* 'master' of git://git.kernel.org/pub/scm/gitk/gitk:
gitk: Add a window to list branches, tags and other references
[PATCH] gitk: Handle 'copy from' and 'copy to' in diff headers.
gitk: Fix bug in fix for warning when removing a branch
gitk: Add a window to list branches, tags and other references
This adds an entry to the File menu labelled "List references" which
pops up a window showing a sorted list of branches, tags, and other
references, with a little icon beside each to indicate what sort it
is. The list only shows refs that point to a commit that is included
in the graph, and if you click on a ref, the corresponding commit
is selected in the main window. The list of refs gets updated
dynamically.
Document what the stage numbers in the :$n:path syntax mean.
The git-rev-parse manpage talks about the :$n:path notation (buried deep in
a list of other syntax) but it just says $n is a "stage number" -- someone
who is not familiar with the internals of git's merge implementation is
never going to be able to figure out that "1", "2", and "3" means.
Don't allow combination of -g and --reverse as it doesn't work
The --walk-reflogs logic and the --reverse logic are completely
incompatible with one another. Attempting to use both at the same
time leads to confusing results that sometimes violates the user's
formatting options or ignores the user's request to see the reflog
message and timestamp.
Unfortunately the implementation of both of these features is glued
onto the side of the revision walking machinary in such a way that
they are probably not going to be easy to make them compatible with
each other. Rather than offering the user confusing results we are
better off bailing out with an error message until such a time as
the implementations can be refactored to be compatible.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
We used to not generate a patch ID for binary diffs, but that means that
some commits may be skipped as being identical to already-applied diffs
when doing a rebase.
So just delete the code that skips the binary diff. At the very least,
we'd want the filenames to be part of the patch ID, but we might also want
to generate some hash for the binary diff itself too.
This fixes an issue noticed by Torgil Svensson.
Tested-by: Torgil Svensson <torgil.svensson@gmail.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When a thin pack wants to send a tree object at "sub/dir", and
the commit that is common between the sender and the receiver
that is used as the base object has a subproject at that path,
we should not try to use the data at "sub/dir" of the base tree
as a tree object. It is not a tree to begin with, and more
importantly, the commit object there does not have to even
exist.
Correct documentation of 'reflog show' to explain it shows HEAD
By default 'git reflog show' will show the reflog of 'HEAD' and not
the reflog of the current branch. This is most likely due to the
work done a while ago as part of the detached HEAD series to allow
HEAD to have its own reflog independent of each branch's reflog.
Since 'git reflog show' is really just an obscure alias for 'git
log -g --abbrev-commit --pretty=oneline' it should behave the same
way and its documentation should match.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
* 'master' of git://git.kernel.org/pub/scm/gitk/gitk:
[PATCH] gitk: Make the date/time display configurable
[PATCH] gitk: Let user easily specify lines of context in diff view
gitk: Fix warning when removing a branch
* skip_optional_lf() decl is old-style -- please say
static skip_optional_lf(void)
{
...
}
* t9300 #14 fails, like this:
* expecting failure: git-fast-import <input
fatal: Branch name doesn't conform to GIT standards: .badbranchname
fast-import: dumping crash report to .git/fast_import_crash_14354
./test-lib.sh: line 143: 14354 Segmentation fault git-fast-import <input
-- >8 --
Subject: [PATCH] fastimport: Fix re-use of va_list
The va_list is designed to be used only once. The current code
reuses va_list argument may cause segmentation fault. Copy and
release the arguments to avoid this problem.
While we are at it, fix old-style function declaration of
skip_optional_lf().
Signed-off-by: Junio C Hamano <gitster@pobox.com> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Include recent command history in fast-import crash reports
When we crash the frontend developer (or end-user) may need to know
roughly around what part of the input stream we had a problem with
and aborted on. Because line numbers aren't very useful in this
sort of application we instead just keep the last 100 commands in
a FIFO queue and print them as part of the crash report.
