"git merge-tree" used to mishandle "both sides added" conflict with
its own "create a fake ancestor file that has the common parts of
what both sides have added and do a 3-way merge" logic; this has
been updated to use the usual "3-way merge with an empty blob as
the fake common ancestor file" approach used in the rest of the
system.
* jk/no-diff-emit-common:
xdiff: drop XDL_EMIT_COMMON
merge-tree: drop generate_common strategy
merge-one-file: use empty blob for add/add base
Merge branch 'nd/dwim-wildcards-as-pathspecs' into maint
"git show 'HEAD:Foo[BAR]Baz'" did not interpret the argument as a
rev, i.e. the object named by the the pathname with wildcard
characters in a tree object.
Handling of errors while writing into our internal asynchronous
process has been made more robust, which reduces flakiness in our
tests.
* jk/epipe-in-async:
t5504: handle expected output from SIGPIPE death
test_must_fail: report number of unexpected signal
fetch-pack: ignore SIGPIPE in sideband demuxer
write_or_die: handle EPIPE in async threads
Many codepaths forget to check return value from git_config_set();
the function is made to die() to make sure we do not proceed when
setting a configuration variable failed.
* ps/config-error:
config: rename git_config_set_or_die to git_config_set
config: rename git_config_set to git_config_set_gently
compat: die when unable to set core.precomposeunicode
sequencer: die on config error when saving replay opts
init-db: die on config errors when initializing empty repo
clone: die on config error in cmd_clone
remote: die on config error when manipulating remotes
remote: die on config error when setting/adding branches
remote: die on config error when setting URL
submodule--helper: die on config error when cloning module
submodule: die on config error when linking modules
branch: die on config error when editing branch description
branch: die on config error when unsetting upstream
branch: report errors in tracking branch setup
config: introduce set_or_die wrappers
Traditionally, the tests that try commands that work on the
contents in the working tree were named with "worktree" in their
filenames, but with the recent addition of "git worktree"
subcommand, whose tests are also named similarly, it has become
harder to tell them apart. The traditional tests have been renamed
to use "work-tree" instead in an attempt to differentiate them.
* mg/work-tree-tests:
tests: rename work-tree tests to *work-tree*
Commit 8bf4bec (add "ok=sigpipe" to test_must_fail and use
it to fix flaky tests, 2015-11-27) taught t5504 to handle
"git push" racily exiting with SIGPIPE rather than failing.
However, one of the tests checks the output of the command,
as well. In the SIGPIPE case, we will not have produced any
output. If we want the test to be truly non-flaky, we have
to accept either output.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
test_must_fail: report number of unexpected signal
If a command is marked as test_must_fail but dies with a
signal, we consider that a problem and report the error to
stderr. However, we don't say _which_ signal; knowing that
can make debugging easier. Let's share as much as we know.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
If the other side feeds us a bogus pack, index-pack (or
unpack-objects) may die early, before consuming all of its
input. As a result, the sideband demuxer may get SIGPIPE
(racily, depending on whether our data made it into the pipe
buffer or not). If this happens and we are compiled with
pthread support, it will take down the main thread, too.
This isn't the end of the world, as the main process will
just die() anyway when it sees index-pack failed. But it
does mean we don't get a chance to say "fatal: index-pack
failed" or similar. And it also means that we racily fail
t5504, as we sometimes die() and sometimes are killed by
SIGPIPE.
So let's ignore SIGPIPE while demuxing the sideband. We are
already careful to check the return value of write(), so we
won't waste time writing to a broken pipe. The caller will
notice the error return from the async thread, though in
practice we don't even get that far, as we die() as soon as
we see that index-pack failed.
The non-sideband case is already fine; we let index-pack
read straight from the socket, so there is no SIGPIPE at
all. Technically the non-threaded async case is also OK
without this (the forked async process gets SIGPIPE), but
it's not worth distinguishing from the threaded case here.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When write_or_die() sees EPIPE, it treats it specially by
converting it into a SIGPIPE death. We obviously cannot
ignore it, as the write has failed and the caller expects us
to die. But likewise, we cannot just call die(), because
printing any message at all would be a nuisance during
normal operations.
However, this is a problem if write_or_die() is called from
a thread. Our raised signal ends up killing the whole
process, when logically we just need to kill the thread
(after all, if we are ignoring SIGPIPE, there is good reason
to think that the main thread is expecting to handle it).
