There's a bug in builtin/am.c in which we take a lock on
MERGE_RR recursively. But rather than fix am.c, this patch
fixes the confusing interface from rerere.c that caused the
bug. Read on for the gory details.
The setup_rerere() function both reads the existing MERGE_RR
file, and takes MERGE_RR.lock. In the rerere() and
rerere_forget() functions, we end up in write_rr(), which
will then commit the lock file.
But for functions like rerere_clear() that do not write to
MERGE_RR, we expect the caller to have handled
setup_rerere(). That caller would then need to release the
lockfile, but it can't; the lock struct is local to
rerere.c.
For builtin/rerere.c, this is OK. We run a single rerere
operation and then exit immediately, which has the side
effect of rolling back the lockfile.
But in builtin/am.c, this is actively wrong. If we run "git
am -3 --skip", we call setup-rerere twice without releasing
the lock:
1. The "--skip" causes us to call am_rerere_clear(), which
calls setup_rerere(), but never drops the lock.
2. We then proceed to the next patch.
3. The "--3way" may cause us to call rerere() to handle
conflicts in that patch, but we are already holding the
lock. The lockfile code dies with:
BUG: prepare_tempfile_object called for active object
We could fix this by having rerere_clear() call
rollback_lock_file(). But it feels a bit odd for it to roll
back a lockfile that it did not itself take. So let's
simplify the interface further, and handle setup_rerere in
the function itself, taking away the question from the
caller over whether they need to do so.
We can give rerere_gc() the same treatment, as well (even
though it doesn't have any callers besides builtin/rerere.c
at this point). Note that these functions don't take flags
from their callers to pass along to setup_rerere; that's OK,
because the flags would not be meaningful for what they are
doing.
Both of those functions need to hold the lock because even
though they do not write to MERGE_RR, they are still writing
and should be protected from a simultaneous "rerere" run.
But rerere_remaining(), "rerere diff", and "rerere status"
are all read-only operations. They want to setup_rerere(),
but do not care about taking the lock in the first place.
Since our update of MERGE_RR is the usual atomic rename done
by commit_lock_file, they can just do a lockless read. For
that, we teach setup_rerere a READONLY flag to avoid the
lock.
As a bonus, this pushes builtin/rerere.c's setup_rerere call
closer to the functions that use it. Which means that "git
rerere totally-bogus-command" will no longer silently
exit(0) in a repository without rerere enabled.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
One of the most common uses of git_path() is to pass a
constant, like git_path("MERGE_MSG"). This has two
drawbacks:
1. The return value is a static buffer, and the lifetime
is dependent on other calls to git_path, etc.
2. There's no compile-time checking of the pathname. This
is OK for a one-off (after all, we have to spell it
correctly at least once), but many of these constant
strings appear throughout the code.
This patch introduces a series of functions to "memoize"
these strings, which are essentially globals for the
lifetime of the program. We compute the value once, take
ownership of the buffer, and return the cached value for
subsequent calls. cache.h provides a helper macro for
defining these functions as one-liners, and defines a few
common ones for global use.
Using a macro is a little bit gross, but it does nicely
document the purpose of the functions. If we need to touch
them all later (e.g., because we learned how to change the
git_dir variable at runtime, and need to invalidate all of
the stored values), it will be much easier to have the
complete list.
Note that the shared-global functions have separate, manual
declarations. We could do something clever with the macros
(e.g., expand it to a declaration in some places, and a
declaration _and_ a definition in path.c). But there aren't
that many, and it's probably better to stay away from
too-magical macros.
Likewise, if we abandon the C preprocessor in favor of
generating these with a script, we could get much fancier.
E.g., normalizing "FOO/BAR-BAZ" into "git_path_foo_bar_baz".
But the small amount of saved typing is probably not worth
the resulting confusion to readers who want to grep for the
function's definition.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
By consistently using "upon failure, set 'ret' and jump to out"
pattern, flatten the function further.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This gives a thin abstraction between the conflict ID that is a hash
value obtained by inspecting the conflicts and the name of the
directory under $GIT_DIR/rr-cache/, in which the previous resolution
is recorded to be replayed. The plan is to make sure that the
presence of the directory does not imply the presense of a previous
resolution and vice-versa, and later allow us to have more than one
pair of <preimage, postimage> for a given conflict ID.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Most places we call conflict IDs "name" and some others we call them
"hex"; update all of them to "id".
