This is the bug that just won't die; there always seems to be another
form of it somewhere. See the commit message of 55f39cf755 ("merge:
fix misleading pre-merge check documentation", 2018-06-30) for a more
detailed explanation), but in short:
<quick summary>
builtin/merge.c contains this important requirement for merge
strategies:
...the index must be in sync with the head commit. The strategies are
responsible to ensure this.
This condition is important to enforce because there are two likely
failure cases when the index isn't in sync with the head commit:
* we silently throw away changes the user had staged before the merge
* we accidentally (and silently) include changes in the merge that
were not part of either of the branches/trees being merged
Discarding users' work and mis-merging are both bad outcomes, especially
when done silently, so naturally this rule was stated sternly -- but,
unfortunately totally ignored in practice unless and until actual bugs
were found. But, fear not: the bugs from this were fixed in commit
ee6566e8d7 ("[PATCH] Rewrite read-tree", 2005-09-05)
through a rewrite of read-tree (again, commit 55f39cf755 has a more
detailed explanation of how this affected merge). And it was fixed
again in commit
160252f816 ("git-merge-ours: make sure our index matches HEAD", 2005-11-03)
...and it was fixed again in commit
3ec62ad9ff ("merge-octopus: abort if index does not match HEAD", 2016-04-09)
...and again in commit
65170c07d4 ("merge-recursive: avoid incorporating uncommitted changes in a merge", 2017-12-21)
...and again in commit
eddd1a411d ("merge-recursive: enforce rule that index matches head before merging", 2018-06-30)
...with multiple testcases added to the testsuite that could be
enumerated in even more commits.
Then, finally, in a patch in the same series as the last fix above, the
documentation about this requirement was fixed in commit 55f39cf755
("merge: fix misleading pre-merge check documentation", 2018-06-30), and
we all lived happily ever after...
</quick summary>
Unfortunately, "ever after" apparently denotes a limited time and it
expired today. The merge-recursive rule to enforce that index matches
head was at the beginning of merge_trees() and would only trigger when
opt->call_depth was 0. Since merge_recursive() doesn't call
merge_trees() until after returning from recursing, this meant that the
check wasn't triggered by merge_recursive() until it had first finished
all the intermediate merges to create virtual merge bases. That is a
potentially HUGE amount of computation (and writing of intermediate
merge results into the .git/objects directory) before it errors out and
says, in effect, "Sorry, I can't do any merging because you have some
local changes that would be overwritten."
Trying to enforce that all of merge_trees(), merge_recursive(), and
merge_recursive_generic() checked the index == head condition earlier
resulted in a bunch of broken tests. It turns out that
merge_recursive() has code to drop and reload the cache while recursing
to create intermediate virtual merge bases, but unfortunately that code
runs even when no recursion is necessary. This unconditional dropping
and reloading of the cache masked a few bugs:
* builtin/merge-recursive.c: didn't even bother loading the index.
* builtin/stash.c: feels like a fake 'builtin' because it repeatedly
invokes git subprocesses all over the place, mixed with other
operations. In particular, invoking "git reset" will reset the
index on disk, but the parent process that invoked it won't
automatically have its in-memory index updated.
* t3030-merge-recursive.h: this test has always been broken in that it
didn't make sure to make index match head before running. But, it
didn't care about the index or even the merge result, just the
verbose output while running. While commit eddd1a411d
("merge-recursive: enforce rule that index matches head before
merging", 2018-06-30) should have uncovered this broken test, it
used a test_must_fail wrapper around the merge-recursive call
because it was known that the merge resulted in a rename/rename
conflict. Thus, that fix only made this test fail for a different
reason, and since the index == head check didn't happen until after
coming all the way back out of the recursion, the testcase had
enough information to pass the one check that it did perform.
So, load the index in builtin/merge-recursive.c, reload the in-memory
index in builtin/stash.c, and modify the t3030 testcase to correctly
setup the index and make sure that the test fails in the expected way
(meaning it reports a rename/rename conflict). This makes sure that
all callers actually make the index match head. The next commit will
then enforce the condition that index matches head earlier so this
problem doesn't return in the future.
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The two test checking 'git mmerge-recursive' in an empty worktree in
't3030-merge-recursive.sh' fail when the test script is run with '-x'
tracing (and using a shell other than a Bash version supporting
BASH_XTRACEFD). The reason for those failures is that the tests check
the emptiness of a subshell's stderr, which includes the trace of
commands executed in that subshell as well, throwing off the emptiness
check.
