Calling cmd_foo() as if it is a general purpose helper function is
a no-no. Correct two instances of such to set an example.
* jc/no-cmd-as-subroutine:
merge-ours: do not use cmd_*() as a subroutine
describe: do not use cmd_*() as a subroutine
An earlier update made it possible to use an on-stack in-core
lockfile structure (as opposed to having to deliberately leak an
on-heap one). Many codepaths have been updated to take advantage
of this new facility.
* ma/lockfile-fixes:
read_cache: roll back lock in `update_index_if_able()`
read-cache: leave lock in right state in `write_locked_index()`
read-cache: drop explicit `CLOSE_LOCK`-flag
cache.h: document `write_locked_index()`
apply: remove `newfd` from `struct apply_state`
apply: move lockfile into `apply_state`
cache-tree: simplify locking logic
checkout-index: simplify locking logic
tempfile: fix documentation on `delete_tempfile()`
lockfile: fix documentation on `close_lock_file_gently()`
treewide: prefer lockfiles on the stack
sha1_file: do not leak `lock_file`
This is the "theoretically more correct" approach of simply
stepping back to the state before plumbing commands started paying
attention to "color.ui" configuration variable.
Let's run with this one.
* jk/ref-filter-colors-fix:
tag: respect color.ui config
Revert "color: check color.ui in git_default_config()"
Revert "t6006: drop "always" color config tests"
Revert "color: make "always" the same as "auto" in config"
Since 11b087adfd (ref-filter: consult want_color() before
emitting colors, 2017-07-13), we expect that setting
"color.ui" to "always" will enable color tag formats even
without a tty. As that commit was built on top of
136c8c8b8f (color: check color.ui in git_default_config(),
2017-07-13) from the same series, we didn't need to touch
tag's config parsing at all.
However, since we reverted 136c8c8b8f, we now need to
explicitly call git_color_default_config() to make this
work.
Let's do so, and also restore the test dropped in 0c88bf5050
(provide --color option for all ref-filter users,
2017-10-03). That commit swapped out our "color.ui=always"
test for "--color" in preparation for "always" going away.
But since it is here to stay, we should test both cases.
Note that for-each-ref also lost its color.ui support as
part of reverting 136c8c8b8f. But as a plumbing command, it
should _not_ respect the color.ui config. Since it also
gained a --color option in 0c88bf5050, that's the correct
way to ask it for color. We'll continue to test that, and
confirm that "color.ui" is not respected.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This reverts commit 136c8c8b8f.
That commit was trying to address a bug caused by 4c7f1819b3
(make color.ui default to 'auto', 2013-06-10), in which
plumbing like diff-tree defaulted to "auto" color, but did
not respect a "color.ui" directive to disable it.
But it also meant that we started respecting "color.ui" set
to "always". This was a known problem, but 4c7f1819b3 argued
that nobody ought to be doing that. However, that turned out
to be wrong, and we got a number of bug reports related to
"add -p" regressing in v2.14.2.
Let's revert 136c8c8b8, fixing the regression to "add -p".
This leaves the problem from 4c7f1819b3 unfixed, but:
1. It's a pretty obscure problem in the first place. I
only noticed it while working on the color code, and we
haven't got a single bug report or complaint about it.
2. We can make a more moderate fix on top by respecting
"never" but not "always" for plumbing commands. This
is just the minimal fix to go back to the working state
we had before v2.14.2.
Note that this isn't a pure revert. We now have a test in
t3701 which shows off the "add -p" regression. This can be
flipped to success.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
* 'jk/ui-color-always-to-auto-maint' (early part):
color: make "always" the same as "auto" in config
provide --color option for all ref-filter users
t3205: use --color instead of color.branch=always
t3203: drop "always" color test
t6006: drop "always" color config tests
t7502: use diff.noprefix for --verbose test
t7508: use test_terminal for color output
t3701: use test-terminal to collect color output
t4015: prefer --color to -c color.diff=always
test-terminal: set TERM=vt100
The call to cmd_diff_index() "git merge-ours" makes has been working
by accident that the function did not call exit(3), and the caller
exited almost immediately after making a call, but it sets a bad
precedent for people to cut and paste.
