Junio C Hamano [Mon, 31 Jul 2023 22:44:09 +0000 (15:44 -0700)]
checkout: allow "checkout -m path" to unmerge removed paths
"git checkout -m -- path" uses the unmerge_marked_index() API, whose
implementation is incapable of unresolving a path that was resolved
as removed. Extend the unmerge_index() API function so that we can
mark the ce_flags member of the cache entries we add to the index as
unmerged, and replace use of unmerge_marked_index() with it.
Now, together with its unmerge_index_entry_at() helper function,
unmerge_marked_index() function is no longer called by anybody, and
can safely be removed.
This makes two known test failures in t2070 and t7201 to succeed.
Junio C Hamano [Mon, 31 Jul 2023 22:44:08 +0000 (15:44 -0700)]
checkout/restore: add basic tests for --merge
Even though "checkout --merge -- paths" had some tests, we never
made sure it worked to recreate the conflicted state _after_ the
resolution was recorded in the index. Also "restore --merge" did
not even have any tests.
Currently these commands use the unmerge_marked_index() interface
that cannot handle paths that have been resolved as removal, and
tests for that case are marked with test_expect_failure; these
should eventually be fixed, but not in this patch.
Junio C Hamano [Mon, 31 Jul 2023 22:44:07 +0000 (15:44 -0700)]
checkout/restore: refuse unmerging paths unless checking out of the index
Recreating unmerged index entries using resolve-undo data,
recreating conflicted working tree files using unmerged index
entries, and writing data out of unmerged index entries, make
sense only when we are checking paths out of the index and not when
we are checking paths out of a tree-ish.
Add an extra check to make sure "--merge" and "--ours/--theirs"
options are rejected when checking out from a tree-ish, update the
document (especially the SYNOPSIS section) to highlight that they
are incompatible, and add a few tests to make sure the combination
fails.
Junio C Hamano [Mon, 31 Jul 2023 22:44:06 +0000 (15:44 -0700)]
update-index: remove stale fallback code for "--unresolve"
The "update-index --unresolve" is a relatively old feature that was
introduced in Git v1.4.1 (June 2006), which predates the
resolve-undo extension introduced in Git v1.7.0 (February 2010).
The original code that was limited only to work during a merge (and
not during a rebase or a cherry-pick) has been kept as the fallback
codepath to be used as a transition measure.
By now, for more than 10 years we have stored resolve-undo extension
in the index file, and the fallback code way outlived its usefulness.
Remove it, together with two file-scope static global variables.
One of these variables is still used by surviving function, but it
does not have to be a global at all, so move it to local to that
function.
Junio C Hamano [Mon, 31 Jul 2023 22:44:05 +0000 (15:44 -0700)]
update-index: use unmerge_index_entry() to support removal
"update-index --unresolve" uses the unmerge_index_entry_at() that
assumes that the path to be unresolved must be in the index, which
makes it impossible to unresolve a path that was resolved as removal.
Rewrite unresolve_one() to use the unmerge_index_entry() to support
unresolving such a path.
Existing tests for "update-index --unresolve" forgot to check one
thing that tests for "checkout --merge -- paths" tested, which is to
make sure that resolve-undo record that has already been used to
recreate higher-stage index entries is removed. Add new invocations
of "ls-files --resolve-undo" after running "update-index --unresolve"
to make sure that unresolving with update-index does remove the used
resolve-undo records.
Junio C Hamano [Mon, 31 Jul 2023 22:44:04 +0000 (15:44 -0700)]
resolve-undo: allow resurrecting conflicted state that resolved to deletion
The resolve-undo index extension records up to three (mode, object
name) tuples for non-zero stages for each path that was resolved,
to be used to recreate the original conflicted state later when the
user requests.
