Recent "rebase --autostash" update made it impossible to recover
with "git am --abort" from a repository where "git am" without mbox
was run by mistake and then was killed with "^C".
* rr/am-quit-empty-then-abort-fix:
t/am: use test_path_is_missing() where appropriate
am: handle stray $dotest directory
Allow various commit objects to be given to "git rebase" by ':/look
for this string' syntax, e.g. "git rebase --onto ':/there'".
* rr/rebase-sha1-by-string-query:
rebase: use peel_committish() where appropriate
sh-setup: add new peel_committish() helper
t/rebase: add failing tests for a peculiar revision
Make it possible to call into copy-notes API from the sequencer code.
* jh/libify-note-handling:
Move create_notes_commit() from notes-merge.c into notes-utils.c
Move copy_note_for_rewrite + friends from builtin/notes.c to notes-utils.c
finish_copy_notes_for_rewrite(): Let caller provide commit message
bash prompt: mention that PROMPT_COMMAND mode is faster
__git_ps1() is usually added to the prompt inside a command
substitution, imposing the overhead of fork()ing a subshell. Using
__git_ps1() for $PROMPT_COMMAND is slightly faster, because it avoids
that command substitution.
Mention this in the comments about setting up the git prompt.
The whole series speeds up the bash prompt on Windows/MSysGit
considerably. Here are some timing results in three scenarios, each
repeated 10 times:
At the top of the work tree, before:
$ time for i in {0..9} ; do prompt="$(__git_ps1)" ; done
real 0m1.716s
user 0m0.301s
sys 0m0.772s
After:
real 0m0.687s
user 0m0.075s
sys 0m0.396s
After, from $PROMPT_COMMAND:
$ time for i in {0..9} ; do __git_ps1 '\h:\w' '$ ' ; done
real 0m0.546s
user 0m0.075s
sys 0m0.181s
At the top of the work tree, detached head, before:
real 0m2.574s
user 0m0.376s
sys 0m1.207s
After:
real 0m1.139s
user 0m0.151s
sys 0m0.500s
After, from $PROMPT_COMMAND:
real 0m1.030s
user 0m0.245s
sys 0m0.336s
In a subdirectory, during rebase, stash status indicator enabled,
before:
real 0m3.557s
user 0m0.495s
sys 0m1.767s
After:
real 0m0.717s
user 0m0.120s
sys 0m0.300s
After, from $PROMPT_COMMAND:
real 0m0.577s
user 0m0.047s
sys 0m0.258s
On Linux the speedup ratio is comparable to Windows, but overall it
was about an order of magnitude faster to begin with. The last case
from above, repeated 100 times, before:
$ time for i in {0..99} ; do prompt="$(__git_ps1)" ; done
bash prompt: avoid command substitution when finalizing gitstring
Before setting $PS1, __git_ps1() uses a command substitution to
redirect the output from a printf into a variable. Spare the overhead
of fork()ing a subshell by using 'printf -v <var>' to directly assign
the output to that variable.
zsh's printf doesn't support the '-v <var>' option, so stick with the
command substitution when under zsh.
bash prompt: avoid command substitution when checking for untracked files
When enabled, the bash prompt can indicate the presence of untracked
files with a '%' sign. __git_ps1() checks for untracked files by running the
'$(git ls-files --others --exclude-standard)' command substitution,
and displays the indicator when there is no output.
Avoid this command substitution by additionally passing
'--error-unmatch *', and checking the command's return value.
bash prompt: use bash builtins to check stash state
When the environment variable $GIT_PS1_SHOWSTASHSTATE is set
__git_ps1() checks the presence of stashes by running 'git rev-parse
--verify refs/stash'. This command not only checks that the
'refs/stash' ref exists but also, well, verifies that it's a valid
ref.
However, we don't need to be that thorough for the bash prompt. We
can omit that verification and only check whether 'refs/stash' exists
or not. Since 'git pack-refs' never packs 'refs/stash', it's a matter
of checking the existence of a ref file. Perform this check using
only bash builtins to spare the overhead of fork()+exec()ing a git
process.
Also run 'git pack-refs --all' in the corresponding test to document
that the prompt script depends on 'git pack-refs' not packing
'refs/stash' and to catch possible breakages should this behavior ever
change.
bash prompt: use bash builtins to check for unborn branch for dirty state
When the dirty work tree and index status indicator is enabled,
__git_ps1() checks for changes in the index by running 'git diff-index
--cached --quiet HEAD --' and looking at its exit code. However, that
makes sense only when HEAD points to a valid commit: on an unborn
branch the failure of said command would be caused by the invalid
HEAD, not by changes in the index. Therefore, __git_ps1() first
checks for a valid HEAD by running 'git rev-parse --quiet --verify
HEAD'.
