In 73320e49ad (builtin/repack.c: only collect fully-formed packs,
2023-06-07), we switched the check for which packs to collect by
starting at the .idx files and looking for matching .pack files. This
avoids trying to repack pack-files that have not had their pack-indexes
installed yet.
However, it does cause maintenance to halt if we find the (problematic,
but not insurmountable) case of a .idx file without a corresponding
.pack file. In an environment where packfile maintenance is a critical
function, such a hard stop is costly and requires human intervention to
resolve (by deleting the .idx file).
This was not the case before. We successfully repacked through this
scenario until the recent change to scan for .idx files.
Further, if we are actually in a case where objects are missing, we
detect this at a different point during the reachability walk.
In other cases, Git prepares its list of packfiles by scanning .idx
files and then only adds it to the packfile list if the corresponding
.pack file exists. It even does so without a warning! (See
add_packed_git() in packfile.c for details.)
This case is much less likely to occur than the failures seen before
73320e49ad. Packfiles are "installed" by writing the .pack file before
the .idx and that process can be interrupted. Packfiles _should_ be
deleted by deleting the .idx first, followed by the .pack file, but
unlink_pack_path() does not do this: it deletes the .pack _first_,
allowing a window where this process could be interrupted. We leave the
consideration of changing this order as a separate concern. Knowing that
this condition is possible from interrupted Git processes and not other
tools lends some weight that Git should be more flexible around this
scenario.
Add a check to see if the .pack file exists before adding it to the list
for repacking. This will stop a number of maintenance failures seen in
production but fixed by deleting the .idx files.
This brings us closer to the case before 73320e49ad in that 'git
repack' will not fail when there is an orphaned .idx file, at least, not
due to the way we scan for packfiles. In the case that the .pack file
was erroneously deleted without copies of its objects in other installed
packfiles, then 'git repack' will fail due to the reachable object walk.
This does resolve the case where automated repacks will no longer be
halted on this case. The tests in t7700 show both these successful
scenarios and the case of failing if the .pack was truly required.
Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
A few places failed to differenciate the case where the index is
truly empty (nothing added) and we haven't yet read from the
on-disk index file, which have been corrected.
* js/empty-index-fixes:
commit -a -m: allow the top-level tree to become empty again
split-index: accept that a base index can be empty
do_read_index(): always mark index as initialized unless erroring out
Reduce reliance on a global state in the config reading API.
* gc/config-context:
config: pass source to config_parser_event_fn_t
config: add kvi.path, use it to evaluate includes
config.c: remove config_reader from configsets
config: pass kvi to die_bad_number()
trace2: plumb config kvi
config.c: pass ctx with CLI config
config: pass ctx with config files
config.c: pass ctx in configsets
config: add ctx arg to config_fn_t
urlmatch.h: use config_fn_t type
config: inline git_color_default_config
During a cherry-pick or revert session that works on multiple
commits, "git status" did not give correct information, which has
been corrected.
* jk/cherry-pick-revert-status:
fix cherry-pick/revert status when doing multiple commits
"git apply" punts when it is fed too large a patch input; the error
message it gives when it happens has been clarified.
* pw/apply-too-large:
apply: improve error messages when reading patch
'git notes append' was taught '--separator' to specify string to insert
between paragraphs.
* tl/notes-separator:
notes: introduce "--no-separator" option
notes.c: introduce "--[no-]stripspace" option
notes.c: append separator instead of insert by pos
notes.c: introduce '--separator=<paragraph-break>' option
t3321: add test cases about the notes stripspace behavior
notes.c: use designated initializers for clarity
notes.c: cleanup 'strbuf_grow' call in 'append_edit'
Partially revert a sanity check that the rest of the config code
was not ready, to avoid triggering it in a corner case.