Inside an async thread, the die() code already does the
right thing, because we use our custom die_async() routine,
which calls pthread_join(). So ideally we would piggy-back
on that, and simply call:
die_quietly_with_code(141);
or similar. But refactoring the die code to do this is
surprisingly non-trivial. The die_routines themselves handle
both printing and the decision of the exit code. Every one
of them would have to be modified to take new parameters for
the code, and to tell us to be quiet.
Instead, we can just teach write_or_die() to check for the
async case and handle it specially. We do have to build an
interface to abstract the async exit, but it's simple and
self-contained. If we had many call-sites that wanted to do
this die_quietly_with_code(), this approach wouldn't scale
as well, but we don't. This is the only place where do this
weird exit trick.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
There are no more callers that use this mode, and none
likely to be added (as our xdl_merge() eliminates the common
use of it for generating 3-way merge bases).
This is effectively a revert of a9ed376 (xdiff: generate
"anti-diffs" aka what is common to two files, 2006-06-28),
though of course trying to revert that ancient commit
directly produces many textual conflicts.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When merge_blobs sees an add/add conflict, it tries to
create a virtual base object for the 3-way merge that
consists of the common lines of each file. It inherited this
strategy from merge-one-file in 0c79938 (Improved three-way
blob merging code, 2006-06-28), and the point is to minimize
the size of the conflict hunks. That commit talks about "if
libxdiff were to ever grow a compatible three-way merge, it
could probably be directly plugged in".
That has long since happened. So as with merge-one-file in
the previous commit, this extra step is no longer necessary.
Our 3-way merge code is smart enough to do the minimizing
itself if we simply feed it an empty base, which is what the
more modern merge-recursive strategy already does.
Not only does this let us drop some code, but it removes an
overflow bug in generate_common_file(). We allocate a buffer
as large as the smallest of the two blobs, under the
assumption that there cannot be more common content than
what is in the smaller blob. However, xdiff may feed us
more: if neither file ends in a newline, it feeds us the
"\nNo newline at end of file" marker as common content, and
we write it into the output. If the differences between the
files are small than that string, we overflow the output
buffer. This patch solves it by simply dropping the buggy
code entirely.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When we see an add/add conflict on a file, we generate the
conflicted content by doing a 3-way merge with a "virtual"
base consisting of the common lines of the two sides. This
strategy dates back to cb93c19 (merge-one-file: use common
as base, instead of emptiness., 2005-11-09).
Back then, the next step was to call rcs merge to generate
the 3-way conflicts. Using the virtual base produced much
better results, as rcs merge does not attempt to minimize
the hunks. As a result, you'd get a conflict with the
entirety of the files on either side.
Since then, though, we've switched to using git-merge-file,
which uses xdiff's "zealous" merge. This will find the
minimal hunks even with just the simple, empty base.
Let's switch to using that empty base. It's simpler, more
efficient, and reduces our dependencies (we no longer need a
working diff binary). It's also how the merge-recursive
strategy handles this same case.
We can almost get rid of git-sh-setup's create_virtual_base,
but we don't here, for two reasons:
1. The functions in git-sh-setup are part of our public
interface, so it's possible somebody is depending on
it. We'd at least need to deprecate it first.
2. It's also used by mergetool's p4merge driver. It's
unknown whether its 3-way merge is as capable as git's;
if not, then it is benefiting from the function.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Paths that have been told the index about with "add -N" are not
quite yet in the index, but a few commands behaved as if they
already are in a harmful way.
* nd/ita-cleanup:
grep: make it clear i-t-a entries are ignored
add and use a convenience macro ce_intent_to_add()
blame: remove obsolete comment
Merge branch 'dw/mergetool-vim-window-shuffle' into maint
The vimdiff backend for "git mergetool" has been tweaked to arrange
and number buffers in the order that would match the expectation of
majority of people who read left to right, then top down and assign
buffers 1 2 3 4 "mentally" to local base remote merge windows based
on that order.
* dw/mergetool-vim-window-shuffle:
mergetool: reorder vim/gvim buffers in three-way diffs
The memory allocation scheme for the textconv interface is a
bit tricky, and not well documented. It was originally
designed as an internal part of diff.c (matching
fill_mmfile), but gradually was made public.
Refactoring it is difficult, but we can at least improve the
situation by documenting the intended flow and enforcing it
with an in-code assertion.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
config: rename git_config_set to git_config_set_gently
The desired default behavior for `git_config_set` is to die
whenever an error occurs. Dying is the default for a lot of
internal functions when failures occur and is in this case the
right thing to do for most callers as otherwise we might run into
inconsistent repositories without noticing.