The unmerge_index_entry_at() function uses the resolve-undo data to
do so, but it assumes that the path for which the conflicted state
needs to be recreated can be specified by the position in the
active_cache[] array. This obviously cannot salvage the state of
conflicted paths that were resolved by removing them. For example,
a delete-modify conflict, in which the change whose "modify" side
made is a trivial typofix, may legitimately be resolved to remove
the path, and resolve-undo extension does record the two (mode,
object name) tuples for the common ancestor version and their
version, lacking our version. But after recording such a removal of
the path, you should be able to use resolve-undo data to recreate
the conflicted state.
Introduce a new unmerge_index_entry() helper function that takes the
path (which does not necessarily have to exist in the active_cache[]
array) and resolve-undo data, and use it to reimplement unmerge_index()
public function that is used by "git rerere".
The limited interface is still kept for now, as it is used by "git
checkout -m" and "git update-index --unmerge", but these two codepaths
will be updated to lift the assumption to allow conflicts that resolved
to deletion can be recreated.
Junio C Hamano [Mon, 31 Jul 2023 22:44:03 +0000 (15:44 -0700)]
update-index: do not read HEAD and MERGE_HEAD unconditionally
When "update-index --unresolve $path" cannot find the resolve-undo
record for the path the user requested to unresolve, it stuffs the
blobs from HEAD and MERGE_HEAD to stage #2 and stage #3 as a
fallback. For this reason, the operation does not even start unless
both "HEAD" and "MERGE_HEAD" exist.
This is suboptimal in a few ways:
* It does not recreate stage #1. Even though it is a correct
design decision not to do so (because it is impossible to
recreate in general cases, without knowing how we got there,
including what merge strategy was used), it is much less useful
not to have that information in the index.
* It limits the "unresolve" operation only during a conflicted "git
merge" and nothing else. Other operations like "rebase",
"cherry-pick", and "switch -m" may result in conflicts, and the
user may want to unresolve the conflict that they incorrectly
resolved in order to redo the resolution, but the fallback would
not kick in.
* Most importantly, the entire "unresolve" operation is disabled
after a conflicted merge is committed and MERGE_HEAD is removed,
even though the index has perfectly usable resolve-undo records.
By lazily reading the HEAD and MERGE_HEAD only when we need to go to
the fallback codepath, we will allow cases where resolve-undo
records are available (which is 100% of the time, unless the user is
reading from an index file created by Git more than 10 years ago) to
proceed even after a conflicted merge was committed, during other
mergy operations that do not use MERGE_HEAD, or after the result of
such mergy operations has been committed.
Adam Majer [Mon, 31 Jul 2023 13:42:02 +0000 (15:42 +0200)]
doc: sha256 is no longer experimental
Remove scary wording that basically stops people using sha256
repositories not because of interoperability issues with sha1
repositories, but from fear that their work will suddenly become
incompatible in some future version of git.
We should be clear that currently sha256 repositories will not work with
sha1 repositories but stop the scary words.
Signed-off-by: Adam Majer <adamm@zombino.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Eric Wong [Mon, 31 Jul 2023 12:08:07 +0000 (12:08 +0000)]
sha256/gcrypt: fix memory leak with SHA-256 repos
`gcry_md_open' needs to be paired with `gcry_md_close' to ensure
resources are released. Since our internal APIs don't have
separate close/release callbacks, sticking it into the finalization
callback seems appropriate.
Building with SANITIZE=leak and running `git fsck' on a SHA-256
repository no longer reports leaks.
Signed-off-by: Eric Wong <e@80x24.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Eric Wong [Mon, 31 Jul 2023 12:08:06 +0000 (12:08 +0000)]
sha256/gcrypt: fix build with SANITIZE=leak
Non-static functions cause `undefined reference' errors when
building with `SANITIZE=leak' due to the lack of prototypes.
Mark all these functions as `static inline' as we do in
sha256/nettle.h to avoid the need to maintain prototypes.