Note that both subshells execute four git commands each, meaning that
checking the emptiness of the whole subshell implicitly ensures that
not only 'git merge-recursive' but none of the other three commands
outputs anything to their stderr. Note also that if one of those
commands were to output anything on its stderr, then the current
combined check would not tell us which one of those four commands the
unexpected output came from.
Save the stderr of those four commands only instead of the whole
subshell, so it remains free from tracing output, and save and check
them individually, so they will show us from which command the
unexpected output came from.
After this change t3030 passes with '-x', even when running with
/bin/sh.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When merging another branch into ours, if their tree is the same as
the common ancestor's, we can declare that our tree represents the
result of three-way merge. In such a case, the recursive merge
backend incorrectly used to create a commit out of our index, even
when the index has changes.
A recent fix attempted to prevent this by adding a comparison
between "our" tree and the index, but forgot that this check must be
restricted only to the outermost merge. Inner merges performed by
the recursive backend across merge bases are by definition made from
scratch without having any local changes added to the index. The
call to index_has_changes() during an inner merge is working on the
index that has no relation to the merge being performed, preventing
legitimate merges from getting carried out.
Fix it by limiting the check to the outermost merge.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When a submodule is being merged or cherry-picked into a working
tree that already contains a corresponding empty directory, do not
record a conflict.
One situation where this bug appears is:
- Commit 1 adds a submodule
- Commit 2 removes that submodule and re-adds it into a subdirectory
(sub1 to sub1/sub1).
- Commit 3 adds an unrelated file.
Now the user checks out commit 1 (first deinitializing the submodule),
and attempts to cherry-pick commit 3. Previously, this would fail,
because the incoming submodule sub1/sub1 would falsely conflict with
the empty sub1 directory.
This patch ignores the empty sub1 directory, fixing the bug. We only
ignore the empty directory if the object being emplaced is a
submodule, which expects an empty directory.
Signed-off-by: David Turner <dturner@twosigma.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
One of the indirect callers of make_virtual_commit() passes the result of
oid_to_hex() as the name, i.e. a pointer to a static buffer. Since the
function uses that string pointer directly in building a struct
merge_remote_desc, multiple entries can end up sharing the same name
inadvertently.
Fix that by calling set_merge_remote_desc(), which creates a copy of the
string, instead of building the struct by hand.
Signed-off-by: Rene Scharfe <l.s.r@web.de>
Reviewed-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The Git CodingGuidelines prefer the $(...) construct for command
substitution instead of using the backquotes `...`.
The backquoted form is the traditional method for command
substitution, and is supported by POSIX. However, all but the
simplest uses become complicated quickly. In particular, embedded
command substitutions and/or the use of double quotes require
careful escaping with the backslash character.
The patch was generated by:
for _f in $(find . -name "*.sh")
do
perl -i -pe 'BEGIN{undef $/;} s/`(.+?)`/\$(\1)/smg' "${_f}"
done
and then carefully proof-read.
Signed-off-by: Elia Pinto <gitter.spiros@gmail.com>
Reviewed-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The t6031 test was introduced to check filemode handling of
merge-recursive. Much later, an unrelated test was tacked on
to look at renames and d/f conflicts. This test does not
depend on anything that happened before (it actually blows
away any existing content in the test repo). Let's move it
to t3030, where there are more related tests.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Teach add_cacheinfo to tell make_cache_entry to skip refreshing stat
information when a file is missing from the work tree. We do not want
the index to be stat-dirty after the merge but also do not want to fail
when a file happens to be missing.
This fixes the 'merge-recursive w/ empty work tree - ours has rename'
case in t3030-merge-recursive.
Suggested-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Brad King <brad.king@kitware.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Sometimes when working with a large repository it can be useful to try
out a merge and only check out conflicting files to disk (for example as
a speed optimization on a server). Until v1.7.7-rc1~28^2~20
(merge-recursive: When we detect we can skip an update, actually skip
it, 2011-08-11), it was possible to do so with the following idiom:
# Prepare a temporary index and empty work tree.
GIT_INDEX_FILE="$PWD/tmp-$$-index" &&
export GIT_INDEX_FILE &&
GIT_WORK_TREE="$PWD/tmp-$$-work" &&
export GIT_WORK_TREE &&
mkdir "$GIT_WORK_TREE" &&
# Convince the index that our side is on disk.
git read-tree -i -m $ours &&
git update-index --ignore-missing --refresh &&
# Merge their side into our side.
bases=$(git merge-base --all $ours $theirs) &&
git merge-recursive $bases -- $ours $theirs &&
tree=$(git write-tree)
Nowadays, that still works and the exit status is the same, but
merge-recursive produces a diagnostic if "our" side renamed a file:
error: addinfo_cache failed for path 'dst'
Add a test to document this regression.
