When doing two-way merge, we failed to invalidate the directory
that a new entry is added (we correctly did so for modified and
deleted entries).
Signed-off-by: Junio C Hamano <junkio@cox.net>
This teaches read-tree to fully populate valid cache-tree when
reading a tree from scratch, or reading a single tree into an
existing index, reusing only the cached stat information (i.e.
one-way merge). We have already taught update-index about cache-tree,
so "git checkout" followed by updates to a few path followed by
a "git commit" would become very efficient.
Signed-off-by: Junio C Hamano <junkio@cox.net>
This teaches one-way and two-way "read-tree -m" (and its special
form, "read-tree --reset" as well) not to discard cache-tree but
invalidate only the changed parts of the tree. When switching
between related branches, this helps the eventual commit
(i.e. write-tree) by keeping cache-tree valid as much as
possible.
This does not prime cache-tree yet, but we ought to be able to
do that for no-merge (i.e. reading from a tree object) case and,
and also perhaps 1 way merge case.
With this patch applied, switching between the tip of Linux 2.6
kernel tree and a branch that touches one path (fs/ext3/Makefile)
from it invalidates only 3 paths out of 1201 cache-tree entries
in the index, and subsequent write-tree takes about a half as
much time as before.
Signed-off-by: Junio C Hamano <junkio@cox.net>
Might as well ape the sigaction change in read-tree.c to avoid
the same potential problems. The fprintf status output will
be overwritten in a second, so don't bother guarding it. Do
move the fputc after disabling SIGALRM to ensure we go to the
next line, though.
Also add a NO_SA_RESTART option in the Makefile in case someone
doesn't have SA_RESTART but does restart (maybe older HP/UX?).
We want the builder to chose this specifically in case the
system both lacks SA_RESTART and does not restart stdio calls;
a compat #define in git-compat-utils.h would silently allow
broken systems.
Signed-off-by: Jason Riedy <ejr@cs.berkeley.edu>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This makes sure that many commands that take refs on the command
line to honor core.warnambiguousrefs configuration. Earlier,
the commands affected by this patch did not read the
configuration file.
Signed-off-by: Junio C Hamano <junkio@cox.net>
Sometimes it is convient for a Porcelain to be able to checkout all
unmerged files in all stages so that an external merge tool can be
executed by the Porcelain or the end-user. Using git-unpack-file
on each stage individually incurs a rather high penalty due to the
need to fork for each file version obtained. git-checkout-index -a
--stage=all will now do the same thing, but faster.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
When both heads deleted, or our side deleted while the other
side did not touch, we did not have to update the working tree.
However, we forgot to remove existing working tree file when we
did not touch and the other side did.
Signed-off-by: Junio C Hamano <junkio@cox.net>
This implements "eye candy" similar to the pack-object/unpack-object
to entertain users while a large tree is being checked out after
a clone or a pull.
Signed-off-by: Junio C Hamano <junkio@cox.net>
This adds "assume unchanged" logic, started by this message in the list
discussion recently:
<Pine.LNX.4.64.0601311807470.7301@g5.osdl.org>
This is a workaround for filesystems that do not have lstat()
that is quick enough for the index mechanism to take advantage
of. On the paths marked as "assumed to be unchanged", the user
needs to explicitly use update-index to register the object name
to be in the next commit.
You can use two new options to update-index to set and reset the
CE_VALID bit:
git-update-index --assume-unchanged path...
git-update-index --no-assume-unchanged path...
These forms manipulate only the CE_VALID bit; it does not change
the object name recorded in the index file. Nor they add a new
entry to the index.
When the configuration variable "core.ignorestat = true" is set,
the index entries are marked with CE_VALID bit automatically
after:
- update-index to explicitly register the current object name to the
index file.
- when update-index --refresh finds the path to be up-to-date.
- when tools like read-tree -u and apply --index update the working
tree file and register the current object name to the index file.
The flag is dropped upon read-tree that does not check out the index
entry. This happens regardless of the core.ignorestat settings.
