very beginning. I.e. "match_line()" would check "opt->extended" to see if it should call "match_expr()", and the first thing we do in that function is assume that we have a "opt->pattern_expression". We'd then call "match_expr_eval()", which would have died if that "opt->pattern_expression" was NULL. The "die" was added in c922b01f54c (grep: fix segfault when "git grep '('" is given, 2009-04-27), and can now be removed as it's now clearly unreachable. We still do the right thing in the case that prompted that fix: git grep '(' fatal: unmatched parenthesis Arguably neither the "--invert-grep" option added in [1] nor the earlier "--all-match" option added in [2] were intended to be used stand-alone, and another approach[3] would be to error out in those cases. But since we've been treating them as a NOOP when given without --grep for a long time let's keep doing that. We could also return in "free_pattern_expr()" if the argument is non-NULL, as an alternative fix for this segfault does [4]. That would be more elegant in making the "free_*()" function behave like "free()", but it would also remove a sanity check: The "free_pattern_expr()" function calls itself recursively, and only the top-level is allowed to be NULL, let's not conflate those two conditions. 1. 22dfa8a23de (log: teach --invert-grep option, 2015-01-12) 2. 0ab7befa31d (grep --all-match, 2006-09-27) 3. https://lore.kernel.org/git/patch-1.1-f4b90799fce-20221010T165711Z-avarab@gmail.com/ 4. http://lore.kernel.org/git/7e094882c2a71894416089f894557a9eae07e8f8.1665423686.git.me@ttaylorr.com Reported-by: orygaw <orygaw@protonmail.com> Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2022-10-11 09:48:45 +00:00
if (opt->pattern_expression)
grep.c: expose {,inverted} match column in match_line() When calling match_line(), callers presently cannot determine the relative offset of the match because match_line() discards the 'regmatch_t' that contains this information. Instead, teach match_line() to take in two 'ssize_t's. Fill the first with the offset of the match produced by the given expression. If extended, fill the later with the offset of the match produced as if --invert were given. For instance, matching "--not -e x" on this line produces a columnar offset of 0, (i.e., the whole line does not contain an x), but "--invert --not -e -x" will fill the later ssize_t of the column containing an "x", because this expression is semantically equivalent to "-e x". To determine the column for the inverted and non-inverted case, do the following: - If matching an atom, the non-inverted column is as given from match_one_pattern(), and the inverted column is unset. - If matching a --not, the inverted column and non-inverted column swap. - If matching an --and, or --or, the non-inverted column is the minimum of the two children. Presently, the existing short-circuiting logic for AND and OR applies as before. This will change in the following commit when we add options to configure the --column flag. Taken together, this and the forthcoming change will always yield the earlier column on a given line. This patch will become useful when we later pick between the two new results in order to display the column number of the first match on a line with --column. Co-authored-by: Jeff King <peff@peff.net> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-06-22 15:49:35 +00:00
return match_expr(opt, bol, eol, ctx, col, icol,
collect_hits);
/* we do not call with collect_hits without being extended */
for (p = opt->pattern_list; p; p = p->next) {
grep.c: expose {,inverted} match column in match_line() When calling match_line(), callers presently cannot determine the relative offset of the match because match_line() discards the 'regmatch_t' that contains this information. Instead, teach match_line() to take in two 'ssize_t's. Fill the first with the offset of the match produced by the given expression. If extended, fill the later with the offset of the match produced as if --invert were given. For instance, matching "--not -e x" on this line produces a columnar offset of 0, (i.e., the whole line does not contain an x), but "--invert --not -e -x" will fill the later ssize_t of the column containing an "x", because this expression is semantically equivalent to "-e x". To determine the column for the inverted and non-inverted case, do the following: - If matching an atom, the non-inverted column is as given from match_one_pattern(), and the inverted column is unset. - If matching a --not, the inverted column and non-inverted column swap. - If matching an --and, or --or, the non-inverted column is the minimum of the two children. Presently, the existing short-circuiting logic for AND and OR applies as before. This will change in the following commit when we add options to configure the --column flag. Taken together, this and the forthcoming change will always yield the earlier column on a given line. This patch will become useful when we later pick between the two new results in order to display the column number of the first match on a line with --column. Co-authored-by: Jeff King <peff@peff.net> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-06-22 15:49:35 +00:00
regmatch_t tmp;
if (match_one_pattern(p, bol, eol, ctx, &tmp, 0)) {
hit |= 1;
if (!opt->columnnum) {
/*
* Without --column, any single match on a line
* is enough to know that it needs to be
* printed. With --column, scan _all_ patterns
* to find the earliest.