Currently one problem with this design is a commit that has
more than 100 modified files in it will flood the FIFO and any
context regarding branch/from/committer/mark/comments will be lost.
We really should save only the last few (10?) file changes for the
current commit, ensuring we have some prior higher level commands
in the FIFO when we crash on a file M/D/C/R command.
Another issue with this approach is the FIFO only includes the
commands, it does not include the commit messages. Yet having a
commit message may be useful to help locate the relevant change in
the source material. In practice I don't think this is going to be a
major concern as the frontend can always embed its own source change
set identifier as a comment (which will appear in the crash report)
and the commit message(s) for the most recent commits of any given
branch should be obtainable from the (packed) commit objects.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
As fast-import is quite strict about its input and die()'s anytime
something goes wrong it can be difficult for a frontend developer
to troubleshoot why fast-import rejected their input, or to even
determine what input command it rejected.
This change introduces a custom handler for Git's die() routine.
When we receive a die() for any reason (fast-import or a lower level
core Git routine we called) the error is first dumped onto stderr
and then a more extensive crash report file is prepared in GIT_DIR.
Finally we exit the process with status 128, just like the stock
builtin die handler.
An internal flag is set to prevent any further die()'s that may be
invoked during the crash report generator from causing us to enter
into an infinite loop. We shouldn't die() from our crash report
handler, but just in case someone makes a future code change we are
prepared to gaurd against small mistakes turning into huge problems
for the end-user.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Allow frontends to bidirectionally communicate with fast-import
The existing checkpoint command is very useful to force fast-import
to dump the branches out to disk so that standard Git tools can
access them and the objects they refer to. However there was not a
way to know when fast-import had finished executing the checkpoint
and it was safe to read those refs.
The progress command can be used to make fast-import output any
message of the frontend's choosing to standard out. The frontend
can scan for these messages using select() or poll() to monitor a
pipe connected to the standard output of fast-import.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Make trailing LF optional for all fast-import commands
For the same reasons as the prior change we want to allow frontends
to omit the trailing LF that usually delimits commands. In some
cases these just make the input stream more verbose looking than
it needs to be, and its just simpler for the frontend developer to
get started if our parser is slightly more lenient about where an
LF is required and where it isn't.
To make this optional LF feature work we now have to buffer up to one
line of input in command_buf. This buffering can happen if we look
at the current input command but don't recognize it at this point
in the code. In such a case we need to "unget" the entire line,
but we cannot depend upon the stdio library to let us do ungetc()
for that many characters at once.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Make trailing LF following fast-import `data` commands optional
A few fast-import frontend developers have found it odd that we
require the LF following a `data` command, especially in the exact
byte count format. Technically we don't need this LF to parse
the stream properly, but having it here does make the stream more
readable to humans. We can easily make the LF optional by peeking
at the next byte available from the stream and pushing it back into
the buffer if its not LF.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Teach fast-import to ignore lines starting with '#'
Several frontend developers have asked that some form of stream
comments be permitted within a fast-import data stream. This way
they can include information from their own frontend program about
where specific data was taken from in the source system, or about
a decision that their frontend may have made while creating the
fast-import data stream.
This change introduces comments in the Bourne-shell/Tcl/Perl style.
Lines starting with '#' are ignored, up to and including the LF.
Unlike the above mentioned three languages however we do not look for
and ignore leading whitespace. This just simplifies the definition
of the comment format and the code that parses them.
To make comments work we had to stop using read_next_command() within
cmd_data() and directly invoke read_line() during the inline variant
of the function. This is necessary to retain any lines of the
input data that might otherwise look like a comment to fast-import.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Use handy ALLOC_GROW macro in fast-import when possible
Instead of growing our buffer by hand during the inline variant of
cmd_data() we can save a few lines of code and just use the nifty
new ALLOC_GROW macro already available to us.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Michael Haggerty <mhagger@alum.mit.edu> noticed while debugging a
Git backend for cvs2svn that fast-import was barfing when he tried
to use "TAG_FIXUP" as a branch name for temporary work needed to
cleanup the tree prior to creating an annotated tag object.