As some code may rely on the actual return values for
`git_config_set` we still require the ability to invoke these
functions without aborting. Rename the existing `git_config_set`
functions to `git_config_set_gently` to keep them available for
those callers.
Signed-off-by: Patrick Steinhardt <ps@pks.im> Signed-off-by: Junio C Hamano <gitster@pobox.com>
compat: die when unable to set core.precomposeunicode
When calling `git_config_set` to set 'core.precomposeunicode' we
ignore the return value of the function, which may indicate that
we were unable to write the value back to disk. As the function
is only called by init-db we can and should die when an error
occurs.
Signed-off-by: Patrick Steinhardt <ps@pks.im> Signed-off-by: Junio C Hamano <gitster@pobox.com>
sequencer: die on config error when saving replay opts
When we start picking a range of revisions we save the replay
options that are required to restore state when interrupting and
later continuing picking the revisions. However, we do not check
the return values of the `git_config_set` functions, which may
lead us to store incomplete information. As this may lead us to
fail when trying to continue the sequence the error can be fatal.
Fix this by dying immediately when we are unable to write back
any replay option.
Signed-off-by: Patrick Steinhardt <ps@pks.im> Signed-off-by: Junio C Hamano <gitster@pobox.com>
init-db: die on config errors when initializing empty repo
When creating an empty repository with `git init-db` we do not
check for error codes returned by `git_config_set` functions.
This may cause the user to end up with an inconsistent repository
without any indication for the user.
Fix this problem by dying early with an error message when we are
unable to write the configuration files to disk.
Signed-off-by: Patrick Steinhardt <ps@pks.im> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The clone command does not check for error codes returned by
`git_config_set` functions. This may cause the user to end up
with an inconsistent repository without any indication with what
went wrong.
Fix this problem by dying with an error message when we are
unable to write the configuration files to disk.
Signed-off-by: Patrick Steinhardt <ps@pks.im> Signed-off-by: Junio C Hamano <gitster@pobox.com>
remote: die on config error when manipulating remotes
When manipulating remotes we try to set various configuration
values without checking if the values were persisted correctly,
possibly leaving the remote in an inconsistent state.
Fix this issue by dying early and notifying the user about the
error.
Signed-off-by: Patrick Steinhardt <ps@pks.im> Signed-off-by: Junio C Hamano <gitster@pobox.com>
remote: die on config error when setting/adding branches
When we add or set new branches (e.g. by `git remote add -f` or
`git remote set-branches`) we do not check for error codes when
writing the branches to the configuration file. When persisting
the configuration failed we are left with a remote that has none
or not all of the branches that should have been set without
notifying the user.
Fix this issue by dying early on configuration error.
Signed-off-by: Patrick Steinhardt <ps@pks.im> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When invoking `git-remote --set-url` we do not check the return
value when writing the actual new URL to the configuration file,
pretending to the user that the configuration has been set while
it was in fact not persisted.
Fix this problem by dying early when setting the config fails.
Signed-off-by: Patrick Steinhardt <ps@pks.im> Signed-off-by: Junio C Hamano <gitster@pobox.com>
submodule--helper: die on config error when cloning module
When setting the 'core.worktree' option for a newly cloned
submodule we ignore the return value of `git_config_set_in_file`.
As this leaves the submodule in an inconsistent state, we instead
want to inform the user that something has gone wrong by printing
an error and aborting the program.
Signed-off-by: Patrick Steinhardt <ps@pks.im> Signed-off-by: Junio C Hamano <gitster@pobox.com>
submodule: die on config error when linking modules
When trying to connect a submodule with its corresponding
repository in '.git/modules' we try to set the core.worktree
setting in the submodule, which may fail due to an error
encountered in `git_config_set_in_file`.
The function is used in the git-mv command when trying to move a
submodule to another location. We already die when renaming a
file fails but do not pay attention to the case where updating
the connection between submodule and its repository fails. As
this leaves the repository in an inconsistent state, as well,
abort the program by dying early and presenting the failure to
the user.
Signed-off-by: Patrick Steinhardt <ps@pks.im> Signed-off-by: Junio C Hamano <gitster@pobox.com>
branch: die on config error when unsetting upstream
When we try to unset upstream configurations we do not check
return codes for the `git_config_set` functions. As those may
indicate that we were unable to unset the respective
configuration we may exit successfully without any error message
while in fact the upstream configuration was not unset.