*/
break;
}
if (*col < 0 || tmp.rm_so < *col)
*col = tmp.rm_so;
grep.c: expose {,inverted} match column in match_line() When calling match_line(), callers presently cannot determine the relative offset of the match because match_line() discards the 'regmatch_t' that contains this information. Instead, teach match_line() to take in two 'ssize_t's. Fill the first with the offset of the match produced by the given expression. If extended, fill the later with the offset of the match produced as if --invert were given. For instance, matching "--not -e x" on this line produces a columnar offset of 0, (i.e., the whole line does not contain an x), but "--invert --not -e -x" will fill the later ssize_t of the column containing an "x", because this expression is semantically equivalent to "-e x". To determine the column for the inverted and non-inverted case, do the following: - If matching an atom, the non-inverted column is as given from match_one_pattern(), and the inverted column is unset. - If matching a --not, the inverted column and non-inverted column swap. - If matching an --and, or --or, the non-inverted column is the minimum of the two children. Presently, the existing short-circuiting logic for AND and OR applies as before. This will change in the following commit when we add options to configure the --column flag. Taken together, this and the forthcoming change will always yield the earlier column on a given line. This patch will become useful when we later pick between the two new results in order to display the column number of the first match on a line with --column. Co-authored-by: Jeff King <peff@peff.net> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-06-22 15:49:35 +00:00
}
}
return hit;
}
static int match_next_pattern(struct grep_pat *p,
const char *bol, const char *eol,
enum grep_context ctx,
regmatch_t *pmatch, int eflags)
{
regmatch_t match;
if (!headerless_match_one_pattern(p, bol, eol, ctx, &match, eflags))
return 0;
if (match.rm_so < 0 || match.rm_eo < 0)
return 0;
if (pmatch->rm_so >= 0 && pmatch->rm_eo >= 0) {
if (match.rm_so > pmatch->rm_so)
return 1;
if (match.rm_so == pmatch->rm_so && match.rm_eo < pmatch->rm_eo)
return 1;
}
pmatch->rm_so = match.rm_so;
pmatch->rm_eo = match.rm_eo;
return 1;
}
int grep_next_match(struct grep_opt *opt,
const char *bol, const char *eol,
enum grep_context ctx, regmatch_t *pmatch,
enum grep_header_field field, int eflags)
{
struct grep_pat *p;
int hit = 0;
pmatch->rm_so = pmatch->rm_eo = -1;
if (bol < eol) {
for (p = ((ctx == GREP_CONTEXT_HEAD)
? opt->header_list : opt->pattern_list);
p; p = p->next) {
switch (p->token) {
case GREP_PATTERN_HEAD:
if ((field != GREP_HEADER_FIELD_MAX) &&
(p->field != field))
continue;
/* fall thru */
case GREP_PATTERN: /* atom */
case GREP_PATTERN_BODY:
hit |= match_next_pattern(p, bol, eol, ctx,
pmatch, eflags);
break;
default:
break;
}
}
}
return hit;
}
static void show_line_header(struct grep_opt *opt, const char *name,
unsigned lno, ssize_t cno, char sign)
{
if (opt->heading && opt->last_shown == 0) {
output_color(opt, name, strlen(name), opt->colors[GREP_COLOR_FILENAME]);
opt->output(opt, "\n", 1);
}
opt->last_shown = lno;
if (!opt->heading && opt->pathname) {
output_color(opt, name, strlen(name), opt->colors[GREP_COLOR_FILENAME]);
output_sep(opt, sign);
}
if (opt->linenum) {
char buf[32];
xsnprintf(buf, sizeof(buf), "%d", lno);
output_color(opt, buf, strlen(buf), opt->colors[GREP_COLOR_LINENO]);
output_sep(opt, sign);
}
/*
* Treat 'cno' as the 1-indexed offset from the start of a non-context
* line to its first match. Otherwise, 'cno' is 0 indicating that we are
* being called with a context line.
*/
if (opt->columnnum && cno) {
char buf[32];
xsnprintf(buf, sizeof(buf), "%"PRIuMAX, (uintmax_t)cno);
output_color(opt, buf, strlen(buf), opt->colors[GREP_COLOR_COLUMNNO]);
output_sep(opt, sign);
}
}
static void show_line(struct grep_opt *opt,
const char *bol, const char *eol,
const char *name, unsigned lno, ssize_t cno, char sign)
{
int rest = eol - bol;
const char *match_color = NULL;
const char *line_color = NULL;
if (opt->file_break && opt->last_shown == 0) {
if (opt->show_hunk_mark)
opt->output(opt, "\n", 1);
} else if (opt->pre_context || opt->post_context || opt->funcbody) {
if (opt->last_shown == 0) {
if (opt->show_hunk_mark) {
output_color(opt, "--", 2, opt->colors[GREP_COLOR_SEP]);
opt->output(opt, "\n", 1);
}
} else if (lno > opt->last_shown + 1) {
output_color(opt, "--", 2, opt->colors[GREP_COLOR_SEP]);
opt->output(opt, "\n", 1);
}
}
if (!opt->only_matching) {
/*
* In case the line we're being called with contains more than
* one match, leave printing each header to the loop below.