* gc/config-partial-submodule-kvi-fix:
config: don't BUG when both kvi and source are set
Move functions that are not about pure string manipulation out of
strbuf.[ch]
* cw/strbuf-cleanup:
strbuf: remove global variable
path: move related function to path
object-name: move related functions to object-name
credential-store: move related functions to credential-store file
abspath: move related functions to abspath
strbuf: clarify dependency
strbuf: clarify API boundary
Add more "git var" for toolsmiths to learn various locations Git is
configured with either via the configuration or hardcoded defaults.
* bc/more-git-var:
var: add config file locations
var: add attributes files locations
attr: expose and rename accessor functions
var: adjust memory allocation for strings
var: format variable structure with C99 initializers
var: add support for listing the shell
t: add a function to check executable bit
var: mark unused parameters in git_var callbacks
The set-up code for the get_revision() API now allows feeding
options like --all and --not in the --stdin mode.
* ps/revision-stdin-with-options:
revision: handle pseudo-opts in `--stdin` mode
revision: small readability improvement for reading from stdin
revision: reorder `read_revisions_from_stdin()`
When the external merge driver is killed by a signal, its output
should not be trusted as a resolution with conflicts that is
proposed by the driver, but the code did.
* jc/abort-ll-merge-with-a-signal:
t6406: skip "external merge driver getting killed by a signal" test on Windows
ll-merge: killing the external merge driver aborts the merge
Header files cleanup.
* en/header-split-cache-h-part-3: (28 commits)
fsmonitor-ll.h: split this header out of fsmonitor.h
hash-ll, hashmap: move oidhash() to hash-ll
object-store-ll.h: split this header out of object-store.h
khash: name the structs that khash declares
merge-ll: rename from ll-merge
git-compat-util.h: remove unneccessary include of wildmatch.h
builtin.h: remove unneccessary includes
list-objects-filter-options.h: remove unneccessary include
diff.h: remove unnecessary include of oidset.h
repository: remove unnecessary include of path.h
log-tree: replace include of revision.h with simple forward declaration
cache.h: remove this no-longer-used header
read-cache*.h: move declarations for read-cache.c functions from cache.h
repository.h: move declaration of the_index from cache.h
merge.h: move declarations for merge.c from cache.h
diff.h: move declaration for global in diff.c from cache.h
preload-index.h: move declarations for preload-index.c from elsewhere
sparse-index.h: move declarations for sparse-index.c from cache.h
name-hash.h: move declarations for name-hash.c from cache.h
run-command.h: move declarations for run-command.c from cache.h
...
We create .pack and then .idx, we consider only packfiles that have
.idx usable (those with only .pack are not ready yet), so we should
remove .idx before removing .pack for consistency.
* ds/remove-idx-before-pack:
packfile: delete .idx files before .pack files
When reporting a problem, `git fsck` emits a message such as:
missing blob 1234abcd (:file)
However, this can be ambiguous when the problem is detected in the index
of a worktree other than the one in which `git fsck` was invoked. To
address this shortcoming, 592ec63b38 (fsck: mention file path for index
errors, 2023-02-24) enhanced the output to mention the path of the index
when the problem is detected in some other worktree:
missing blob 1234abcd (.git/worktrees/wt/index:file)
Unfortunately, the variable in fsck_index() which controls whether the
index path should be shown is misleadingly named "is_main_index" which
can be misunderstood as referring to the main worktree (i.e. the one
housing the .git/ repository) rather than to the current worktree (i.e.
the one in which `git fsck` was invoked). Avoid such potential confusion
by choosing a name more reflective of its actual purpose.
Signed-off-by: Eric Sunshine <sunshine@sunshineco.com>
Acked-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The pretty format %(describe:abbrev=<number>) tells describe to use
at least <number> digits of the oid to generate the human-readable
format of the commit-ish.
There are three things to test here:
- Check that we can describe a commit that is not tagged (that is,
for example our HEAD is at least one commit ahead of some reachable
commit which is tagged) with at least <number> digits of the oid
being used for describing it.
- Check that when using such a commit-ish, we always use at least
<number> digits of the oid to describe it.