The reason we were rejecting the branch name was check_ref_format()
returns -2 when there are less than 2 '/' characters in the input
name. TAG_FIXUP has 0 '/' characters, but is technically just as
valid of a ref as HEAD and MERGE_HEAD, so we really should permit it
(and any other similar looking name) during import.
New test cases have been added to make sure we still detect very
wrong branch names (e.g. containing [ or starting with .) and yet
still permit reasonable names (e.g. TAG_FIXUP).
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
[PATCH] gitk: Handle 'copy from' and 'copy to' in diff headers.
If a commit contained a copy operation, the file name was not correctly
determined, and the corresponding part of the patch could not be
navigated to from the list of files.
Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at> Signed-off-by: Paul Mackerras <paulus@samba.org>
gitk: Fix bug in fix for warning when removing a branch
My fix in commit b1054ac985aebc90c0a78202dab8477b74d7818a was only
half-right, since it ignored the case where the descendent heads of
the head being removed correspond to two or more different commits.
This fixes it. Reported by Mark Levedahl.
This is a follow-up to f34f2b0b; list_tree() function is where it
first notices that the command line fed too many trees for us to
handle, so move the error exit message to there, and raise the
MAX_TREES to 8 (not that it matters very much in practice).
Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The t1301-shared-repo.sh testscript uses /usr/bin/stat to get the file
mode, which isn't portable. Implement the test in shell using 'ls' as
shown by Junio.
Signed-off-by: Arjen Laarhoven <arjen@yaph.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Apply "Subproject commit HEX" changes produced by git-diff.
As usual in the current git, only the superproject itself is actually
modified (possibly creating empty directories for new submodules).
Any checked-out submodule is left untouched and is not required to
be up-to-date.
With clean-ups from Junio C Hamano.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Fix read-tree merging more than 3 trees using 3-way merge
For multi-base merges, we allowed read-tree -m to take more than
three trees (the last two are our and their branches, and all the
earlier ones, typically one but potentially more, are used as the
merge base). Unfortunately, the conversion done by commit 933bf40
broke this.
git-add -u also takes the path limiters, but unlike the
command without the -u option, the code forgot that it
could be invoked from a subdirectory, and did not correctly
handle the prefix.
Signed-off-by: Salikh Zakirov <salikh@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
This applies to 'maint' to fix a rather serious data corruption
issue. When "git add -u" affects a subdirectory in such a way
that the only changes to its contents are path removals, the
next tree object written out of that index was bogus, as the
remove codepath forgot to invalidate the cache-tree entry.
Always free .paths if .strdup_paths is set, no matter if the
parameter free_items is set or not, plugging a minor memory leak.
And to clarify the meaning of the flag, rename it to free_util,
since it now only affects the freeing of the .util field.
Signed-off-by: Rene Scharfe <rene.scharfe@lsrfire.ath.cx> Signed-off-by: Junio C Hamano <gitster@pobox.com>
This makes .gitattributes files to be read from the index when
they are not checked out to the work tree. This is in line with
the way we always allowed low-level tools to operate in sparsely
checked out work tree in a reasonable way.
It swaps the order of new file creation and converting the blob
to work tree representation; otherwise when we are in the middle
of checking out .gitattributes we would notice an empty but
unwritten .gitattributes file in the work tree and will ignore
the copy in the index.
This splits out a common routine that parses a single line of
attribute file and adds it to the attr_stack. It should not
change any behaviour, other than attrs array in the attr_stack
structure is now grown with alloc_nr() macro, instead of one by
one, which relied on xrealloc() to give enough slack to be
efficient enough.
When we compare two non-tracked files, or explicitly
specify --no-index, the suggestion to run git-status
is not helpful.
