| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
* rs/allocate-cache-entry-individually:
cache.h: put single NUL at end of struct cache_entry
read-cache.c: allocate index entries individually
Conflicts:
read-cache.c
|
|
If you have a deleted file and a porcelain refreshes the
cache, we print:
Unstaged changes after reset:
M file
This is technically correct, in that the file is modified,
but it's friendlier to the user if we further differentiate
the case of a deleted file (especially because this output
looks a lot like "diff --name-status", which would also make
the distinction).
Similarly, we can distinguish typechanges ("T") and
intent-to-add files ("A"), both of which appear as just "M"
in the current output.
The plumbing output for all cases remains "needs update" for
historical compatibility.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
When refreshing the index, for modified (or unmerged) files we will print
"needs update" (or "needs merge") for plumbing, or line similar to the
output from "diff --name-status" for porcelain.
The variables holding which type of message to show are named after the
plumbing messages. However, as we begin to differentiate more cases at the
porcelain level (with the plumbing message staying the same), that naming
scheme will become awkward.
Instead, name the variables after which case we found (modified or
unmerged), not what we will output.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
This will enable refresh_cache to differentiate more cases
of modification (such as typechange) when telling the user
what isn't fresh.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The code to estimate the in-memory size of the index based on its on-disk
representation is subtly wrong for certain architecture-dependent struct
layouts. Instead of fixing it, replace the code to keep the index entries
in a single large block of memory and allocate each entry separately
instead. This is both simpler and more flexible, as individual entries
can now be freed. Actually using that added flexibility is left for a
later patch.
Suggested-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Rene Scharfe <rene.scharfe@lsrfire.ath.cx>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
estimate_cache_size() tries to guess how much memory is needed for the
in-memory representation of an index file. It does that by using the
file size, the number of entries and the difference of the sizes of the
on-disk and in-memory structs -- without having to check the length of
the name of each entry, which varies for each entry, but their sums are
the same no matter the representation.
Except there can be a difference. First of all, the size is really
calculated by ce_size and ondisk_ce_size based on offsetof(..., name),
not sizeof, which can be different. And entries are padded with 1 to 8
NULs at the end (after the variable name) to make their total length a
multiple of eight.
So in order to allocate enough memory to hold the index, change the
delta calculation to be based on offsetof(..., name) and round up to
the next multiple of eight.
On a 32-bit Linux, this delta was used before:
sizeof(struct cache_entry) == 72
sizeof(struct ondisk_cache_entry) == 64
---
8
The actual difference for an entry with a filename length of one was,
however (find the definitions are in cache.h):
offsetof(struct cache_entry, name) == 72
offsetof(struct ondisk_cache_entry, name) == 62
ce_size == (72 + 1 + 8) & ~7 == 80
ondisk_ce_size == (62 + 1 + 8) & ~7 == 64
---
16
So eight bytes less had been allocated for such entries. The new
formula yields the correct delta:
(72 - 62 + 7) & ~7 == 16
Reported-by: John Hsing <tsyj2007@gmail.com>
Signed-off-by: Rene Scharfe <rene.scharfe@lsrfire.ath.cx>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
* maint:
whitespace: have SP on both sides of an assignment "="
update-ref: whitespace fix
|
|
I've deliberately excluded the borrowed code in compat/nedmalloc
directory.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
* ef/maint-win-verify-path:
verify_dotfile(): do not assume '/' is the path seperator
verify_path(): simplify check at the directory boundary
verify_path: consider dos drive prefix
real_path: do not assume '/' is the path seperator
A Windows path starting with a backslash is absolute
|
|
verify_dotfile() currently assumes that the path seperator is '/', but on
Windows it can also be '\\', so use is_dir_sep() instead.
Signed-off-by: Theo Niessink <theo@taletn.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
We simply want to say "At a directory boundary, be careful with a name
that begins with a dot, forbid a name that ends with the boundary
character or has duplicated bounadry characters".
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
If someone manage to create a repo with a 'C:' entry in the
root-tree, files can be written outside of the working-dir. This
opens up a can-of-worms of exploits.