- Check that we can describe a tag. This just gives the name of the
tag irrespective of abbrev (abbrev doesn't make sense here).
Do this, instead of the current test which only tests the last case.
Helped-by: Junio C Hamano <gitster@pobox.com>
Mentored-by: Christian Couder <christian.couder@gmail.com>
Mentored-by: Hariom Verma <hariom18599@gmail.com>
Signed-off-by: Kousik Sanagavarapu <five231003@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In 03267e8656 (commit: discard partial cache before (re-)reading it,
2022-11-08), a memory leak was plugged by discarding any partial index
before re-reading it.
The problem with this memory leak fix is that it was based on an
incomplete understanding of the logic introduced in 7168624c35 (Do not
generate full commit log message if it is not going to be used,
2007-11-28).
That logic was introduced to add a shortcut when committing without
editing the commit message interactively. A part of that logic was to
ensure that the index was read into memory:
if (!active_nr && read_cache() < 0)
die(...)
Translation to English: If the index has not yet been read, read it, and
if that fails, error out.
That logic was incorrect, though: It used `!active_nr` as an indicator
that the index was not yet read. Usually this is not a problem because
in the vast majority of instances, the index contains at least one
entry.
And it was natural to do it this way because at the time that condition
was introduced, the `index_state` structure had no explicit flag to
indicate that it was initialized: This flag was only introduced in
913e0e99b6 (unpack_trees(): protect the handcrafted in-core index from
read_cache(), 2008-08-23), but that commit did not adjust the code path
where no index file was found and a new, pristine index was initialized.
Now, when the index does not contain any entry (which is quite
common in Git's test suite because it starts quite a many repositories
from scratch), subsequent calls to `do_read_index()` will mistake the
index not to be initialized, and read it again unnecessarily.
This is a problem because after initializing the empty index e.g. the
`cache_tree` in that index could have been initialized before a
subsequent call to `do_read_index()` wants to ensure an initialized
index. And if that subsequent call mistakes the index not to have been
initialized, it would lead to leaked memory.
The correct fix for that memory leak is to adjust the condition so that
it does not mistake `active_nr == 0` to mean that the index has not yet
been read.
Using the `initialized` flag instead, we avoid that mistake, and as a
bonus we can fix a bug at the same time that was introduced by the
memory leak fix: When deleting all tracked files and then asking `git
commit -a -m ...` to commit the result, Git would internally update the
index, then discard and re-read the index undoing the update, and fail
to commit anything.
This fixes https://github.com/git-for-windows/git/issues/4462
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We are about to fix an ancient bug where `do_read_index()` pretended
that the index was not initialized when there are no index entries.
Before the `index_state` structure gained the `initialized` flag in
913e0e99b6 (unpack_trees(): protect the handcrafted in-core index from
read_cache(), 2008-08-23), that was the best we could do (even if it was
incorrect: it is totally possible to read a Git index file that contains
no index entries).
This pattern was repeated also in 998330ac2e (read-cache: look for
shared index files next to the index, too, 2021-08-26), which we fix
here by _not_ mistaking an empty base index for a missing
`sharedindex.*` file.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In 913e0e99b6 (unpack_trees(): protect the handcrafted in-core index
from read_cache(), 2008-08-23) a flag was introduced into the
`index_state` structure to indicate whether it had been initialized (or
more correctly: read and parsed).
There was one code path that was not handled, though: when the index
file does not yet exist (but the `must_exist` parameter is set to 0 to
indicate that that's okay). In this instance, Git wants to go forward
with a new, pristine Git index, almost as if the file had existed and
contained no index entries or extensions.
Since Git wants to handle this situation the same as if an "empty" Git
index file existed, let's set the `initialized` flag also in that case.
This is necessary to prepare for fixing the bug where the condition
`cache_nr == 0` is incorrectly used as an indicator that the index was
already read, and the condition `initialized != 0` needs to be used
instead.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The summary under the NAME section for git hash-object can mislead
readers to conclude that the command can only be used to create blobs,
whereas the description makes it clear that it can be used to create
objects, not just blobs. Let's clarify the one-line summary.