For finding out if the index exactly matches the HEAD (or a given
tree-ish), there is index_differs_from() which is exactly written
for that purpose.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The cmd_foo() function is a moral equivalent of 'main' for a Git
subcommand 'git foo', and as such, it is allowed to do many things
that make it unsuitable to be called as a subroutine, including
- call exit(3) to terminate the process;
- allocate resource held and used throughout its lifetime, without
releasing it upon return/exit;
- rely on global variables being initialized at program startup,
and update them as needed, making another clean invocation of the
function impossible.
The call to cmd_diff_index() "git describe" makes has been working
by accident that the function did not call exit(3); it sets a bad
precedent for people to cut and paste.
We could invoke it via the run_command() interface, but the diff
family of commands have helper functions in diff-lib.c that are
meant to be usable as subroutines, and using the latter does not
make the resulting code all that longer. Use it.
Note that there is also an invocation of cmd_name_rev() at the end;
"git describe --contains" massages its command line arguments to be
suitable for "git name-rev" invocation and jumps to it, never to
regain control. This call is left as-is as an exception to the
rule. When we start to allow calling name-rev repeatedly as a
helper function, we would be able to remove this call as well.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
A common mistake when writing binary search is to allow possible
integer overflow by using the simple average:
mid = (min + max) / 2;
Instead, use the overflow-safe version:
mid = min + (max - min) / 2;
This translation is safe since the operation occurs inside a loop
conditioned on "min < max". The included changes were found using
the following git grep:
git grep '/ *2;' '*.c'
Making this cleanup will prevent future review friction when a new
binary search is contructed based on existing code.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Reviewed-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Fix regression of "git add -p" for users with "color.ui = always"
in their configuration, by merging the topic below and adjusting it
for the 'master' front.
* jk/ui-color-always-to-auto:
t7301: use test_terminal to check color
t4015: use --color with --color-moved
color: make "always" the same as "auto" in config
provide --color option for all ref-filter users
t3205: use --color instead of color.branch=always
t3203: drop "always" color test
t6006: drop "always" color config tests
t7502: use diff.noprefix for --verbose test
t7508: use test_terminal for color output
t3701: use test-terminal to collect color output
t4015: prefer --color to -c color.diff=always
test-terminal: set TERM=vt100
Many variables that points at a region of memory that will live
throughout the life of the program have been marked with UNLEAK
marker to help the leak checkers concentrate on real leaks..
* ma/builtin-unleak:
builtin/: add UNLEAKs
If the original version of `write_locked_index()` returned with an
error, it didn't roll back the lockfile unless the error occured at the
very end, during closing/committing. See commit 03b866477 (read-cache:
new API write_locked_index instead of write_index/write_cache,
2014-06-13).
In commit 9f41c7a6b (read-cache: close index.lock in do_write_index,
2017-04-26), we learned to close the lock slightly earlier in the
callstack. That was mostly a side-effect of lockfiles being implemented
using temporary files, but didn't cause any real harm.
Recently, commit 076aa2cbd (tempfile: auto-allocate tempfiles on heap,
2017-09-05) introduced a subtle bug. If the temporary file is deleted
(i.e., the lockfile is rolled back), the tempfile-pointer in the `struct
lock_file` will be left dangling. Thus, an attempt to reuse the
lockfile, or even just to roll it back, will induce undefined behavior
-- most likely a crash.
Besides not crashing, we clearly want to make things consistent. The
guarantees which the lockfile-machinery itself provides is A) if we ask
to commit and it fails, roll back, and B) if we ask to close and it
fails, do _not_ roll back. Let's do the same for consistency.