The patch adds a new diff_options bitfield member, no_index, that
is used instead of the special value of -2 of the rev_info field
max_count to indicate that the index is not to be used. This makes
it possible to pass that flag down to diffcore_skip_stat_unmatch(),
which only has one diff_options parameter.
This could even become a cleanup if we removed all assignments of
max_count to a value of -2 (viz. replacement of a magic value with
a self-documenting field name) but I didn't dare to do that so late
in the rc game..
The no_index bit, if set, then tells diffcore_skip_stat_unmatch()
to not account for any skipped stat-mismatches, which avoids the
suggestion to run git-status.
Signed-off-by: Rene Scharfe <rene.scharfe@lsfire.ath.cx> Signed-off-by: Junio C Hamano <gitster@pobox.com>
merge-recursive: do not rudely die on binary merge
When you try to merge a path that involves binary file-level
merge, merge-recursive died rudely without cleaning up its own
mess. A files added by the merge were left in the working tree,
but the index was not written out (because it just punted and
died), so it was cumbersome for the user to retry it by first
running "git reset --hard".
This changes merge-recursive to still warn but do the "binary"
merge for such a path; leave the "our" version in the working
tree, but still keep the path unmerged so that the user can sort
it out.
[PATCH] gitk: Let user easily specify lines of context in diff view
More lines of context sometimes help to better understand a diff.
This patch introduces a text field above the box displaying the
blobdiffs. You can type in the number of lines of context that
you wish to view. The number of lines of context is saved to
~/.gitk.
Signed-off-by: Steffen Prohaska <prohaska@zib.de> Signed-off-by: Paul Mackerras <paulus@samba.org>
When we had two heads on the same commit, and the user tried to remove
one of them, gitk was sometimes incorrectly saying that the commits
on that branch weren't on any other branch. This fixes it.
After starting to edit a working tree file but later when your edit ends
up identical to the original (this can also happen when you ran a
wholesale regexp replace with something like "perl -i" that does not
actually modify many of the paths), "git diff" between the index and the
working tree outputs many "empty" diffs that show "diff --git" headers
and nothing else, because these paths are stat-dirty. While it was a
way to warn the user that the earlier action of the user made the index
ineffective as an optimization mechanism, it was felt too loud for the
purpose of warning even to experienced users, and also resulted in
confusing people new to git.
This replaces the "empty" diffs with a single warning message at the
end. Having many such paths hurts performance, and you can run
"git-update-index --refresh" to update the lstat(2) information recorded
in the index in such a case. "git-status" does so as a side effect, and
that is more familiar to the end-user, so we recommend it to them.
The change affects only "git diff" that outputs patch text, because that
is where the annoyance of too many "empty" diff is most strongly felt,
and because the warning message can be safely ignored by downstream
tools without getting mistaken as part of the patch. For the low-level
"git diff-files" and "git diff-index", the traditional behaviour is
retained.
The first thing we teach in the tutorial is to set the default
identity in $HOME/.gitconfig using "git config --global". The
suggestion in the error message should match the order, while
hinting that per repository identity can later be configured
differently.
Signed-off-by: Steffen Prohaska <prohaska@zib.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Quite a few of the scripts are rather careless about using GIT_DIR
while changing directories.
Some try their hands (with different likelihood of success) in making
GIT_DIR absolute.
This patch lets git-sh-setup.sh cater for absolute directories (in a
way that should work reliably also with non-Unix path names) and
removes the respective kludges in git-filter-branch.sh and
git-instaweb.sh.
Signed-off-by: David Kastrup <dak@gnu.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Allow git-svnimport to take "" as the trunk directory.
Some repositories started with the trunk in "/" and then moved it to the
standard "trunk/" location.
On these repositories, the correct thing would be to call git-svnimport -T "",
but because of the way the options are handled, it uses the default "trunk"
instead of the given empty string. This patch fixes that behaviour.
Reported by Leandro Lucarella <llucax@gmail.com>.
Signed-off-by: Alberto Bertogli <albertito@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
git-p4: Fix the sorting of changelists when cloning a Perforce repository.
