Prime example of where the raw tree parser is easier for everybody.
[jc: "Aieee" one-liner fix from the list applied. ]
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Instead, just use the tree buffer directly, and use the tree-walk
infrastructure to walk the buffers instead of the tree-entry list.
The tree-entry list is inefficient, and generates tons of small
allocations for no good reason. The tree-walk infrastructure is
generally no harder to use than following a linked list, and allows
us to do most tree parsing in-place.
Some programs still use the old tree-entry lists, and are a bit
painful to convert without major surgery. For them we have a helper
function that creates a temporary tree-entry list on demand.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This is preparatory work for further cleanups, where we try to make
tree_entry look more like the more efficient tree-walk descriptor.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This allows us to avoid allocating information for names etc, because
we can just use the information from the tree buffer directly.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
There are times when gitk needs to know that the commits it has sent
to git-diff-tree --stdin did not match, and it needs to know in a
timely fashion even if none of them match. At the moment,
git-diff-tree outputs nothing for non-matching commits, so it is
impossible for gitk to distinguish between git-diff-tree being slow
and git-diff-tree saying no.
This makes git-diff-tree flush its output and echo back the
input line when it is not a valid-looking object name. Gitk, or
other users of git-diff-tree --stdin, can use a blank line or
any other "marker line" to indicate that git-diff-tree has
processed all the commits on its input up to the echoed back
marker line, and any commits that have not been output do not
match.
[jc: re-done after a couple of back-and-forth discussion on the list.]
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Moved the setup commands into test_expect_success blocks so their
output is hidden unless -v is used. This makes the test suite look
a little cleaner when the rm test-file setup step fails (and was
expected to fail for most cases).
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
* jc/cache-tree: (26 commits)
builtin-rm: squelch compiler warnings.
git-write-tree writes garbage on sparc64
Fix crash when reading the empty tree
fsck-objects: do not segfault on missing tree in cache-tree
cache-tree: a bit more debugging support.
read-tree: invalidate cache-tree entry when a new index entry is added.
Fix test-dump-cache-tree in one-tree disappeared case.
fsck-objects: mark objects reachable from cache-tree
cache-tree: replace a sscanf() by two strtol() calls
cache-tree.c: typefix
test-dump-cache-tree: validate the cached data as well.
cache_tree_update: give an option to update cache-tree only.
read-tree: teach 1-way merege and plain read to prime cache-tree.
read-tree: teach 1 and 2 way merges about cache-tree.
update-index: when --unresolve, smudge the relevant cache-tree entries.
test-dump-cache-tree: report number of subtrees.
cache-tree: sort the subtree entries.
Teach fsck-objects about cache-tree.
index: make the index file format extensible.
cache-tree: protect against "git prune".
...
Conflicts:
Makefile, builtin.h, git.c: resolved the same way as in next.
* jc/dirwalk-n-cache-tree: (212 commits)
builtin-rm: squelch compiler warnings.
Add builtin "git rm" command
Move pathspec matching from builtin-add.c into dir.c
Prevent bogus paths from being added to the index.
builtin-add: fix unmatched pathspec warnings.
Remove old "git-add.sh" remnants
builtin-add: warn on unmatched pathspecs
Do "git add" as a builtin
Clean up git-ls-file directory walking library interface
libify git-ls-files directory traversal
Add a conversion tool to migrate remote information into the config
fetch, pull: ask config for remote information
Fix build procedure for builtin-init-db
read-tree -m -u: do not overwrite or remove untracked working tree files.
apply --cached: do not check newly added file in the working tree
Implement a --dry-run option to git-quiltimport
Implement git-quiltimport
Revert "builtin-grep: workaround for non GNU grep."
builtin-grep: workaround for non GNU grep.
builtin-grep: workaround for non GNU grep.
...
Doing an oprofile run on the result of my git rev-list memory leak fixes
and tree parsing cleanups, I was surprised by the third-highest entry
being
samples % image name app name symbol name
179751 2.7163 libc-2.4.so libc-2.4.so _IO_vfscanf@@GLIBC_2.4
where that 2.7% is actually more than 5% of one CPU, because this was run
on a dual CPU setup with the other CPU just being idle.
That seems to all be from the use of 'sscanf(tree, "%o", &mode)' for the
tree buffer parsing.
So do the trivial octal parsing by hand, which also gives us where the
first space in the string is (and thus where the pathname starts) so we
can get rid of the "strchr(tree, ' ')" call too.
This brings the "git rev-list --all --objects" time down from 63 seconds
to 55 seconds on the historical kernel archive for me, so it's quite
noticeable - tree parsing is a lot of what we end up doing when following
all the objects.
[ I also see a 5% speedup on a full "git fsck-objects" on the current
kernel archive, so that sscanf() really does seem to have hurt our
performance by a surprising amount ]
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
NetBSD ash chokes on the optional open parenthesis for case arms. Inside
$(command substitution), however, bash barfs without. So adjust things
accordingly.
Originally pointed out by Dennis Stosberg.
