This updates git-rev-list --objects to be a bit more careful
when listing a blob object to make sure the blob actually
exists, and uses it to make sure the quick-fetch optimization we
introduced earlier is not fooled by a previous incomplete fetch.
The quick-fetch optimization works by running this command:
git rev-list --objects <<commit-list>> --not --all
where <<commit-list>> is a list of commits that we are going to
fetch from the other side. If there is any object missing to
complete the <<commit-list>>, the rev-list would fail and die
(say, the commit was in our repository, but its tree wasn't --
then it will barf while trying to list the blobs the tree
contains because it cannot read that tree).
Usually we do not have the objects (otherwise why would we
fetching?), but in one important special case we do: when the
remote repository is used as an alternate object store
(i.e. pointed by .git/objects/info/alternates). We could check
.git/objects/info/alternates to see if the remote we are
interacting with is one of them (or is used as an alternate,
recursively, by one of them), but that check is more cumbersome
than it is worth.
The above check however did not catch missing blob, because
object listing code did not read nor check blob objects, knowing
that blobs do not contain any further references to other
objects. This commit fixes it with practically unmeasurable
overhead.
I've benched this with
git rev-list --objects --all >/dev/null
in the kernel repository, with three different implementations
of the "check-blob".
- Checking with has_sha1_file() has negligible (unmeasurable)
performance penalty.
- Checking with sha1_object_info() makes it somewhat slower,
perhaps by 5%.
- Checking with read_sha1_file() to cause a fully re-validation
is prohibitively expensive (about 4 times as much runtime).
In my original patch, I had this as a command line option, but
the overhead is small enough that it is not really worth it.
Signed-off-by: Junio C Hamano <junkio@cox.net>
* maint:
Fix lseek(2) calls with args 2 and 3 swapped
Honor -p<n> when applying git diffs
Fix dependency of common-cmds.h
Fix renaming branch without config file
DESTDIR support for git/contrib/emacs
gitweb: Fix bug in "blobdiff" view for split (e.g. file to symlink) patches
Document --left-right option to rev-list.
Revert "builtin-archive: use RUN_SETUP"
rename contrib/hooks/post-receieve-email to contrib/hooks/post-receive-email.
rerere: make sorting really stable.
Fix t4200-rerere for white-space from "wc -l"
The trick is to give a child commit that is not tree-changing
the same depth as its parent, so that the depth is propagated
properly along strand of pearls.
Signed-off-by: Junio C Hamano <junkio@cox.net>
If you have 5 commits in the set, commits that reach 2 or 3
commits are at halfway. If you have 6 commits, only commits
that reach exactly 3 commits are at halfway. The earlier one is
completely botched the math.
Signed-off-by: Junio C Hamano <junkio@cox.net>
In a path-limited bisection, when the $bad commit is not
changing the limited path, and the number of suspects is 1, the
code miscounted and returned $bad from find_bisection(), which
is not marked with TREECHANGE. This is of course filtered by
the output routine, resulting in an empty output, in turn
causing git-bisect driver to say "$bad was both good and bad".
Illustration. Suppose you have these four commits, and only C
changes path P. You know D is bad and A is good.
A---B---C*--D
git-bisect driver runs this to find a bisection point:
$ git rev-list --bisect A..D -- P
which calls find_bisection() with B, C and D. The set of
commits that is given to this function is the same set of
commits as rev-list without --bisect option and pathspec
returns. Among them, only C is marked with TREECHANGE. Let's
call the set of commits given to find_bisection() that are
marked with TREECHANGE (or all of them if no path limiter is in
effect) "the bisect set". In the above example, the size of the
bisect set is 1 (contains only "C").
For each commit in its input, find_bisection() computes the
number of commits it can reach in the bisect set. For a commit
in the bisect set, this number includes itself, so the number is
1 or more. This number is called "depth", and computed by
count_distance() function.
When you have a bisect set of N commits, and a commit has depth
D, how good is your bisection if you returned that commit? How
good this bisection is can be measured by how many commits are
effectively tested "together" by testing one commit.
Currently you have (N-1) untested commits (the tip of the bisect
set, although it is included in the bisect set, is already known
to be bad). If the commit with depth D turns out to be bad,
then your next bisect set will have D commits and you will have
(D-1) untested commits left, which means you tested (N-1)-(D-1)
= (N-D) commits with this bisection. If it turns out to be good, then
your next bisect set will have (N-D) commits, and you will have
(N-D-1) untested commits left, which means you tested
(N-1)-(N-D-1) = D commits with this bisection.
Therefore, the goodness of this bisection is is min(N-D, D), and
find_bisection() function tries to find a commit that maximizes
this, by initializing "closest" variable to 0 and whenever a
commit with the goodness that is larger than the current
"closest" is found, that commit and its goodness are remembered
by updating "closest" variable. The "the commit with the best
goodness so far" is kept in "best" variable, and is initialized
to a commit that happens to be at the beginning of the list of
commits given to this function (which may or may not be in the
bisect set when path-limit is in use).