Fix this by dying with an error message when we cannot unset the
configuration.
Signed-off-by: Patrick Steinhardt <ps@pks.im> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When setting up a new tracking branch fails due to issues with
the configuration file we do not report any errors to the user
and pretend setting the tracking branch succeeded.
Setting up the tracking branch is handled by the
`install_branch_config` function. We do not want to simply die
there as the function is not only invoked when explicitly setting
upstream information with `git branch --set-upstream-to=`, but
also by `git push --set-upstream` and `git clone`. While it is
reasonable to die in the explict first case, we would lose
information in the latter two cases, so we only print the error
message but continue the program as usual.
Signed-off-by: Patrick Steinhardt <ps@pks.im> Signed-off-by: Junio C Hamano <gitster@pobox.com>
"Work tree" or "working tree" is the name of a checked out tree,
"worktree" the name of the command which manages several working trees.
The naming of tests mixes these two, currently:
Merge branch 'js/test-lib-windows-emulated-yes' into maint
The emulated "yes" command used in our test scripts has been
tweaked not to spend too much time generating unnecessary output
that is not used, to help those who test on Windows where it would
not stop until it fills the pipe buffer due to lack of SIGPIPE.
* js/test-lib-windows-emulated-yes:
test-lib: limit the output of the yes utility
Merge branch 'nd/do-not-move-worktree-manually' into maint
"git worktree" had a broken code that attempted to auto-fix
possible inconsistency that results from end-users moving a
worktree to different places without telling Git (the original
repository needs to maintain backpointers to its worktrees, but
"mv" run by end-users who are not familiar with that fact will
obviously not adjust them), which actually made things worse
when triggered.
The low-level merge machinery has been taught to use CRLF line
termination when inserting conflict markers to merged contents that
are themselves CRLF line-terminated.
* js/xmerge-marker-eol:
merge-file: ensure that conflict sections match eol style
merge-file: let conflict markers match end-of-line style of the context
The 'v'iew subcommand of the interactive mode of "git am -i" was
broken by the rewrite to C we did at around 2.6.0 timeframe at 7ff26832 (builtin-am: implement -i/--interactive, 2015-08-04); we
used to spawn the pager via the shell, accepting things like
PAGER='less -S'
in the environment, but the rewrite forgot and tried to directly
spawn a command whose name is the entire string.
The previous refactoring of the new helper function makes it easier
for us to do the right thing.
pager: factor out a helper to prepare a child process to run the pager
When running a pager, we need to run the program git_pager() gave
us, but we need to make sure we spawn it via the shell (i.e. it is
valid to say PAGER='less -S', for example) and give default values
to $LESS and $LV environment variables. Factor out these details
to a separate helper function.
A lot of call-sites for the existing family of `git_config_set`
functions do not check for errors that may occur, e.g. when the
configuration file is locked. In many cases we simply want to die
when such a situation arises.
Introduce wrappers that will cause the program to die in those
cases. These wrappers are temporary only to ease the transition
to let `git_config_set` die by default. They will be removed
later on when `git_config_set` itself has been replaced by
`git_config_set_gently`.
Signed-off-by: Patrick Steinhardt <ps@pks.im> Signed-off-by: Junio C Hamano <gitster@pobox.com>
remote-curl: include curl_errorstr on SSL setup failures
For curl error 35 (CURLE_SSL_CONNECT_ERROR) users need the
additional text stored in CURLOPT_ERRORBUFFER to debug why
the connection did not start. This is curl_errorstr inside
of http.c, so include that in the message if it is non-empty.
Sometimes HTTP response codes aren't yet available, such as
when the SSL setup fails. Don't include HTTP 0 in the message.
Signed-off-by: Shawn Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
rev-parse: take prefix into account in --git-common-dir
Most of the time, get_git_common_dir() returns an absolute path so
prefix is irrelevant. If it returns a relative path (e.g. from the
main worktree) then prefixing is required.
Noticed-by: Mike Hommey <mh@glandium.org> Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
mergetool: reorder vim/gvim buffers in three-way diffs
When invoking default (g)vimdiff three-way merge, the merged file is
loaded as the first buffer but moved to the bottom as the fourth window.
This causes a disconnect between vim commands that operate on window
positions (e.g. CTRL-W_w) and those that operate on buffer index (e.g.
do/dp).
This change reorders the buffers to have the same index as windows while
keeping the cursor default to the merged result as the bottom window.