When performing a git-p4 clone operation on a Perforce repository,
where the changelists change in order of magnitude (e.g. 100 to 1000),
the set of changes to import from is not sorted properly. This is
because the data in the list is strings not integers. The other place
where this is done already converts the value to an integer, so it is
not affected.
diff: don't run pager if user asked for a diff style exit code
As Wincent Colaiuta found out, it's a bit unexpected for git diff to
start a pager even when the --quiet option is specified. The problem
is that the pager hides the return code -- which is the only output
we're interested in in this case.
Push pager setup down into builtin-diff.c and don't start the pager
if --exit-code or --quiet (which implies --exit-code) was specified.
Signed-off-by: Rene Scharfe <rene.scharfe@lsrfire.ath.cx> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Here-text to create fake-editor did not use <<\EOF but <<EOF,
but there was no point doing so, as it quoted all the variables
anyway. Simplify it.
Also futureproof the special mode to edit COMMIT_EDITMSG file;
it is interested in editing the COMMIT_EDITMSG file in any
GIT_DIR; GIT_DIR may be given as an absolute path.
* 'master' of git://git.kernel.org/pub/scm/gitk/gitk:
gitk: Fix bug causing Tcl error when updating graph
gitk: Fix bug introduced in commit 67a4f1a7
[PATCH] gitk: Show an error and exit if no .git could be found
[PATCH] gitk: Continue and show error message in new repos
[PATCH] gitk: Handle MouseWheel events on Windows
[PATCH] gitk: Enable selected patch text on Windows
gitk: Fix bug causing the "can't unset idinlist(...)" error
gitk: Add a context menu for file list entries
gitk: Fix bug causing Tcl error when updating graph
If "Show nearby tags" is turned off, selecting "Update" from the File
menu will cause a Tcl error. This fixes it. The problem was that
we were calling regetallcommits unconditionally, but it assumed that
getallcommits had been called previously. This also restructures
{re,}getallcommits to be a bit simpler.
In fixing the "can't unset idinlist" error, I moved the setting of
idinlist into the loop that splits the parents into "new" parents
(i.e. those of which this is the first child) and "old" parents.
Unfortunately this is incorrect in the case where we hit the break
statement a few lines further down, since when we come back in,
we'll see idinlist($p) set for some parents that aren't in the list.
This fixes it by moving the loop that sets up newolds and oldolds
further down.
[PATCH] gitk: Continue and show error message in new repos
If there is no commit made yet, gitk just dumps a Tcl error on stderr,
which sometimes is hard to see. Noticed when gitk was run from Xfce
file manager (thunar's custom action).
Signed-off-by: Alex Riesen <raa.lkml@gmail.com> Signed-off-by: Paul Mackerras <paulus@samba.org>
Windows, unlike X-Windows, sends mousewheel events by default to the
window that has keyboard focus and uses the MouseWheel event to do so.
The window to be scrolled must be able to take focus, but gitk's panels
are disabled so cannot take focus. For all these reasons, a different
design is needed to use the mousewheel on Windows. The approach here is
to bind the mousewheel events to the top level window and redirect them
based upon the current mouse position.
Signed-off-by: Mark Levedahl <mdl123@verizon.net> Signed-off-by: Paul Mackerras <paulus@samba.org>
[PATCH] gitk: Enable selected patch text on Windows
On windows, mouse input follows the keyboard focus, so to allow selecting
text from the patch canvas we must not shift focus back to the top level.
This change has no negative impact on X, so we don't explicitly test
for Win32 on this change. This provides similar selection capability
as already available using X-Windows.
Signed-off-by: Mark Levedahl <mdl123@verizon.net> Signed-off-by: Paul Mackerras <paulus@samba.org>
gitk: Fix bug causing the "can't unset idinlist(...)" error
Under some circumstances, having duplicate parents in a commit could
trigger a "can't unset idinlist" Tcl error. This fixes the cause
(the logic in layoutrows could end up putting the same commit into
rowidlist twice) and also puts a catch around the unset to ignore
the error.