Signed-off-by: Eric Wong <e@80x24.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
René Scharfe [Sat, 29 Jul 2023 20:40:27 +0000 (22:40 +0200)]
bundle: use OPT_PASSTHRU_ARGV
"git bundle" passes the progress control options to "git pack-objects"
by parsing and then recreating them explicitly. Simplify that process
by using OPT_PASSTHRU_ARGV instead.
This also fixes --no-quiet, which has been doing the same as --quiet
since its introduction by 79862b6b77 (bundle-create: progress output
control, 2019-11-10) because it had been defined using OPT_SET_INT with
a value of 0, which sets 0 when negated as well.
Signed-off-by: René Scharfe <l.s.r@web.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Junio C Hamano [Fri, 28 Jul 2023 00:43:17 +0000 (17:43 -0700)]
MyFirstContribution: refrain from self-iterating too much
Finding mistakes in and improving your own patches is a good idea,
but doing so too quickly is being inconsiderate to reviewers who
have just seen the initial iteration and taking their time to review
it. Encourage new developers to perform such a self review before
they send out their patches, not after. After sending a patch that
they immediately found mistakes in, they are welcome to comment on
them, mentioning what and how they plan to improve them in an
updated version, before sending out their updates.
Helped-by: Torsten Bögershausen <tboegi@web.de> Helped-by: Linus Arver <linusa@google.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Junio C Hamano [Wed, 26 Jul 2023 05:21:12 +0000 (22:21 -0700)]
SubmittingPatches: use of older maintenance tracks is an exception
While we could technically fix each and every bug on top of the
commit that introduced it, it is not necessarily practical. For
trivial and low-value bugfixes, it often is simpler and sufficient
to just fix it in the current maintenance track, leaving the bug
unfixed in the older maintenance tracks.
Demote the "use older maintenance track to fix old bugs" as a side
note, and explain that the choice is used only in exceptional cases.
Junio C Hamano [Wed, 26 Jul 2023 05:17:31 +0000 (22:17 -0700)]
SubmittingPatches: explain why 'next' and above are inappropriate base
The 'next' branch is primarily meant to be a testing ground to make
sure that topics that are reasonably well done work well together.
Building a new work on it would mean everything that was already in
'next' must have graduated to 'master' before the new work can also
be merged to 'master', and that is why we do not encourage basing
new work on 'next'.
Helped-by: Linus Arver <linusa@google.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
M Hickford [Wed, 26 Jul 2023 19:51:22 +0000 (19:51 +0000)]
credential/wincred: erase matching creds only
The credential erase request typically includes protocol, host, username
and password.
credential-wincred erases stored credentials that match protocol,
host and username, regardless of password.
This is confusing in the case the stored password differs from that
in the request. This case can occur when multiple credential helpers are
configured.
Only erase credential if stored password matches request (or request
omits password).
This fixes test "helper (wincred) does not erase a password distinct
from input" when t0303 is run with GIT_TEST_CREDENTIAL_HELPER set to
"wincred". This test was added in aeb21ce22e (credential: avoid
erasing distinct password, 2023-06-13).
Signed-off-by: M Hickford <mirth.hickford@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
M Hickford [Wed, 26 Jul 2023 19:46:54 +0000 (19:46 +0000)]
credential/libsecret: erase matching creds only
The credential erase request typically includes protocol, host, username
and password.
credential-libsecret erases a stored credential if it matches protocol,
host and username, regardless of password.
This is confusing in the case the stored password differs from that
in the request. This case can occur when multiple credential helpers are
configured.
Only erase credential if stored password matches request (or request
omits password).
This fixes test "helper (libsecret) does not erase a password distinct
from input" when t0303 is run with GIT_TEST_CREDENTIAL_HELPER set to
"libsecret". This test was added in aeb21ce22e (credential: avoid
erasing distinct password, 2023-06-13).