Signed-off-by: Dickson Wong <dicksonwong@gmail.com> Tested-by: Michael J Gruber <git@drmicha.warpmail.net> Acked-by: David Aguilar <davvid@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The get_sha1() function generally returns an error code
rather than dying, and we sometimes speculatively call it
with something that may be a revision or a pathspec, in
order to see which one it might be.
If it sees a bogus ":/" search string, though, it complains,
without giving the caller the opportunity to recover. We can
demonstrate this in t6133 by looking for ":/*.t", which
should mean "*.t at the root of the tree", but instead dies
because of the invalid regex (the "*" has nothing to operate
on).
We can fix this by returning an error rather than calling
die(). Unfortunately, the tradeoff is that the error message
is slightly worse in cases where we _do_ know we have a rev.
E.g., running "git log ':/*.t' --" before yielded:
fatal: Invalid search pattern: *.t
and now we get only:
fatal: bad revision ':/*.t'
There's not a simple way to fix this short of passing a
"quiet" flag all the way through the get_sha1() stack.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When specifying both revisions and pathnames, we allow
"<rev> -- <pathspec>" to be spelled without the "--" as long
as it is not ambiguous. The original logic was something
like:
1. Resolve each item with get_sha1(). If successful,
we know it can be a <rev>. Verify that it _isn't_ a
filename, using verify_non_filename(), and complain of
ambiguity otherwise.
2. If get_sha1() didn't succeed, make sure that it _is_
a file, using verify_filename(). If not, complain
that it is neither a <rev> nor a <pathspec>.
Both verify_filename() and verify_non_filename() rely on
check_filename(), which definitely said "yes, this is a
file" or "no, it is not" using lstat().
Commit 28fcc0b (pathspec: avoid the need of "--" when
wildcard is used, 2015-05-02) introduced a convenience
feature: check_filename() will consider anything with
wildcard meta-characters as a possible filename, without
even checking the filesystem.
This works well for case 2. For such a wildcard, we would
previously have died and said "it is neither". Post-28fcc0b,
we assume it's a pathspec and proceed.
But it makes some instances of case 1 worse. We may have an
extended sha1 expression that contains meta-characters
(e.g., "HEAD^{/foo.*bar}"), and we now complain that it's
also a filename, due to the wildcard characters (even though
that wildcard would not match anything in the filesystem).
One solution would be to actually expand the pathname and
see if it matches anything on the filesystem. But that's
potentially expensive, and we do not have to be so rigorous
for this DWIM magic (if you want rigor, use "--").
Instead, we can just use different rules for cases 1 and 2.
When we know something is a rev, we will complain only if it
meets a much higher standard for "this is also a file";
namely that it actually exists in the filesystem. Case 2
remains the same: we use the looser "it could be a filename"
standard introduced by 28fcc0b.
We can accomplish this by pulling the wildcard logic out of
check_filename() and putting it into verify_filename(). Its
partner verify_non_filename() does not need a change, since
check_filename() goes back to implementing the "higher
standard".
Besides these two callers of check_filename(), there is one
other: git-checkout does a similar DWIM itself. It hits this
code path only after get_sha1() has returned failure, making
it case 2, which gets the special wildcard treatment.
Note that we drop the tests in t2019 in favor of a more
complete set in t6133. t2019 was not the right place for
them (it's about refname ambiguity, not dwim parsing
ambiguity), and the second test explicitly checked for the
opposite result of the case we are fixing here (which didn't
really make any sense; as shown by the test_must_fail in the
test, it would only serve to annoy people).
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
If we have a "--" flag, we should not be doing DWIM magic
based on whether arguments can be filenames. Reorder the
conditional to avoid the check_filename() call entirely in
this case. The outcome is the same, but the short-circuit
makes the dependency more clear.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Documentation/git-clean.txt: don't mention deletion of .git/modules/*
The latter half of this sentence, the removal of the submodules, was
never done with (or without) double -f back when it was written, and
we still do not do so.
Signed-off-by: Matt McCutchen <matt@mattmccutchen.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Merge branch 'ss/clone-depth-single-doc' into maint
Documentation for "git fetch --depth" has been updated for clarity.
* ss/clone-depth-single-doc:
docs: clarify that --depth for git-fetch works with newly initialized repos
docs: say "commits" in the --depth option wording for git-clone
docs: clarify that passing --depth to git-clone implies --single-branch
Drop a few old "todo" items by deciding that the change one of them
suggests is not such a good idea, and doing the change the other
one suggested to do.