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Extract the body of a loop that attempts to replay recorded
resolution for each conflicted path into a helper function, not
because I want to call it from multiple places later, but because
the logic has become too deeply nested and hard to read.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Explain the internals of rerere as in-code comments, while
sprinkling "NEEDSWORK" comment to highlight iffy bits and
questionable assumptions.
This covers the codepath that implements "rerere gc" and "rerere
clear".
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Explain the internals of rerere as in-code comments, while
sprinkling "NEEDSWORK" comment to highlight iffy bits and
questionable assumptions.
This covers the codepath that implements "rerere forget".
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Explain the internals of rerere as in-code comments, while
sprinkling "NEEDSWORK" comment to highlight iffy bits and
questionable assumptions.
This one covers the codepath reached from rerere(), the primary
interface to the subsystem.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Explain the internals of rerere as in-code comments, while
sprinkling "NEEDSWORK" comment to highlight iffy bits and
questionable assumptions.
This one covers the "$GIT_DIR/MERGE_RR" file and in-core merge_rr
that are used to keep track of the status of "rerere" session in
progress.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
rerere_io_putconflict() wants to use a limited fixed-sized buf[] on
stack repeatedly to formulate a longer string, but its implementation
is doubly confusing:
* When it knows that the whole thing fits in buf[], it wants to
fill early part of buf[] with conflict marker characters,
followed by a LF and a NUL. It miscounts the size of the buffer
by 1 and does not use the last byte of buf[].
* When it needs to show only the early part of a long conflict
marker string (because the whole thing does not fit in buf[]), it
adjusts the number of bytes shown in the current round in a
strange-looking way. It makes sure that this round does not emit
all bytes and leaves at least one byte to the next round, so that
"it all fits" case will pick up the rest and show the terminating
LF. While this is correct, one needs to stop and think for a
while to realize why it is correct without an explanation.
Fix the benign off-by-one, and add comments to explain the
strange-looking size adjustment.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Explain the internals of rerere as in-code comments.
This one covers our thin I/O abstraction to read from either
a file or a memory while optionally writing out to a file.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
A conflicted index can have multiple stage #1 entries when dealing
with a criss-cross merge and using the "resolve" merge strategy.
Plug the leak by reading only the first one of the same stage
entries.
Strictly speaking, this fix does change the semantics, in that we
used to use the last stage #1 entry as the common ancestor when
doing the plain-vanilla three-way merge, but with the leak fix, we
will use the first stage #1 entry. But it is not a grave backward
compatibility breakage. Either way, we are arbitrarily picking one
of multiple stage #1 entries and using it, ignoring others, and
there is no meaning in the ordering of these stage #1 entries.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
handle_cache() loops 3 times starting from an index entry that is
unmerged, while ignoring an entry for a path that is different from
what we are looking for.
As the index is sorted, once we see a different path, we know we saw
all stages for the path we are interested in. Just loop while we
see the same path and then break, instead of continuing for 3 times.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
As the nature of the conflict marker line determines if there should
be a SP and label after it, the caller shouldn't have to pass the
parameter redundantly.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Instead of writing the hash for a conflict, a HT, and the path
with three separate write_in_full() calls, format them into a
single record into a strbuf and write it out in one go.
As a more recent "rerere remaining" codepath abuses the .util field
of the merge_rr data to store a sentinel token, make sure that
codepath does not call into this function (of course, "remaining" is
a read-only operation and currently does not call it).
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The MERGE_RR file records a collection of NUL-terminated entries,
each of which consists of
- a hash that identifies the conflict
- a HT
- the pathname
We used to read this piece-by-piece, and worse yet, read the
pathname part a byte at a time into a fixed buffer of size PATH_MAX.