Fix it by explicitly checking for a dos drive prefix when verifying
a paht. While we're at it, make sure that paths beginning with '\' is
considered absolute as well.
Noticed-by: Theo Niessink <theo@taletn.com>
Signed-off-by: Erik Faye-Lund <kusmabite@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The "format_check" parameter tucked after the existing parameters is too
ugly an afterthought to live in any reasonable API.
Combine it with the other boolean parameter "write_object" into a single
"flags" parameter.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
* jc/index-update-if-able:
update $GIT_INDEX_FILE when there are racily clean entries
diff/status: refactor opportunistic index update
|
|
* jc/index-update-if-able:
update $GIT_INDEX_FILE when there are racily clean entries
diff/status: refactor opportunistic index update
|
|
Traditional "opportunistic index update" done by read-only "diff" and
"status" was about updating cached lstat(2) information in the index for
the next round. We missed another obvious optimization opportunity: when
there are racily clean entries that will cease to be racily clean by
updating $GIT_INDEX_FILE. Detect that case and write $GIT_INDEX_FILE out
to give it a newer timestamp.
Noticed by Lasse Makholm by stracing "git status" in a fresh checkout and
counting the number of open(2) calls.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
When we had to refresh the index internally before running diff or status,
we opportunistically updated the $GIT_INDEX_FILE so that later invocation
of git can use the lstat(2) we already did in this invocation.
Make them share a helper function to do so.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
* nd/hash-object-sanity:
Make hash-object more robust against malformed objects
Conflicts:
cache.h
|
|
* nd/struct-pathspec: (22 commits)
t6004: add pathspec globbing test for log family
t7810: overlapping pathspecs and depth limit
grep: drop pathspec_matches() in favor of tree_entry_interesting()
grep: use writable strbuf from caller for grep_tree()
grep: use match_pathspec_depth() for cache/worktree grepping
grep: convert to use struct pathspec
Convert ce_path_match() to use match_pathspec_depth()
Convert ce_path_match() to use struct pathspec
struct rev_info: convert prune_data to struct pathspec
pathspec: add match_pathspec_depth()
tree_entry_interesting(): optimize wildcard matching when base is matched
tree_entry_interesting(): support wildcard matching
tree_entry_interesting(): fix depth limit with overlapping pathspecs
tree_entry_interesting(): support depth limit
tree_entry_interesting(): refactor into separate smaller functions
diff-tree: convert base+baselen to writable strbuf
glossary: define pathspec
Move tree_entry_interesting() to tree-walk.c and export it
tree_entry_interesting(): remove dependency on struct diff_options
Convert struct diff_options to use struct pathspec
...
|
|
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Commits, trees and tags have structure. Don't let users feed git
with malformed ones. Sooner or later git will die() when
encountering them.
Note that this patch does not check semantics. A tree that points
to non-existent objects is perfectly OK (and should be so, users
may choose to add commit first, then its associated tree for example).
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
* jj/icase-directory:
Support case folding in git fast-import when core.ignorecase=true
Support case folding for git add when core.ignorecase=true
Add case insensitivity support when using git ls-files
Add case insensitivity support for directories when using git status
Case insensitivity support for .gitignore via core.ignorecase
Add string comparison functions that respect the ignore_case variable.
Makefile & configure: add a NO_FNMATCH_CASEFOLD flag
Makefile & configure: add a NO_FNMATCH flag
Conflicts:
Makefile
config.mak.in
configure.ac
fast-import.c
|
|
When MyDir/ABC/filea.txt is added to Git, the disk directory MyDir/ABC/
is renamed to mydir/aBc/, and then mydir/aBc/fileb.txt is added, the
index will contain MyDir/ABC/filea.txt and mydir/aBc/fileb.txt. Although
the earlier portions of this patch series account for those differences
in case, this patch makes the pathing consistent by folding the case of
newly added files against the first file added with that path.
In read-cache.c's add_to_index(), the index_name_exists() support used
for git status's case insensitive directory lookups is used to find the
proper directory case according to what the user already checked in.
