There were few places where merge errors detected deeper in the call chain
were ignored and not propagated up the callchain to the caller.
Most notably, this caused switching branches with "git checkout" to ignore
a path modified in a work tree are different between the HEAD version and
the commit being switched to, which it internally notices but ignores it,
resulting in an incorrect two-way merge and loss of the change in the work
tree.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This adds a "discard_index(&o->result)" to the failure path, to reclaim
memory from an in-core index we built but ended up not using.
The *big* memory leak comes from the fact that we leak the cache_entry
things left and right. That's a very traditional and deliberate leak:
because we used to build up the cache entries by just mapping them
directly in from the index file (and we emulate that in modern times
by allocating them from one big array), we can't actually free them
one-by-one.
So doing the "discard_index()" will free the hash tables etc, which is
good, and it will free the "istate->alloc" but that is never set on the
result because we don't get the result from the index read. So we don't
actually free the individual cache entries themselves that got created
from the trees.
That's not something new, btw. We never did. But some day we should just
add a flag to the cache_entry() that it's a "free one by one" kind, and
then we could/should do it. In the meantime, this one-liner will fix
*some* of the memory leaks, but not that old traditional one.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We will always unpack into our own internal index, but we will take the
source from wherever specified, and we will optionally write the result
to a specified index (optionally, because not everybody even _wants_ any
result: the index diffing really wants to just walk the tree and index
in parallel).
This ends up removing a fair number more lines than it adds, for the
simple reason that we can now skip all the crud that tried to be
oh-so-careful about maintaining our position in the index as we were
traversing and modifying it. Since we don't actually modify the source
index any more, we can just update the 'o->pos' pointer without worrying
about whether an index entry got removed or replaced or added to.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This is just a very mechanical conversion, and makes everybody set it to
'&the_index' before calling, but at least it makes it more explicit
where we work with the index.
The next stage would be to split that index usage up into a 'source' and
a 'destination' index, so that we can unpack into a different index than
we started out from.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This is in an effort to make the source index of 'unpack_trees()' as
being const, and thus making the compiler help us verify that we only
access it for reading.
The constification also extended to some of the hashing helpers that get
called indirectly.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When we make the "root" tree-walk info entry have a pathname in it, we
need to have a ->prev pointer so that compare_entry will actually notice
and traverse into the root.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This not only deletes more code than it adds, it gets rid of a
singularly hard-to-understand function (unpack_trees_rec()), and
replaces it with a set of smaller and simpler functions that use the
generic tree traversal mechanism to walk over one or more git trees in
parallel.
It's still not the most wonderful interface, and by no means is the new
code easy to understand either, but it's at least a bit less opaque.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This makes the traverse_trees() entry comparator routine use the more
relaxed form of name comparison that considers files and directories
with the same name identical.
We pass in a separate mask for just the directory entries, so that the
callback routine can decide (if it wants to) to only handle one or the
other type, but generally most (all?) users are expected to really want
to see the case of a name 'foo' showing up in one tree as a file and in
another as a directory at the same time.
In particular, moving 'unpack_trees()' over to use this tree traversal
mechanism requires this.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This allows the callback to return an error value, but it can also
specify which of the tree entries that it actually used up by returning
a positive mask value.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This makes the calling convention a bit less obvious, but a lot more
flexible. Instead of allocating and extending a new 'base' string, we
just link the top-most name into a linked list of the 'info' structure
when traversing a subdirectory, and we can generate the basename by
following the list.
Perhaps even more importantly, the linked list of info structures also
gives us a place to naturally save off other information than just the
directory name.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This new helper is identical to base_name_compare(), except it compares
conflicting directory/file entries as equal in order to help handling DF
conflicts (thus the name).
Note that while a directory name compares as equal to a regular file
with the new helper, they then individually compare _differently_ to a
filename that has a dot after the basename (because '\0' < '.' < '/').
So a directory called "foo/" will compare equal to a file "foo", even
though "foo.c" will compare after "foo" and before "foo/"
This will be used by routines that want to traverse the git namespace
but then handle conflicting entries together when possible.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
* ph/parseopt:
parse-options: new option type to treat an option-like parameter as an argument.
parse-opt: bring PARSE_OPT_HIDDEN and NONEG to git-rev-parse --parseopt
* dp/clean-fix:
git-clean: add tests for relative path
git-clean: correct printing relative path
Make private quote_path() in wt-status.c available as quote_path_relative()
Revert part of d089eba (setup: sanitize absolute and funny paths in get_pathspec())
Revert part of 1abf095 (git-add: adjust to the get_pathspec() changes)
Revert part of 744dacd (builtin-mv: minimum fix to avoid losing files)
get_pathspec(): die when an out-of-tree path is given
* sp/fetch-optim:
Teach git-fetch to exploit server side automatic tag following
Teach fetch-pack/upload-pack about --include-tag
git-pack-objects: Automatically pack annotated tags if object was packed
Teach git-fetch to grab a tag at the same time as a commit
Make git-fetch follow tags we already have objects for sooner
Teach upload-pack to log the received need lines to an fd
Free the path_lists used to find non-local tags in git-fetch
Allow builtin-fetch's find_non_local_tags to append onto a list
Ensure tail pointer gets setup correctly when we fetch HEAD only
Remove unnecessary delaying of free_refs(ref_map) in builtin-fetch
Remove unused variable in builtin-fetch find_non_local_tags
* maint:
GIT 1.5.4.4
ident.c: reword error message when the user name cannot be determined
Fix dcommit, rebase when rewriteRoot is in use
Really make the LF after reset in fast-import optional
The "config --global" suggested in the message is a valid one-shot fix,
and hopefully one-shot across machines that NFS mounts the home directories.