Index entries marked with CE_VALID bit are assumed to be
unchanged most of the time. However, there are cases that
CE_VALID bit is ignored for the sake of safety and usability:
- while "git-read-tree -m" or git-apply need to make sure
that the paths involved in the merge do not have local
modifications. This sacrifices performance for safety.
- when git-checkout-index -f -q -u -a tries to see if it needs
to checkout the paths. Otherwise you can never check
anything out ;-).
- when git-update-index --really-refresh (a new flag) tries to
see if the index entry is up to date. You can start with
everything marked as CE_VALID and run this once to drop
CE_VALID bit for paths that are modified.
Most notably, "update-index --refresh" honours CE_VALID and does
not actively stat, so after you modified a file in the working
tree, update-index --refresh would not notice until you tell the
index about it with "git-update-index path" or "git-update-index
--no-assume-unchanged path".
This version is not expected to be perfect. I think diff
between index and/or tree and working files may need some
adjustment, and there probably needs other cases we should
automatically unmark paths that are marked to be CE_VALID.
But the basics seem to work, and ready to be tested by people
who asked for this feature.
Signed-off-by: Junio C Hamano <junkio@cox.net>
A new flag --aggressive resolves what we traditionally resolved
with external git-merge-one-file inside index while read-tree
3-way merge works.
git-merge-octopus and git-merge-resolve use this flag before
running git-merge-index with git-merge-one-file.
Signed-off-by: Junio C Hamano <junkio@cox.net>
Avoid asking for zero bytes when that change simplifies overall
logic. Later we would change the wrapper to ask for 1 byte on
platforms that return NULL for zero byte request.
Signed-off-by: Junio C Hamano <junkio@cox.net>
These commands are converted to run from a subdirectory.
commit-tree convert-objects merge-base merge-index mktag
pack-objects pack-redundant prune-packed read-tree tar-tree
unpack-file unpack-objects update-server-info write-tree
Signed-off-by: Junio C Hamano <junkio@cox.net>
ls-files.c and read-tree.c miss the default configuration, in
particular the filemode=false part. The recent +x bit flip made me
notice that, because git-merge refused to merge anything saying that
git-pull.sh is not up to date.
Signed-off-by: Junio C Hamano <junkio@cox.net>
This fixes everybody's favorite gripe that switching branche with
'git checkout' leaves empty directories.
Signed-off-by: Junio C Hamano <junkio@cox.net>
This adds an option --trivial to restrict 3-way 'read-tree -m -u'
to happen only if there is no file-level merging required.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Initially it was to allow specifying more than one remote to
allow creation of an Octopus, but it is not being used.
Signed-off-by: Junio C Hamano <junkio@cox.net>
Somehow I missed it when we updated read-tree to support the recursive
merge strategy. Also -i should require -m as well, which the command
did not check.
Signed-off-by: Junio C Hamano <junkio@cox.net>
We wanted to detect case #16 which should be rare, but botched the
case when some paths are missing, causing a segfault. My fault.
Signed-off-by: Junio C Hamano <junkio@cox.net>
I really wanted to try this out, instead of asking for an adjustment
to the 'git merge' driver and waiting. For now the new strategy is
called 'fredrik' and not in the list of default strategies to be tried.
The script wants Python 2.4 so this commit also adjusts Debian and RPM
build procecure files.
Signed-off-by: Junio C Hamano <junkio@cox.net>
This will help us detect if real-world example merges have multiple
merge-base candidates and one of them matches one head while another
matches the other head.
Signed-off-by: Junio C Hamano <junkio@cox.net>
Adds support for multiple ancestors, removes --emu23, much simplification.
Signed-off-by: Daniel Barkalow <barkalow@iabervon.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
All usage strings are now declared as static const char [].
This is carried over from my old git-pb branch.
Signed-off-by: Petr Baudis <pasky@ucw.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This patch teaches read-tree 3-way merge that, when only "the
other tree" changed a path, and if the index file already has
the same change, we are not in a situation that would clobber
the index and the work tree, and lets the merge succeed; this is
case #14ALT in t1000 test. It does not change the result of the
merge, but prevents it from failing when it does not have to.
Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This adds #3ALT rule (and #2ALT rule for symmetry) to the
read-tree 3-way merge logic that collapses paths that are added
only in one branch and not in the other internally.
This makes --emu23 to succeed in the last remaining case where
the pure 2-way merge succeeded and earlier one failed. Running
diff between t1001 and t1005 test scripts shows that the only
difference between the two is that --emu23 can leave the states
into separate stages so that the user can use usual 3-way merge
resolution techniques to carry forward the local changes when
pure 2-way merge would have refused to run.
Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This fixes too strong index requirement 3-way merge enforces in
one case: the same file is added in both branches.
In this case, the original code insisted that if the index file
has that path, it must match our branch and be up-to-date.
However in this particular case, it only has to match it, and
can be dirty. We just need to make sure that we keep the
work-tree copy instead of checking out the merge result.
The resolution of such a path, however, cannot be left to
outside script, because we will not keep the original stage0
entries for unmerged paths when read-tree finishes, and at that
point, the knowledge of "if we resolve it to match the new file
added in both branches, the merge succeeds and the work tree
would not lose information, but we should _not_ update the work
tree from the resulting index file" is lost. For this reason,
the now code needs to resolve this case (#5ALT) internally.
This affects some existing tests in the test suite, but all in
positive ways. In t1000 (3-way test), this #5ALT case now gets
one stage0 entry, instead of an identical stage2 and stage3
entry pair, for such a path, and one test that checked for merge
failure (because the test assumed the "stricter-than-necessary"
behaviour) does not have to fail anymore. In t1005 (emu23
test), two tests that involves a case where the work tree
already had a change introduced in the upstream (aka "merged
head"), the merge succeeds instead of failing.
Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This new flag causes two-way fast forward to internally use the
three-way merge mechanism. This behaviour is intended to offer
a better fast forward semantics when used in a dirty work tree.
The new test t1005 is parallel to the existing t1001 "pure
2-way" tests, but some parts that are commented out would fail.
These failures are due to three-way merge enforcing too strict
index requirements for cases that could succeed. This problem
will be addressed by later patches.
Without even changing three-way mechanism, the --emu23 two-way
fast forward already gives the user an easier-to-handle merge
result when a file that "merged head" updates has local
modifications. This is demonstrated as "case 16" test in t1005.
Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This is the same as "-m", but it will silently ignore any unmerged
entries, which makes it useful for efficiently forcing a new position
regardless of the state of the current index file.
IOW, to reset to a previous HEAD (in case you have had a failed
merge, for example), you'd just do
git-read-tree -u --reset HEAD
which will also update your working tree to the right state.
NOTE! The "update" will not remove files that may have been added by the
merge. Yet.
This implements the "never lose the current cache information or
the work tree state, but favor a successful merge over merge
failure" principle in the fast-forward two-tree merge operation.
It comes with a set of tests to cover all the cases described in
the case matrix found in the new documentation.
Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Looking good, but hey, it's not like I even have a real testcase for any
of this. But unlike the mess that this was yerstday, today read-cache
is pretty readable and understandable. Which is always a good sign.
Stop trying to haev this stateful thing that keeps track of what it has
seen, and use a much simpler "gather all the different stages with the
same name together and just merge them in one go" approach.
Makes it a lot more understandable, and allows the different merge
algorithms to share the basic merge loop.
This patch teaches read-tree how to use the index file locking
helpers the same way "checkout-cache -u" and "update-cache" do.
Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
That's the final one ("Yeah, sure, we believe you").
Anyway, at least the tests pass, which is not saying a lot, since they
don't end up testing all the new the things that the new merge world
order tries to do. But hopefully we're now at least not any worse off
than we were before the rewrite.
Making the main loop look more like the one- and two-way cases
introduced a bug where "src" had been updated early, but later
users hadn't been adjusted to match.
