Currently we treat "*.c" and "path/to/*.c" the same way. Which means
we check all possible paths in repo against "path/to/*.c". One could
see that "path/elsewhere/foo.c" obviously cannot match "path/to/*.c"
and we only need to check all paths _inside_ "path/to/" against that
pattern.
This patch checks the leading fixed part of a pathspec against base
directory and exit early if possible. We could even optimize further
in "path/to/something*.c" case (i.e. check the fixed part against
name_entry as well) but that's more complicated and probably does not
gain us much.
-O2 build on linux-2.6, without and with this patch respectively:
$ time git rev-list --quiet HEAD -- 'drivers/*.c'
real 1m9.484s
user 1m9.128s
sys 0m0.181s
$ time ~/w/git/git rev-list --quiet HEAD -- 'drivers/*.c'
real 0m15.710s
user 0m15.564s
sys 0m0.107s
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When a pattern contains only a single asterisk as wildcard,
e.g. "foo*bar", after literally comparing the leading part "foo" with
the string, we can compare the tail of the string and make sure it
matches "bar", instead of running fnmatch() on "*bar" against the
remainder of the string.
-O2 build on linux-2.6, without the patch:
$ time git rev-list --quiet HEAD -- '*.c'
real 0m40.770s
user 0m40.290s
sys 0m0.256s
With the patch
$ time ~/w/git/git rev-list --quiet HEAD -- '*.c'
real 0m34.288s
user 0m33.997s
sys 0m0.205s
The above command is not supposed to be widely popular. It's chosen
because it exercises pathspec matching a lot. The point is it cuts
down matching time for popular patterns like *.c, which could be used
as pathspec in other places.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We mark pathspec with wildcards with the field use_wildcard. We
could do better by saving the length of the non-wildcard part, which
can be used for optimizations such as f9f6e2c (exclude: do strcmp as
much as possible before fnmatch - 2012-06-07).
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Commit d688cf0 (tree_entry_interesting(): give meaningful names to
return values - 2011-10-24) converts most of the tree_entry_interesting
values to the new enum, except "never_interesting". This completes the
conversion.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
It's actually unlimited recursion if wildcards are active regardless
--max-depth
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
It is a basic code hygiene to avoid magic constants that are unnamed.
Besides, this helps extending the value later on for "interesting, but
cannot decide if the entry truely matches yet" (ie. prefix matches)
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We know we will find nothing.
This incidentally squelches false warning from gcc about potentially
uninitialized usage of t.entry fields. For an empty tree, it is true that
init_tree_desc() does not call decode_tree_entry() and the tree_desc is
left uninitialized, but find_tree_entry() only calls tree_entry_extract()
that uses the tree_desc while it has more things to read from the tree, so
the uninitialized t.entry fields are never used in such a case anyway.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
tree_entry_len() does not simply take two random arguments and return
a tree length. The two pointers must point to a tree item structure,
or struct name_entry. Passing random pointers will return incorrect
value.
Force callers to pass struct name_entry instead of two pointers (with
hope that they don't manually construct struct name_entry themselves)
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In particular, gcc complains as follows:
CC tree-walk.o
tree-walk.c: In function `traverse_trees':
tree-walk.c:347: warning: 'e' might be used uninitialized in this \
function
CC builtin/revert.o
builtin/revert.c: In function `verify_opt_mutually_compatible':
builtin/revert.c:113: warning: 'opt2' might be used uninitialized in \
this function
Signed-off-by: Ramsay Jones <ramsay@ramsay1.demon.co.uk>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In the case of a wide breadth top-level tree (~2400 entries, all trees
in this case), we can see a noticeable cost in the profiler calling
strncmp() here. Most of the time we are at the base level of the
repository, so base is "" and baselen == 0, which means we will always
test true. Break out this one tiny case so we can short circuit the
strncmp() call.
Test cases are as follows. packages.git is the Arch Linux git-svn clone
of the packages repository which has the characteristics above.