Instead, read a whole entry using strbuf_getwholeline() and parse
out the fields. This way, we issue fewer read(2) calls and more
importantly we do not have to limit the pathname to PATH_MAX.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The merge_rr string list stores the conflict ID (a hexadecimal
string that is used to index into $GIT_DIR/rr-cache) in the .util
field of its elements, and when do_plain_rerere() resolves a
conflict, the field is cleared. Also, when rerere_forget()
recomputes the conflict ID to updates the preimage file, the
conflict ID for the path is updated.
We forgot to free the existing conflict ID when we did these two
operations.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
A conflicted index can have multiple stage #1 entries when dealing
with a criss-cross merge and using the "resolve" merge strategy.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When ac49f5ca (rerere "remaining", 2011-02-16) split out a new
helper function check_one_conflict() out of find_conflict()
function, so that the latter will use the returned value from the
new helper to update the loop control variable that is an index into
active_cache[], the new variable incremented the index by one too
many when it found a path with only stage #1 entry at the very end
of active_cache[].
This "strange" return value does not have any effect on the loop
control of two callers of this function, as they all notice that
active_nr+2 is larger than active_nr just like active_nr+1 is, but
nevertheless it puzzles the readers when they are trying to figure
out what the function is trying to do.
In fact, there is no need to do an early return. The code that
follows after skipping the stage #1 entry is fully prepared to
handle a case where the entry is at the very end of active_cache[].
Help future readers from unnecessary confusion by dropping an early
return. We skip the stage #1 entry, and if there are stage #2 and
stage #3 entries for the same path, we diagnose the path as
THREE_STAGED (otherwise we say PUNTED), and then we skip all entries
for the same path.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If you run "git rerere forget foo" in a repository that does
not have rerere enabled, git hits an internal error:
$ git init -q
$ git rerere forget foo
fatal: BUG: attempt to commit unlocked object
The problem is that setup_rerere() will not actually take
the lock if the rerere system is disabled. We should notice
this and return early. We can return with a success code
here, because we know there is nothing to forget.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We have been silently tolerating errors by returning early with an
error that the caller ignores since rerere.autoupdate was introduced
in v1.6.0-rc0~120^2 (2008-06-22). So on error (for example if the
index is already locked), rerere can return success silently without
updating the index or with only some items in the index updated.
Better to treat such failures as a fatal error so the operator can
figure out what is wrong and fix it.
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Move the interface declaration for the functions in lockfile.c from
cache.h to a new file, lockfile.h. Add #includes where necessary (and
remove some redundant includes of cache.h by files that already
include builtin.h).
Move the documentation of the lock_file state diagram from lockfile.c
to the new header file.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Use `git_config_get_*()` family instead of `git_config()` to take advantage of
the config-set API which provides a cleaner control flow.
Signed-off-by: Tanay Abhra <tanayabh@gmail.com>
Reviewed-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This patch activates the DO_MATCH_DIRECTORY code in m_p_i(), which
makes "git diff HEAD submodule/" and "git diff HEAD submodule" produce
the same output. Previously only the version without trailing slash
returns the difference (if any).
That's the effect of new ce_path_match(). dir_path_match() is not
executed by the new tests. And it should not introduce regressions.
Previously if path "dir/" is passed in with pathspec "dir/", they
obviously match. With new dir_path_match(), the path becomes
_directory_ "dir" vs pathspec "dir/", which is not executed by the old
code path in m_p_i(). The new code path is executed and produces the
same result.
The other case is pathspec "dir" and path "dir/" is now turned to
"dir" (with DO_MATCH_DIRECTORY). Still the same result before or after
the patch.
So why change? Because of the next patch about clean.c.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
A long time ago, for some reason I was not happy with
match_pathspec(). I created a better version, match_pathspec_depth()
that was suppose to replace match_pathspec()
eventually. match_pathspec() has finally been gone since 6 months
ago. Use the shorter name for match_pathspec_depth().