Since the previous patch we implicitly check HEAD's validity by
running 'git rev-parse ... --short HEAD', making the dirty status
indicator's 'git rev-parse' check redundant. It's sufficient to check
for non-emptyness of the variable holding the abbreviated commit
object name, thereby sparing the overhead of fork()+exec()ing a git
process.
bash prompt: combine 'git rev-parse' for detached head
When describing a detached HEAD according to the $GIT_PS1_DESCRIBE
environment variable fails, __git_ps1() now runs the '$(git rev-parse
--short HEAD)' command substitution to get the abbreviated detached
HEAD commit object name. This imposes the overhead of fork()ing a
subshell and fork()+exec()ing a git process.
Avoid this overhead by combining this command substitution with the
"main" 'git rev-parse' execution for getting the path to the .git
directory & co. This means that we'll look for the abbreviated commit
object name even when it's not necessary, because we're on a branch or
the detached HEAD can be described. It doesn't matter, however,
because once 'git rev-parse' is up and running to fulfill all those
other queries, the additional overhead of looking for the abbreviated
commit object name is not measurable because it's lost in the noise.
There is a caveat, however, when we are on an unborn branch, because
in that case HEAD doesn't point to a valid commit, hence the query for
the abbreviated commit object name fails. Therefore, '--short HEAD'
must be the last options to 'git rev-parse' in order to get all the
other necessary information for the prompt even on an unborn branch.
Furthermore, in that case, and in that case only, 'git rev-parse'
doesn't output the last line containing the abbreviated commit object
name, obviously, so we have to take care to only parse it if 'git
rev-parse' exited without any error.
Although there are tests already excercising __git_ps1() on unborn
branches, they all do so implicitly. Add a test that checks this
explicitly.
bash prompt: combine 'git rev-parse' executions in the main code path
There are a couple of '$(git rev-parse --<opt>)' command substitutions
in __git_ps1() and three of them are executed in the main code path:
- the first to get the path to the .git directory ('--git-dir'),
- the second to check whether we're inside the .git directory
('--is-inside-git-dir'),
- and the last, depending on the results of the second, either
* to check whether it's a bare repo ('--is-bare-repository'), or
* to check whether inside a work tree ('--is-inside-work-tree').
Naturally, this imposes the overhead of fork()ing three subshells and
fork()+exec()ing three git commands.
Combine these four 'git rev-parse' queries into a single one and use
bash parameter expansions to parse the combined output, i.e. to
separate the path to the .git directory from the true/false of
'--is-inside-git-dir', etc. This way we can eliminate two of the
three subshells and git commands.
bash prompt: use bash builtins to find out current branch
__git_ps1() runs the '$(git symbolic-ref HEAD)' command substitution
to find out whether we are on a branch and to find out the name of
that branch. This imposes the overhead of fork()ing a subshell and
fork()+exec()ing a git process.
Since HEAD is in most cases a single-line file and the symbolic ref
format is quite simple to recognize and parse, read and parse it using
only bash builtins, thereby sparing all that fork()+exec() overhead.
Don't display the git prompt if reading HEAD fails, because a readable
HEAD is required for a git repository. HEAD can also be a symlink
symbolic ref (due to 'core.preferSymlinkRefs'), so use bash builtins
for reading HEAD only when HEAD is not a symlink.
bash prompt: use bash builtins to find out rebase state
During an ongoing interactive rebase __git_ps1() finds out the name of
the rebased branch, the total number of patches and the number of the
current patch by executing a '$(cat .git/rebase-merge/<FILE>)' command
substitution for each. That is not quite the most efficient way to
read single line single word files, because it imposes the overhead of
fork()ing a subshell and fork()+exec()ing 'cat' several times.
Use the 'read' bash builtin instead to avoid those overheads.
bash prompt: run 'git rev-parse --git-dir' directly instead of __gitdir()
__git_ps1() finds out the path to the repository by using the
__gitdir() helper function. __gitdir() is basically just a wrapper
around 'git rev-parse --git-dir', extended with support for
recognizing a remote repository given as argument, to use the path
given on the command line, and with a few shortcuts to recognize a git
repository in cwd or at $GIT_DIR quickly without actually running 'git
rev-parse'. However, the former two is only necessary for the
completion script but makes no sense for the bash prompt, while the
latter shortcuts are performance optimizations __git_ps1() can do
without (they just avoid the overhead of fork()+exec()ing a git
process).