Commands:
[1] packages.git, /usr/bin/time git log >/dev/null
[2] packages.git, /usr/bin/time git log -- autogen/trunk pacman/trunk wget/trunk >/dev/null
[3] linux.git, /usr/bin/time git log >/dev/null
[4] linux.git, /usr/bin/time git log -- drivers/ata drivers/uio tools >/dev/null
Results:
before after %faster
[1] 2.56 2.55 0.4%
[2] 51.82 48.66 6.5%
[3] 5.58 5.61 -0.5%
[4] 1.55 1.51 0.2%
The takeaway here is this doesn't matter in many operations, but it does
for a certain style of repository and operation where it nets a 6.5%
measured improvement. The other changes are likely not significant by
reasonable statistics methods.
Note: the measured improvement when originally submitted was ~11% (43 to
38 secs) for operation [2]. At the time, the repository had 117220
commits; it now has 137537 commits.
Signed-off-by: Dan McGee <dpmcgee@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The traverse_trees() machinery is primarily meant for merging two (or
more) trees, and because a merge is a full tree operation, it doesn't
support any pruning with pathspec.
Since d1f2d7e (Make run_diff_index() use unpack_trees(), not read_tree(),
2008-01-19), however, we use unpack_trees() to traverse_trees() callchain
to perform "diff-index", which could waste a lot of work traversing trees
outside the user-supplied pathspec, only to discard at the blob comparison
level in diff-lib.c::oneway_diff() which is way too late.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
As the point of the last change is to allow use of strings as
literals no matter what characters are in them, "has_wildcard"
does not match what we use this field for anymore.
It is used to decide if the wildcard matching should be used, so
rename it to match the usage better.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If base is already matched, skip that part when calling
fnmatch(). This happens quite often if users start a command from
worktree's subdirectory and prefix is usually prepended to all
pathspecs.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
never_interesting optimization is disabled if there is any wildcard
pathspec, even if it only matches exactly on trees.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Suppose we have two pathspecs 'a' and 'a/b' (both are dirs) and depth
limit 1. In current code, pathspecs are checked in input order. When
'a/b' is checked against pathspec 'a', it fails depth limit and
therefore is excluded, although it should match 'a/b' pathspec.
This patch reorders all pathspecs alphabetically, then teaches
tree_entry_interesting() to check against the deepest pathspec first,
so depth limit of a shallower pathspec won't affect a deeper one.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This is needed to replace pathspec_matches() in builtin/grep.c.
max_depth == -1 means infinite depth. Depth limit is only effective
when pathspec.recursive == 1. When pathspec.recursive == 0, the
behavior depends on match functions: non-recursive for
tree_entry_interesting() and recursive for match_pathspec{,_depth}
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In traversing trees, a full path is splitted into two parts: base
directory and entry. They are however quite often concatenated
whenever a full path is needed. Current code allocates a new buffer,
do two memcpy(), use it, then release.
Instead this patch turns "base" to a writable, extendable buffer. When
a concatenation is needed, the callee only needs to append "entry" to
base, use it, then truncate the entry out again. "base" must remain
unchanged before and after entering a function.
This avoids quite a bit of malloc() and memcpy().
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When an error is encountered, it calls add_rejected_file() which either
- directly displays the error message and stops if in plumbing mode
(i.e. if show_all_errors is not initialized at 1)
- or stores it so that it will be displayed at the end with display_error_msgs(),
Storing the files by error type permits to have a list of files for
which there is the same error instead of having a serie of almost
identical errors.
As each bind_overlap error combines a file and an old file, a list cannot be
done, therefore, theses errors are not stored but directly displayed.
Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
traverse_trees() is supposed to call its callback with all the matching
entries from the given trees. The current algorithm keeps a pointer to
each of the tree being traversed, and feeds the entry with the earliest
name to the callback.
This breaks down if the trees being traversed looks like this:
A B
t-1 t
t-2 u
t/a v
When we are currently looking at an entry "t-1" in tree A, and tree B has
returned "t", feeding "t" from the B and not feeding anything from A, only
because "t-1" sorts later than "t", will miss an entry for a subtree "t"
behind the current entry in tree A.