Signed-off-by: Brad King <brad.king@kitware.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The test cases include many corner-cases of merge-recursive's behavior,
some of them involve type changes and symbolic links. All cases, including
those that are protected by SYMLINKS check only whether the result of
merge-recursive is correctly stored in the database and the index; the
file system is not investigated. Use test_ln_s_add to enter a symbolic
link in the index in the test setup and run the tests without the
SYMLINKS prerequisite.
Notice that one test that has the SYMLINKS protection removed is an
expect_failure. There is a possibility that the test fails differently
depending on whether SYMLINKS is present or not; but this is not the case
presently.
Signed-off-by: Johannes Sixt <j6t@kdbg.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Use test_expect_code in preference to repeatedly checking exit codes
by hand.
Signed-off-by: Ramkumar Ramachandra <artagnon@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In this test, we have merge two branches. On one branch, we
renamed "a" to "e". On the other, we renamed "a" to "e" and
then added a symlink pointing at "a" pointing to "e".
The results for the test indicate that the merge should
succeed, but also that "a" should no longer exist. Since
both sides renamed "a" to the same destination, we will end
up comparing those destinations for content.
But what about what's left? One side (the rename only),
replaced "a" with nothing. The other side replaced it with a
symlink. The common base must also be nothing, because any
"a" before this was meaningless (it was totally unrelated
content that ended up getting renamed).
The only sensible resolution is to keep the symlink. The
rename-only side didn't touch the content versus the common
base, and the other side added content. The 3-way merge
dictates that we take the side with a change.
And this gives the overall merge an intuitive result. One
side made one change (a rename), and the other side made two
changes: an identical rename, and an addition (that just
happened to be at the same spot). The end result should
contain both changes.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Gettextize the "You have not concluded your merge messages. A test in
t3030-merge-recursive.sh explicitly checked for this message. Change
it to skip the test under GETTEXT_POISON=YesPlease.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
* en/merge-recursive: (41 commits)
t6022: Use -eq not = to test output of wc -l
merge-recursive:make_room_for_directories - work around dumb compilers
merge-recursive: Remove redundant path clearing for D/F conflicts
merge-recursive: Make room for directories in D/F conflicts
handle_delete_modify(): Check whether D/F conflicts are still present
merge_content(): Check whether D/F conflicts are still present
conflict_rename_rename_1to2(): Fix checks for presence of D/F conflicts
conflict_rename_delete(): Check whether D/F conflicts are still present
merge-recursive: Delay modify/delete conflicts if D/F conflict present
merge-recursive: Delay content merging for renames
merge-recursive: Delay handling of rename/delete conflicts
merge-recursive: Move handling of double rename of one file to other file
merge-recursive: Move handling of double rename of one file to two
merge-recursive: Avoid doubly merging rename/add conflict contents
merge-recursive: Update merge_content() call signature
merge-recursive: Update conflict_rename_rename_1to2() call signature
merge-recursive: Structure process_df_entry() to handle more cases
merge-recursive: Have process_entry() skip D/F or rename entries
merge-recursive: New function to assist resolving renames in-core only
merge-recursive: New data structures for deferring of D/F conflicts
...
Conflicts:
t/t6020-merge-df.sh
t/t6036-recursive-corner-cases.sh
Breaks in a test assertion's && chain can potentially hide
failures from earlier commands in the chain.
Commands intended to fail should be marked with !, test_must_fail, or
test_might_fail. The examples in this patch do not require that.
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
d5af510 (RE: [PATCH] Avoid rename/add conflict when contents are identical
2010-09-01) avoided erroring out in a rename/add conflict when the contents
were identical. A simpler fix could have handled that particular testcase,
but it would not correctly handle the case where a symlink is involved.
Add another testcase using symlinks, to avoid breaking that case.
Signed-off-by: Ken Schalk <ken.schalk@intel.com>
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
>Due to this this (and maybe all the tests) need to depend on the
>SYMLINKS prereq.
Here's a third attempt with no use of symlinks in the test:
Skip the entire rename/add conflict case if the file added on the
other branch has the same contents as the file being renamed. This
avoids giving the user an extra copy of the same file and presenting a
conflict that is confusing and pointless.
A simple test of this case has been added in
t/t3030-merge-recursive.sh.