Further, the description for the option -t does not list out other types
that can be used when creating objects. Let's make this explicit by
listing out the different object types.
Signed-off-by: John Cai <johncai86@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
..so that the callback can use a "struct config_source" parameter
instead of "config_reader.source". "struct config_source" is internal to
config.c, so we are adding a pointer to a struct defined in config.c
into a public function signature defined in config.h, but this is okay
because this function has only ever been (and probably ever will be)
used internally by config.c.
As a result, the_reader isn't used anywhere, so "struct config_reader"
is obsolete (it was only intended to be used with the_reader). Remove
them.
Signed-off-by: Glen Choo <chooglen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Include directives are evaluated using the path of the config file. To
reduce the dependence on "config_reader.source", add a new
"key_value_info.path" member and use that instead of
"config_source.path". This allows us to remove a "struct config_reader
*" field from "struct config_include_data", which will subsequently
allow us to remove "struct config_reader" entirely.
Signed-off-by: Glen Choo <chooglen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Remove the last usage of "struct config_reader" from configsets by
copying the "kvi" arg instead of recomputing "kvi" from
config_reader.source. Since we no longer need to pass both "struct
config_reader" and "struct config_set" in a single "void *cb", remove
"struct configset_add_data" too.
Signed-off-by: Glen Choo <chooglen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Plumb "struct key_value_info" through all code paths that end in
die_bad_number(), which lets us remove the helper functions that read
analogous values from "struct config_reader". As a result, nothing reads
config_reader.config_kvi any more, so remove that too.
In config.c, this requires changing the signature of
git_configset_get_value() to 'return' "kvi" in an out parameter so that
git_configset_get_<type>() can pass it to git_config_<type>(). Only
numeric types will use "kvi", so for non-numeric types (e.g.
git_configset_get_string()), pass NULL to indicate that the out
parameter isn't needed.
Outside of config.c, config callbacks now need to pass "ctx->kvi" to any
of the git_config_<type>() functions that parse a config string into a
number type. Included is a .cocci patch to make that refactor.
The only exceptional case is builtin/config.c, where git_config_<type>()
is called outside of a config callback (namely, on user-provided input),
so config source information has never been available. In this case,
die_bad_number() defaults to a generic, but perfectly descriptive
message. Let's provide a safe, non-NULL for "kvi" anyway, but make sure
not to change the message.
Signed-off-by: Glen Choo <chooglen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There is a code path starting from trace2_def_param_fl() that eventually
calls current_config_scope(), and thus it needs to have "kvi" plumbed
through it. Additional plumbing is also needed to get "kvi" to
trace2_def_param_fl(), which gets called by two code paths:
- Through tr2_cfg_cb(), which is a config callback, so it trivially
receives "kvi" via the "struct config_context ctx" parameter.
- Through tr2_list_env_vars_fl(), which is a high level function that
lists environment variables for tracing. This has been secretly
behaving like git_config_from_parameters() (in that it parses config
from environment variables/the CLI), but does not set config source
information.
Teach tr2_list_env_vars_fl() to be well-behaved by using
kvi_from_param(), which is used elsewhere for CLI/environment
variable-based config.
As a result, current_config_scope() has no more callers, so remove it.
Signed-off-by: Glen Choo <chooglen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Pass config_context when parsing CLI config. To provide the .kvi member,
refactor out kvi_from_param() from the logic that caches CLI config in
configsets. Now that config_context and config_context.kvi is always
present when config machinery calls config callbacks, plumb "kvi" so
that we can remove all calls of current_config_scope() except for
trace2/*.c (which will be handled in a later commit), and remove all
other current_config_*() (the functions themselves and their calls).
Note that this results in .kvi containing a different, more complete
set of information than the mocked up "struct config_source" in
git_config_from_parameters().