Signed-off-by: M Hickford <mirth.hickford@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Junio C Hamano [Wed, 26 Jul 2023 05:16:49 +0000 (22:16 -0700)]
SubmittingPatches: choice of base for fixing an older maintenance track
When working on an high-value bugfix that must be given to ancient
maintenance tracks, a starting point that is older than `maint` may
have to be chosen.
Helped-by: Linus Arver <linusa@google.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Petar Vutov [Tue, 25 Jul 2023 21:22:18 +0000 (23:22 +0200)]
doc: highlight that .gitmodules does not support !command
Bugfix for fc01a5d2 (submodule update documentation: don't repeat
ourselves, 2016-12-27).
The `custom command` and `none` options are described as sharing the
same limitations, but one is allowed in .gitmodules and the other is
not.
Rewrite the description for custom commands to be more precise,
and make it easier for readers to notice that custom commands cannot
be used in the .gitmodules file.
Signed-off-by: Petar Vutov <pvutov@imap.cc> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Junio C Hamano [Tue, 25 Jul 2023 19:05:24 +0000 (12:05 -0700)]
Merge branch 'jk/nested-points-at'
"git tag --list --points-at X" showed tags that directly refers to
object X, but did not list a tag that points at such a tag, which
has been corrected.
* jk/nested-points-at:
ref-filter: simplify return type of match_points_at
ref-filter: avoid parsing non-tags in match_points_at()
ref-filter: avoid parsing tagged objects in match_points_at()
ref-filter: handle nested tags in --points-at option
Junio C Hamano [Tue, 25 Jul 2023 19:05:24 +0000 (12:05 -0700)]
Merge branch 'jk/unused-parameter'
Mark-up unused parameters in the code so that we can eventually
enable -Wunused-parameter by default.
* jk/unused-parameter:
t/helper: mark unused callback void data parameters
tag: mark unused parameters in each_tag_name_fn callbacks
rev-parse: mark unused parameter in for_each_abbrev callback
replace: mark unused parameter in each_mergetag_fn callback
replace: mark unused parameter in ref callback
merge-tree: mark unused parameter in traverse callback
fsck: mark unused parameters in various fsck callbacks
revisions: drop unused "opt" parameter in "tweak" callbacks
count-objects: mark unused parameter in alternates callback
am: mark unused keep_cr parameters
http-push: mark unused parameter in xml callback
http: mark unused parameters in curl callbacks
do_for_each_ref_helper(): mark unused repository parameter
test-ref-store: drop unimplemented reflog-expire command
Junio C Hamano [Tue, 25 Jul 2023 19:05:23 +0000 (12:05 -0700)]
Merge branch 'mh/mingw-case-sensitive-build'
Names of MinGW header files are spelled in mixed case in some
source files, but the build host can be using case sensitive
filesystem with header files with their name spelled in all
lowercase.
* mh/mingw-case-sensitive-build:
mingw: use lowercase includes for some Windows headers
Various offset computation in the code that accesses the packfiles
and other data in the object layer has been hardened against
arithmetic overflow, especially on 32-bit systems.
* tb/object-access-overflow-protection:
commit-graph.c: prevent overflow in `verify_commit_graph()`
commit-graph.c: prevent overflow in `write_commit_graph()`
commit-graph.c: prevent overflow in `merge_commit_graph()`
commit-graph.c: prevent overflow in `split_graph_merge_strategy()`
commit-graph.c: prevent overflow in `load_tree_for_commit()`
commit-graph.c: prevent overflow in `fill_commit_in_graph()`
commit-graph.c: prevent overflow in `fill_commit_graph_info()`
commit-graph.c: prevent overflow in `load_oid_from_graph()`
commit-graph.c: prevent overflow in add_graph_to_chain()
commit-graph.c: prevent overflow in `write_commit_graph_file()`
pack-bitmap.c: ensure that eindex lookups don't overflow
midx.c: prevent overflow in `fill_included_packs_batch()`
midx.c: prevent overflow in `write_midx_internal()`
midx.c: store `nr`, `alloc` variables as `size_t`'s
midx.c: prevent overflow in `nth_midxed_offset()`
midx.c: prevent overflow in `nth_midxed_object_oid()`
midx.c: use `size_t`'s for fanout nr and alloc
packfile.c: use checked arithmetic in `nth_packed_object_offset()`
packfile.c: prevent overflow in `load_idx()`
packfile.c: prevent overflow in `nth_packed_object_id()`
Junio C Hamano [Tue, 25 Jul 2023 19:05:23 +0000 (12:05 -0700)]
Merge branch 'ah/advise-force-pushing'
Help newbies by suggesting that there are cases where force-pushing
is a valid and sensible thing to update a branch at a remote
repository, rather than reconciling with merge/rebase.