*/
show_line_header(opt, name, lno, cno, sign);
}
if (opt->color || opt->only_matching) {
regmatch_t match;
enum grep_context ctx = GREP_CONTEXT_BODY;
int eflags = 0;
if (opt->color) {
if (sign == ':')
match_color = opt->colors[GREP_COLOR_MATCH_SELECTED];
else
match_color = opt->colors[GREP_COLOR_MATCH_CONTEXT];
if (sign == ':')
line_color = opt->colors[GREP_COLOR_SELECTED];
else if (sign == '-')
line_color = opt->colors[GREP_COLOR_CONTEXT];
else if (sign == '=')
line_color = opt->colors[GREP_COLOR_FUNCTION];
}
while (grep_next_match(opt, bol, eol, ctx, &match,
GREP_HEADER_FIELD_MAX, eflags)) {
if (match.rm_so == match.rm_eo)
break;
if (opt->only_matching)
show_line_header(opt, name, lno, cno, sign);
else
output_color(opt, bol, match.rm_so, line_color);
output_color(opt, bol + match.rm_so,
match.rm_eo - match.rm_so, match_color);
if (opt->only_matching)
opt->output(opt, "\n", 1);
bol += match.rm_eo;
cno += match.rm_eo;
rest -= match.rm_eo;
eflags = REG_NOTBOL;
}
}
if (!opt->only_matching) {
output_color(opt, bol, rest, line_color);
opt->output(opt, "\n", 1);
}
}
grep: make locking flag global The low-level grep code traditionally didn't care about threading, as it doesn't do any threading itself and didn't call out to other non-thread-safe code. That changed with 0579f91 (grep: enable threading with -p and -W using lazy attribute lookup, 2011-12-12), which pushed the lookup of funcname attributes (which is not thread-safe) into the low-level grep code. As a result, the low-level code learned about a new global "grep_attr_mutex" to serialize access to the attribute code. A multi-threaded caller (e.g., builtin/grep.c) is expected to initialize the mutex and set "use_threads" in the grep_opt structure. The low-level code only uses the lock if use_threads is set. However, putting the use_threads flag into the grep_opt struct is not the most logical place. Whether threading is in use is not something that matters for each call to grep_buffer, but is instead global to the whole program (i.e., if any thread is doing multi-threaded grep, every other thread, even if it thinks it is doing its own single-threaded grep, would need to use the locking). In practice, this distinction isn't a problem for us, because the only user of multi-threaded grep is "git-grep", which does nothing except call grep. This patch turns the opt->use_threads flag into a global flag. More important than the nit-picking semantic argument above is that this means that the locking functions don't need to actually have access to a grep_opt to know whether to lock. Which in turn can make adding new locks simpler, as we don't need to pass around a grep_opt. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2012-02-02 08:18:29 +00:00
int grep_use_locks;
/*
* This lock protects access to the gitattributes machinery, which is
* not thread-safe.
*/
pthread_mutex_t grep_attr_mutex;
grep: make locking flag global The low-level grep code traditionally didn't care about threading, as it doesn't do any threading itself and didn't call out to other non-thread-safe code. That changed with 0579f91 (grep: enable threading with -p and -W using lazy attribute lookup, 2011-12-12), which pushed the lookup of funcname attributes (which is not thread-safe) into the low-level grep code. As a result, the low-level code learned about a new global "grep_attr_mutex" to serialize access to the attribute code. A multi-threaded caller (e.g., builtin/grep.c) is expected to initialize the mutex and set "use_threads" in the grep_opt structure. The low-level code only uses the lock if use_threads is set. However, putting the use_threads flag into the grep_opt struct is not the most logical place. Whether threading is in use is not something that matters for each call to grep_buffer, but is instead global to the whole program (i.e., if any thread is doing multi-threaded grep, every other thread, even if it thinks it is doing its own single-threaded grep, would need to use the locking). In practice, this distinction isn't a problem for us, because the only user of multi-threaded grep is "git-grep", which does nothing except call grep. This patch turns the opt->use_threads flag into a global flag. More important than the nit-picking semantic argument above is that this means that the locking functions don't need to actually have access to a grep_opt to know whether to lock. Which in turn can make adding new locks simpler, as we don't need to pass around a grep_opt. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2012-02-02 08:18:29 +00:00
static inline void grep_attr_lock(void)
{
grep: make locking flag global The low-level grep code traditionally didn't care about threading, as it doesn't do any threading itself and didn't call out to other non-thread-safe code. That changed with 0579f91 (grep: enable threading with -p and -W using lazy attribute lookup, 2011-12-12), which pushed the lookup of funcname attributes (which is not thread-safe) into the low-level grep code. As a result, the low-level code learned about a new global "grep_attr_mutex" to serialize access to the attribute code. A multi-threaded caller (e.g., builtin/grep.c) is expected to initialize the mutex and set "use_threads" in the grep_opt structure. The low-level code only uses the lock if use_threads is set. However, putting the use_threads flag into the grep_opt struct is not the most logical place. Whether threading is in use is not something that matters for each call to grep_buffer, but is instead global to the whole program (i.e., if any thread is doing multi-threaded grep, every other thread, even if it thinks it is doing its own single-threaded grep, would need to use the locking). In practice, this distinction isn't a problem for us, because the only user of multi-threaded grep is "git-grep", which does nothing except call grep. This patch turns the opt->use_threads flag into a global flag. More important than the nit-picking semantic argument above is that this means that the locking functions don't need to actually have access to a grep_opt to know whether to lock. Which in turn can make adding new locks simpler, as we don't need to pass around a grep_opt. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2012-02-02 08:18:29 +00:00