Plumbing "kvi" reveals a few places where we've been doing the wrong
thing:
* git_config_parse_parameter() hasn't been setting config source
information, so plumb "kvi" there too.
* Several sites in builtin/config.c have been calling current_config_*()
functions outside of config callbacks (indirectly, via the
format_config() helper), which means they're reading state that isn't
set correctly:
* "git config --get-urlmatch --show-scope" iterates config to collect
values, but then attempts to display the scope after config
iteration, causing the "unknown" scope to be shown instead of the
config file's scope. It's clear that this wasn't intended: we knew
that "--get-urlmatch" couldn't show config source metadata, which is
why "--show-origin" was marked incompatible with "--get-urlmatch"
when it was introduced [1]. It was most likely a mistake that we
allowed "--show-scope" to sneak through.
Fix this by copying the "kvi" value in the collection phase so that
it can be read back later. This means that we can now support "git
config --get-urlmatch --show-origin", but that is left unchanged
for now.
* "git config --default" doesn't have config source metadata when
displaying the default value, so "--show-scope" also results in
"unknown", and "--show-origin" results in a BUG(). Fix this by
treating the default value as if it came from the command line (e.g.
like we do with "git -c" or "git config --file"), using
kvi_from_param().
[1] https://lore.kernel.org/git/20160205112001.GA13397@sigill.intra.peff.net/
Signed-off-by: Glen Choo <chooglen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Pass config_context to config_callbacks when parsing config files. To
provide the .kvi member, refactor out the configset logic that caches
"struct config_source" and "enum config_scope" as a "struct
key_value_info". Make the "enum config_scope" available to the config
file machinery by plumbing an additional arg through
git_config_from_file_with_options().
We do not exercise ctx yet because the remaining current_config_*()
callers may be used with config_with_options(), which may read config
from parameters, but parameters don't pass ctx yet.
Signed-off-by: Glen Choo <chooglen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Pass config_context to config callbacks in configset_iter(), trivially
setting the .kvi member to the cached key_value_info. Then, in config
callbacks that are only used with configsets, use the .kvi member to
replace calls to current_config_*(), and delete current_config_line()
because it has no remaining callers.
This leaves builtin/config.c and config.c as the only remaining users of
current_config_*().
Signed-off-by: Glen Choo <chooglen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
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>
These are actually used as config callbacks, so use the typedef-ed type
and make future refactors easier.
Signed-off-by: Glen Choo <chooglen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
git_color_default_config() is a shorthand for calling two other config
callbacks. There are no other non-static functions that do this and it
will complicate our refactoring of config_fn_t so inline it instead.
Signed-off-by: Glen Choo <chooglen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The status report for an in-progress cherry-pick does not show the
current commit if the cherry-pick happens as part of a series of
multiple commits:
$ git cherry-pick <commit1> <commit2>
< one of the cherry-picks fails to merge clean >
Cherry-pick currently in progress.
(run "git cherry-pick --continue" to continue)
(use "git cherry-pick --skip" to skip this patch)
(use "git cherry-pick --abort" to cancel the cherry-pick operation)
$ git status
On branch <branch>
Your branch is ahead of '<upstream>' by 1 commit.
(use "git push" to publish your local commits)
Cherry-pick currently in progress.
(run "git cherry-pick --continue" to continue)
(use "git cherry-pick --skip" to skip this patch)
(use "git cherry-pick --abort" to cancel the cherry-pick operation)
The show_cherry_pick_in_progress() function prints "Cherry-pick
currently in progress". That function does have a more verbose print
based on whether the cherry_pick_head_oid is null or not. If it is not
null, then a more helpful message including which commit is actually
being picked is displayed.
The introduction of the "Cherry-pick currently in progress" message
comes from 4a72486de9 ("fix cherry-pick/revert status after commit",
2019-04-17). This commit modified wt_status_get_state() in order to
detect that a cherry-pick was in progress even if the user has used `git
commit` in the middle of the sequence.