Signed-off-by: Ken Schalk <ken.schalk@intel.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When an error is encountered, it calls add_rejected_file() which either
- directly displays the error message and stops if in plumbing mode
(i.e. if show_all_errors is not initialized at 1)
- or stores it so that it will be displayed at the end with display_error_msgs(),
Storing the files by error type permits to have a list of files for
which there is the same error instead of having a serie of almost
identical errors.
As each bind_overlap error combines a file and an old file, a list cannot be
done, therefore, theses errors are not stored but directly displayed.
Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When one side of a merge turns a directory into a submodule, and the other
side does not touch that directory (but has other non-conflicting changes),
then a merge should succeed. But currently, it does not; it rather fails
with a file/directory conflict.
Signed-off-by: Johannes Sixt <j6t@kdbg.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Various commands refuse to run in the presence of conflicts (commit,
merge, pull, cherry-pick/revert). They all used to provide rough, and
inconsistant error messages.
A new variable advice.resolveconflict is introduced, and allows more
verbose messages, pointing the user to the appropriate solution.
For commit, the error message used to look like this:
$ git commit
foo.txt: needs merge
foo.txt: unmerged (c34a92682e0394bc0d6f4d4a67a8e2d32395c169)
foo.txt: unmerged (3afcd75de8de0bb5076942fcb17446be50451030)
foo.txt: unmerged (c9785d77b76dfe4fb038bf927ee518f6ae45ede4)
error: Error building trees
The "need merge" line is given by refresh_cache. We add the IN_PORCELAIN
option to make the output more consistant with the other porcelain
commands, and catch the error in return, to stop with a clean error
message. The next lines were displayed by a call to cache_tree_update(),
which is not reached anymore if we noticed the conflict.
The new output looks like:
U foo.txt
fatal: 'commit' is not possible because you have unmerged files.
Please, fix them up in the work tree, and then use 'git add/rm <file>' as
appropriate to mark resolution and make a commit, or use 'git commit -a'.
Pull is slightly modified to abort immediately if $GIT_DIR/MERGE_HEAD
exists instead of waiting for merge to complain.
The behavior of merge and the test-case are slightly modified to reflect
the usual flow: start with conflicts, fix them, and afterwards get rid of
MERGE_HEAD, with different error messages at each stage.
Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The following is an easy mistake to make for users coming from version
control systems with an "update and commit"-style workflow.
1. git pull
2. resolve conflicts
3. git pull
Step 3 overrides MERGE_HEAD, starting a new merge with dirty index.
IOW, probably not what the user intended. Instead, refuse to merge
again if a merge is in progress.
Reported-by: Dave Olszewski <cxreg@pobox.com>
Signed-off-by: Clemens Buchacher <drizzd@aon.at>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The code was actually supposed to do that, but was accidentally broken.
Noticed by Anders Melchiorsen.
Signed-off-by: Alex Riesen <raa.lkml@gmail.com>
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
The "trivial merge" codepath wants to optimize itself by making an
internal call to the read-tree machinery, but it does not read the index
before doing so, and the codepath is never exercised. Incidentally, this
failure to read the index upfront means that the safety to refuse doing
anything when the index is unmerged does not kick in, either.
These two problem are fixed by using read_cache_unmerged() that does read
the index before checking if it is unmerged at the beginning of
cmd_merge().
The primary logic of the merge, however, assumes that the process never
reads the index in-core, and the call to write_cache_as_tree() it makes
from write_tree_trivial() will always read from the on-disk index that is
prepared the strategy back-ends. This assumption is now broken by the
above fix. To fix this issue, we now call discard_cache() before calling
write_tree_trivial() when it wants to write the on-disk index as a tree.
When multiple strategies are tried, their results are evaluated by reading
the resulting index and inspecting it. The codepath needs to make a call
to read_cache() for each successful strategy, and for that to work, they
need to discard_cache() the one read by the previous round.
Also the "trivial merge" forgot that the current commit is one of the
parents of the resulting commit.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
As a general principle, we should not use "git diff" to validate the
results of what git command that is being tested has done. We would not
know if we are testing the command in question, or locating a bug in the
cute hack of "git diff --no-index".
Rather use test_cmp for that purpose.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This uses "git-apply --whitespace=strip" to fix whitespace errors that have
crept in to our source files over time. There are a few files that need
to have trailing whitespaces (most notably, test vectors). The results
still passes the test, and build result in Documentation/ area is unchanged.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We have fairly extensive coverage of read-tree 3-way machinery,
and many Porcelain-ish tests use git-merge front-end tests, but
we did not have good basic test for merge-recursive, which made
it very hard to hack on it.
I used this during the recent work to teach D/F conflicts to
merge-recursive.
Signed-off-by: Junio C Hamano <junkio@cox.net>