That is, MyDir/ABC/'s case is used to alter the stored path for
fileb.txt to MyDir/ABC/fileb.txt (instead of mydir/aBc/fileb.txt).
This is especially important when cloning a repository to a case
sensitive file system. MyDir/ABC/ and mydir/aBc/ exist in the same
directory on a Windows machine, but on Linux, the files exist in two
separate directories. The update to add_to_index(), in effect, treats a
Windows file system as case sensitive by making path case consistent.
Signed-off-by: Joshua Jensen <jjensen@workspacewhiz.com>
Signed-off-by: Johannes Sixt <j6t@kdbg.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Noticed with valgrind.
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The new dircache extension CACHE_EXT_RESOLVE_UNDO, whose value is
0x52455543, is actually the ASCII sequence 'REUC', not the ASCII
sequence 'REUN'.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The rule has always been that a cache entry that is ce_uptodate(ce)
means that we already have checked the work tree entity and we know
there is no change in the work tree compared to the index, and nobody
should have to double check. Note that false ce_uptodate(ce) does not
mean it is known to be dirty---it only means we don't know if it is
clean.
There are a few codepaths (refresh-index and preload-index are among
them) that mark a cache entry as up-to-date based solely on the return
value from ie_match_stat(); this function uses lstat() to see if the
work tree entity has been touched, and for a submodule entry, if its
HEAD points at the same commit as the commit recorded in the index of
the superproject (a submodule that is not even cloned is considered
clean).
A submodule is no longer considered unmodified merely because its HEAD
matches the index of the superproject these days, in order to prevent
people from forgetting to commit in the submodule and updating the
superproject index with the new submodule commit, before commiting the
state in the superproject. However, the patch to do so didn't update
the codepath that marks cache entries up-to-date based on the updated
definition and instead worked it around by saying "we don't trust the
return value of ce_uptodate() for submodules."
This makes ce_uptodate() trustworthy again by not marking submodule
entries up-to-date.
The next step _could_ be to introduce a few "in-core" flag bits to
cache_entry structure to record "this entry is _known_ to be dirty",
call is_submodule_modified() from ie_match_stat(), and use these new
bits to avoid running this rather expensive check more than once, but
that can be a separate patch.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Exal Sibeaz pointed out that some git files are way too big, and that
add_files_to_cache() brings in all the diff machinery to any git binary
that needs the basic git SHA1 object operations from read-cache.c. Which
is pretty much all of them.
It's doubly silly, since add_files_to_cache() is only used by builtin
programs (add, checkout and commit), so it's fairly easily fixed by just
moving the thing to builtin-add.c, and avoiding the dependency entirely.
I initially argued to Exal that it would probably be best to try to depend
on smart compilers and linkers, but after spending some time trying to
make -ffunction-sections work and giving up, I think Exal was right, and
the fix is to just do some trivial cleanups like this.
This trivial cleanup results in pretty stunning file size differences.
The diff machinery really is mostly used by just the builtin programs, and
you have things like these trivial before-and-after numbers:
-rwxr-xr-x 1 torvalds torvalds 1727420 2010-01-21 10:53 git-hash-object
-rwxrwxr-x 1 torvalds torvalds 940265 2010-01-21 11:16 git-hash-object
Now, I'm not saying that 940kB is good either, but that's mostly all the
debug information - you can see the real code with 'size':
text data bss dec hex filename
418675 3920 127408 550003 86473 git-hash-object (before)
230650 2288 111728 344666 5425a git-hash-object (after)
ie we have a nice 24% size reduction from this trivial cleanup.
It's not just that one file either. I get:
[torvalds@nehalem git]$ du -s /home/torvalds/libexec/git-core
45640 /home/torvalds/libexec/git-core (before)
33508 /home/torvalds/libexec/git-core (after)
so we're talking 12MB of diskspace here.