* ah/advise-force-pushing:
push: don't imply that integration is always required before pushing
remote: don't imply that integration is always required before pushing
wt-status: don't show divergence advice when committing
Junio C Hamano [Mon, 24 Jul 2023 23:11:03 +0000 (16:11 -0700)]
hex: retire get_sha1_hex()
The naming convention around get_sha1_hex() and its friends is
awkward these days, after "struct object_id" was introduced.
There are three public functions around this area:
* get_sha1_hex() - use the implied the_hash_algo, fill uchar *
* get_oid_hex() - use the implied the_hash_algo, fill oid *
* get_oid_hex_algop() - use the passed algop, fill oid *
Between the latter two, the "_algop" suffix signals whether the
the_hash_algo is used as the implied algorithm or the caller should
pass an algorithm explicitly. That is very much understandable and
is a good convention.
Between the former two, however, the "SHA1" vs "OID" in the names
differentiate in what type of variable the result is stored.
We could argue that it makes sense to use "SHA1" to mean "flat byte
buffer" to honor the historical practice in the days before "struct
object_id" was invented, but the natural fourth friend of the above
group would take an algop and fill a flat byte buffer, and it would
be strange to name it get_sha1_hex_algop(). Do we use the passed in
algo, or are we limited to SHA-1 ;-)?
In fact, such a function exists, albeit as a private helper function
used by the implementation of these functions, and is named a lot
more sensibly: get_hash_hex_algop().
Correct the misnomer of get_sha1_hex() and use "hash", instead of
"sha1", as "flat byte buffer that stores binary (as opposed to
hexadecimal) representation of the hash".
The four (2x2) friends now become:
* get_hash_hex() - use the implied the_hash_algo, fill uchar *
* get_oid_hex() - use the implied the_hash_algo, fill oid *
* get_hash_hex_algop() - use the passed algop, fill uchar *
* get_oid_hex_algop() - use the passed algop, fill oid *
As there are only two remaining calls to get_sha1_hex() in the
codebase right now, the blast radious of this change is fairly
small.
Taylor Blau [Mon, 24 Jul 2023 16:39:34 +0000 (12:39 -0400)]
t/lib-commit-graph.sh: avoid sub-shell in `graph_git_behavior()`
In a previous commit, we introduced a sub-shell in the implementation of
`graph_git_behavior()`, in order to allow us to pass `-C "$DIR"`
directly to the git processes spawned by `graph_git_two_modes()`.
Now that its callers are always operating from the "$TRASH_DIRECTORY"
instead of one of its sub-directories, we can drop the inner sub-shell,
as it is no longer required.
Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Taylor Blau [Mon, 24 Jul 2023 16:39:28 +0000 (12:39 -0400)]
t5318: avoid top-level directory changes
Avoid changing the current working directory from outside of a sub-shell
during the tests in t5318.
Each test has mostly straightforward changes, either:
- Removing the top-level `cd "$TRASH_DIRECTORY/full"`, which is
unnecessary after ensuring that other tests don't change their
working directory outside of a sub-shell.
- Changing any Git invocations which want to be in a sub-directory by
either (a) adding a "-C $DIR" argument, or (b) moving the whole test
into a sub-shell.