* ss/user-manual:
user-manual: add addition gitweb information
user-manual: add section documenting shallow clones
glossary: define the term shallow clone
user-manual: remove temporary branch entry from todo list
Merge branch 'jk/ref-cache-non-repository-optim' into maint
The underlying machinery used by "ls-files -o" and other commands
have been taught not to create empty submodule ref cache for a
directory that is not a submodule. This removes a ton of wasted
CPU cycles.
* jk/ref-cache-non-repository-optim:
resolve_gitlink_ref: ignore non-repository paths
clean: make is_git_repository a public function
dirname() emulation has been added, as Msys2 lacks it.
* js/dirname-basename:
mingw: avoid linking to the C library's isalpha()
t0060: loosen overly strict expectations
t0060: verify that basename() and dirname() work as expected
compat/basename.c: provide a dirname() compatibility function
compat/basename: make basename() conform to POSIX
Refactor skipping DOS drive prefixes
Merge branch 'jk/list-tag-2.7-regression' into maint
"git tag" started listing a tag "foo" as "tags/foo" when a branch
named "foo" exists in the same repository; remove this unnecessary
disambiguation, which is a regression introduced in v2.7.0.
* jk/list-tag-2.7-regression:
tag: do not show ambiguous tag names as "tags/foo"
t6300: use test_atom for some un-modern tests
Merge branch 'js/close-packs-before-gc' into maint
Many codepaths that run "gc --auto" before exiting kept packfiles
mapped and left the file descriptors to them open, which was not
friendly to systems that cannot remove files that are open. They
now close the packs before doing so.
* js/close-packs-before-gc:
receive-pack: release pack files before garbage-collecting
merge: release pack files before garbage-collecting
am: release pack files before garbage-collecting
fetch: release pack files before garbage-collecting
Merge branch 'ho/gitweb-squelch-undef-warning' into maint
Asking gitweb for a nonexistent commit left a warning in the server
log.
Somebody may want to follow this up with a new test, perhaps?
IIRC, we do test that no Perl warnings are given to the server log,
so this should have been caught if our test coverage were good.
Some codepaths used fopen(3) when opening a fixed path in $GIT_DIR
(e.g. COMMIT_EDITMSG) that is meant to be left after the command is
done. This however did not work well if the repository is set to
be shared with core.sharedRepository and the umask of the previous
user is tighter. They have been made to work better by calling
unlink(2) and retrying after fopen(3) fails with EPERM.
* js/fopen-harder:
Handle more file writes correctly in shared repos
commit: allow editing the commit message even in shared repos
Merge branch 'nd/exclusion-regression-fix' into maint
The ignore mechanism saw a few regressions around untracked file
listing and sparse checkout selection areas in 2.7.0; the change
that is responsible for the regression has been reverted.
* nd/exclusion-regression-fix:
Revert "dir.c: don't exclude whole dir prematurely if neg pattern may match"
The "exclude_list" structure has the usual "alloc, nr" pair of
fields to be used by ALLOC_GROW(), but clear_exclude_list() forgot
to reset 'alloc' to 0 when it cleared 'nr' to discard the managed
array.
* nd/dir-exclude-cleanup:
dir.c: clean the entire struct in clear_exclude_list()
Merge branch 'nd/stop-setenv-work-tree' into maint
An earlier change in 2.5.x-era broke users' hooks and aliases by
exporting GIT_WORK_TREE to point at the root of the working tree,
interfering when they tried to use a different working tree without
setting GIT_WORK_TREE environment themselves.
* nd/stop-setenv-work-tree:
Revert "setup: set env $GIT_WORK_TREE when work tree is set, like $GIT_DIR"
On Windows, there is no SIGPIPE. A consequence of this is that the
upstream process of a pipe does not notice the death of the downstream
process until the pipe buffer is full and writing more data returns an
error. This behavior is the reason for an annoying delay during the
execution of t7610-mergetool.sh: There are a number of test cases where
'yes' is invoked upstream. Since the utility is basically an endless
loop it runs, on Windows, until the pipe buffer is full. This does take
a few seconds.
The test suite has its own implementation of 'yes'. Modify it to produce
only a limited amount of output that is sufficient for the test suite.
The amount chosen should be sufficiently high for any test case, assuming
that future test cases will not exaggerate their demands of input from
an upstream 'yes' invocation.
[j6t: commit message]
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Johannes Sixt <j6t@kdbg.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
push: fix ref status reporting for --force-with-lease
The --force--with-lease push option leads to less
detailed status information than --force. In particular,
the output indicates that a reference was fast-forwarded,
even when it was force-updated.