(Of course, stripping all the binaries brings the 33MB down to 9MB, so the
whole debug information thing is still the bulk of it all, but that's a
separate issue entirely)
Now, I'm sure there are other things we should do, and changing our
compiler flags from -O2 to -Os would bring the text size down by an
additional almost 20%, but this thing Exal pointed out seems to be some
good low-hanging fruit.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
* jc/cache-unmerge:
rerere forget path: forget recorded resolution
rerere: refactor rerere logic to make it independent from I/O
rerere: remove silly 1024-byte line limit
resolve-undo: teach "update-index --unresolve" to use resolve-undo info
resolve-undo: "checkout -m path" uses resolve-undo information
resolve-undo: allow plumbing to clear the information
resolve-undo: basic tests
resolve-undo: record resolved conflicts in a new index extension section
builtin-merge.c: use standard active_cache macros
Conflicts:
builtin-ls-files.c
builtin-merge.c
builtin-rerere.c
|
|
* jc/symbol-static:
date.c: mark file-local function static
Replace parse_blob() with an explanatory comment
symlinks.c: remove unused functions
object.c: remove unused functions
strbuf.c: remove unused function
sha1_file.c: remove unused function
mailmap.c: remove unused function
utf8.c: mark file-local function static
submodule.c: mark file-local function static
quote.c: mark file-local function static
remote-curl.c: mark file-local function static
read-cache.c: mark file-local functions static
parse-options.c: mark file-local function static
entry.c: mark file-local function static
http.c: mark file-local functions static
pretty.c: mark file-local function static
builtin-rev-list.c: mark file-local function static
bisect.c: mark file-local function static
|
|
* cc/reset-more:
t7111: check that reset options work as described in the tables
Documentation: reset: add some missing tables
Fix bit assignment for CE_CONFLICTED
"reset --merge": fix unmerged case
reset: use "unpack_trees()" directly instead of "git read-tree"
reset: add a few tests for "git reset --merge"
Documentation: reset: add some tables to describe the different options
reset: improve mixed reset error message when in a bare repo
|
|
* nd/sparse: (25 commits)
t7002: test for not using external grep on skip-worktree paths
t7002: set test prerequisite "external-grep" if supported
grep: do not do external grep on skip-worktree entries
commit: correctly respect skip-worktree bit
ie_match_stat(): do not ignore skip-worktree bit with CE_MATCH_IGNORE_VALID
tests: rename duplicate t1009
sparse checkout: inhibit empty worktree
Add tests for sparse checkout
read-tree: add --no-sparse-checkout to disable sparse checkout support
unpack-trees(): ignore worktree check outside checkout area
unpack_trees(): apply $GIT_DIR/info/sparse-checkout to the final index
unpack-trees(): "enable" sparse checkout and load $GIT_DIR/info/sparse-checkout
unpack-trees.c: generalize verify_* functions
unpack-trees(): add CE_WT_REMOVE to remove on worktree alone
Introduce "sparse checkout"
dir.c: export excluded_1() and add_excludes_from_file_1()
excluded_1(): support exclude files in index
unpack-trees(): carry skip-worktree bit over in merged_entry()
Read .gitignore from index if it is skip-worktree
Avoid writing to buffer in add_excludes_from_file_1()
...
Conflicts:
.gitignore
Documentation/config.txt
Documentation/git-update-index.txt
Makefile
entry.c
t/t7002-grep.sh
|
|
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Commit 9e8ecea (Add 'merge' mode to 'git reset', 2008-12-01) disallowed
"git reset --merge" when there was unmerged entries. But it wished if
unmerged entries were reset as if --hard (instead of --merge) has been
used. This makes sense because all "mergy" operations makes sure that
any path involved in the merge does not have local modifications before
starting, so resetting such a path away won't lose any information.
The previous commit changed the behavior of --merge to accept resetting
unmerged entries if they are reset to a different state than HEAD, but it
did not reset the changes in the work tree, leaving the conflict markers
in the resulting file in the work tree.
Fix it by doing three things:
- Update the documentation to match the wish of original "reset --merge"
better, namely, "An unmerged entry is a sign that the path didn't have
any local modification and can be safely resetted to whatever the new
HEAD records";
- Update read_index_unmerged(), which reads the index file into the cache
while dropping any higher-stage entries down to stage #0, not to copy
the object name from the higher stage entry. The code used to take the
object name from the a stage entry ("base" if you happened to have
stage #1, or "ours" if both sides added, etc.), which essentially meant
that you are getting random results depending on what the merge did.