While we're here, remove any explicit "git config core.commitGraph true"
invocations which were designed to enable use of the commit-graph. These
are unnecessary following 31b1de6a09b (commit-graph: turn on
commit-graph by default, 2019-08-13).
Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Taylor Blau [Mon, 24 Jul 2023 16:39:25 +0000 (12:39 -0400)]
t/lib-commit-graph.sh: avoid directory change in `graph_git_behavior()`
The `graph_git_behavior()` helper asserts that a number of common Git
operations (such as `git log --oneline`, `git log --topo-order`, etc.)
produce identical output regardless of whether or not a commit-graph is
in use.
This helper takes as its second argument the location (relative to the
`$TRASH_DIRECTORY`) of the Git repostiory under test. In order to run
each of its commands within that repository, it first changes into that
directory, without the use of a sub-shell.
This pollutes future tests which expect to be run in the top-level
`$TRASH_DIRECTORY` as usual. We could wrap `graph_git_behavior()` in a
sub-shell, like:
, but since we're invoking git directly, we can pass along a "-C $DIR"
when "$DIR" is non-empty.
Note, however, that until the remaining callers are cleaned up to avoid
changing working directories outside of a sub-shell, that we need to
ensure that we are operating in the top-level $TRASH_DIRECTORY. The
inner-subshell will go away in a future commit once it is no longer
necessary.
Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Taylor Blau [Mon, 24 Jul 2023 16:39:22 +0000 (12:39 -0400)]
t/lib-commit-graph.sh: allow `graph_read_expect()` in sub-directories
The `graph_read_expect()` function is used to ensure that the output of
the "read-graph" test helper matches certain parameters (e.g., how many
commits are in the graph, which chunks were written, etc.).
It expects the Git repository being tested to be at the current working
directory. However, a handful of t5318 tests use different repositories
stored in sub-directories. To work around this, several tests in t5318
change into the relevant repository outside of a sub-shell, altering the
context for the rest of the suite.
Prepare to remove these globally-scoped directory changes by teaching
`graph_read_expect()` to take an optional "-C dir" to specify where the
repository containing the commit-graph being tested is.
Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Duplicate the logic of %(describe) and friends from pretty to
ref-filter. In the future, this change helps in unifying both the
formats as ref-filter will be able to do everything that pretty is doing
and we can have a single interface.
The new atom "describe" and its friends are equivalent to the existing
pretty formats with the same name.
Helped-by: Junio C Hamano <gitster@pobox.com> Mentored-by: Christian Couder <christian.couder@gmail.com> Mentored-by: Hariom Verma <hariom18599@gmail.com> Signed-off-by: Kousik Sanagavarapu <five231003@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
were added in pretty 4f732e0fd7 (pretty: allow %(trailers) options
with explicit value, 2019-01-29) to parse multiple options in an
argument to --pretty. For example,
will output all the trailers matching the key and seperates them by
a comma followed by a space per commit.
Add similar functions,
match_atom_arg_value()
match_atom_bool_arg()
in ref-filter.
There is no atom yet that can use these functions in ref-filter, but we
are going to add a new %(describe) atom in a subsequent commit where we
parse options like tags=<bool-value> or match=<pattern> given to it.
Helped-by: Junio C Hamano <gitster@pobox.com> Mentored-by: Christian Couder <christian.couder@gmail.com> Mentored-by: Hariom Verma <hariom18599@gmail.com> Signed-off-by: Kousik Sanagavarapu <five231003@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Alex Henrie [Sat, 22 Jul 2023 21:28:25 +0000 (15:28 -0600)]
sequencer: finish parsing the todo list despite an invalid first line
Before the todo list is edited it is rewritten to shorten the OIDs of
the commits being picked and to append advice about editing the list.