Run 'git rev-parse --git-dir' directly in __git_ps1(), because it will
allow this patch series to combine several $(git rev-parse ...)
command substitutions in the main code path, and the overall
performance benefit will far outweigh the loss of those few shortcuts
in __gitdir(). Furthermore, since __gitdir() is not needed anymore
for the prompt, remove it from the prompt script finally eliminating
its duplication between the prompt and completion scripts. Also
remove the comment from the completion script warning about this code
duplication.
bash prompt: print unique detached HEAD abbreviated object name
When describing a detached HEAD according to the $GIT_PS1_DESCRIBE
environment variable fails, __git_ps1() runs 'cut -c1-7 .git/HEAD' to
show the 7 hexdigits abbreviated commit object name in the prompt.
Obviously, this neither respects core.abbrev nor produces a unique
object name.
Fix this by using 'git rev-parse --short HEAD' instead and adjust the
corresponding test to use non-standard number of hexdigits.
completion, bash prompt: move __gitdir() tests to completion test suite
Currently __gitdir() is duplicated in the git completion and prompt
scripts, while its tests are in the prompt test suite. This patch
series is about to change __git_ps1() in a way that it won't need
__gitdir() anymore and __gitdir() will be removed from the prompt
script.
So move all __gitdir() tests from the prompt test suite over to the
completion test suite. Update the setup tests so that they perform
only those steps that are necessary for each test suite.
* mz/rebase-tests:
rebase topology tests: fix commit names on case-insensitive file systems
tests: move test for rebase messages from t3400 to t3406
t3406: modernize style
add tests for rebasing merged history
add tests for rebasing root
add tests for rebasing of empty commits
add tests for rebasing with patch-equivalence present
add simple tests of consistency across rebase types
* jk/apache-test-for-2.4:
lib-httpd/apache.conf: check version only after mod_version loads
t/lib-httpd/apache.conf: configure an MPM module for apache 2.4
t/lib-httpd/apache.conf: load compat access module in apache 2.4
t/lib-httpd/apache.conf: load extra auth modules in apache 2.4
t/lib-httpd/apache.conf: do not use LockFile in apache >= 2.4
* cm/remote-mediawiki-perlcritique: (31 commits)
git-remote-mediawiki: make error message more precise
git-remote-mediawiki: add a perlcritic rule in Makefile
git-remote-mediawiki: add a .perlcriticrc file
git-remote-mediawiki: clearly rewrite double dereference
git-remote-mediawiki: fix a typo ("mediwiki" instead of "mediawiki")
git-remote-mediawiki: put non-trivial numeric values in constants.
git-remote-mediawiki: don't use quotes for empty strings
git-remote-mediawiki: replace "unless" statements with negated "if" statements
git-remote-mediawiki: brace file handles for print for more clarity
git-remote-mediawiki: modify strings for a better coding-style
git-remote-mediawiki: put long code into a subroutine
git-remote-mediawiki: remove import of unused open2
git-remote-mediawiki: check return value of open
git-remote-mediawiki: assign a variable as undef and make proper indentation
git-remote-mediawiki: rename a variable ($last) which has the name of a keyword
git-remote-mediawiki: remove unused variable $entry
git-remote-mediawiki: turn double-negated expressions into simple expressions
git-remote-mediawiki: change the name of a variable
git-remote-mediawiki: add newline in the end of die() error messages
git-remote-mediawiki: change style in a regexp
...
* rr/rebase-autostash:
rebase: finish_rebase() in noop rebase
rebase: finish_rebase() in fast-forward rebase
rebase: guard against missing files in read_basic_state()
"git status" learned status.branch and status.short configuration
variables to use --branch and --short options by default (override
with --no-branch and --no-short options from the command line).
* jg/status-config:
status: introduce status.branch to enable --branch by default
status: introduce status.short to enable --short by default
add -i: add extra options at the right place in "diff" command line
Appending "--diff-algorithm=histogram" at the end of canned command
line for various modes of "diff" is correct for most of them but not
for "stash" that has a non-option already wired in, like so:
'stash' => {
DIFF => 'diff-index -p HEAD',
Appending an extra option after non-option may happen to work due to
overly lax command line parser, but that is not something we should
rely on. Instead, splice in the extra argument immediately after the
command name (i.e. 'diff-index', 'diff-files', etc.).
test-lib: support running tests under valgrind in parallel
With the new --valgrind-parallel=<n> option, we support running the
tests in a single test script under valgrind in parallel using 'n'
processes.