The _only_ reason we want to keep a previously unmerged entry in the
index at stage #0 is so that we don't forget the fact that we have
corresponding file in the work tree in order to be able to remove it
when the tree we are resetting to does not have the path. In order to
differentiate such an entry from ordinary cache entry, the cache entry
added by read_index_unmerged() is marked as CE_CONFLICTED.
- Update merged_entry() and deleted_entry() so that they pay attention to
cache entries marked as CE_CONFLICTED. They are previously unmerged
entries, and the files in the work tree that correspond to them are
resetted away by oneway_merge() to the version from the tree we are
resetting to.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
* nf/maint-fix-index-ext-len-on-be64:
read_index(): fix reading extension size on BE 64-bit archs
|
|
On big endian platforms with 8-byte unsigned long, the code reads the
size of the index extension section (which is a 4-byte network byte
order integer) incorrectly.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
When resolving a conflict using "git add" to create a stage #0 entry, or
"git rm" to remove entries at higher stages, remove_index_entry_at()
function is eventually called to remove unmerged (i.e. higher stage)
entries from the index. Introduce a "resolve_undo_info" structure and
keep track of the removed cache entries, and save it in a new index
extension section in the index_state.
Operations like "read-tree -m", "merge", "checkout [-m] <branch>" and
"reset" are signs that recorded information in the index is no longer
necessary. The data is removed from the index extension when operations
start; they may leave conflicted entries in the index, and later user
actions like "git add" will record their conflicted states afresh.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Previously CE_MATCH_IGNORE_VALID flag is used by both valid and
skip-worktree bits. While the two bits have similar behaviour, sharing
this flag means "git update-index --really-refresh" will ignore
skip-worktree while it should not. Instead another flag is
introduced to ignore skip-worktree bit, CE_MATCH_IGNORE_VALID only
applies to valid bit.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
grep: turn on --cached for files that is marked skip-worktree
ls-files: do not check for deleted file that is marked skip-worktree
update-index: ignore update request if it's skip-worktree, while still allows removing
diff*: skip worktree version
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
git reset without argument displays a summary of the local modification,
like this:
$ git reset
Makefile: locally modified
Some people have problems with this; they look like an error message.
This patch makes its output mimic how "git checkout $another_branch"
reports the paths with local modifications. "git add --refresh --verbose"
is changed in the same way.
It also adds a header to make it clear that the output is informative,
and not an error.
Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr>
|
|
The change in the output is going to become more general than just saying
"changed", so let's make the variable name more general too.
Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Lots of die() calls did not actually report the kind of error, which
can leave the user confused as to the real problem. Use die_errno()
where we check a system/library call that sets errno on failure, or
one of the following that wrap such calls:
Function Passes on error from
-------- --------------------
odb_pack_keep open
read_ancestry fopen
read_in_full xread
strbuf_read xread
strbuf_read_file open or strbuf_read_file
strbuf_readlink readlink
write_in_full xwrite
Signed-off-by: Thomas Rast <trast@student.ethz.ch>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Change calls to die(..., strerror(errno)) to use the new die_errno().
In the process, also make slight style adjustments: at least state
_something_ about the function that failed (instead of just printing
the pathname), and put paths in single quotes.
Signed-off-by: Thomas Rast <trast@student.ethz.ch>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Commit e1afca4fd "write_index(): update index_state->timestamp after
flushing to disk" on 2009-02-23 used stat.ctime to record the
timestamp of the index-file. This is wrong, so fix this and use the
correct stat.mtime timestamp instead.
Commit 110c46a909 "Not all systems use st_[cm]tim field for ns
resolution file timestamp" on 2009-03-08, has a similar bug for the
builtin-fetch-pack.c file.
Signed-off-by: Kjetil Barvik <barvik@broadpark.no>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Some codepaths do not still use the ST_[CM]TIME_NSEC() pair of macros
introduced by the previous commit but assumes all systems use st_mtim
and st_ctim fields in "struct stat" to record nanosecond resolution part
of the file timestamps.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Traditionally, the lack of USE_NSEC meant "do not record nor use the
nanosecond resolution part of the file timestamps". To avoid problems on
filesystems that lose the ns part when the metadata is flushed to the disk
and then later read back in, disabling USE_NSEC has been a good idea in
general.