The exact advice depends on whether the todo list is being edited for
the first time or not. After the todo list has been edited it is
rewritten to lengthen the OIDs of the commits being picked and to remove
the advice. If the edited list cannot be parsed then this last step is
skipped.
Prior to db81e50724 (rebase-interactive: use todo_list_write_to_file()
in edit_todo_list(), 2019-03-05) if the existing todo list could not be
parsed then the initial rewrite was skipped as well. This had the
unfortunate consequence that if the list could not be parsed after the
initial edit the advice given to the user was wrong when they re-edited
the list. This change relied on todo_list_parse_insn_buffer() returning
the whole todo list even when it cannot be parsed. Unfortunately if the
list starts with a "fixup" command then it will be truncated and the
remaining lines are lost. Fix this by continuing to parse after an
initial "fixup" commit as we do when we see any other invalid line.
Signed-off-by: Alex Henrie <alexhenrie24@gmail.com>
[jc: removed an apparently unneeded subshell around the test body] Signed-off-by: Junio C Hamano <gitster@pobox.com>
Junio C Hamano [Fri, 21 Jul 2023 21:53:12 +0000 (14:53 -0700)]
branch: update the message to refuse touching a branch in-use
The "git branch -f" command can refuse to force-update a branch that
is used by another worktree. The original rationale for this
behaviour was that updating a branch that is checked out in another
worktree, without making a matching change to the index and the
working tree files in that worktree, will lead to a very confused
user. "git diff HEAD" will no longer give a useful patch, because
HEAD is a commit unrelated to what the index and the working tree in
the worktree were based on, for example.
These days, the same mechanism also protects branches that are being
rebased or bisected, and the same machanism is expected to be the
right place to add more checks, when we decide to protect branches
undergoing other kinds of operations. We however forgot to rethink
the messaging, which originally said that we are refusing to touch
the branch because it is "checked out" elsewhere, when d2ba271a
(branch: check for bisects and rebases, 2022-06-14) started to
protect branches that are being rebased or bisected.
The spirit of the check has always been that we do not want to
disrupt the use of the same branch in other worktrees. Let's reword
the message slightly to say that the branch is "used by" another
worktree, instead of "checked out".
We could teach the branch.c:prepare_checked_out_branches() function
to remember why it decided that a particular branch needs protecting
(i.e. was it because it was checked out? being bisected? something
else?) in addition to which worktree the branch was in use, and use
that in the error message to say "you cannot force update this
branch because it is being bisected in the worktree X", etc., but it
is dubious that such extra complexity is worth it. The message
already tells which directory the worktree in question is, and it
should be just a "chdir" away for the user to find out what state it
is in, if the user felt curious enough. So let's not go there yet.
Helped-by: Josh Sref <jsoref@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Junio C Hamano [Fri, 21 Jul 2023 20:47:26 +0000 (13:47 -0700)]
Merge branch 'tb/refs-exclusion-and-packed-refs'
Enumerating refs in the packed-refs file, while excluding refs that
match certain patterns, has been optimized.
* tb/refs-exclusion-and-packed-refs:
ls-refs.c: avoid enumerating hidden refs where possible
upload-pack.c: avoid enumerating hidden refs where possible
builtin/receive-pack.c: avoid enumerating hidden references
refs.h: implement `hidden_refs_to_excludes()`
refs.h: let `for_each_namespaced_ref()` take excluded patterns
revision.h: store hidden refs in a `strvec`
refs/packed-backend.c: add trace2 counters for jump list
refs/packed-backend.c: implement jump lists to avoid excluded pattern(s)
refs/packed-backend.c: refactor `find_reference_location()`
refs: plumb `exclude_patterns` argument throughout
builtin/for-each-ref.c: add `--exclude` option
ref-filter.c: parameterize match functions over patterns
ref-filter: add `ref_filter_clear()`
ref-filter: clear reachable list pointers after freeing
ref-filter.h: provide `REF_FILTER_INIT`
refs.c: rename `ref_filter`
René Scharfe [Fri, 21 Jul 2023 12:41:53 +0000 (14:41 +0200)]
pack-objects: fix --no-quiet
Since 99fb6e04cb (pack-objects: convert to use parse_options(),
2012-02-01) git pack-objects has accepted the option --no-quiet, but it
does the same as --quiet. That's because it's defined using OPT_SET_INT
with a value of 0, which sets 0 when negated, too.