However, when N is 1, then the sole tree-changing commit has
depth of 1, and min(N-D, D) evaluates to 0. This is not larger
than the initial value of "closest", and the "so far the best
one" commit is never replaced in the loop.
When path-limit is not in use, this is not a problem, as any
commit in the input set is tree-changing. But when path-limit
is in use, and when the starting "bad" commit does not change
the specified path, it is not correct to return it.
Signed-off-by: Junio C Hamano <junkio@cox.net>
This improves the performance of revision bisection.
The idea is to avoid rather expensive count_distance() function,
which counts the number of commits that are reachable from any
given commit (including itself) in the set. When a commit has
only one relevant parent commit, the number of commits the
commit can reach is exactly the number of commits that the
parent can reach plus one; instead of running count_distance()
on commits that are on straight single strand of pearls, we can
just add one to the parents' count.
On the other hand, for a merge commit, because the commits
reachable from one parent can be reachable from another parent,
you cannot just add the parents' counts up plus one for the
commit itself; that would overcount ancestors that are reachable
from more than one parents.
The algorithm used in the patch runs count_distance() on merge
commits, and uses the util field of commit objects to remember
them. After that, the number of commits reachable from each of
the remaining commits is counted by finding a commit whose count
is not yet known but the count for its (sole) parent is known,
and adding one to the parent's count, until we assign numbers to
everybody.
Another small optimization is whenever we find a half-way commit
(that is, a commit that can reach exactly half of the commits),
we stop giving counts to remaining commits, as we will not find
any better commit than we just found.
The performance to bisect between v1.0.0 and v1.5.0 in git.git
repository was improved by saying good and bad in turns from
3.68 seconds down to 1.26 seconds. Bisecting the kernel between
v2.6.18 and v2.6.20 was sped up from 21.84 seconds down to 4.22
seconds.
Signed-off-by: Junio C Hamano <junkio@cox.net>
This adds --bisect-vars option to rev-list. The output is suitable
for `eval` in shell and defines five variables:
- bisect_rev is the next revision to test.
- bisect_nr is the expected number of commits to test after
bisect_rev is tested.
- bisect_good is the expected number of commits to test
if bisect_rev turns out to be good.
- bisect_bad is the expected number of commits to test
if bisect_rev turns out to be bad.
- bisect_all is the number of commits we are bisecting right now.
The documentation text was partly stolen from Johannes
Schindelin's patch.
Signed-off-by: Junio C Hamano <junkio@cox.net>
Otherwise "git rev-list --header HEAD" will not do the right
thing if i18n.commitencoding is set.
Signed-off-by: Fredrik Kuivinen <frekui@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
The output from "symmetric diff", i.e. A...B, does not
distinguish between commits that are reachable from A and the
ones that are reachable from B. In this picture, such a
symmetric diff includes commits marked with a and b.
x---b---b branch B
/ \ /
/ .
/ / \
o---x---a---a branch A
However, you cannot tell which ones are 'a' and which ones are
'b' from the output. Sometimes this is frustrating. This adds
an output option, --left-right, to rev-list.
rev-list --left-right A...B
would show ones reachable from A prefixed with '<' and the ones
reachable from B prefixed with '>'.
When combined with --boundary, boundary commits (the ones marked
with 'x' in the above picture) are shown with prefix '-', so you
would see list that looks like this:
git rev-list --left-right --boundary --pretty=oneline A...B
>bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb 3rd on b
>bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb 2nd on b
<aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa 3rd on a
<aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa 2nd on a
-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 1st on b
-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 1st on a
Signed-off-by: Junio C Hamano <junkio@cox.net>
Now we can tell the built-in grep to grep only in head or in
body, use that to update --author, --committer, and --grep.
Unfortunately, to make --and, --not and other grep boolean
expressions useful, as in:
# Things written by Junio committed and by Linus and log
# does not talk about diff.
git log --author=Junio --and --committer=Linus \
--grep-not --grep=diff
we will need to do another round of built-in grep core
enhancement, because grep boolean expressions are designed to
work on one line at a time.
Signed-off-by: Junio C Hamano <junkio@cox.net>
We need to save the commit buffer if we're going to match against it.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This teaches the internal rev-list logic to understand options
that are needed for pack handling: --all, --unpacked, and --thin.
It also moves two functions from builtin-rev-list to list-objects
so that the two programs can share more code.
Signed-off-by: Junio C Hamano <junkio@cox.net>
Create a separate file, list-objects.c, and move object listing
routines from rev-list to it. The next round will use it in
pack-objects directly.
Signed-off-by: Junio C Hamano <junkio@cox.net>
When --stdin option is given, in addition to the <rev>s listed
on the command line, the command can read one rev parameter per
line from the standard input. The list of revs ends at the
first empty line or EOF.
Note that you still have to give all the flags from the command
line; only rev arguments (including A..B, A...B, and A^@ notations)
can be give from the standard input.
Signed-off-by: Junio C Hamano <junkio@cox.net>
Like xmalloc and xrealloc xstrdup dies with a useful message if
the native strdup() implementation returns NULL rather than a
valid pointer.