Do not delete the temporary file in `do_write_index()`. One of its
callers, `write_locked_index()` will thereby avoid rolling back the
lock. The other caller, `write_shared_index()`, will delete its
temporary file anyway. Both of these callers will avoid undefined
behavior (crashing).
Teach `write_locked_index(..., COMMIT_LOCK)` to roll back the lock
before returning. If we have already succeeded and committed, it will be
a noop. Simplify the existing callers where we now have a superfluous
call to `rollback_lockfile()`. That should keep future readers from
wondering why the callers are inconsistent.
Signed-off-by: Martin Ågren <martin.agren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
`write_locked_index()` takes two flags: `COMMIT_LOCK` and `CLOSE_LOCK`.
At most one is allowed. But it is also possible to use no flag, i.e.,
`0`. But when `write_locked_index()` calls `do_write_index()`, the
temporary file, a.k.a. the lockfile, will be closed. So passing `0` is
effectively the same as `CLOSE_LOCK`, which seems like a bug.
We might feel tempted to restructure the code in order to close the file
later, or conditionally. It also feels a bit unfortunate that we simply
"happen" to close the lock by way of an implementation detail of
lockfiles. But note that we need to close the temporary file before
`stat`-ing it, at least on Windows. See 9f41c7a6b (read-cache: close
index.lock in do_write_index, 2017-04-26).
Drop `CLOSE_LOCK` and make it explicit that `write_locked_index()`
always closes the lock. Whether it is also committed is governed by the
remaining flag, `COMMIT_LOCK`.
This means we neither have nor suggest that we have a mode to write the
index and leave the file open. Whatever extra contents we might
eventually want to write, we should probably write it from within
`write_locked_index()` itself anyway.
Signed-off-by: Martin Ågren <martin.agren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We have two users of `struct apply_state` and the related functionality
in apply.c. Each user sets up its `apply_state` by handing over a
pointer to its static `lock_file`. (Before 076aa2cbd (tempfile:
auto-allocate tempfiles on heap, 2017-09-05), we could never free
lockfiles, so making them static was a reasonable approach.)
Other than that, they never directly access their `lock_file`s, which
are instead handled by the functionality in apply.c.
To make life easier for the caller and to make it less tempting for a
future caller to mess with the lock, make apply.c fully responsible for
setting up the `lock_file`. As mentioned above, it is now safe to free a
`lock_file`, so we can make the `struct apply_state` contain an actual
`struct lock_file` instead of a pointer to one.
The user in builtin/apply.c is rather simple. For builtin/am.c, we might
worry that the lock state is actually meant to be inherited across
calls. But the lock is only taken as `apply_all_patches()` executes, and
code inspection shows that it will always be released.
Alternatively, we can observe that the lock itself is never queried
directly. When we decide whether we should lock, we check a related
variable `newfd`. That variable is not inherited, so from the point of
view of apply.c, the state machine really is reset with each call to
`init_apply_state()`. (It would be a bug if `newfd` and the lock status
were not in sync. The duplication of information in `newfd` and the lock
will be addressed in the next patch.)
Signed-off-by: Martin Ågren <martin.agren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
`newfd` starts out negative. If we then take the lock, `newfd` will
become non-negative. We later check for exactly that property before
calling `write_locked_index()`. That is, we are simply using `newfd` as
a boolean to keep track of whether we took the lock or not. (We always
use `newfd` and `lock_file` together, so they really are mirroring each
other.)
Drop `newfd` and check with `is_lock_file_locked()` instead. While at
it, move the `static struct lock_file` into `cmd_checkout_index()` and
make it non-static. It is only used in this function, and after
076aa2cbd (tempfile: auto-allocate tempfiles on heap, 2017-09-05), we
can have lockfiles on the stack.
Signed-off-by: Martin Ågren <martin.agren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There is no longer any need to allocate and leak a `struct lock_file`.
The previous patch addressed an instance where we needed a minor tweak
alongside the trivial changes.
Deal with the remaining instances where we allocate and leak a struct
within a single function. Change them to have the `struct lock_file` on
the stack instead.