This knowledge can hopefully be reused when you are forced to use git on
Windows, but the fix based on GECOS would not be applicable, so
it is not such a useful hint to mention the exact reason why the
name cannot be determined.
Signed-off-by: Santi Béjar <sbejar@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When the rewriteRoot setting is used with git-svn, it causes the svn
IDs added to commit messages to bear a different URL than is actually
used to retrieve Subversion data.
It is common for Subversion repositories to be available multiple
ways: for instance, HTTP to the public, and svn+ssh to people with
commit access. The need to switch URLs for access is fairly common as
well -- perhaps someone was just given commit access. To switch URLs
without having to rewrite history, one can use the old url as a
rewriteRoot, and use the new one in the svn-remote url setting.
This works well for svn fetching and general git commands.
However, git-svn dcommit, rebase, and perhaps other commands do not
work in this scenario. They scan the svn ID lines in commit messages
and attempt to match them up with url lines in [svn-remote] sections
in the git config.
This patch allows them to match rewriteRoot options, if such options
are present.
Signed-off-by: John Goerzen <jgoerzen@complete.org>
Acked-by: Eric Wong <normalperson@yhbt.net>
The subdirectory filter had a bug to notice that the commit in question
did not have anything in the path-limited part of the tree. $commit:$path
does not name an empty tree when $path does not appear in $commit.
This should fix it. The additional test in t7003 is originally from Kevin
Ballard but with fixups.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The configuration variables for custom merge tools were documented
only in config.txt but there was no reference to the functionality in
git-mergetool.txt.
Signed-off-by: Charles Bailey <charles@hashpling.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
cmd_from() ends with a call to read_next_command(), which is needed
when using cmd_from() from commands where from is not the last element.
With reset, however, "from" is the last command, after which the flow
returns to the main loop, which calls read_next_command() again.
Because of this, always set unread_command_buf in cmd_reset_branch(),
even if cmd_from() was successful.
Add a test case for this in t9300-fast-import.sh.
Signed-off-by: Adeodato Simó <dato@net.com.org.es>
Acked-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
* maint:
unquote_c_style: fix off-by-one.
test-lib: fix TERM to dumb for test repeatability
config.txt: refer to --upload-pack and --receive-pack instead of --exec
git-gui: Gracefully fall back to po2msg.sh if msgfmt --tcl fails
The logic to countermand suppression of Cc to the signers with a more
explicit --signed-off-by option done in 6564828 (git-send-email:
Generalize auto-cc recipient mechanism) suffers from a double-negation
error.
A --signed-off-cc option, when false, should actively suppress CC: to be
generated out of S-o-b lines, and it should refrain from suppressing when
it is true.
It also fixes "(sob) Adding cc:" status output; earlier it included the
line terminator LF inside '%s', which was totally bogus.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
* js/reflog-delete:
t3903-stash.sh: Add tests for new stash commands drop and pop
git-reflog.txt: Document new commands --updateref and --rewrite
t3903-stash.sh: Add missing '&&' to body of testcase
git-stash: add new 'pop' subcommand
git-stash: add new 'drop' subcommand
git-reflog: add option --updateref to write the last reflog sha1 into the ref
refs.c: make close_ref() and commit_ref() non-static
git-reflog: add option --rewrite to update reflog entries while expiring
reflog-delete: parse standard reflog options
builtin-reflog.c: fix typo that accesses an unset variable
Teach "git reflog" a subcommand to delete single entries
* cb/mergetool:
Add a very basic test script for git mergetool
Teach git mergetool to use custom commands defined at config time
Changed an internal variable of mergetool to support custom commands
Tidy up git mergetool's backup file behaviour
This adds tests for recent change by Dmitry to fix the report "git
clean" gives on removed paths, and also makes sure the command detects
paths that is outside working tree.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When the given path contains '..' then git-clean incorrectly printed names
of files. This patch changes cmd_clean to use quote_path_relative().
Also, "failed to remove ..." message used absolutely path, but not it is
corrected to use relative path.
Signed-off-by: Dmitry Potapov <dpotapov@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Move quote_path() from wt-status.c to quote.c and rename it as
quote_path_relative(), because it is a better name for a public function.
Also, instead of handcrafted quoting, quote_c_style_counted() is now used,
to make its quoting more consistent with the rest of the system, also
honoring core.quotepath specified in configuration.
Signed-off-by: Dmitry Potapov <dpotapov@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The optional endp parameter to unquote_c_style() was supposed to point at
a location past the closing double quote, but it was going one beyond it.
git-fast-import used this function heavily and the bug caused it to
misparse the input stream, especially when parsing a rename command:
R "filename that needs quoting" rename-target-name
Because the function erroneously ate the whitespace after the closing dq,
this triggered "Missing space after source" error when it shouldn't.
Thanks to Adeodato Simò for having caught this.
Signed-off-by: Pierre Habouzit <madcoder@debian.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Dscho noticed that Term::ReadLine (used by send-email) colorized its
output for his TERM settings, inside t9001 tests.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The options --upload-pack (of git-fetch-pack) and --receive-pack (of
git-push) do the same as --exec (for both commands). But the former options
have the more descriptive name.
Signed-off-by: Uwe Kleine-König <Uwe.Kleine-Koenig@digi.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When get_pathspec() was originally made absolute-path capable,
we botched the interface to it, without dying inside the function
when given a path that is outside the work tree, and made it the
responsibility of callers to check the condition in a roundabout
way. This is made unnecessary with the previous patch.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Junio C Hamano