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
I attempted to make index_state->cache[] a "const struct cache_entry **"
to find out how existing entries in index are modified and where. The
question I have is what do we do if we really need to keep track of on-disk
changes in the index. The result is
- diff-lib.c: setting CE_UPTODATE
- name-hash.c: setting CE_HASHED
- preload-index.c, read-cache.c, unpack-trees.c and
builtin/update-index: obvious
- entry.c: write_entry() may refresh the checked out entry via
fill_stat_cache_info(). This causes "non-const struct cache_entry
*" in builtin/apply.c, builtin/checkout-index.c and
builtin/checkout.c
- builtin/ls-files.c: --with-tree changes stagemask and may set
CE_UPDATE
Of these, write_entry() and its call sites are probably most
interesting because it modifies on-disk info. But this is stat info
and can be retrieved via refresh, at least for porcelain
commands. Other just uses ce_flags for local purposes.
So, keeping track of "dirty" entries is just a matter of setting a
flag in index modification functions exposed by read-cache.c. Except
unpack-trees, the rest of the code base does not do anything funny
behind read-cache's back.
The actual patch is less valueable than the summary above. But if
anyone wants to re-identify the above sites. Applying this patch, then
this:
diff --git a/cache.h b/cache.h
index 430d021..1692891 100644
--- a/cache.h
+++ b/cache.h
@@ -267,7 +267,7 @@ static inline unsigned int canon_mode(unsigned int mode)
#define cache_entry_size(len) (offsetof(struct cache_entry,name) + (len) + 1)
struct index_state {
- struct cache_entry **cache;
+ const struct cache_entry **cache;
unsigned int version;
unsigned int cache_nr, cache_alloc, cache_changed;
struct string_list *resolve_undo;
will help quickly identify them without bogus warnings.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The loop that fills in the buffers that are later passed to the merge
driver exits early when not all stages of a path are present in the index.
But since the buffer pointers are not initialized in advance, the
subsequent accesses are undefined.
Initialize buffer pointers in advance to avoid undefined behavior later.
That is not sufficient, though, to get correct operation of handle_cache().
The function replays a conflicted merge to extract the part inside the
conflict markers. As written, the loop exits early when a stage is missing.
Consequently, the buffers for later stages that would be present in the
index are not filled in and the merge is replayed with incomplete data.
Fix it by investigating all stages of the given path.
Signed-off-by: Johannes Sixt <j6t@kdbg.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Using 'git rerere forget .' after a merge that involved binary files
runs into an infinite loop if the binary file contains a zero byte.
Replace a strchrnul by memchr because the former does not make progress
as soon as the NUL is encountered.
Signed-off-by: Johannes Sixt <j6t@kdbg.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This is the last remaining call to mkdir(2) that restricts the permission
bits by passing 0755. Just use the same mkdir_in_gitdir() used to create
the leaf directories.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If we reach EOF after the SHA1-then-TAB, yet before the NUL that
terminates each file name, we would fill the file name buffer with \255
bytes resulting from the repeatedly-failing fgetc (returns EOF/-1) and
ultimately complain about "filename too long", because no NUL was
encountered.
Signed-off-by: Jim Meyering <jim@meyering.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This moves the two features from builtin/rerere.c to a more library-ish
portion of the codebase. No behaviour change.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
* load_file() returns a void pointer but is using 0 for the return
value
* builtin/receive-pack.c forgot to include builtin.h
* packet_trace_prefix can be marked static
* ll_merge takes a pointer for its last argument, not an int
* crc32 expects a pointer as the second argument but Z_NULL is defined
to be 0 (see 38f4d13 sparse fix: Using plain integer as NULL pointer,
2006-11-18 for more info)
Signed-off-by: Stephen Boyd <bebarino@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The git-new-workdir script in contrib/ makes a new work tree by sharing
many subdirectories of the .git directory with the original repository.
When rerere.enabled is set in the original repository, but the user has
not encountered any conflicts yet, the original repository may not yet
have .git/rr-cache directory.
When rerere wants to run in a new work tree created from such a young
original repository, it fails to mkdir(2) .git/rr-cache that is a symlink
to a yet-to-be-created directory.