I just tried to use xstrdup in new code and found it to be missing.
However I expected it to be present as xmalloc and xrealloc are
already commonly used throughout the code.
[jc: removed the part that deals with last_XXX, which I am
finding more and more dubious these days.]
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
[jc: I needed to hand merge the changes to the updated codebase,
so the result needs to be checked.]
Signed-off-by: David Rientjes <rientjes@google.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This changes the calling convention of built-in commands and
passes the "prefix" (i.e. pathname of $PWD relative to the
project root level) down to them.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Any git command that expects to work in a subdirectory of a project, and
that reads the git config files (which is just about all of them) needs to
make sure that it does the "setup_git_directory()" call before it tries to
read the config file.
This means, among other things, that we need to move the call out of
"init_revisions()", and into the caller.
This does the mostly trivial conversion to do that.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This updates the type-enumeration constants introduced to reduce
the memory footprint of "struct object" to match the type bits
already used in the packfile format, by removing the former
(i.e. TYPE_* constant macros) and using the latter (i.e. enum
object_type) throughout the code for consistency.
Eventually we can stop passing around the "type strings"
entirely, and this will help - no confusion about two different
integer enumeration.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
We've had this notion of a "object_list" for a long time, which eventually
grew a "name" member because some users (notably git-rev-list) wanted to
name each object as it is generated.
That object_list is great for some things, but it isn't all that wonderful
for others, and the "name" member is generally not used by everybody.
This patch splits the users of the object_list array up into two: the
traditional list users, who want the list-like format, and who don't
actually use or want the name. And another class of users that really used
the list as an extensible array, and generally wanted to name the objects.
The patch is fairly straightforward, but it's also biggish. Most of it
really just cleans things up: switching the revision parsing and listing
over to the array makes things like the builtin-diff usage much simpler
(we now see exactly how many members the array has, and we don't get the
objects reversed from the order they were on the command line).
One of the main reasons for doing this at all is that the malloc overhead
of the simple object list was actually pretty high, and the array is just
a lot denser. So this patch brings down memory usage by git-rev-list by
just under 3% (on top of all the other memory use optimizations) on the
mozilla archive.
It does add more lines than it removes, and more importantly, it adds a
whole new infrastructure for maintaining lists of objects, but on the
other hand, the new dynamic array code is pretty obvious. The change to
builtin-diff-tree.c shows a fairly good example of why an array interface
is sometimes more natural, and just much simpler for everybody.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This is really the dregs of my effort to not waste memory in git-rev-list,
and makes barely one percent of a difference in the memory footprint, but
hey, it's also a pretty small patch.
It discards the parent lists and the commit buffer after the commit has
been shown by git-rev-list (and "git log" - which already did the commit
buffer part), and frees the commit list entry that was used by the
revision walker.
The big win would be to get rid of the "refs" pointer in the object
structure (another 5%), because it's only used by fsck. That would require
some pretty major surgery to fsck, though, so I'm timid and did the less
interesting but much easier part instead.
This (percentually) makes a bigger difference to "git log" and friends,
since those are walking _just_ commits, and thus the list entries tend to
be a bigger percentage of the memory use. But the "list all objects" case
does improve too.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This shrinks "struct object" by a small amount, by getting rid of the
"struct type *" pointer and replacing it with a 3-bit bitfield instead.
In addition, we merge the bitfields and the "flags" field, which
incidentally should also remove a useless 4-byte padding from the object
when in 64-bit mode.
Now, our "struct object" is still too damn large, but it's now less
obviously bloated, and of the remaining fields, only the "util" (which is
not used by most things) is clearly something that should be eventually
discarded.
This shrinks the "git-rev-list --all" memory use by about 2.5% on the
kernel archive (and, perhaps more importantly, on the larger mozilla
archive). That may not sound like much, but I suspect it's more on a
64-bit platform.
There are other remaining inefficiencies (the parent lists, for example,
probably have horrible malloc overhead), but this was pretty obvious.
Most of the patch is just changing the comparison of the "type" pointer
from one of the constant string pointers to the appropriate new TYPE_xxx
small integer constant.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
The tree-walking conversion of the "process_tree()" function
broke packing by using an unrelated variable from outer scope.
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>
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>
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 patch touches a couple of files, because it adds options to print a
custom text just after the subject of a commit, and just after the
diffstat.
[jc: made "many dashes" used as the boundary leader into a single
variable, to reduce the possibility of later tweaks to miscount the
number of dashes to break it.]
Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
* master: (119 commits)
diff family: add --check option
Document that "git add" only adds non-ignored files.
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.
git-am: use apply --cached
apply --cached: apply a patch without using working tree.
apply --numstat: show new name, not old name.
Documentation/Makefile: create tarballs for the man pages and html files
Allow pickaxe and diff-filter options to be used by git log.
Libify the index refresh logic
Builtin git-init-db
Remove unnecessary local in get_ref_sha1.
...
This was surprisingly easy. The diff is truly minimal: rename "main()" to
"cmd_rev_list()" in rev-list.c, and rename the whole file to reflect its
new built-in status.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