This really follows the dumbest approach possible, as follows:
* We spawn the test script 'n' times, using a throw-away
TEST_OUTPUT_DIRECTORY. Each of the instances is given options that
ensures that it only runs every n-th test under valgrind, but
together they cover the entire range.
* We add up the numbers from the individual tests, and provide the
usual output.
This is really a gross hack at this point, and should be improved. In
particular we should keep the actual outputs somewhere more easily
discoverable, and summarize them to the user.
Nevertheless, this is already workable and gives a speedup of more
than 2 on a dual-core (hyperthreaded) machine, using n=4. This is
expected since the overhead of valgrind is so big (on the order of 20x
under good conditions, and a large startup overhead at every git
invocation) that redundantly running the non-valgrind tests in between
is not that expensive.
Helped-by: Jeff King <peff@peff.net> Signed-off-by: Thomas Rast <trast@inf.ethz.ch> Signed-off-by: Junio C Hamano <gitster@pobox.com>
test-lib: allow prefixing a custom string before "ok N" etc.
This is not really meant for external use, and thus not documented. It
allows the next commit to neatly distinguish between sub-tests and the
main run.
The format is intentionally not valid TAP. The use in the next commit
would not result in anything valid either way, and it seems better to
make it obvious.
Signed-off-by: Thomas Rast <trast@inf.ethz.ch> Signed-off-by: Junio C Hamano <gitster@pobox.com>
test-lib: valgrind for only tests matching a pattern
With the new --valgrind-only=<pattern> option, one can enable
--valgrind at a per-test granularity, exactly analogous to
--verbose-only from the previous commit.
The options are wired such that --valgrind implies --verbose (as
before), but --valgrind-only=<pattern> implies
--verbose-only=<pattern> unless --verbose is also in effect.
Signed-off-by: Thomas Rast <trast@inf.ethz.ch> Signed-off-by: Junio C Hamano <gitster@pobox.com>
test-lib: verbose mode for only tests matching a pattern
With the new --verbose-only=<pattern> option, one can enable --verbose
at a per-test granularity. The pattern is matched against the test
number, e.g.
./t0000-basic.sh --verbose-only='2[0-2]'
to see only the full output of test 20-22, while showing the rest in the
one-liner format.
As suggested by Jeff King, this takes care to wrap the entire
test_expect_* block, but nothing else, in the verbose toggling. We
can use the test_start/end functions from the previous commit for the
purpose.
This is arguably not *too* useful on its own, but makes the next patch
easier to follow.
Helped-by: Jeff King <peff@peff.net> Signed-off-by: Thomas Rast <trast@inf.ethz.ch> Signed-off-by: Junio C Hamano <gitster@pobox.com>
t0000 contains some light self-tests of test-lib.sh, but --verbose was
not covered. Add a test.
The only catch is that the presence of a test harness influences the
output (specifically, the presence of some empty lines). So we need
to unset TEST_HARNESS or set it to a known value. Leaving it unset
leads to spurious test failures in the final summary, which come from
the subtest. So we always set it.
Signed-off-by: Thomas Rast <trast@inf.ethz.ch> Signed-off-by: Junio C Hamano <gitster@pobox.com>
user-manual: Update download size for Git and the kernel
They've grown since d19fbc3 (Documentation: add git user's manual,
2007-01-07) when the stats were initially added. I've rounded
download sizes up to the nearest multiple of ten MiB to decrease the
precision and give a bit of growing room. Exact sizes:
t/t5505-remote: test multiple push/pull in remotes-file
Extend the test "migrate a remote from named file in $GIT_DIR/remotes"
to test that multiple "Push:" and "Pull:" lines in the remotes-file
works as expected.
Signed-off-by: Ramkumar Ramachandra <artagnon@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Replace the <repository> paragraph containing specific references to
$GIT_DIR/branches and "." with a generic urls-or-remotes paragraph
referencing the relevant sections in the git-fetch(1) manpage.
Signed-off-by: Ramkumar Ramachandra <artagnon@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Under the EXAMPLES section, there is one invocation on the git.git
repository that attempts to list the refs master, pu, and rc. The ref
rc does not exist in today's repository, so remove it. Among other
things, this example demonstrates that the "<refs>..." argument is
simply a filter; requesting a non-existent ref is not an error.
