The logic in 83ae209 (checkout branch: prime cache-tree fully,
2009-04-20) is bogus; checkout can switch branches with a dirty
index and in such a case the tree won't match HEAD.
According to POSIX, tv_sec is supposed to be a time_t, but OpenBSD
(and FreeBSD, too) defines it to be a long, which triggers a type
mismatch when a pointer to it is given to localtime_r().
Acked-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Getopt-Long v2.38 is much stricter about sloppy getopt usage. The
trailing pipe causes git-svn testcases to fail for all of the --stdin
argument calls.
Signed-off-by: Robin H. Johnson <robbat2@gentoo.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Merge branch 'jc/maint-read-tree-multi' into maint
* jc/maint-read-tree-multi:
checkout branch: prime cache-tree fully
read-tree -m A B: prime cache-tree from the switched-to tree
Move prime_cache_tree() to cache-tree.c
read-tree A B: do not corrupt cache-tree
* mk/maint-apply-swap:
tests: make test-apply-criss-cross-rename more robust
builtin-apply: keep information about files to be deleted
tests: test applying criss-cross rename patch
Merge branch 'jc/maint-1.6.0-keep-pack' into maint-1.6.1
* jc/maint-1.6.0-keep-pack:
pack-objects: don't loosen objects available in alternate or kept packs
t7700: demonstrate repack flaw which may loosen objects unnecessarily
Remove --kept-pack-only option and associated infrastructure
pack-objects: only repack or loosen objects residing in "local" packs
git-repack.sh: don't use --kept-pack-only option to pack-objects
t7700-repack: add two new tests demonstrating repacking flaws
is_kept_pack(): final clean-up
Simplify is_kept_pack()
Consolidate ignore_packed logic more
has_sha1_kept_pack(): take "struct rev_info"
has_sha1_pack(): refactor "pretend these packs do not exist" interface
git-repack: resist stray environment variable
Merge branch 'bs/maint-1.6.0-tree-walk-prefix' into maint-1.6.1
* bs/maint-1.6.0-tree-walk-prefix:
match_tree_entry(): a pathspec only matches at directory boundaries
tree_entry_interesting: a pathspec only matches at directory boundary
honor repack.usedeltabaseoffset when fetching packs
If the local receiving repository has disabled the use of delta base
offset, for example to retain compatibility with older versions of
Git that predate OFS_DELTA, we shouldn't ask for ofs-delta support
when we obtain a pack from the remote server.
[ issue noticed by Shawn Pearce ]
Signed-off-by: Nicolas Pitre <nico@cam.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The combine diff logic knew only about blobs (and their checked-out form
in the work tree, either regular files or symlinks), and barfed when fed
submodules. This "externalizes" gitlinks in the same way as the normal
patch generation codepath does (i.e. "Subproject commit Xxx\n") to fix the
issue.
* maint-1.6.1:
grep: fix segfault when "git grep '('" is given
Documentation: fix a grammatical error in api-builtin.txt
builtin-merge: fix a typo in an error message
* maint-1.6.0:
grep: fix segfault when "git grep '('" is given
Documentation: fix a grammatical error in api-builtin.txt
builtin-merge: fix a typo in an error message
builtin-apply: keep information about files to be deleted
Example correct diff generated by `diff -M -B' might look like this:
diff --git a/file1 b/file2
similarity index 100%
rename from file1
rename to file2
diff --git a/file2 b/file1
similarity index 100%
rename from file2
rename to file1
Information about removing `file2' comes after information about creation
of new `file2' (renamed from `file1'). Existing implementation isn't able to
apply such patch, because it has to know in advance which files will be
removed.
This patch populates fn_table with information about removal of files
before calling check_patch() for each patch to be applied.
Signed-off-by: Michał Kiedrowicz <michal.kiedrowicz@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When switching to another branch, the earlier code relied on incremental
invalidation of cache-tree entries to degrade it. While it is not wrong
per-se, we know that the resulting index must fully match the branch we
are switching to unless the -m (merge) option is used.
We should simply fully re-prime the cache-tree using the new tree object
in such a case. And for safety, invalidate the cache-tree as a whole in
other cases.