There are three possible approaches to this:
- A naive solution is not to create a symlink in the git-new-workdir
script to a directory the original does not have (yet). This is not a
solution, as we tend to lazily create subdirectories of .git/, and
having rerere.enabled configuration set is a strong indication that the
user _wants_ to have this lazy creation to happen;
- We could always create .git/rr-cache upon repository creation. This is
tempting but will not help people with existing repositories.
- Detect this case by seeing that mkdir(2) failed with EEXIST, checking
that the path is a symlink, and try running mkdir(2) on the link
target.
This patch solves the issue by doing the third one.
Strictly speaking, this is incomplete. It does not attempt to handle
relative symbolic link that points into the original repository, but this
is good enough to help people who use contrib/workdir/git-new-workdir
script.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
After "rerere" resolves conflicts by reusing old resolution, there would
be three kinds of paths with conflict in the index:
* paths that have been resolved in the working tree by rerere;
* paths that need further work whose resolution could be recorded;
* paths that need resolving that rerere won't help.
When the user wants a list of paths that need hand-resolving, output from
"rerere status" does not help, as it shows only the second category, but
the paths in the third category still needs work (rerere only makes sense
for regular files that have both our side and their side, and does not
help other kinds of conflicts, e.g. "we modified, they deleted").
The new subcommand "rerere remaining" can be used to show both. As
opposed to "rerere status", this subcommand also skips printing paths
that have been added to the index, since these paths are already
resolved and are no longer "remaining".
Initial patch provided by Junio. Refactored and modified to skip
resolved paths by Martin. Commit message mostly by Junio.
Helped-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Martin von Zweigbergk <martin.von.zweigbergk@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Keeping track of the flag bits is proving more trouble than it's
worth. Instead, use a pointer to an options struct like most similar
APIs do.
Callers with no special requests can pass NULL to request the default
options.
Cc: Bert Wesarg <bert.wesarg@googlemail.com>
Cc: Avery Pennarun <apenwarr@gmail.com>
Helped-by: Justin Frankel <justin@cockos.com>
Helped-by: Bert Wesarg <bert.wesarg@googlemail.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
plain rerere performs three tasks; let us consider how the new
merge.renormalize option should apply to each.
After an unsuccessful merge, rerere records conflict hunks from the
work tree under .git/rr-cache. If the merge was performed with
merge.renormalize enabled, both sides of the conflict hunk use the
current work tree’s end-of-line and smudge rules; there is not really
much of a choice.
After a successful manual resolution, rerere records the postimage.
Here, also, the file will be in the current work tree’s canonical
format and there is not much to do about it.
When encountering that conflict again, merge looks up the preimage
and postimage using the conflict hunk as a key and runs a three-way
merge to apply that resolution to the work tree. Since the conflict
hunk used the current work tree’s canonical format, chances are the
version in the work tree, the preimage, and the postimage will, too.
In fact using the merge.renormalize machinery is exactly the wrong
thing to do, since its result has been run through convert_to_git
and therefore is not suitable for writing to the work tree.
The only affected caller is "git merge".
NEEDSWORK: lacks test
Cc: Eyvind Bernhardsen <eyvind.bernhardsen@gmail.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add a “renormalize” bit to the ll-merge options word so callers can
decide on a case-by-case basis whether the merge is likely to have
overlapped with a change in smudge/clean rules.
This reveals a few commands that have not been taking that situation
into account, though it does not fix them.
No functional change intended.
Cc: Eyvind Bernhardsen <eyvind.bernhardsen@gmail.com>
Improved-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
'rerere gc' prunes resolutions of conflicted merges that occurred long
time ago, and when doing so it takes the creation time of the
conflicted automerge results into account. This can cause the loss of
frequently used conflict resolutions (e.g. long-living topic branches
are merged into a regularly rebuilt integration branch (think of git's
pu)) when they become old enough to exceed 'rerere gc's threshold.
To prevent the loss of valuable merge resolutions 'rerere' will (1)
update the timestamp of the recorded conflict resolution (i.e.
'postimage') each time when encountering and resolving the same merge
conflict, and (2) take this timestamp, i.e. the time of the last usage
into account when gc'ing.
Signed-off-by: SZEDER Gábor <szeder@ira.uka.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Jeff King