The check used to detect this is the call to sequencer_get_last_command.
If the sequencer indicates that the lass command was a REPLAY_PICK, then
the state->cherry_pick_in_progress is set to 1 and the
cherry_pick_head_oid is initialized to the null_oid. Similar behavior is
done for the case of REPLAY_REVERT.
It happens that this call of sequencer_get_last_command will always
report the action even if the user hasn't interrupted anything. Thus,
during a range of cherry-picks or reverts, the cherry_pick_head_oid and
revert_head_oid will always be overwritten and initialized to the null
oid.
This results in status always displaying the terse message which does
not include commit information.
Fix this by adding an additional check so that we do not re-initialize
the cherry_pick_head_oid or revert_head_oid if we have already set the
cherry_pick_in_progress or revert_in_progress bits. This ensures that
git status will display the more helpful information when its available.
Add a test case covering this behavior.
Signed-off-by: Jacob Keller <jacob.keller@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Much like with attributes files, sometimes programs would like to know
the location of configuration files at the global or system levels.
However, it isn't always clear where these may live, especially for the
system file, which may have been hard-coded at compile time or computed
dynamically based on the runtime prefix.
Since other parties cannot intuitively know how Git was compiled and
where it looks for these files, help them by providing variables that
can be queried. Because we have multiple paths for global config
values, print them in order from highest to lowest priority, and be sure
to split on newlines so that "git var -l" produces two entries for the
global value.
However, be careful not to split all values on newlines, since our
editor values could well contain such characters, and we don't want to
split them in such a case.
Note in the documentation that some values may contain multiple paths
and that callers should be prepared for that fact. This helps people
write code that will continue to work in the event we allow multiple
items elsewhere in the future.
Signed-off-by: brian m. carlson <bk2204@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Currently, there are some programs which would like to read and parse
the gitattributes files at the global or system levels. However, it's
not always obvious where these files live, especially for the system
file, which may have been hard-coded at compile time or computed
dynamically based on the runtime prefix.
It's not reasonable to expect all callers of Git to intuitively know
where the Git distributor or user has configured these locations to
be, so add some entries to allow us to determine their location. Honor
the GIT_ATTR_NOSYSTEM environment variable if one is specified. Expose
the accessor functions in a way that we can reuse them from within the
var code.
In order to make our paths consistent on Windows and also use the same
form as paths use in "git rev-parse", let's normalize the path before we
return it. This results in Windows-style paths that use slashes, which
is convenient for making our tests function in a consistent way across
platforms. Note that this requires that some of our values be freed, so
let's add a flag about whether the value needs to be freed and use it
accordingly.
Signed-off-by: brian m. carlson <bk2204@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Right now, the functions which determine the current system and global
gitattributes files are not exposed. We'd like to use them in a future
commit, but they're not ideally named. Rename them to something more
suitable as a public interface, expose them, and document them.
Signed-off-by: brian m. carlson <bk2204@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Right now, all of our values are constants whose allocation is managed
elsewhere. However, in the future, we'll have some variables whose
memory we will need to free. To keep things consistent, let's make each
of our functions allocate its own memory and make the caller responsible
for freeing it.
Signed-off-by: brian m. carlson <bk2204@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Right now, we have only two items in our variable struct. However, in
the future, we're going to add two more items. To help keep our diffs
nice and tidy and make this structure easier to read, switch to use
C99-style initializers for our data.
Signed-off-by: brian m. carlson <bk2204@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
On most Unix systems, finding a suitable shell is easy: one simply uses
"sh" with an appropriate PATH value. However, in many Windows
environments, the shell is shipped alongside Git, and it may or may not
be in PATH, even if Git is.
In such an environment, it can be very helpful to query Git for the
shell it's using, since other tools may want to use the same shell as
well. To help them out, let's add a variable, GIT_SHELL_PATH, that
points to the location of the shell.