These instances were identified by running `git grep "^\s*struct
lock_file\s*\*"`.
Signed-off-by: Martin Ågren <martin.agren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Code clean-up.
* rs/cleanup-strbuf-users:
graph: use strbuf_addchars() to add spaces
use strbuf_addstr() for adding strings to strbufs
path: use strbuf_add_real_path()
Code clean-up.
* rs/resolve-ref-optional-result:
refs: pass NULL to resolve_refdup() if hash is not needed
refs: pass NULL to refs_resolve_refdup() if hash is not needed
Our documentation advises to not re-use a strbuf, after strbuf_release
has been called on it. Use the proper reset instead.
Currently 'strbuf_release' releases and re-initializes the strbuf, so it
is safe, but slow. 'strbuf_reset' only resets the internal length variable,
such that this could also be accounted for as a micro-optimization.
Reviewed-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The error messages shown when the branch command is misused
by supplying it wrong number of parameters wasn't meaningful.
That's because it used the the phrase "too many branches"
assuming all parameters to be "valid" branch names. It's not
always the case as exemplified below,
$ git branch
foo
* master
$ git branch -m foo foo old
fatal: too many branches for a rename operation
Change the messages to be more general thus making no assumptions
about the "parameters".
Signed-off-by: Kaartic Sivaraam <kaarticsivaraam91196@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
* jk/ui-color-always-to-auto-maint:
color: make "always" the same as "auto" in config
provide --color option for all ref-filter users
t3205: use --color instead of color.branch=always
t3203: drop "always" color test
t6006: drop "always" color config tests
t7502: use diff.noprefix for --verbose test
t7508: use test_terminal for color output
t3701: use test-terminal to collect color output
t4015: prefer --color to -c color.diff=always
test-terminal: set TERM=vt100
When ref-filter learned about want_color() in 11b087adfd
(ref-filter: consult want_color() before emitting colors,
2017-07-13), it became useful to be able to turn colors off
and on for specific commands. For git-branch, you can do so
with --color/--no-color.
But for git-for-each-ref and git-tag, the other users of
ref-filter, you have no option except to tweak the
"color.ui" config setting. Let's give both of these commands
the usual color command-line options.
This is a bit more obvious as a method for overriding the
config. And it also prepares us for the behavior of "always"
changing (so that we are still left with a way of forcing
color when our output goes to a non-terminal).
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Code clean-up to prevent future mistakes by copying and pasting
code that checks the result of read_in_full() function.
* jk/read-in-full:
worktree: check the result of read_in_full()
worktree: use xsize_t to access file size
distinguish error versus short read from read_in_full()
avoid looking at errno for short read_in_full() returns
prefer "!=" when checking read_in_full() result
notes-merge: drop dead zero-write code
files-backend: prefer "0" for write_in_full() error check
Some commands (most notably "git status") makes an opportunistic
update when performing a read-only operation to help optimize later
operations in the same repository. The new "--no-optional-locks"
option can be passed to Git to disable them.
* jk/no-optional-locks:
git: add --no-optional-locks option
"git branch" learned "-c/-C" to create a new branch by copying an
existing one.
* sd/branch-copy:
branch: fix "copy" to never touch HEAD
branch: add a --copy (-c) option to go with --move (-m)
branch: add test for -m renaming multiple config sections
config: create a function to format section headers
Recent versions of "git rev-parse --parseopt" did not parse the
option specification that does not have the optional flags (*=?!)
correctly, which has been corrected.
* bc/rev-parse-parseopt-fix:
parse-options: only insert newline in help text if needed
parse-options: write blank line to correct output stream
t0040,t1502: Demonstrate parse_options bugs
git-rebase: don't ignore unexpected command line arguments
rev-parse parseopt: interpret any whitespace as start of help text
rev-parse parseopt: do not search help text for flag chars
t1502: demonstrate rev-parse --parseopt option mis-parsing
The final batch to "git rebase -i" updates to move more code from
the shell script to C.