if (grep_use_locks)
pthread_mutex_lock(&grep_attr_mutex);
}
grep: make locking flag global The low-level grep code traditionally didn't care about threading, as it doesn't do any threading itself and didn't call out to other non-thread-safe code. That changed with 0579f91 (grep: enable threading with -p and -W using lazy attribute lookup, 2011-12-12), which pushed the lookup of funcname attributes (which is not thread-safe) into the low-level grep code. As a result, the low-level code learned about a new global "grep_attr_mutex" to serialize access to the attribute code. A multi-threaded caller (e.g., builtin/grep.c) is expected to initialize the mutex and set "use_threads" in the grep_opt structure. The low-level code only uses the lock if use_threads is set. However, putting the use_threads flag into the grep_opt struct is not the most logical place. Whether threading is in use is not something that matters for each call to grep_buffer, but is instead global to the whole program (i.e., if any thread is doing multi-threaded grep, every other thread, even if it thinks it is doing its own single-threaded grep, would need to use the locking). In practice, this distinction isn't a problem for us, because the only user of multi-threaded grep is "git-grep", which does nothing except call grep. This patch turns the opt->use_threads flag into a global flag. More important than the nit-picking semantic argument above is that this means that the locking functions don't need to actually have access to a grep_opt to know whether to lock. Which in turn can make adding new locks simpler, as we don't need to pass around a grep_opt. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2012-02-02 08:18:29 +00:00
static inline void grep_attr_unlock(void)
{
grep: make locking flag global The low-level grep code traditionally didn't care about threading, as it doesn't do any threading itself and didn't call out to other non-thread-safe code. That changed with 0579f91 (grep: enable threading with -p and -W using lazy attribute lookup, 2011-12-12), which pushed the lookup of funcname attributes (which is not thread-safe) into the low-level grep code. As a result, the low-level code learned about a new global "grep_attr_mutex" to serialize access to the attribute code. A multi-threaded caller (e.g., builtin/grep.c) is expected to initialize the mutex and set "use_threads" in the grep_opt structure. The low-level code only uses the lock if use_threads is set. However, putting the use_threads flag into the grep_opt struct is not the most logical place. Whether threading is in use is not something that matters for each call to grep_buffer, but is instead global to the whole program (i.e., if any thread is doing multi-threaded grep, every other thread, even if it thinks it is doing its own single-threaded grep, would need to use the locking). In practice, this distinction isn't a problem for us, because the only user of multi-threaded grep is "git-grep", which does nothing except call grep. This patch turns the opt->use_threads flag into a global flag. More important than the nit-picking semantic argument above is that this means that the locking functions don't need to actually have access to a grep_opt to know whether to lock. Which in turn can make adding new locks simpler, as we don't need to pass around a grep_opt. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2012-02-02 08:18:29 +00:00
if (grep_use_locks)
pthread_mutex_unlock(&grep_attr_mutex);
}
static int match_funcname(struct grep_opt *opt, struct grep_source *gs,
const char *bol, const char *eol)
{
xdemitconf_t *xecfg = opt->priv;
if (xecfg && !xecfg->find_func) {
grep_source_load_driver(gs, opt->repo->index);
if (gs->driver->funcname.pattern) {
const struct userdiff_funcname *pe = &gs->driver->funcname;
xdiff_set_find_func(xecfg, pe->pattern, pe->cflags);
} else {
xecfg = opt->priv = NULL;
}
}
if (xecfg) {
char buf[1];
return xecfg->find_func(bol, eol - bol, buf, 1,
xecfg->find_func_priv) >= 0;
}
if (bol == eol)
return 0;
if (isalpha(*bol) || *bol == '_' || *bol == '$')
return 1;
return 0;
}
static void show_funcname_line(struct grep_opt *opt, struct grep_source *gs,
const char *bol, unsigned lno)
{
while (bol > gs->buf) {
const char *eol = --bol;
while (bol > gs->buf && bol[-1] != '\n')
bol--;
lno--;
if (lno <= opt->last_shown)
break;
if (match_funcname(opt, gs, bol, eol)) {
show_line(opt, bol, eol, gs->name, lno, 0, '=');
break;
}
}
}
static int is_empty_line(const char *bol, const char *eol);
static void show_pre_context(struct grep_opt *opt, struct grep_source *gs,
const char *bol, const char *end, unsigned lno)
{
unsigned cur = lno, from = 1, funcname_lno = 0, orig_from;
int funcname_needed = !!opt->funcname, comment_needed = 0;
if (opt->pre_context < lno)
from = lno - opt->pre_context;
if (from <= opt->last_shown)
from = opt->last_shown + 1;
orig_from = from;
if (opt->funcbody) {
if (match_funcname(opt, gs, bol, end))
comment_needed = 1;
else
funcname_needed = 1;
from = opt->last_shown + 1;
}
/* Rewind. */
while (bol > gs->buf && cur > from) {
const char *next_bol = bol;
const char *eol = --bol;
while (bol > gs->buf && bol[-1] != '\n')
bol--;
cur--;
if (comment_needed && (is_empty_line(bol, eol) ||
match_funcname(opt, gs, bol, eol))) {
comment_needed = 0;
from = orig_from;
if (cur < from) {
cur++;
bol = next_bol;
break;
}
}
if (funcname_needed && match_funcname(opt, gs, bol, eol)) {
funcname_lno = cur;
funcname_needed = 0;
if (opt->funcbody)
comment_needed = 1;
else
from = orig_from;
}
}
/* We need to look even further back to find a function signature. */
if (opt->funcname && funcname_needed)
show_funcname_line(opt, gs, bol, cur);
/* Back forward. */
while (cur < lno) {
const char *eol = bol, sign = (cur == funcname_lno) ? '=' : '-';