On Unix, we know our shell must be executable to be functional, so
assume that the distributor has correctly configured their environment,
and use that as a basic test. On Git for Windows, we know that our
shell will be one of a few fixed values, all of which end in "sh" (such
as "bash"). This seems like it might be a nice test on Unix as well,
since it is customary for all shells to end in "sh", but there probably
exist such systems that don't have such a configuration, so be careful
here not to break them.
Signed-off-by: brian m. carlson <bk2204@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In line with our other helper functions for paths, let's add a function
to check whether a path is executable, and if not, print a suitable
error message. Document this function, and note that it must only be
used under the POSIXPERM prerequisite, since it doesn't otherwise work
on Windows.
Signed-off-by: brian m. carlson <bk2204@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We abstract the set of variables into a table, with a "read" callback to
provide the value of each. Each callback takes a "flag" argument, but
most callbacks don't make use of it.
This flag is a bit odd. It may be set to IDENT_STRICT, which make sense
for ident-based callbacks, but is just confusing for things like
GIT_EDITOR.
At first glance, it seems like this is just a hack to let us directly
stick the generic git_committer_info() and git_author_info() functions
into our table. And we'd be better off to wrap them with local functions
which pass IDENT_STRICT, and have our callbacks take no option at all.
But that doesn't quite work. We pass IDENT_STRICT when the caller asks
for a specific variable, but otherwise do not (so that "git var -l" does
not bail if the committer ident cannot be formed).
So we really do need to pass in the flag to each invocation, even if the
individual callback doesn't care about it. Let's mark the unused ones so
that -Wunused-parameter does not complain. And while we're here, let's
rename them so that it's clear that the flag values we get will be from
the IDENT_* set. That may prevent confusion for future readers of the
code.
Another option would be to define our own local "strict" flag for the
callbacks, and then have wrappers that translate that to IDENT_STRICT
where it matters. But that would be more boilerplate for little gain
(most functions would still ignore the "strict" flag anyway).
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: brian m. carlson <bk2204@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When iterating through config, we read config source metadata from
global values - either a "struct config_source + enum config_scope"
or a "struct key_value_info", using the current_config* functions. Prior
to the series starting from 0c60285147 (config.c: create config_reader
and the_reader, 2023-03-28), we weren't very picky about which values we
should read in which situation; we did note that both groups of values
generally shouldn't be set together, but if both were set,
current_config* preferentially reads key_value_info. When that series
added more structure, we enforced that either the former (when parsing a
config source) can be set, or the latter (when iterating a config set),
but *never* both at the same time. See 9828453ff0 (config.c: remove
current_config_kvi, 2023-03-28) and 5cdf18e7cd (config.c: remove
current_parsing_scope, 2023-03-28).
That was a good simplifying constraint that helped us reason about the
global state, but it turns out that there is at least one situation
where we need both to be set at the same time: in a blobless partial
clone where .gitmodules is missing. "git fetch" in such a repo will
start a config parse over .gitmodules (setting the config_source), and
Git will attempt to lazy-fetch it from the promisor remote. However,
when we try to read the promisor configuration, we start iterating a
config set (setting the key_value_info), and we BUG() out because that's
not allowed any more.
Teaching config_reader to gracefully handle this is somewhat
complicated, but fortunately, there are proposed changes to the config.c
machinery to get rid of this global state, and make the BUG() obsolete
[1]. We should rely on that as the eventual solution, and avoid doing
yet another refactor in the meantime.
Therefore, fix the bug by removing the BUG() check. We're reverting to
an older, less safe state, but that's generally okay since
key_value_info is always preferentially read, so we'd always read the
correct values when we iterate a config set in the middle of a config
parse (like we are here). The reverse would be wrong, but extremely
unlikely to happen since very few callers parse config without going
through a config set.
[1] https://lore.kernel.org/git/pull.1497.v3.git.git.1687290231.gitgitgadget@gmail.com
Signed-off-by: Glen Choo <chooglen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