Signed-off-by: Ramkumar Ramachandra <artagnon@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add one more test similar to "migrate a remote from named file in
$GIT_DIR/branches" to check that a url with a # can be used to specify
the branch name (as opposed to the constant "master").
Signed-off-by: Ramkumar Ramachandra <artagnon@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The first line of the function checks that the remote-name contains a
slash ('/'), and sets the "slash" variable accordingly. The only caller
of read_branches_file() is remote_get_1(); the calling codepath is
guarded by valid_remote_nick(), which checks that the remote does not
contain a slash. Therefore, the "slash" variable can never be set:
remove the dead code that assumes otherwise.
Signed-off-by: Ramkumar Ramachandra <artagnon@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
t/t5505-remote: test push-refspec in branches-file
The test "migrate a remote from named file in $GIT_DIR/branches" reads
the branches-file, but only checks that the url and fetch-refspec are
set correctly. Check that the push-refspec is also set correctly.
Signed-off-by: Ramkumar Ramachandra <artagnon@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Modernize the style of all tests throughout the file:
- Remove spurious blank lines.
- Indent the test body.
- Make sure that all lines end with &&, to make it easier to spot
breaks in the chain.
- When executing something in a subshell, put the parenthesis on
separate lines and indent the body. Also make sure that the
first statement in the subshell is a 'cd'.
- When redirecting input or output, do not use SP between
redirection operator and the target filename.
- Use the <<-\EOF and <<-EOF forms of heredoc, not <<EOF, when the
command is indented and the heredoc text itself does not have to
have a leading tab.
Signed-off-by: Ramkumar Ramachandra <artagnon@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The 3.x tree has been out for a while now. The -2.6 repository name
survived the initial release [1], but kernel.org now only lists
'linux.git' (for aegl as well as torvalds) [2].
[1]: http://article.gmane.org/gmane.linux.kernel/1147422
On 2011-05-30 01:47:57 GMT, Linus Torvalds wrote:
> ... yes, that means that my git tree is still called
> "linux-2.6.git" on kernel.org.
[2]: http://git.kernel.org/cgit/
Signed-off-by: W. Trevor King <wking@tremily.us> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Documentation: Update the NFS remote examples to use the staging repo
linux-nfs.org seems to have restructured their repository layout since 8391c60 (git-remote.txt: fix example url, 2007-11-02), and Bruce's
repo is now at git://git.linux-nfs.org/projects/bfields/linux.git.
Bruce also has a more richer internal branch structure (master,
everything, for-3.1, ...), so updating the existing example to use his
current repo may be confusing.
To simplify, I've replaced the NFS repo with Greg's staging repo.
I've also updated the output of the surrounding commands to match the
output of a current run through.
Signed-off-by: W. Trevor King <wking@tremily.us> Signed-off-by: Junio C Hamano <gitster@pobox.com>
doc/clone: Pick more compelling paths for the --reference example
There may be times when using one of your local repositories as a
reference for a new clone make sense, but the implied version-bump in
the old example isn't one of them. I think a more intuitive example
is multi-user system with a central reference clone, and the new paths
hint at this use case.
Signed-off-by: W. Trevor King <wking@tremily.us> Signed-off-by: Junio C Hamano <gitster@pobox.com>
lib-httpd/apache.conf: check version only after mod_version loads
Commit 0442743 introduced an <IfVersion> directive near the
top of the apache config file. However, at that point we
have not yet checked for and loaded the mod_version module.
This means that the directive will behave oddly if
mod_version is dynamically loaded, failing to match when it
should.
We can fix this by moving the whole block below the
LoadModule directive for mod_version.
Reported-by: Brian Gernhardt <brian@gernhardtsoftware.com> Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Tweak the --topo/date-order test vector a bit and mark the author
dates of two commits (a2 and a3) earlier than their own committer
dates, making them much older than other commits that are made on
parallel branches to simulate the case where a long running topic
was rebased recently.
They will show up as recent in the --date-order output due to their
timestamps, but they appear a lot later in the --author-date-order
output, even though their committer timestamp says otherwise.
The "--date-order" output is a slight twist of "--topo-order" in
that commits from parallel histories are shown in their committer
date order without an attempt to clump commits from a single line
of history together like --topo-order does.
topology tests: teach a helper to take abbreviated timestamps
The on_committer_date helper in t/lib-t6000 is used in t6002 and
t6003 with timestamps on a single day within a single minute
(i.e. 1971-08-16 00:00) and the tests repeat this over and over.