Thanks to Jeff King for coming up with a test script to generate a
repo that shows the problem.
* cr/tag:
Teach "git stripspace" the --strip-comments option
Make verify-tag a builtin.
builtin-tag.c: Fix two memory leaks and minor notation changes.
launch_editor(): Heed GIT_EDITOR and core.editor settings
Make git tag a builtin.
Add support for an info version of the user manual
These patches use docbook2x in order to create an info version of the
git user manual. No existing Makefile targets (including "all") are
touched, so you need to explicitly say
make info
sudo make install-info
to get git.info created and installed. If the info target directory
does not already contain a "dir" file, no directory entry is created.
This facilitates $(DESTDIR)-based installations. The same could be
achieved with
sudo make INSTALL_INFO=: install-info
explicitly.
perl is used for patching up sub-par file and directory information in
the Texinfo file. It would be cleaner to place the respective info
straight into user-manual.txt or the conversion configurations, but I
find myself unable to find out how to do this with Asciidoc/Texinfo.
As mentioned, the three-way case *should* be as trivial as the
following. It passes all the tests, and I verified that a conflicting
merge in the 100,000 file horror-case merged correctly (with the conflict
markers) in 0.687 seconds with this, so it works, but I'm lazy and
somebody else should double-check it [jc: followed all three-way merge
codepaths and verified it removes when it should].
Without this patch, the merge took 8.355 seconds, so this patch
really does make a huge difference for merge performance with lots and
lots of files, and we're not talking percentages, we're talking
orders-of-magnitude differences!
Now "unpack_trees()" is just fast enough that we don't need to avoid it
(although it's probably still a good idea to eventually convert it to use
the traverse_trees() infrastructure some day - just to avoid having
extraneous tree traversal functions).
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
builtin-bundle - use buffered reads for bundle header
This eliminates all use of byte-at-a-time reading of data in this
function: as Junio noted, a bundle file is seekable so we can
reset the file position to the first part of the pack-file using lseek
after reading the header.
Signed-off-by: Mark Levedahl <mdl123@verizon.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
builtin-bundle.c - use stream buffered input for rev-list
git-bundle create on cygwin was nearly unusable due to 1 character
at a time (unbuffered) reading from an exec'ed process. Fix by using
fdopen to get a buffered stream.
Results for "time git bundle create test.bdl v1.0.3..v1.5.2" are:
before this patch:
cygwin linux
real 1m38.828s 0m3.578s
user 0m12.122s 0m2.896s
sys 1m28.215s 0m0.692s
after this patch:
real 0m3.688s 0m2.835s
user 0m3.075s 0m2.731s
sys 0m1.075s 0m0.149s
Signed-off-by: Mark Levedahl <mdl123@verizon.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
> Junio C Hamano wrote:
>> While "git bundle" was a useful way to sneakernet incremental
>> changes, we did not allow:
>>
> Thanks - I've been thinking for months I could fix this bug, never
> figured it out and didn't want to nag Dscho one more time. I confirm
> that this allows creation of bundles with arbitrary refs, not just
> those under refs/heads. Yahoo!
the bundle records that it requires v2.6.20^0 commit (correct)
and gives you tag v2.6.22 (incorrect); the bug is that the
object it lists in fact is the commit v2.6.22^0, not the tag.
This is because the revision range operation .. is always about
set of commits, but the code near where my patch touches does
not validate that the sha1 value obtained from dwim_ref()
against the commit object name e->item->sha1 before placing the
head information in the commit.
While "git bundle" was a useful way to sneakernet incremental
changes, we did not allow:
$ git bundle create v2.6.20.bndl v2.6.20
to create a bundle that contains the whole history to a
well-known good revision. Such a bundle can be mirrored
everywhere, and people can prime their repository with it to
reduce the load on the repository that serves near the tip of
the development.