An earlier commit aab3b9a (read-tree A B C: do not create a bogus index
and do not segfault, 2009-03-12) resurrected the support for an obscure
(but useful) feature to read and overlay more than one tree into the index
without the -m (merge) option. But the fix was not enough.
Exercising this feature exposes a longstanding bug in the code that primes
the cache-tree in the index from the tree that was read. The intention
was that when we know that the index must exactly match the tree we just
read, we prime the entire cache-tree with it.
However, the logic to detect that case incorrectly triggered if you read
two trees without -m. This resulted in a corrupted cache-tree, and
write-tree would have produced an incorrect tree object out of such an
index.
* ef/maint-fast-export:
builtin-fast-export.c: handle nested tags
builtin-fast-export.c: fix crash on tagged trees
builtin-fast-export.c: turn error into warning
test-suite: adding a test for fast-export with tag variants
* mm/maint-add-p-quit:
Update git-add.txt according to the new possibilities of 'git add -p'.
add-interactive: refactor mode hunk handling
git add -p: new "quit" command at the prompt.
The original implementation considered the mode separately
from the rest of the hunks, asking about it outside the main
hunk-selection loop. This patch instead places a mode change
as the first hunk in the loop. This has two advantages:
1. less duplicated code (since we use the main selection
loop). This also cleans up an inconsistency, which is
that the main selection loop separates options with a
comma, whereas the mode prompt used slashes.
2. users can now skip the mode change and come back to it,
search for it (via "/mode"), etc, as they can with other
hunks.
To facilitate this, each hunk is now marked with a "type".
Mode hunks are not considered for splitting (which would
make no sense, and also confuses the split_hunk function),
nor are they editable. In theory, one could edit the mode
lines and change to a new mode. In practice, there are only
two modes that git cares about (0644 and 0755), so either
you want to move from one to the other or not (and you can
do that by staging or not staging).
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
There's already 'd' to stop staging hunks in a file, but no explicit
command to stop the interactive staging (for the current files and the
remaining ones). Of course you can do 'd' and then ^C, but it would be
more intuitive to allow 'quit' action.
Instead of doing the (potentially very expensive) "in_merge_base()"
check for each commit that might be pruned if it is unreachable, do a
preparatory reachability graph of the commit space, so that the common
case of being reachable can be tested directly.
[ Cleaned up a bit and tweaked to actually work. - Linus ] Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When tags that points to tags are passed to fast-export, an error is given,
saying "Tag [TAGNAME] points nowhere?". This fix calls parse_object() on the
object before referencing it's tag, to ensure the tag-info is fully initialized.
In addition, it inserts a comment to point out where nested tags are handled.
This is consistent with the comment for signed tags.
Signed-off-by: Erik Faye-Lund <kusmabite@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
If a tag object points to a tree (or another unhandled type), the commit-
pointer is left uninitialized and later dereferenced. This patch adds a
default case to the switch that issues a warning and skips the object.
Signed-off-by: Erik Faye-Lund <kusmabite@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
fast-import doesn't have a syntax to support tree-objects (and some other
object-types), so fast-export shouldn't handle them. However, aborting the
operation is a bit drastic. This patch turns the error into a warning instead.
Signed-off-by: Erik Faye-Lund <kusmabite@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Merge branch 'bs/maint-1.6.0-tree-walk-prefix' into maint
* bs/maint-1.6.0-tree-walk-prefix:
match_tree_entry(): a pathspec only matches at directory boundaries
tree_entry_interesting: a pathspec only matches at directory boundary
Merge branch 'cb/maint-merge-recursive-submodule-fix' into maint
* cb/maint-merge-recursive-submodule-fix:
simplify output of conflicting merge
update cache for conflicting submodule entries
add tests for merging with submodules
Update docs on behaviour of 'core.sharedRepository' and 'git init --shared'
This documentation update is needed to reflect the recent changes where
"core.sharedRepository = 0mode" was changed to set, not loosen, the
repository permissions.
Signed-off-by: Johan Herland <johan@herland.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
doc/gitattributes: clarify location of config text
The gitattributes documentation has a section on the "diff"
attribute, with subsections for each of the things you might
want to configure in your diff config section (external
diff, hunk headers, etc). The first such subsection
specifically notes that the definition of the diff driver
should go into $GIT_DIR/config, but subsequent sections do
not.