This introduces extended_entry_extract() helper function that gives what
name is expected from the tree, and implements a mechanism to look-ahead
in the tree object using it, to make sure such a case is handled sanely.
Traversal in tree A in the above example will first return "t" to match
that of B, and then the next request for an entry to A then returns "t-1".
This roughly corresponds to what Linus's "prepare for one-entry lookahead"
wanted to do, but because this does implement look ahead, t6035 and one more
test in t1012 reveal that the approach would not work without adjusting the
side that walks the index in unpack_trees() as well.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When we make the "root" tree-walk info entry have a pathname in it, we
need to have a ->prev pointer so that compare_entry will actually notice
and traverse into the root.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This makes the traverse_trees() entry comparator routine use the more
relaxed form of name comparison that considers files and directories
with the same name identical.
We pass in a separate mask for just the directory entries, so that the
callback routine can decide (if it wants to) to only handle one or the
other type, but generally most (all?) users are expected to really want
to see the case of a name 'foo' showing up in one tree as a file and in
another as a directory at the same time.
In particular, moving 'unpack_trees()' over to use this tree traversal
mechanism requires this.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This allows the callback to return an error value, but it can also
specify which of the tree entries that it actually used up by returning
a positive mask value.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This makes the calling convention a bit less obvious, but a lot more
flexible. Instead of allocating and extending a new 'base' string, we
just link the top-most name into a linked list of the 'info' structure
when traversing a subdirectory, and we can generate the basename by
following the list.
Perhaps even more importantly, the linked list of info structures also
gives us a place to naturally save off other information than just the
directory name.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The current code can access memory outside of the tree buffer in the
case of malformed tree entries.
This patch prevents this by:
* The rest of the buffer must be at least 24 bytes (at least 1 byte
mode, 1 blank, at least one byte path name, 1 NUL, 20 bytes sha1).
* Check that the last NUL (21 bytes before the end) is present.
This ensures that strlen() and get_mode() calls stay within the
buffer.
* The mode may not be empty. We have only to reject a blank at the
begin, as the rest is handled by if (c < '0' || c > '7').
* The blank is ensured by get_mode().
* The path must contain at least one character.
Signed-off-by: Martin Koegler <mkoegler@auto.tuwien.ac.at>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This uses "git-apply --whitespace=strip" to fix whitespace errors that have
crept in to our source files over time. There are a few files that need
to have trailing whitespaces (most notably, test vectors). The results
still passes the test, and build result in Documentation/ area is unchanged.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This makes the tree descriptor contain a "struct name_entry" as part of
it, and it gets filled in so that it always contains a valid entry. On
some benchmarks, it improves performance by up to 15%.
That makes tree entry "extract" trivial, and means that we only actually
need to decode each tree entry just once: we decode the first one when
we initialize the tree descriptor, and each subsequent one when doing
"update_tree_entry()". In particular, this means that we don't need to
do strlen() both at extract time _and_ at update time.
Finally, it also allows more sharing of code (entry_extract(), that
wanted a "struct name_entry", just got totally trivial, along with the
"tree_entry()" function).
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This removes slightly more lines than it adds, but the real reason for
doing this is that future optimizations will require more setup of the
tree descriptor, and so we want to do it in one place.
Also renamed the "desc.buf" field to "desc.buffer" just to trigger
compiler errors for old-style manual initializations, making sure I
didn't miss anything.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Since we have the "tree_entry_len()" helper function these days, and
don't need to do a full strlen(), there's no point in saving the path
length - it's just redundant information.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This is a micro-optimization that grew out of the mailing list discussion
about "strlen()" showing up in profiles.
We used to pass regular C strings around to the low-level tree walking
routines, and while this worked fine, it meant that we needed to call
strlen() on strings that the caller always actually knew the size of
anyway.