while (*eol != '\n')
eol++;
show_line(opt, bol, eol, gs->name, cur, 0, sign);
bol = eol + 1;
cur++;
}
}
static int should_lookahead(struct grep_opt *opt)
{
struct grep_pat *p;
grep.c: remove "extended" in favor of "pattern_expression", fix segfault Since 79d3696cfb4 (git-grep: boolean expression on pattern matching., 2006-06-30) the "pattern_expression" member has been used for complex queries (AND/OR...), with "pattern_list" being used for the simple OR queries. Since then we've used both "pattern_expression" and its associated boolean "extended" member to see if we have a complex expression. Since f41fb662f57 (revisions API: have release_revisions() release "grep_filter", 2022-04-13) we've had a subtle bug relating to that: If we supplied options that were only used for "complex queries", but didn't supply the query itself we'd set "opt->extended", but would have a NULL "pattern_expression". As a result these would segfault as we tried to call "free_grep_patterns()" from "release_revisions()": git -P log -1 --invert-grep git -P log -1 --all-match The root cause of this is that we were conflating the state management we needed in "compile_grep_patterns()" itself with whether or not we had an "opt->pattern_expression" later on. In this cases as we're going through "compile_grep_patterns()" we have no "opt->pattern_list" but have "opt->no_body_match" or "opt->all_match". So we'd set "opt->extended = 1", but not "return" on "opt->extended" as that's an "else if" in the same "if" statement. That behavior is intentional and required, as the common case is that we have an "opt->pattern_list" that we're about to parse into the "opt->pattern_expression". But we don't need to keep track of this "extended" flag beyond the state management in compile_grep_patterns() itself. It needs it, but once we're out of that function we can rely on "opt->pattern_expression" being non-NULL instead for using these extended patterns. As 79d3696cfb4 itself shows we've assumed that there's a one-to-one mapping between the two since the very beginning. I.e. "match_line()" would check "opt->extended" to see if it should call "match_expr()", and the first thing we do in that function is assume that we have a "opt->pattern_expression". We'd then call "match_expr_eval()", which would have died if that "opt->pattern_expression" was NULL. The "die" was added in c922b01f54c (grep: fix segfault when "git grep '('" is given, 2009-04-27), and can now be removed as it's now clearly unreachable. We still do the right thing in the case that prompted that fix: git grep '(' fatal: unmatched parenthesis Arguably neither the "--invert-grep" option added in [1] nor the earlier "--all-match" option added in [2] were intended to be used stand-alone, and another approach[3] would be to error out in those cases. But since we've been treating them as a NOOP when given without --grep for a long time let's keep doing that. We could also return in "free_pattern_expr()" if the argument is non-NULL, as an alternative fix for this segfault does [4]. That would be more elegant in making the "free_*()" function behave like "free()", but it would also remove a sanity check: The "free_pattern_expr()" function calls itself recursively, and only the top-level is allowed to be NULL, let's not conflate those two conditions. 1. 22dfa8a23de (log: teach --invert-grep option, 2015-01-12) 2. 0ab7befa31d (grep --all-match, 2006-09-27) 3. https://lore.kernel.org/git/patch-1.1-f4b90799fce-20221010T165711Z-avarab@gmail.com/ 4. http://lore.kernel.org/git/7e094882c2a71894416089f894557a9eae07e8f8.1665423686.git.me@ttaylorr.com Reported-by: orygaw <orygaw@protonmail.com> Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2022-10-11 09:48:45 +00:00
if (opt->pattern_expression)
return 0; /* punt for too complex stuff */
if (opt->invert)
return 0;
for (p = opt->pattern_list; p; p = p->next) {
if (p->token != GREP_PATTERN)
return 0; /* punt for "header only" and stuff */
}
return 1;
}
static int look_ahead(struct grep_opt *opt,
unsigned long *left_p,
unsigned *lno_p,
const char **bol_p)
{
unsigned lno = *lno_p;
const char *bol = *bol_p;
struct grep_pat *p;
const char *sp, *last_bol;
regoff_t earliest = -1;
for (p = opt->pattern_list; p; p = p->next) {
int hit;
regmatch_t m;
hit = patmatch(p, bol, bol + *left_p, &m, 0);
if (!hit || m.rm_so < 0 || m.rm_eo < 0)
continue;
if (earliest < 0 || m.rm_so < earliest)
earliest = m.rm_so;
}
if (earliest < 0) {
*bol_p = bol + *left_p;
*left_p = 0;
return 1;
}
for (sp = bol + earliest; bol < sp && sp[-1] != '\n'; sp--)
; /* find the beginning of the line */
last_bol = sp;
for (sp = bol; sp < last_bol; sp++) {
if (*sp == '\n')
lno++;
}
*left_p -= last_bol - bol;
*bol_p = last_bol;
*lno_p = lno;
return 0;
}
static int fill_textconv_grep(struct repository *r,
struct userdiff_driver *driver,
struct grep_source *gs)
{
struct diff_filespec *df;
char *buf;
size_t size;
if (!driver || !driver->textconv)
return grep_source_load(gs);
/*
* The textconv interface is intimately tied to diff_filespecs, so we
* have to pretend to be one. If we could unify the grep_source
* and diff_filespec structs, this mess could just go away.
*/
df = alloc_filespec(gs->path);
switch (gs->type) {
case GREP_SOURCE_OID:
fill_filespec(df, gs->identifier, 1, 0100644);
break;
case GREP_SOURCE_FILE:
fill_filespec(df, null_oid(), 0, 0100644);
break;
default:
BUG("attempt to textconv something without a path?");
}
/*
* fill_textconv is not remotely thread-safe; it modifies the global
* diff tempfile structure, writes to the_repo's odb and might
* internally call thread-unsafe functions such as the
* prepare_packed_git() lazy-initializator. Because of the last two, we
* must ensure mutual exclusion between this call and the object reading
* API, thus we use obj_read_lock() here.
*
* TODO: allowing text conversion to run in parallel with object
* reading operations might increase performance in the multithreaded
* non-worktreee git-grep with --textconv.