The actual value of the timestamp, however, does not matter very
much; only their relative ordering does.
Introduce another helper to expand only the suffix of the timestamp
to a full timestamp to make the lines shorter, and use it in this
helper. Also, because all the commits in the test are made with
specific GIT_COMMITTER_DATE, stop unsetting it at the end of the
helper.
We'll be specifying the author timestamp to these test commits in a
later patch, which will be helped with this change.
Also remove a test that was commented-out from t6003; it used to
test a commit with the same parent listed twice, which was allowed
by mistake but was fixed long time ago.
While both GUI and console Cygwin browsers do exist, anecdotal evidence
suggests most users rely on their native Windows browser. cygstart,
which is a long-standing part of the base Cygwin installation, will
cause the page to be opened in the default Windows browser (the one
registered to open .html files).
Signed-off-by: Yaakov Selkowitz <yselkowitz@users.sourceforge.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
transport-helper: be quiet on read errors from helpers
Prior to commit 81d340d4, we did not print any error message
if a remote transport helper died unexpectedly. If a helper
did not print any error message (e.g., because it crashed),
the user could be left confused. That commit tried to
rectify the situation by printing a note that the helper
exited unexpectedly.
However, this makes a much more common case worse: when a
helper does die with a useful message, we print the extra
"Reading from 'git-remote-foo failed" message. This can also
end up confusing users, as they may not even know what
remote helpers are (e.g., the fact that http support comes
through git-remote-https is purely an implementation detail
that most users do not know or care about).
Since we do not have a good way of knowing whether the
helper printed a useful error, and since the common failure
mode is for it to do so, let's default to remaining quiet.
Debuggers can dig further by setting GIT_TRANSPORT_HELPER_DEBUG.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2901bbe (apply: free patch->{def,old,new}_name fields, 2012-03-21)
cleaned up the memory management of filenames in the patches, but
forgot that find_name_traditional() can return NULL as a way of saying
"I couldn't find a name".
That NULL unfortunately gets passed into xstrdup() next, resulting in
a segfault. Use null_strdup() so as to safely propagate the null,
which will let us emit the correct error message.
Reported-by: DevHC on #git Signed-off-by: Thomas Rast <trast@inf.ethz.ch> Signed-off-by: Junio C Hamano <gitster@pobox.com>
* fc/makefile:
Makefile: use $^ to avoid listing prerequisites on the command line
build: do not install git-remote-testgit
build: generate and clean test scripts
Many tests that check the behaviour of symbolic links stored in the
index or the tree objects do not have to be skipped on a filesystem
that lack symbolic link support.
* js/test-ln-s-add:
t4011: remove SYMLINKS prerequisite
t6035: use test_ln_s_add to remove SYMLINKS prerequisite
t3509, t4023, t4114: use test_ln_s_add to remove SYMLINKS prerequisite
t3100: use test_ln_s_add to remove SYMLINKS prerequisite
t3030: use test_ln_s_add to remove SYMLINKS prerequisite
t0000: use test_ln_s_add to remove SYMLINKS prerequisite
tests: use test_ln_s_add to remove SYMLINKS prerequisite (trivial cases)
tests: introduce test_ln_s_add
t3010: modernize style
test-chmtime: Fix exit code on Windows
pull: merge into unborn by fast-forwarding from empty tree
The logic for pulling into an unborn branch was originally
designed to be used on a newly-initialized repository
(d09e79c, git-pull: allow pulling into an empty repository,
2006-11-16). It thus did not initially deal with
uncommitted changes in the unborn branch. The case of an
_unstaged_ untracked file was fixed by 4b3ffe5 (pull: do not
clobber untracked files on initial pull, 2011-03-25).
However, it still clobbered existing staged files, both when
the file exists in the merged commit (it will be
overwritten), and when it does not (it will be deleted).
We fix this by doing a two-way merge, where the "current"
side of the merge is an empty tree, and the "target" side is
HEAD (already updated to FETCH_HEAD at this point). This
amounts to claiming that all work in the index was done vs.
an empty tree, and thus all content of the index is
precious.
Note that this use of read-tree just gives us protection
against overwriting index and working tree changes. It will
not actually result in a 3-way merge conflict in the index.
This is fine, as this is a rare situation, and the conflict
would not be interesting anyway (it must, by definition, be
an add/add conflict with the whole content conflicting). And
it makes it simpler for the user to recover, as they have no
HEAD to "git reset" back to.