So pass the length of the string down wih the string, and avoid
unnecessary calls to strlen(). Also, when extracting a pathname from a
tree entry, use "tree_entry_len()" instead of strlen(), since the length
of the pathname is directly calculable from the decoded tree entry itself
without having to actually do another strlen().
This shaves off another ~5-10% from some loads that are very tree
intensive (notably doing commit filtering by a pathspec).
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>"
Signed-off-by: Junio C Hamano <junkio@cox.net>
This means that
git show HEAD:
will now return HEAD^{tree}, which is logically consistent with
git show HEAD:Documentation
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
The variable named entry is allocated using malloc() and then
forgotten, it being shadowed by an automatic variable of the
same name. Fixing the array size at 3 worked so far because
the only caller of traverse_trees() needed only as much
entries. Simply remove the shadowing varaible and we're able
to traverse more than three trees and save stack space at the
same time!
Signed-off-by: Rene Scharfe <rene.scharfe@lsrfire.ath.cx>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This abstracts away the size of the hash values when copying them
from memory location to memory location, much as the introduction
of hashcmp abstracted away hash value comparsion.
A few call sites were using char* rather than unsigned char* so
I added the cast rather than open hashcpy to be void*. This is a
reasonable tradeoff as most call sites already use unsigned char*
and the existing hashcmp is also declared to be unsigned char*.
[jc: Splitted the patch to "master" part, to be followed by a
patch for merge-recursive.c which is not in "master" yet.
Fixed the cast in the latter hunk to combine-diff.c which was
wrong in the original.
Also converted ones left-over in combine-diff.c, diff-lib.c and
upload-pack.c ]
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
ANSI C99 doesn't allow void-pointer arithmetic. This patch fixes this in
various ways. Usually the strategy that required the least changes was used.
Signed-off-by: Florian Forster <octo@verplant.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This adds a "tree_entry()" function that combines the common operation of
doing a "tree_entry_extract()" + "update_tree_entry()".
It also has a simplified calling convention, designed for simple loops
that traverse over a whole tree: the arguments are pointers to the tree
descriptor and a name_entry structure to fill in, and it returns a boolean
"true" if there was an entry left to be gotten in the tree.
This allows tree traversal with
struct tree_desc desc;
struct name_entry entry;
desc.buf = tree->buffer;
desc.size = tree->size;
while (tree_entry(&desc, &entry) {
... use "entry.{path, sha1, mode, pathlen}" ...
}
which is not only shorter than writing it out in full, it's hopefully less
error prone too.
[ It's actually a tad faster too - we don't need to recalculate the entry
pathlength in both extract and update, but need to do it only once.
Also, some callers can avoid doing a "strlen()" on the result, since
it's returned as part of the name_entry structure.
However, by now we're talking just 1% speedup on "git-rev-list --objects
--all", and we're definitely at the point where tree walking is no
longer the issue any more. ]
NOTE! Not everybody wants to use this new helper function, since some of
the tree walkers very much on purpose do the descriptor update separately
from the entry extraction. So the "extract + update" sequence still
remains as the core sequence, this is just a simplified interface.
We should probably add a silly two-line inline helper function for
initializing the descriptor from the "struct tree" too, just to cut down
on the noise from that common "desc" initializer.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
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>
This replaces occurences of "blob", "commit", "tag", and "tree",
where they're really used as type specifiers, which we already
have defined global constants for.
Signed-off-by: Peter Eriksen <s022018@student.dtu.dk>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Introduce tree-walk.[ch] and move "struct tree_desc" and
associated functions from various places.
Rename DIFF_FILE_CANON_MODE(mode) macro to canon_mode(mode) and
move it to cache.h. This macro returns the canonicalized
st_mode value in the host byte order for files, symlinks and
directories -- to be compared with a tree_desc entry.
create_ce_mode(mode) in cache.h is similar but is intended to be
used for index entries (so it does not work for directories) and
returns the value in the network byte order.
Signed-off-by: Junio C Hamano <junkio@cox.net>