*/
obj_read_lock();
size = fill_textconv(r, driver, df, &buf);
obj_read_unlock();
free_filespec(df);
/*
* The normal fill_textconv usage by the diff machinery would just keep
* the textconv'd buf separate from the diff_filespec. But much of the
* grep code passes around a grep_source and assumes that its "buf"
* pointer is the beginning of the thing we are searching. So let's
* install our textconv'd version into the grep_source, taking care not
* to leak any existing buffer.
*/
grep_source_clear_data(gs);
gs->buf = buf;
gs->size = size;
return 0;
}
static int is_empty_line(const char *bol, const char *eol)
{
while (bol < eol && isspace(*bol))
bol++;
return bol == eol;
}
static int grep_source_1(struct grep_opt *opt, struct grep_source *gs, int collect_hits)
{
const char *bol;
const char *peek_bol = NULL;
unsigned long left;
unsigned lno = 1;
unsigned last_hit = 0;
int binary_match_only = 0;
unsigned count = 0;
int try_lookahead = 0;
int show_function = 0;
struct userdiff_driver *textconv = NULL;
enum grep_context ctx = GREP_CONTEXT_HEAD;
xdemitconf_t xecfg;
grep: fail if call could output and name is null grep_source(), which performs much of the work for Git's grep library, allows passing an arbitrary struct grep_source which represents the text which grep_source() should search to match a pattern in the provided struct grep_opt. In most callers, the grep_source::name field is set to an appropriate prefix to print before a colon when a result matches: README:Git is an Open Source project covered by the GNU General One caller, grep_buffer(), leaves the grep_source::name field set to NULL because there isn't enough context to determine an appropriate name for this kind of output line. In practice, this has been fine: the only caller of grep_buffer() is "git log --grep", and that caller sets grep_opt::status_only, which disables output and only checks whether a match exists. But this is brittle: a future caller can call grep_buffer() without grep_opt::status_only set, and as soon as it hits a match, grep_source() will try to print the match and segfault: (null):Git is an Open Source project covered by the GNU General For example, a future caller might want to print all matching lines from commits which match a regex. Futureproof by diagnosing early a use of the API that could trigger that condition, before we know whether the pattern matches: BUG: grep.c:1783: grep call which could print a name requires grep_source.name be non-NULL Aborted This way, the caller's author gets an indication of how to fix the issue - by providing grep_source::name or setting grep_opt::status_only - and they are warned of the potential for a segfault unconditionally, rather than only if there is a match. Noticed while adding such a call to a tutorial on revision walks. Signed-off-by: Emily Shaffer <emilyshaffer@google.com> Reviewed-by: Jonathan Nieder <jrnieder@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-23 20:23:56 +00:00
if (!opt->status_only && gs->name == NULL)
BUG("grep call which could print a name requires "
"grep_source.name be non-NULL");
if (!opt->output)
opt->output = std_output;
if (opt->pre_context || opt->post_context || opt->file_break ||
opt->funcbody) {
/* Show hunk marks, except for the first file. */
if (opt->last_shown)
opt->show_hunk_mark = 1;
/*
* If we're using threads then we can't easily identify
* the first file. Always put hunk marks in that case
* and skip the very first one later in work_done().
*/
if (opt->output != std_output)
opt->show_hunk_mark = 1;
}
opt->last_shown = 0;
if (opt->allow_textconv) {
grep_source_load_driver(gs, opt->repo->index);
/*
* We might set up the shared textconv cache data here, which
* is not thread-safe. Also, get_oid_with_context() and
* parse_object() might be internally called. As they are not
* currently thread-safe and might be racy with object reading,
* obj_read_lock() must be called.
*/
grep_attr_lock();
obj_read_lock();
textconv = userdiff_get_textconv(opt->repo, gs->driver);
obj_read_unlock();
grep_attr_unlock();
}
/*
* We know the result of a textconv is text, so we only have to care
* about binary handling if we are not using it.
*/
if (!textconv) {
switch (opt->binary) {
case GREP_BINARY_DEFAULT:
if (grep_source_is_binary(gs, opt->repo->index))
binary_match_only = 1;
break;
case GREP_BINARY_NOMATCH:
if (grep_source_is_binary(gs, opt->repo->index))
return 0; /* Assume unmatch */
break;
case GREP_BINARY_TEXT:
break;
default:
BUG("unknown binary handling mode");
}
}
memset(&xecfg, 0, sizeof(xecfg));
opt->priv = &xecfg;
try_lookahead = should_lookahead(opt);
if (fill_textconv_grep(opt->repo, textconv, gs) < 0)
return 0;
bol = gs->buf;
left = gs->size;
while (left) {
const char *eol;
int hit;
ssize_t cno;
grep.c: expose {,inverted} match column in match_line() When calling match_line(), callers presently cannot determine the relative offset of the match because match_line() discards the 'regmatch_t' that contains this information. Instead, teach match_line() to take in two 'ssize_t's. Fill the first with the offset of the match produced by the given expression. If extended, fill the later with the offset of the match produced as if --invert were given. For instance, matching "--not -e x" on this line produces a columnar offset of 0, (i.e., the whole line does not contain an x), but "--invert --not -e -x" will fill the later ssize_t of the column containing an "x", because this expression is semantically equivalent to "-e x". To determine the column for the inverted and non-inverted case, do the following: - If matching an atom, the non-inverted column is as given from match_one_pattern(), and the inverted column is unset. - If matching a --not, the inverted column and non-inverted column swap. - If matching an --and, or --or, the non-inverted column is the minimum of the two children. Presently, the existing short-circuiting logic for AND and OR applies as before. This will change in the following commit when we add options to configure the --column flag. Taken together, this and the forthcoming change will always yield the earlier column on a given line. This patch will become useful when we later pick between the two new results in order to display the column number of the first match on a line with --column. Co-authored-by: Jeff King <peff@peff.net> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-06-22 15:49:35 +00:00
ssize_t col = -1, icol = -1;
/*
* look_ahead() skips quickly to the line that possibly
* has the next hit; don't call it if we need to do
* something more than just skipping the current line
* in response to an unmatch for the current line. E.g.