This location is implied if you are reading the
documentation sequentially, but it is not uncommon for a new
user to jump to (or be referred to) a specific section. For
a new user who does not know git well enough to recognize
the config syntax, it is not clear that those directives
don't also go into the gitattributes file.
This patch just mentions the config file in each subsection,
similar to the way it is mentioned in the first.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When interpreting a config value, the config parser reads in 1+ space
character(s) and puts -one- space character in the buffer as soon as
the first non-space character is encountered (if not inside quotes).
Unfortunately the buffer size check lacks the extra space character
which gets inserted at the next non-space character, resulting in
a crash with a specially crafted config entry.
The unit test now uses Java to compile a platform independent
.NET framework to output the test string in C# :o)
Read: Thanks to Johannes Sixt for the correct printf call
which replaces the perl invocation.
Signed-off-by: Thomas Jarosch <thomas.jarosch@intra2net.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
In particular, pushing the "path_name()" call _into_ the show() function
would seem to allow
- more clarity into who "owns" the name (ie now when we free the name in
the show_object callback, it's because we generated it ourselves by
calling path_name())
- not calling path_name() at all, either because we don't care about the
name in the first place, or because we are actually happy walking the
linked list of "struct name_path *" and the last component.
Now, I didn't do that latter optimization, because it would require some
more coding, but especially looking at "builtin-pack-objects.c", we really
don't even want the whole pathname, we really would be better off with the
list of path components.
Why? We use that name for two things:
- add_preferred_base_object(), which actually _wants_ to traverse the
path, and now does it by looking for '/' characters!
- for 'name_hash()', which only cares about the last 16 characters of a
name, so again, generating the full name seems to be just unnecessary
work.
Anyway, so I didn't look any closer at those things, but it did convince
me that the "show_object()" calling convention was crazy, and we're
actually better off doing _less_ in list-objects.c, and giving people
access to the internal data structures so that they can decide whether
they want to generate a path-name or not.
This patch does that, and then for people who did use the name (even if
they might do something more clever in the future), it just does the
straightforward "name = path_name(path, component); .. free(name);" thing.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
process_{tree,blob}: show objects without buffering
Here's a less trivial thing, and slightly more dubious one.
I was looking at that "struct object_array objects", and wondering why we
do that. I have honestly totally forgotten. Why not just call the "show()"
function as we encounter the objects? Rather than add the objects to the
object_array, and then at the very end going through the array and doing a
'show' on all, just do things more incrementally.
Now, there are possible downsides to this:
- the "buffer using object_array" _can_ in theory result in at least
better I-cache usage (two tight loops rather than one more spread out
one). I don't think this is a real issue, but in theory..
- this _does_ change the order of the objects printed. Instead of doing a
"process_tree(revs, commit->tree, &objects, NULL, "");" in the loop
over the commits (which puts all the root trees _first_ in the object
list, this patch just adds them to the list of pending objects, and
then we'll traverse them in that order (and thus show each root tree
object together with the objects we discover under it)
I _think_ the new ordering actually makes more sense, but the object
ordering is actually a subtle thing when it comes to packing
efficiency, so any change in order is going to have implications for
packing. Good or bad, I dunno.
- There may be some reason why we did it that odd way with the object
array, that I have simply forgotten.
Anyway, now that we don't buffer up the objects before showing them
that may actually result in lower memory usage during that whole
traverse_commit_list() phase.
This is seriously not very deeply tested. It makes sense to me, it seems
to pass all the tests, it looks ok, but...
Does anybody remember why we did that "object_array" thing? It used to be
an "object_list" a long long time ago, but got changed into the array due
to better memory usage patterns (those linked lists of obejcts are
horrible from a memory allocation standpoint). But I wonder why we didn't
do this back then. Maybe there's a reason for it.
Or maybe there _used_ to be a reason, and no longer is.
On Wed, 8 Apr 2009, Björn Steinbrink wrote:
>
> The name of the processed object was duplicated for passing it to
> add_object(), but that already calls path_name, which allocates a new
> string anyway. So the memory allocated by the xstrdup calls just went
> nowhere, leaking memory.