If you are on a filesystem without such an issue, it does not hurt to read
and store them in the cached stat data in the index entries even if your
git is compiled without USE_NSEC. The index left with such a version of
git can be read by git compiled with USE_NSEC and it can make use of the
nanosecond part to optimize the check to see if the path on the filesystem
hsa been modified since we last looked at.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Since this timestamp is used to check for racy-clean files, it is
important to keep it uptodate.
For the 'git checkout' command without the '-q' option, this make a
huge difference. Before, each and every file which was updated, was
racy-clean after the call to unpack_trees() and write_index() but
before the GIT process ended.
And because of the call to show_local_changes() in builtin-checkout.c,
we ended up reading those files back into memory, doing a SHA1 to
check if the files was really different from the index. And, of
course, no file was different.
With this fix, 'git checkout' without the '-q' option should now be
almost as fast as with the '-q' option, but not quite, as we still do
some few lstat(2) calls more without the '-q' option.
Below is some average numbers for 10 checkout's to v2.6.27 and 10 to
v2.6.25 of the Linux kernel, to show the difference:
before (git version 1.6.2.rc1.256.g58a87):
7.860 user 2.427 sys 19.465 real 52.8% CPU faults: 0 major 95331 minor
after:
6.184 user 2.160 sys 17.619 real 47.4% CPU faults: 0 major 38994 minor
Signed-off-by: Kjetil Barvik <barvik@broadpark.no>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Since the filesystem ext4 is now defined as stable in Linux v2.6.28,
and ext4 supports nanonsecond resolution timestamps natively, it is
time to make USE_NSEC work as expected.
This will make racy git situations less likely to happen. For 'git
checkout' this means it will be less likely that we have to open, read
the contents of the file into RAM, and check if file is really
modified or not. The result sould be a litle less used CPU time, less
pagefaults and a litle faster program, at least for 'git checkout'.
Since the number of possible racy git situations would increase when
disks gets faster, this patch would be more and more helpfull as times
go by. For a fast Solid State Disk, this patch should be helpfull.
Note that, when file operations starts to take less than 1 nanosecond,
one would again start to get more racy git situations.
For more info on racy git, see Documentation/technical/racy-git.txt
For more info on ext4, see http://kernelnewbies.org/Ext4
Signed-off-by: Kjetil Barvik <barvik@broadpark.no>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Below is oprofile output from GIT command 'git chekcout -q my-v2.6.25'
(move from tag v2.6.27 to tag v2.6.25 of the Linux kernel):
CPU: Core 2, speed 1999.95 MHz (estimated)
Counted CPU_CLK_UNHALTED events (Clock cycles when not halted) with a unit
mask of 0x00 (Unhalted core cycles) count 20000
Counted INST_RETIRED_ANY_P events (number of instructions retired) with a
unit mask of 0x00 (No unit mask) count 20000
CPU_CLK_UNHALT...|INST_RETIRED:2...|
samples| %| samples| %|
------------------------------------
409247 100.000 342878 100.000 git
CPU_CLK_UNHALT...|INST_RETIRED:2...|
samples| %| samples| %|
------------------------------------
260476 63.6476 257843 75.1996 libz.so.1.2.3
100876 24.6492 64378 18.7758 kernel-2.6.28.4_2.vmlinux
30850 7.5382 7874 2.2964 libc-2.9.so
14775 3.6103 8390 2.4469 git