* js/rebase-i-final:
rebase -i: rearrange fixup/squash lines using the rebase--helper
t3415: test fixup with wrapped oneline
rebase -i: skip unnecessary picks using the rebase--helper
rebase -i: check for missing commits in the rebase--helper
t3404: relax rebase.missingCommitsCheck tests
rebase -i: also expand/collapse the SHA-1s via the rebase--helper
rebase -i: do not invent onelines when expanding/collapsing SHA-1s
rebase -i: remove useless indentation
rebase -i: generate the script via rebase--helper
t3415: verify that an empty instructionFormat is handled as before
Use strbuf_addstr() instead of strbuf_addf() for adding strings. That's
simpler and makes the intent clearer.
Patch generated by Coccinelle and contrib/coccinelle/strbuf.cocci;
adjusted indentation in refs/packed-backend.c manually.
Signed-off-by: Rene Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add some UNLEAKs where we are about to return from `cmd_*`. UNLEAK the
variables in the same order as we've declared them. While addressing
`msg` in builtin/tag.c, convert the existing `strbuf_release()` calls as
well.
Signed-off-by: Martin Ågren <martin.agren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This allows us to get rid of several write-only variables.
Signed-off-by: Rene Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Memory leaks in various codepaths have been plugged.
* ma/leakplugs:
pack-bitmap[-write]: use `object_array_clear()`, don't leak
object_array: add and use `object_array_pop()`
object_array: use `object_array_clear()`, not `free()`
leak_pending: use `object_array_clear()`, not `free()`
commit: fix memory leak in `reduce_heads()`
builtin/commit: fix memory leak in `prepare_index()`
"git fast-export" with -M/-C option issued "copy" instruction on a
path that is simultaneously modified, which was incorrect.
* jt/fast-export-copy-modify-fix:
fast-export: do not copy from modified file
"git describe --match <pattern>" has been taught to play well with
the "--all" option.
* mk/describe-match-with-all:
describe: teach --match to handle branches and remotes
Code clean-up.
* rs/resolve-ref-optional-result:
refs: pass NULL to resolve_ref_unsafe() if hash is not needed
refs: pass NULL to refs_resolve_ref_unsafe() if hash is not needed
refs: make sha1 output parameter of refs_resolve_ref_unsafe() optional
Many codepaths have been updated to squelch -Wimplicit-fallthrough
warnings from Gcc 7 (which is a good code hygiene).
* jk/fallthrough:
consistently use "fallthrough" comments in switches
curl_trace(): eliminate switch fallthrough
test-line-buffer: simplify command parsing
"git describe --match" learned to take multiple patterns in v2.13
series, but the feature ignored the patterns after the first one
and did not work at all. This has been fixed.
* jk/describe-omit-some-refs:
describe: fix matching to actually match all patterns
Some tools like IDEs or fancy editors may periodically run
commands like "git status" in the background to keep track
of the state of the repository. Some of these commands may
refresh the index and write out the result in an
opportunistic way: if they can get the index lock, then they
update the on-disk index with any updates they find. And if
not, then their in-core refresh is lost and just has to be
recomputed by the next caller.
But taking the index lock may conflict with other operations
in the repository. Especially ones that the user is doing
themselves, which _aren't_ opportunistic. In other words,
"git status" knows how to back off when somebody else is
holding the lock, but other commands don't know that status
would be happy to drop the lock if somebody else wanted it.
There are a couple possible solutions:
1. Have some kind of "pseudo-lock" that allows other
commands to tell status that they want the lock.
This is likely to be complicated and error-prone to
implement (and maybe even impossible with just
dotlocks to work from, as it requires some
inter-process communication).
2. Avoid background runs of commands like "git status"
that want to do opportunistic updates, preferring
instead plumbing like diff-files, etc.