Reported-by: Stefan Schüßler <mail@stefanschuessler.de> Signed-off-by: Thomas Rast <trast@inf.ethz.ch> Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When commit d09e79c taught git to pull into an unborn
branch, it first updated the unborn branch to point at the
pulled commit, and then used read-tree to update the index
and working tree. That ordering made sense, since any
failure of the latter step would be due to filesystem
errors, and one could then recover with "git reset --hard".
Later, commit 4b3ffe5 added extra safety for existing files
in the working tree by asking read-tree to bail out when it
would overwrite such a file. This error mode is much less
"your pull failed due to random errors" and more like "we
reject this pull because it would lose data". In that case,
it makes sense not to update the HEAD ref, just as a regular
rejected merge would do.
This patch reverses the order of the update-ref and
read-tree calls, so that we do not touch the HEAD ref at all if a
merge is rejected. This also means that we would not update
HEAD in case of a transient filesystem error, but those are
presumably less rare (and one can still recover by repeating
the pull, or by accessing FETCH_HEAD directly).
While we're reorganizing the code, we can drop the "exit 1"
from the end of our command chain. We exit immediately
either way, and just calling exit without an argument will
use the exit code from the last command.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
for_each_ref: load all loose refs before packed refs
If we are iterating through the refs using for_each_ref (or
any of its sister functions), we can get into a race
condition with a simultaneous "pack-refs --prune" that looks
like this:
0. We have a large number of loose refs, and a few packed
refs. refs/heads/z/foo is loose, with no matching entry
in the packed-refs file.
1. Process A starts iterating through the refs. It loads
the packed-refs file from disk, then starts lazily
traversing through the loose ref directories.
2. Process B, running "pack-refs --prune", writes out the
new packed-refs file. It then deletes the newly packed
refs, including refs/heads/z/foo.
3. Meanwhile, process A has finally gotten to
refs/heads/z (it traverses alphabetically). It
descends, but finds nothing there. It checks its
cached view of the packed-refs file, but it does not
mention anything in "refs/heads/z/" at all (it predates
the new file written by B in step 2).
The traversal completes successfully without mentioning
refs/heads/z/foo at all (the name, of course, isn't
important; but the more refs you have and the farther down
the alphabetical list a ref is, the more likely it is to hit
the race). If refs/heads/z/foo did exist in the packed refs
file at state 0, we would see an entry for it, but it would
show whatever sha1 the ref had the last time it was packed
(which could be an arbitrarily long time ago).
This can be especially dangerous when process A is "git
prune", as it means our set of reachable tips will be
incomplete, and we may erroneously prune objects reachable
from that tip (the same thing can happen if "repack -ad" is
used, as it simply drops unreachable objects that are
packed).
This patch solves it by loading all of the loose refs for
our traversal into our in-memory cache, and then refreshing
the packed-refs cache. Because a pack-refs writer will
always put the new packed-refs file into place before
starting the prune, we know that any loose refs we fail to
see will either truly be missing, or will have already been
put in the packed-refs file by the time we refresh.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
get_packed_ref_cache: reload packed-refs file when it changes
Once we read the packed-refs file into memory, we cache it
to save work on future ref lookups. However, our cache may
be out of date with respect to what is on disk if another
process is simultaneously packing the refs. Normally it
is acceptable for us to be a little out of date, since there
is no guarantee whether we read the file before or after the
simultaneous update. However, there is an important special
case: our packed-refs file must be up to date with respect
to any loose refs we read. Otherwise, we risk the following
race condition:
0. There exists a loose ref refs/heads/master.
1. Process A starts and looks up the ref "master". It
first checks $GIT_DIR/master, which does not exist. It
then loads (and caches) the packed-refs file to see if
"master" exists in it, which it does not.
2. Meanwhile, process B runs "pack-refs --all --prune". It
creates a new packed-refs file which contains
refs/heads/master, and removes the loose copy at
$GIT_DIR/refs/heads/master.
3. Process A continues its lookup, and eventually tries
$GIT_DIR/refs/heads/master. It sees that the loose ref
is missing, and falls back to the packed-refs file. But
it examines its cached version, which does not have
refs/heads/master. After trying a few other prefixes,
it reports master as a non-existent ref.