Modify the --force-with-lease ref status logic to leverage
the --force ref status logic when the "lease" conditions
are met.
Also, enhance tests to validate output status reporting.
Signed-off-by: Andrew Wheeler <awheeler@motorola.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
merge-file: ensure that conflict sections match eol style
In the previous patch, we made sure that the conflict markers themselves
match the end-of-line style of the input files. However, this still left
out the conflicting text itself: if it lacks a trailing newline, we
add one, and should add a carriage return when appropriate, too.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
merge-file: let conflict markers match end-of-line style of the context
When merging files with CR/LF line endings, the conflict markers should
match those, lest the output file has mixed line endings.
This is particularly of interest on Windows, where some editors get
*really* confused by mixed line endings.
The original version of this patch by Beat Bolli respected core.eol, and
a subsequent improvement by this developer also respected gitattributes.
This approach was suboptimal, though: `git merge-file` was invented as a
drop-in replacement for GNU merge and as such has no problem operating
outside of any repository at all!
Another problem with the original approach was pointed out by Junio
Hamano: legacy repositories might have their text files committed using
CR/LF line endings (and core.eol and the gitattributes would give us a
false impression there). Therefore, the much superior approach is to
simply match the context's line endings, if any.
We actually do not have to look at the *entire* context at all: if the
files are all LF-only, or if they all have CR/LF line endings, it is
sufficient to look at just a *single* line to match that style. And if
the line endings are mixed anyway, it is *still* okay to imitate just a
single line's eol: we will just add to the pile of mixed line endings,
and there is nothing we can do about that.
So what we do is: we look at the line preceding the conflict, falling
back to the line preceding that in case it was the last line and had no
line ending, falling back to the first line, first in the first
post-image, then the second post-image, and finally the pre-image.
If we find consistent CR/LF (or undecided) end-of-line style, we match
that, otherwise we use LF-only line endings for the conflict markers.
Note that while it is true that there have to be at least two lines we
can look at (otherwise there would be no conflict), the same is not true
for line *endings*: the three files in question could all consist of a
single line without any line ending, each. In this case we fall back to
using LF-only.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
git-svn: fix auth parameter handling on SVN 1.9.0+
For users with "store-passwords = no" set in the "[auth]" section of
their ~/.subversion/config, SVN 1.9.0+ would fail with the
following message when attempting to call svn_auth_set_parameter:
Value is not a string (or undef) at Git/SVN/Ra.pm
Ironically, this breakage was caused by r1553823 in subversion:
"Make svn_auth_set_parameter() usable from Perl bindings."
Since 2007 (602015e0e6ec), git-svn has used a workaround to make
svn_auth_set_parameter usable internally. However this workaround
breaks under SVN 1.9+, which deals properly with the type mapping
and fails to recognize our workaround.
For pre-1.9.0 SVN, we continue to use the existing workaround for
the lack of proper type mapping in the bindings.
Tested under subversion 1.6.17 and 1.9.3.
I've also verified r1553823 was not backported to SVN 1.8.x:
tag: do not show ambiguous tag names as "tags/foo"
Since b7cc53e9 (tag.c: use 'ref-filter' APIs, 2015-07-11),
git-tag has started showing tags with ambiguous names (i.e.,
when both "heads/foo" and "tags/foo" exists) as "tags/foo"
instead of just "foo". This is both:
- pointless; the output of "git tag" includes only
refs/tags, so we know that "foo" means the one in
"refs/tags".
and
- ambiguous; in the original output, we know that the line
"foo" means that "refs/tags/foo" exists. In the new
output, it is unclear whether we mean "refs/tags/foo" or
"refs/tags/tags/foo".
The reason this happens is that commit b7cc53e9 switched
git-tag to use ref-filter's "%(refname:short)" output
formatting, which was adapted from for-each-ref. This more
general code does not know that we care only about tags, and
uses shorten_unambiguous_ref to get the short-name. We need
to tell it that we care only about "refs/tags/", and it
should shorten with respect to that value.
In theory, the ref-filter code could figure this out by us
passing FILTER_REFS_TAGS. But there are two complications
there:
1. The handling of refname:short is deep in formatting
code that does not even have our ref_filter struct, let
alone the arguments to the filter_ref struct.
2. In git v2.7.0, we expose the formatting language to the
user. If we follow this path, it will mean that
"%(refname:short)" behaves differently for "tag" versus
"for-each-ref" (including "for-each-ref refs/tags/"),
which can lead to confusion.