* inside a post-context window, we will show the current
* line as a context around the previous hit when it
* doesn't hit.
*/
if (try_lookahead
&& !(last_hit
&& (show_function ||
lno <= last_hit + opt->post_context))
&& look_ahead(opt, &left, &lno, &bol))
break;
eol = end_of_line(bol, &left);
if ((ctx == GREP_CONTEXT_HEAD) && (eol == bol))
ctx = GREP_CONTEXT_BODY;
grep.c: expose {,inverted} match column in match_line() When calling match_line(), callers presently cannot determine the relative offset of the match because match_line() discards the 'regmatch_t' that contains this information. Instead, teach match_line() to take in two 'ssize_t's. Fill the first with the offset of the match produced by the given expression. If extended, fill the later with the offset of the match produced as if --invert were given. For instance, matching "--not -e x" on this line produces a columnar offset of 0, (i.e., the whole line does not contain an x), but "--invert --not -e -x" will fill the later ssize_t of the column containing an "x", because this expression is semantically equivalent to "-e x". To determine the column for the inverted and non-inverted case, do the following: - If matching an atom, the non-inverted column is as given from match_one_pattern(), and the inverted column is unset. - If matching a --not, the inverted column and non-inverted column swap. - If matching an --and, or --or, the non-inverted column is the minimum of the two children. Presently, the existing short-circuiting logic for AND and OR applies as before. This will change in the following commit when we add options to configure the --column flag. Taken together, this and the forthcoming change will always yield the earlier column on a given line. This patch will become useful when we later pick between the two new results in order to display the column number of the first match on a line with --column. Co-authored-by: Jeff King <peff@peff.net> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-06-22 15:49:35 +00:00
hit = match_line(opt, bol, eol, &col, &icol, ctx, collect_hits);
if (collect_hits)
goto next_line;
/* "grep -v -e foo -e bla" should list lines
* that do not have either, so inversion should
* be done outside.
*/
if (opt->invert)
hit = !hit;
if (opt->unmatch_name_only) {
if (hit)
return 0;
goto next_line;
}
if (hit && (opt->max_count < 0 || count < opt->max_count)) {
count++;
if (opt->status_only)
return 1;
if (opt->name_only) {
show_name(opt, gs->name);
return 1;
}
if (opt->count)
goto next_line;
if (binary_match_only) {
opt->output(opt, "Binary file ", 12);
output_color(opt, gs->name, strlen(gs->name),
opt->colors[GREP_COLOR_FILENAME]);
opt->output(opt, " matches\n", 9);
return 1;
}
/* Hit at this line. If we haven't shown the
* pre-context lines, we would need to show them.
*/
if (opt->pre_context || opt->funcbody)
show_pre_context(opt, gs, bol, eol, lno);
else if (opt->funcname)
show_funcname_line(opt, gs, bol, lno);
cno = opt->invert ? icol : col;
if (cno < 0) {
/*
* A negative cno indicates that there was no
* match on the line. We are thus inverted and
* being asked to show all lines that _don't_
* match a given expression. Therefore, set cno
* to 0 to suggest the whole line matches.
*/
cno = 0;
}
show_line(opt, bol, eol, gs->name, lno, cno + 1, ':');
last_hit = lno;
if (opt->funcbody)
show_function = 1;
goto next_line;
}
if (show_function && (!peek_bol || peek_bol < bol)) {
unsigned long peek_left = left;
const char *peek_eol = eol;
/*
* Trailing empty lines are not interesting.
* Peek past them to see if they belong to the
* body of the current function.
*/
peek_bol = bol;
while (is_empty_line(peek_bol, peek_eol)) {
peek_bol = peek_eol + 1;
peek_eol = end_of_line(peek_bol, &peek_left);
}
if (match_funcname(opt, gs, peek_bol, peek_eol))
show_function = 0;
}
if (show_function ||
(last_hit && lno <= last_hit + opt->post_context)) {
/* If the last hit is within the post context,
* we need to show this line.
*/
show_line(opt, bol, eol, gs->name, lno, col + 1, '-');
}
next_line:
bol = eol + 1;
if (!left)
break;
left--;
lno++;
}
if (collect_hits)
return 0;
if (opt->status_only)
return opt->unmatch_name_only;
if (opt->unmatch_name_only) {
/* We did not see any hit, so we want to show this */
show_name(opt, gs->name);
return 1;
}
xdiff_clear_find_func(&xecfg);
opt->priv = NULL;
/* NEEDSWORK:
* The real "grep -c foo *.c" gives many "bar.c:0" lines,
* which feels mostly useless but sometimes useful. Maybe
* make it another option? For now suppress them.