There are many variants (e.g., step 1 may involve process A
looking up another ref entirely, so even a fully qualified
refname can fail). One of the most interesting ones is if
"refs/heads/master" is already packed. In that case process
A will not see it as missing, but rather will report
whatever value happened to be in the packed-refs file before
process B repacked (which might be an arbitrarily old
value).
We can fix this by making sure we reload the packed-refs
file from disk after looking at any loose refs. That's
unacceptably slow, so we can check its stat()-validity as a
proxy, and read it only when it appears to have changed.
Reading the packed-refs file after performing any loose-ref
system calls is sufficient because we know the ordering of
the pack-refs process: it always makes sure the newly
written packed-refs file is installed into place before
pruning any loose refs. As long as those operations by B
appear in their executed order to process A, by the time A
sees the missing loose ref, the new packed-refs file must be
in place.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
It can sometimes be useful to know whether a path in the
filesystem has been updated without going to the work of
opening and re-reading its content. We trust the stat()
information on disk already to handle index updates, and we
can use the same trick here.
This patch introduces a "stat_validity" struct which
encapsulates the concept of checking the stat-freshness of a
file. It is implemented on top of "struct stat_data" to
reuse the logic about which stat entries to trust for a
particular platform, but hides the complexity behind two
simple functions: check and update.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add public functions fill_stat_data() and match_stat_data() to work
with it. This infrastructure will later be used to check the validity
of other types of file.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
do_for_each_entry(): increment the packed refs cache refcount
This function calls a user-supplied callback function which could do
something that causes the packed refs cache to be invalidated. So
acquire a reference count on the data structure to prevent our copy
from being freed while we are iterating over it.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
refs: manage lifetime of packed refs cache via reference counting
In struct packed_ref_cache, keep a count of the number of users of the
data structure. Only free the packed ref cache when the reference
count goes to zero rather than when the packed ref cache is cleared.
This mechanism will be used to prevent the cache data structure from
being freed while it is being iterated over.
So far, only the reference in struct ref_cache::packed is counted;
other users will be adjusted in separate commits.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
refs: implement simple transactions for the packed-refs file
Handle simple transactions for the packed-refs file at the
packed_ref_cache level via new functions lock_packed_refs(),
commit_packed_refs(), and rollback_packed_refs().
Only allow the packed ref cache to be modified (via add_packed_ref())
while the packed refs file is locked.
Change clone to add the new references within a transaction.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
refs: wrap the packed refs cache in a level of indirection
As we know, we can solve any problem in this manner. In this case,
the problem is to avoid freeing a packed refs cache while somebody is
using it. So add a level of indirection as a prelude to
reference-counting the packed refs cache.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
repack_without_ref(): split list curation and entry writing
The repack_without_ref() function first removes the deleted ref from
the internal packed-refs list, then writes the packed-refs list to
disk, omitting any broken or stale entries. This patch splits that
second step into multiple passes:
* collect the list of refnames that should be deleted from packed_refs
* delete those refnames from the cache
* write the remainder to the packed-refs file
The purpose of this change is to make the "write the remainder" part
reusable.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
send-email: sanitize author when writing From line
sender is now sanitized, but we didn't sanitize author when checking
whether From: line is needed in the message body.
As a result git started writing duplicate From: lines when author
matched sender and has utf8 characters.
Reported-by: SZEDER Gábor <szeder@ira.uka.de> Signed-off-by: Michael S. Tsirkin <mst@redhat.com> Tested-by: SZEDER Gábor <szeder@ira.uka.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Verify that author name is not duplicated if it matches sender, even
if it is in utf8 (the test expects a failure that will be fixed in
the next patch).
Signed-off-by: Michael S. Tsirkin <mst@redhat.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The goal of the patch is to introduce the GNU diff
-B/--ignore-blank-lines as closely as possible. The short option is not
available because it's already used for "break-rewrites".
When this option is used, git-diff will not create hunks that simply
add or remove empty lines, but will still show empty lines
addition/suppression if they are close enough to "valuable" changes.
There are two differences between this option and GNU diff -B option:
- GNU diff doesn't have "--inter-hunk-context", so this must be handled
- The following sequence looks like a bug (context is displayed twice):
So here is a more thorough description of the option:
- real changes are interesting
- blank lines that are close enough (less than context size) to
interesting changes are considered interesting (recursive definition)
- "context" lines are used around each hunk of interesting changes
- If two hunks are separated by less than "inter-hunk-context", they
will be merged into one.
The implementation does the "interesting changes selection" in a single
pass.
Signed-off-by: Antoine Pelisse <apelisse@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>