This is awkward for a couple of reasons. One is that
"status --porcelain" reports a lot more about the
repository state than is available from individual
plumbing commands. And two is that we actually _do_
want to see the refreshed index. We just don't want to
take a lock or write out the result. Whereas commands
like diff-files expect us to refresh the index
separately and write it to disk so that they can depend
on the result. But that write is exactly what we're
trying to avoid.
3. Ask "status" not to lock or write the index.
This is easy to implement. The big downside is that any
work done in refreshing the index for such a call is
lost when the process exits. So a background process
may end up re-hashing a changed file multiple times
until the user runs a command that does an index
refresh themselves.
This patch implements the option 3. The idea (and the test)
is largely stolen from a Git for Windows patch by Johannes
Schindelin, 67e5ce7f63 (status: offer *not* to lock the
index and update it, 2016-08-12). The twist here is that
instead of making this an option to "git status", it becomes
a "git" option and matching environment variable.
The reason there is two-fold:
1. An environment variable is carried through to
sub-processes. And whether an invocation is a
background process or not should apply to the whole
process tree. So you could do "git --no-optional-locks
foo", and if "foo" is a script or alias that calls
"status", you'll still get the effect.
2. There may be other programs that want the same
treatment.
I've punted here on finding more callers to convert,
since "status" is the obvious one to call as a repeated
background job. But "git diff"'s opportunistic refresh
of the index may be a good candidate.
The test is taken from 67e5ce7f63, and it's worth repeating
Johannes's explanation:
Note that the regression test added in this commit does
not *really* verify that no index.lock file was written;
that test is not possible in a portable way. Instead, we
verify that .git/index is rewritten *only* when `git
status` is run without `--no-optional-locks`.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We try to read "len" bytes into a buffer and just assume
that it happened correctly. In practice this should usually
be the case, since we just stat'd the file to get the
length. But we could be fooled by transient errors or by
other processes racily truncating the file.
Let's be more careful. There's a slim chance this could
catch a real error, but it also prevents people and tools
from getting worried while reading the code.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
To read the "gitdir" file into memory, we stat the file and
allocate a buffer. But we store the size in an "int", which
may be truncated. We should use a size_t and xsize_t(),
which will detect truncation.
An overflow is unlikely for a "gitdir" file, but it's a good
practice to model.
Signed-off-by: Jeff King <peff@peff.net>
Reviewed-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Many callers of read_in_full() expect to see the exact
number of bytes requested, but their error handling lumps
together true read errors and short reads due to unexpected
EOF.
We can give more specific error messages by separating these
cases (showing errno when appropriate, and otherwise
describing the short read).
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Comparing the result of read_in_full() using less-than is
potentially dangerous, as a negative return value may be
converted to an unsigned type and be considered a success.
This is discussed further in 561598cfcf (read_pack_header:
handle signed/unsigned comparison in read result,
2017-09-13).
Each of these instances is actually fine in practice:
- in get-tar-commit-id, the HEADERSIZE macro expands to a
signed integer. If it were switched to an unsigned type
(e.g., a size_t), then it would be a bug.
- the other two callers check for a short read only after
handling a negative return separately. This is a fine
practice, but we'd prefer to model "!=" as a general
rule.
So all of these cases can be considered cleanups and not
actual bugfixes.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
"git rev-parse" learned "--is-shallow-repository", that is to be
used in a way similar to existing "--is-bare-repository" and
friends.
* ow/rev-parse-is-shallow-repo:
rev-parse: rev-parse: add --is-shallow-repository
"git gc" tries to avoid running two instances at the same time by
reading and writing pid/host from and to a lock file; it used to
use an incorrect fscanf() format when reading, which has been
corrected.
* aw/gc-lockfile-fscanf-fix:
gc: call fscanf() with %<len>s, not %<len>c, when reading hostname
Step #0 of a planned & larger series to make the in-core object
store per in-core repository object.
* jn/per-repo-object-store-fixes:
replace-objects: evaluate replacement refs without using the object store
push, fetch: error out for submodule entries not pointing to commits
pack: make packed_git_mru global a value instead of a pointer