This trivially optimizes the two-way merge case of git-read-tree too,
which affects switching branches.
When you have tons and tons of files in your repository, but there are
only small differences in the branches (maybe just a couple of files
changed), the biggest cost of the branch switching was actually just the
index calculations.
This fixes it (timings for switching between the "testing" and "master"
branches in the 100,000 file testing-repo-from-hell, where the branches
only differ in one small file).
Before:
[torvalds@woody bummer]$ time git checkout master
real 0m9.919s
user 0m8.461s
sys 0m0.264s
After:
[torvalds@woody bummer]$ time git checkout testing
real 0m0.576s
user 0m0.348s
sys 0m0.228s
so it's easily an order of magnitude different.
This concludes the series. I think we could/should do the three-way merge
too (to speed up merges), but I'm lazy. Somebody else can do it.
The rule is very simple: you need to remove the old entry if:
- you want to remove the file entirely
- you replace it with a "merge conflict" entry (ie a non-stage-0 entry)
and you can avoid removing it if you either
- keep the old one
- or resolve it to a new one.
and these rules should all be valid for the three-way case too.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
This optimizes bind_merge() and oneway_merge() to not unnecessarily
remove and re-add the old index entries when they can just get replaced
by updated ones.
This makes these operations much faster for large trees (where "large"
is in the 50,000+ file range), because we don't unnecessarily move index
entries around in the index array all the time.
Using the "bummer" tree (a test-tree with 100,000 files) we get:
Before:
[torvalds@woody bummer]$ time git commit -m"Change one file" 50/500
real 0m9.470s
user 0m8.729s
sys 0m0.476s
After:
[torvalds@woody bummer]$ time git commit -m"Change one file" 50/500
real 0m1.173s
user 0m0.720s
sys 0m0.452s
so for large trees this is easily very noticeable indeed.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Move old index entry removal from "unpack_trees()" into the individual functions
This makes no changes to current code, but it allows the individual merge
functions to decide what to do about the old entry. They might decide to
update it in place, rather than force them to always delete and re-add it.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
This trivial patch avoids re-hashing files that are already clean in the
index. This mirrors what commit 0781b8a9b2fe760fc4ed519a3a26e4b9bd6ccffe
did for "git add .", only for "git commit ." instead.
This improves the cold-cache case immensely, since we don't need to bring
in all the file contents, just the index and any files dirty in the index.
Before:
[torvalds@woody linux]$ time git commit .
real 1m49.537s
user 0m3.892s
sys 0m2.432s
After:
[torvalds@woody linux]$ time git commit .
real 0m14.273s
user 0m1.312s
sys 0m0.516s
(both after doing a "echo 3 > /proc/sys/vm/drop_caches" to get cold-cache
behaviour - even with the index optimization git still has to "lstat()"
all the files, so with a truly cold cache, bringing all the inodes in
will take some time).
[jc: trivial "return 0;" fixed]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The read_tree() function is called only from the call chain to
run "git diff --cached" (this includes the internal call made by
git-runstatus to run_diff_index()). The function vacates stage
without any funky "merge" magic. The caller then goes and
compares stage #1 entries from the tree with stage #0 entries
from the original index.
When adding the cache entries this way, it used the general
purpose add_cache_entry(). This function looks for an existing
entry to replace or if there is none to find where to insert the
new entry, resolves D/F conflict and all the other things.
For the purpose of reading entries into an empty stage, none of
that processing is needed. We can instead append everything and
then sort the result at the end.
This commit changes read_tree() to first make sure that there is
no existing cache entries at specified stage, and if that is the
case, it runs add_cache_entry() with ADD_CACHE_JUST_APPEND flag
(new), and then sort the resulting cache using qsort().
This new flag tells add_cache_entry() to omit all the checks
such as "Does this path already exist? Does adding this path
remove other existing entries because it turns a directory to a
file?" and instead append the given cache entry straight at the
end of the active cache. The caller of course is expected to
sort the resulting cache at the end before using the result.