*/
if (opt->count && count) {
char buf[32];
if (opt->pathname) {
output_color(opt, gs->name, strlen(gs->name),
opt->colors[GREP_COLOR_FILENAME]);
output_sep(opt, ':');
}
xsnprintf(buf, sizeof(buf), "%u\n", count);
opt->output(opt, buf, strlen(buf));
return 1;
}
return !!last_hit;
}
static void clr_hit_marker(struct grep_expr *x)
{
/* All-hit markers are meaningful only at the very top level
* OR node.
*/
while (1) {
x->hit = 0;
if (x->node != GREP_NODE_OR)
return;
x->u.binary.left->hit = 0;
x = x->u.binary.right;
}
}
static int chk_hit_marker(struct grep_expr *x)
{
/* Top level nodes have hit markers. See if they all are hits */
while (1) {
if (x->node != GREP_NODE_OR)
return x->hit;
if (!x->u.binary.left->hit)
return 0;
x = x->u.binary.right;
}
}
int grep_source(struct grep_opt *opt, struct grep_source *gs)
{
/*
* we do not have to do the two-pass grep when we do not check
* buffer-wide "all-match".
*/
if (!opt->all_match && !opt->no_body_match)
return grep_source_1(opt, gs, 0);
/* Otherwise the toplevel "or" terms hit a bit differently.
* We first clear hit markers from them.
*/
clr_hit_marker(opt->pattern_expression);
opt->body_hit = 0;
grep_source_1(opt, gs, 1);
if (opt->all_match && !chk_hit_marker(opt->pattern_expression))
return 0;
if (opt->no_body_match && opt->body_hit)
return 0;
return grep_source_1(opt, gs, 0);
}
static void grep_source_init_buf(struct grep_source *gs,
const char *buf,
unsigned long size)
{
gs->type = GREP_SOURCE_BUF;
gs->name = NULL;
gs->path = NULL;
gs->buf = buf;
gs->size = size;
gs->driver = NULL;
gs->identifier = NULL;
}
int grep_buffer(struct grep_opt *opt, const char *buf, unsigned long size)
{
struct grep_source gs;
int r;
grep_source_init_buf(&gs, buf, size);
r = grep_source(opt, &gs);
grep_source_clear(&gs);
return r;
}
void grep_source_init_file(struct grep_source *gs, const char *name,
const char *path)
{
gs->type = GREP_SOURCE_FILE;
gs->name = xstrdup_or_null(name);
gs->path = xstrdup_or_null(path);
gs->buf = NULL;
gs->size = 0;
gs->driver = NULL;
gs->identifier = xstrdup(path);
}
void grep_source_init_oid(struct grep_source *gs, const char *name,
const char *path, const struct object_id *oid,
struct repository *repo)
{
gs->type = GREP_SOURCE_OID;
gs->name = xstrdup_or_null(name);
gs->path = xstrdup_or_null(path);
gs->buf = NULL;
gs->size = 0;
gs->driver = NULL;
gs->identifier = oiddup(oid);
gs->repo = repo;
}
void grep_source_clear(struct grep_source *gs)
{
FREE_AND_NULL(gs->name);
FREE_AND_NULL(gs->path);
FREE_AND_NULL(gs->identifier);
grep_source_clear_data(gs);
}
void grep_source_clear_data(struct grep_source *gs)
{
switch (gs->type) {
case GREP_SOURCE_FILE:
case GREP_SOURCE_OID:
/* these types own the buffer */
free((char *)gs->buf);
gs->buf = NULL;
gs->size = 0;
break;
case GREP_SOURCE_BUF:
/* leave user-provided buf intact */
break;
}
}
static int grep_source_load_oid(struct grep_source *gs)
{
enum object_type type;
gs->buf = repo_read_object_file(gs->repo, gs->identifier, &type,
&gs->size);
if (!gs->buf)
return error(_("'%s': unable to read %s"),
gs->name,
oid_to_hex(gs->identifier));
return 0;
}
static int grep_source_load_file(struct grep_source *gs)
{
const char *filename = gs->identifier;
struct stat st;
char *data;
size_t size;
int i;
if (lstat(filename, &st) < 0) {
err_ret:
if (errno != ENOENT)
error_errno(_("failed to stat '%s'"), filename);
return -1;
}
if (!S_ISREG(st.st_mode))
return -1;
size = xsize_t(st.st_size);
i = open(filename, O_RDONLY);
if (i < 0)
goto err_ret;
data = xmallocz(size);
if (st.st_size != read_in_full(i, data, size)) {
error_errno(_("'%s': short read"), filename);
close(i);
free(data);
return -1;
}
close(i);
gs->buf = data;
gs->size = size;
return 0;
}
static int grep_source_load(struct grep_source *gs)
{
if (gs->buf)
return 0;
switch (gs->type) {
case GREP_SOURCE_FILE:
return grep_source_load_file(gs);
case GREP_SOURCE_OID:
return grep_source_load_oid(gs);
case GREP_SOURCE_BUF:
return gs->buf ? 0 : -1;
}
BUG("invalid grep_source type to load");
}
void grep_source_load_driver(struct grep_source *gs,
struct index_state *istate)
{
if (gs->driver)
return;
grep_attr_lock();
if (gs->path)
gs->driver = userdiff_find_by_path(istate, gs->path);
if (!gs->driver)
gs->driver = userdiff_find_by_name("default");
grep_attr_unlock();
}
static int grep_source_is_binary(struct grep_source *gs,
struct index_state *istate)
{
grep_source_load_driver(gs, istate);
if (gs->driver->binary != -1)
return gs->driver->binary;
if (!grep_source_load(gs))
return buffer_is_binary(gs->buf, gs->size);
return 0;
}