Signed-off-by: Junio C Hamano <junkio@cox.net>
* eb/mailinfo:
mailinfo: More carefully parse header lines in read_one_header_line()
Allow in body headers beyond the in body header prefix.
More accurately detect header lines in read_one_header_line
In handle_body only read a line if we don't already have one.
Refactor commit messge handling.
Move B and Q decoding into check header.
Make read_one_header_line return a flag not a length.
In the "next" branch, write_index_ext_header() writes garbage on a
64-bit big-endian machine; the written index file will be unreadable.
I noticed this on NetBSD/sparc64. Reproducible with:
$ git init-db
$ :>file
$ git-update-index --add file
$ git-write-tree
$ git-update-index
error: index uses extension, which we do not understand
fatal: index file corrupt
Signed-off-by: Dennis Stosberg <dennis@stosberg.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Martin Langhoff points out that "git repack -a" ends up using up a lot of
memory for big archives, and that git cvsimport probably should do only
incremental repacks in order to avoid having repacking flush all the
caches.
The big majority of the memory usage of repacking is from git rev-list
tracking all objects, and this patch should go a long way in avoiding the
excessive memory usage: the bulk of it was due to the object names being
leaked from the tree parser.
For the historic Linux kernel archive, this simple patch does:
Before:
/usr/bin/time git-rev-list --all --objects > /dev/null
72.45user 0.82system 1:13.55elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+125376minor)pagefaults 0swaps
After:
/usr/bin/time git-rev-list --all --objects > /dev/null
75.22user 0.48system 1:16.34elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+43921minor)pagefaults 0swaps
where we do end up wasting a bit of time on some extra strdup()s (which
could be avoided, but that would require tracking where the pathnames came
from), but we avoid a lot of memory usage.
Minor page faults track maximum RSS very closely (each page fault maps in
one page into memory), so the reduction from 125376 page faults to 43921
means a rough reduction of VM footprint from almost half a gigabyte to
about a third of that. Those numbers were also double-checked by looking
at "top" while the process was running.
(Side note: at least part of the remaining VM footprint is the mapping of
the 177MB pack-file, so the remaining memory use is at least partly "well
behaved" from a project caching perspective).
For the current git archive itself, the memory usage for a "--all
--objects" rev-list invocation dropped from 7128 pages to 2318 (27MB to
9MB), so the reduction seems to hold for much smaller projects too.
For regular "git-rev-list" usage (ie without the "--objects" flag) this
patch has no impact.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This is a quick port of my initial patch for 1.0.7, that I had forgotten to
post. Possibly needs some testing before applying to master.
Signed-off-by: Yann Dirson <ydirson@altern.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Since there is no bugtracker that I know of, let's just use the scripts
themselves to document their limitations.
Signed-off-by: Yann Dirson <ydirson@altern.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
I cannot find when that flag was removed if it ever existed, I can find
nothing about it in the ChangeLog and NEWS file of GNU diff. The current
flag is -s aka --quiet aka --silent, so let's use -s, assuming it is a
portable flag. Feel free to lart me with a POSIX bible if needed.
Signed-off-by: Yann Dirson <ydirson@altern.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
The revision argument parsing was happily parsing "--abbrev", but it
didn't parse "--abbrev=<n>".
Which was hidden by the fact that the diff options _would_ parse
--abbrev=<n>, so it would actually silently parse it, it just
wouldn't use it for the same things that a plain "--abbrev" was
used for.
Which seems a bit insane.
With this patch, if you do "git log --abbrev=10" it will abbreviate the
merge parent commit ID's to ten hex characters, which was probably what
you expected.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Using "-U0" is definitely more portable than using "--unified=0",
so we should do that regardless.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
- The --start-number handling introduced breakage in the normal
code path. It started numbering at 0 when not --numbered,
for example.
- When generating one file per patch, we needlessly added an
extra blank line in front for second and subsequent files.
Signed-off-by: Junio C Hamano <junkio@cox.net>
New users can be irritated by the git status text in their editor.
Let's give them a short help.
Signed-off-by: Martin Waitz <tali@admingilde.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
We exited prematurely from header parsing loop when the header
field did not have a space after the colon but we insisted on
it, and we got the check wrong because we forgot that we strip
the trailing whitespace before we do the check.
The space after the colon is not even required by RFC2822, so
stop requiring it. While we are at it, the header line is
specified to be more strict than "anything with a colon in it"
(there must be one or more characters before the colon, and they
must not be controls, SP or non US-ASCII), so implement that
check as well, lest we mistakenly think something like:
Bogus not a header line: this is not.
as a header line.
Signed-off-by: Junio C Hamano <junkio@cox.net>
Since the "a..b c..d" syntax is interpreted as "b ^a d ^c" as other
range-ish commands, if you want to format a..b and then c..d and end
up with files consecutively numbered, the second run needs to be able
to tell the command what number to start from.
This does not imply --numbered (which gives [PATCH n/m] to the subject).
Signed-off-by: Junio C Hamano <junkio@cox.net>
Otherwise, if make is suspended, or killed with prejudice, or if the
system crashes, you could be left with an up-to-date, yet corrupt,
generated file.