2020 0.4936 4325 1.2614 libcrypto.so.0.9.8
191 0.0467 32 0.0093 libpthread-2.9.so
58 0.0142 36 0.0105 ld-2.9.so
1 2.4e-04 0 0 libldap-2.3.so.0.2.31
Detail list of the top 20 function entries (libz counted in one blob):
CPU_CLK_UNHALTED INST_RETIRED_ANY_P
samples % samples % image name symbol name
260476 63.6862 257843 75.2725 libz.so.1.2.3 /lib/libz.so.1.2.3
16587 4.0555 3636 1.0615 libc-2.9.so memcpy
7710 1.8851 277 0.0809 libc-2.9.so memmove
3679 0.8995 1108 0.3235 kernel-2.6.28.4_2.vmlinux d_validate
3546 0.8670 2607 0.7611 kernel-2.6.28.4_2.vmlinux __getblk
3174 0.7760 1813 0.5293 libc-2.9.so _int_malloc
2396 0.5858 3681 1.0746 kernel-2.6.28.4_2.vmlinux copy_to_user
2270 0.5550 2528 0.7380 kernel-2.6.28.4_2.vmlinux __link_path_walk
2205 0.5391 1797 0.5246 kernel-2.6.28.4_2.vmlinux ext4_mark_iloc_dirty
2103 0.5142 1203 0.3512 kernel-2.6.28.4_2.vmlinux find_first_zero_bit
2077 0.5078 997 0.2911 kernel-2.6.28.4_2.vmlinux do_get_write_access
2070 0.5061 514 0.1501 git cache_name_compare
2043 0.4995 1501 0.4382 kernel-2.6.28.4_2.vmlinux rcu_irq_exit
2022 0.4944 1732 0.5056 kernel-2.6.28.4_2.vmlinux __ext4_get_inode_loc
2020 0.4939 4325 1.2626 libcrypto.so.0.9.8 /usr/lib/libcrypto.so.0.9.8
1965 0.4804 1384 0.4040 git patch_delta
1708 0.4176 984 0.2873 kernel-2.6.28.4_2.vmlinux rcu_sched_grace_period
1682 0.4112 727 0.2122 kernel-2.6.28.4_2.vmlinux sysfs_slab_alias
1659 0.4056 290 0.0847 git find_pack_entry_one
1480 0.3619 1307 0.3816 kernel-2.6.28.4_2.vmlinux ext4_writepage_trans_blocks
Notice the memmove line, where the CPU did 7710 / 277 = 27.8 cycles
per instruction, and compared to the total cycles spent inside the
source code of GIT for this command, all the memmove() calls
translates to (7710 * 100) / 14775 = 52.2% of this.
Retesting with a GIT program compiled for gcov usage, I found out that
the memmove() calls came from remove_index_entry_at() in read-cache.c,
where we have:
memmove(istate->cache + pos,
istate->cache + pos + 1,
(istate->cache_nr - pos) * sizeof(struct cache_entry *));
remove_index_entry_at() is called 4902 times from check_updates() in
unpack-trees.c, and each time called we move each cache_entry pointers
(from the removed one) one step to the left.
Since we have 28828 entries in the cache this time, and if we on
average move half of them each time, we in total move approximately
4902 * 0.5 * 28828 * 4 = 282 629 712 bytes, or twice this amount if
each pointer is 8 bytes (64 bit).
OK, is seems that the function check_updates() is called 28 times, so
the estimated guess above had been more correct if check_updates() had
been called only once, but the point is: we get lots of bytes moved.
To fix this, and use an O(N) algorithm instead, where N is the number
of cache_entries, we delete/remove all entries in one loop through all
entries.
From a retest, the new remove_marked_cache_entries() from the patch
below, ended up with the following output line from oprofile:
46 0.0105 15 0.0041 git remove_marked_cache_entries
If we can trust the numbers from oprofile in this case, we saved
approximately ((7710 - 46) * 20000) / (2 * 1000 * 1000 * 1000) = 0.077
seconds CPU time with this fix for this particular test. And notice
that now the CPU did only 46 / 15 = 3.1 cycles/instruction.
Signed-off-by: Kjetil Barvik <barvik@broadpark.no>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
Junio C Hamano
Jeff King
René Scharfe
Theo Niessink
Erik Faye-Lund
Jonathan Nieder
Nguyễn Thái Ngọc Duy
Joshua Jensen
Shawn O. Pearce
Linus Torvalds
Nathaniel W Filardo
Matthieu Moy
Thomas Rast
Kjetil Barvik