Ack, ack.
There's another easy 5% or so for the built-in object walker: once we've
created the hash from the name, the name isn't interesting any more, and
so something trivial like this can help a bit.
Does it matter? Probably not on its own. But a few more memory saving
tricks and it might all make a difference.
This test was added recently (5a688fe, "core.sharedrepository = 0mode"
should set, not loosen; 2009-03-28). It checked the result of a sed
invocation for emptyness, but in some cases it forgot to print anything
at all, so that those checks would never be false.
Due to this mistake, it went unnoticed that the files in objects/info are
not necessarily 0440, but can also be 0660. Because the 0mode setting
tries to guarantee that the files are accessible only to the people they
are meant to be used by, we should only make sure that they are readable
by the user and the group when the configuration is set to 0660. It is a
separate matter from the core.shredrepository settings that w-bit from
immutable object files under objects/[0-9a-f][0-9a-f] directories should
be dropped.
COMMIT_EDITMSG is still world-readable, but it (and any transient files
that are meant for repositories with a work tree) does not matter. If you
are working on a shared machine and on a sekrit stuff, the root of the
work tree would be with mode 0700 (or 0750 to allow peeking by other
people in the group), and that would mean that .git/COMMIT_EDITMSG in such
a repository would not be readable by the strangers anyway.
Also, in the real-world use case, .git/COMMIT_EDITMSG will be given to an
arbitrary editor the user happens to use, and we have no guarantee what it
does (e.g. it may create a new file with umask and replace, it may rewrite
in place, it may leave an editor backup file but use umask to create it,
etc.), and the protection of the file lies majorly on the protection of
the root of the work tree.
This test cannot be run on Windows; it requires POSIXPERM when merged to
'master'.
Signed-off-by: Johannes Sixt <j6t@kdbg.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
* jc/shared-literally:
t1301: loosen test for forced modes
set_shared_perm(): sometimes we know what the final mode bits should look like
move_temp_to_file(): do not forget to chmod() in "Coda hack" codepath
Move chmod(foo, 0444) into move_temp_to_file()
"core.sharedrepository = 0mode" should set, not loosen
Merge branch 'jc/maint-1.6.0-keep-pack' into maint
* jc/maint-1.6.0-keep-pack:
pack-objects: don't loosen objects available in alternate or kept packs
t7700: demonstrate repack flaw which may loosen objects unnecessarily
Remove --kept-pack-only option and associated infrastructure
pack-objects: only repack or loosen objects residing in "local" packs
git-repack.sh: don't use --kept-pack-only option to pack-objects
t7700-repack: add two new tests demonstrating repacking flaws
is_kept_pack(): final clean-up
Simplify is_kept_pack()
Consolidate ignore_packed logic more
has_sha1_kept_pack(): take "struct rev_info"
has_sha1_pack(): refactor "pretend these packs do not exist" interface
git-repack: resist stray environment variable
The name of the processed object was duplicated for passing it to
add_object(), but that already calls path_name, which allocates a new
string anyway. So the memory allocated by the xstrdup calls just went
nowhere, leaking memory.
This reduces the RSS usage for a "rev-list --all --objects" by about 10% on
the gentoo repo (fully packed) as well as linux-2.6.git:
gentoo:
| old | new
----------------|-------------------------------
RSS | 1537284 | 1388408
VSZ | 1816852 | 1667952
time elapsed | 1:49.62 | 1:48.99
min. page faults| 417178 | 379919
linux-2.6.git:
| old | new
----------------|-------------------------------
RSS | 324452 | 292996
VSZ | 491792 | 460376
time elapsed | 0:14.53 | 0:14.28
min. page faults| 89360 | 81613
Signed-off-by: Björn Steinbrink <B.Steinbrink@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
git-pull.sh: better warning message for "git pull" on detached head.
Otherwise, git complains about not finding a branch to pull from in
'branch..merge', which is hardly understandable. While we're there,
reword the sentences slightly.
Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr> Signed-off-by: Junio C Hamano <gitster@pobox.com>
* maint-1.6.1:
Documentation: clarify .gitattributes search
git-checkout.txt: clarify that <branch> applies when no path is given.
git-checkout.txt: fix incorrect statement about HEAD and index