I left off the `clean' addition, because I believe "make clean" should
not remove wildcard patterns like "*+", on the off-chance that someone
uses names like that for files they care about. Besides, in practice,
those temporary files are left behind so rarely that they're not a bother,
and they're removed again as part of the next build.
[jc: sign-off?]
Signed-off-by: Junio C Hamano <junkio@cox.net>
I was just testing that "git ls-remote" change by Junio, and when you're
not in a git repository, it gives this totally bogus warning. The _target_
obviously has to be a git repository, but there's no reason why you'd have
to be in a local git repo when doing an ls-remote.
The reason is commit 73136b2e8a by Dscho: it
adds calls to git-repo-config in git-parse-remote.sh to get the remote
shorthands etc.
Now, either we should just hide and ignore the error from git-repo-config
(probably bad, because some errors _are_ valid - like git-repo-config
failing due to bad syntax in the config file), or we should just make
git-repo-config quietly handle the case of not being in a git repository.
This does the latter: just quietly accepting (and doing nothing - trying
to set a value will result in the lock-file failing) our lot in life
sounds better than dying with a bogus error message.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Acked-By: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
As both DESTDIR and the prefix are supposed to be absolute pathnames
they can simply be concatenated without an extra / (like in the main Makefile).
The extra slash may even break installation on Windows.
[jc: adjusted an earlier workaround for this problem in the dist-doc
target in the main Makefile as well. ]
Signed-off-by: Martin Waitz <tali@admingilde.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Avoid "use POSIX qw(strftime dup2 :errno_h)"; it was reported
that a Perl installations on Mandrake 9.1 did not like it, even
though it understood "use POSIX qw(:errno_h)". Funny.
Signed-off-by: Junio C Hamano <junkio@cox.net>
Funnily enough, this variable was never assigned ever since it
was introduced, and has been protecting some code that has never
been executed.
Signed-off-by: Junio C Hamano <junkio@cox.net>
* master: (40 commits)
Clean up sha1 file writing
Builtin git-cat-file
builtin format-patch: squelch content-type for 7-bit ASCII
CMIT_FMT_EMAIL: Q-encode Subject: and display-name part of From: fields.
add more informative error messages to git-mktag
remove the artificial restriction tagsize < 8kb
git-rebase: use canonical A..B syntax to format-patch
git-format-patch: now built-in.
fmt-patch: Support --attach
fmt-patch: understand old <his> notation
Teach fmt-patch about --keep-subject
Teach fmt-patch about --numbered
fmt-patch: implement -o <dir>
fmt-patch: output file names to stdout
Teach fmt-patch to write individual files.
Use RFC2822 dates from "git fmt-patch".
git-fmt-patch: thinkofix to show [PATCH] properly.
rename internal format-patch wip
Minor tweak on subject line in --pretty=email
Tentative built-in format-patch.
...
This cleans up and future-proofs the sha1 file writing in sha1_file.c.
In particular, instead of doing a simple "write()" call and just verifying
that it succeeds (or - as in one place - just assuming it does), it uses
"write_buffer()" to write data to the file descriptor while correctly
checking for partial writes, EINTR etc.
It also splits up write_sha1_to_fd() to be a lot more readable: if we need
to re-create the compressed object, we do so in a separate helper
function, making the logic a whole lot more modular and obvious.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
When there is no leading context, the patch must match at the
beginning of preimage; otherwise there is a "patch adds these
lines while the other lines were added to the original file"
conflict.
This is the opposite of match_end fix earlier in this series.
Unlike matching at the end case, we can additionally check the
preimage line number recorded in the patch, so the change is not
symmetrical with the earlier one.
Signed-off-by: Junio C Hamano <junkio@cox.net>
This makes "git format-patch" a built-in.
* js/fmt-patch:
git-rebase: use canonical A..B syntax to format-patch
git-format-patch: now built-in.
fmt-patch: Support --attach
fmt-patch: understand old <his> notation
Teach fmt-patch about --keep-subject
Teach fmt-patch about --numbered
fmt-patch: implement -o <dir>
fmt-patch: output file names to stdout
Teach fmt-patch to write individual files.
Use RFC2822 dates from "git fmt-patch".
git-fmt-patch: thinkofix to show [PATCH] properly.
rename internal format-patch wip
Minor tweak on subject line in --pretty=email
Tentative built-in format-patch.
This makes 'git add' and 'git rm' built-ins.
* lt/dirwalk:
Add builtin "git rm" command
Move pathspec matching from builtin-add.c into dir.c
Prevent bogus paths from being added to the index.
builtin-add: fix unmatched pathspec warnings.
Remove old "git-add.sh" remnants
builtin-add: warn on unmatched pathspecs
Do "git add" as a builtin
Clean up git-ls-file directory walking library interface
libify git-ls-files directory traversal
git-apply adding an ending line doesn't seem to fail if the ending line is
already present in the patched file.
Signed-off-by: Catalin Marinas <catalin.marinas@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>