Instead, let's add a new modifier to the formatting
language, "strip", to remove a specific set of prefix
components. This fixes "git tag", and lets users invoke the
same behavior from their own custom formats (for "tag" or
"for-each-ref") while leaving ":short" with its same
consistent meaning in all places.
We introduce a test in t7004 for "git tag", which fails
without this patch. We also add a similar test in t3203 for
"git branch", which does not actually fail. But since it is
likely that "branch" will eventually use the same formatting
code, the test helps defend against future regressions.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add --all and --include-untracked to the git stash save completions.
Add --quiet to the git stash drop completions.
Update git stash branch so that the first argument expands out to the
possible branch names, and the other arguments expand to the stash
names.
Signed-off-by: Paul Wagland <paul@kungfoocoder.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The implementation of mingw_skip_dos_drive_prefix() calls isalpha() via
has_dos_drive_prefix(). Since the definition occurs long before isalpha()
is defined in git-compat-util.h, my build environment reports:
CC alloc.o
In file included from git-compat-util.h:186,
from cache.h:4,
from alloc.c:12:
compat/mingw.h: In function 'mingw_skip_dos_drive_prefix':
compat/mingw.h:365: warning: implicit declaration of function 'isalpha'
Dscho does not see a similar warning in his build and suspects that
ctype.h is included somehow behind the scenes. This implies that his build
links to the C library's isalpha() and does not use git's isalpha().
To fix both the warning in my build and the inconsistency in Dscho's
build, move the function definition to mingw.c. Then it picks up git's
isalpha() because git-compat-util.h is included at the top of the file.
Signed-off-by: Johannes Sixt <j6t@kdbg.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Because this script has to test so many formatters, we have
the nice "test_atom" helper, but we don't use it
consistently. Let's do so. This is shorter, gets rid of some
tests that have their "expected" setup outside of a
test_expect_success block, and lets us organize the changes
better (e.g., putting "refname:short" near "refname").
We also expand the "%(push)" tests a little to match the
"%(upstream)" ones.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When we want to look up a submodule ref, we use
get_ref_cache(path) to find or auto-create its ref cache.
But if we feed a path that isn't actually a git repository,
we blindly create the ref cache, and then may die deeper in
the code when we try to access it. This is a problem because
many callers speculatively feed us a path that looks vaguely
like a repository, and expect us to tell them when it is
not.
This patch teaches resolve_gitlink_ref to reject
non-repository paths without creating a ref_cache. This
avoids the die(), and also performs better if you have a
large number of these faux-submodule directories (because
the ref_cache lookup is linear, under the assumption that
there won't be a large number of submodules).
To accomplish this, we also break get_ref_cache into two
pieces: the lookup and auto-creation (the latter is lumped
into create_ref_cache). This lets us first cheaply ask our
cache "is it a submodule we know about?" If so, we can avoid
repeating our filesystem lookup. So lookups of real
submodules are not penalized; they examine the submodule's
.git directory only once.
The test in t3000 demonstrates a case where this improves
correctness (we used to just die). The new perf case in
p7300 shows off the speed improvement in an admittedly
pathological repository:
Test HEAD^ HEAD
----------------------------------------------------------------
7300.4: ls-files -o 66.97(66.15+0.87) 0.33(0.08+0.24) -99.5%
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
We have always had is_git_directory(), for looking at a
specific directory to see if it contains a git repo. In 0179ca7 (clean: improve performance when removing lots of
directories, 2015-06-15), we added is_git_repository() which
checks for a non-bare repository by looking at its ".git"
entry.
However, the fix in 0179ca7 needs to be applied other
places, too. Let's make this new helper globally available.
We need to give it a better name, though, to avoid confusion
with is_git_directory(). This patch does that, documents
both functions with a comment to reduce confusion, and
removes the clean-specific references in the comments.
Based-on-a-patch-by: Andreas Krey <a.krey@gmx.de> Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The current update_linked_gitdir() has a bug that can create "gitdir"
file in non-multi-worktree setup. Worse, sometimes it can write relative
path to "gitdir" file, which will not work (e.g. "git worktree list"
will display the worktree's location incorrectly)
Instead of fixing this, we step back a bit. The original design was
probably not well thought out. For now, if the user manually moves a
worktree, they have to fix up "gitdir" file manually or the worktree
will get pruned.
Helped-by: Eric Sunshine <sunshine@sunshineco.com> Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>