Make --no-quiet equivalent to --progress and ignore it if --all-progress
was given.
Signed-off-by: René Scharfe <l.s.r@web.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
René Scharfe [Fri, 21 Jul 2023 12:41:56 +0000 (14:41 +0200)]
pack-objects: fix --no-keep-true-parents
Since 99fb6e04cb (pack-objects: convert to use parse_options(),
2012-02-01) git pack-objects has accepted --no-keep-true-parents, but
this option does the same as --keep-true-parents. That's because it's
defined using OPT_SET_INT with a value of 0, which sets 0 when negated
as well.
Turn --no-keep-true-parents into the opposite of --keep-true-parents by
using OPT_BOOL and storing the option's status directly in a variable
named "grafts_keep_true_parents" instead of in negative form in
"grafts_replace_parents".
Signed-off-by: René Scharfe <l.s.r@web.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
René Scharfe [Fri, 21 Jul 2023 13:41:33 +0000 (15:41 +0200)]
describe: fix --no-exact-match
Since 2c33f75754 (Teach git-describe --exact-match to avoid expensive
tag searches, 2008-02-24) git describe accepts --no-exact-match, but it
does the same as --exact-match, an alias for --candidates=0. That's
because it's defined using OPT_SET_INT with a value of 0, which sets 0
when negated as well.
Let --no-exact-match set the number of candidates to the default value
instead. Users that need a more specific lack of exactitude can specify
their preferred value using --candidates, as before.
The "--no-exact-match" option was not covered in the tests, so let's
add a few. Also add a case where --exact-match option is used on a
commit that cannot be described without distance from tags and make
sure the command fails.
Signed-off-by: René Scharfe <l.s.r@web.de>
[jc: added trivial tests] Signed-off-by: Junio C Hamano <gitster@pobox.com>
Han Young [Fri, 21 Jul 2023 03:57:58 +0000 (11:57 +0800)]
blame: allow --contents to work with bare repo
The --contents option can be used with git blame to blame the file
as if it had the contents from the specified file. Since 1a3119ed
(blame: allow --contents to work with non-HEAD commit, 2023-03-24),
the --contents option can work with non-HEAD commit. However, if you
try to use --contents in a bare repository, you get the following
error:
fatal: this operation must be run in a work tree
This is because before trying to generate a fake working tree
commit, we always call setup_work_tree(). But in a bare repo,
working tree is not available. The call to setup_work_tree is used
to prepare the reading of the blamed file in the working tree, which
isn't necessary if we are reading the contents from the specific
file instead of the file in the working tree.
Add a check in setup_scoreboard to skip setup_work_tree if we are
reading from the file specified in --contents.
This enables us to use --contents in a bare repo. This is a nice
addition on top of 1a3119ed, having a working tree to use --contents
is optional.
Add test for the --contents option with bare repo to the
annotate-tests.sh test script.
Signed-off-by: Han Young <hanyang.tony@bytedance.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
wrapper: use trace2 counters to collect fsync stats
As mentioned in the thread starting at [1], trace2 counters should be
used to count events instead of ad-hoc static variables.
Convert the two fsync static variables to trace2 counters, reducing the
coupling between wrapper.c and the trace2 subsystem. Adjust t/t5351 to
match the trace2 counter output format.
The counters are not per-thread because the ones being replaced also
were not.
"git reset --no-mixed" behaved exactly like "git reset --mixed",
which was nonsense.
If there were only two kinds, e.g. "mixed" vs "separate", it might
have made sense to make "git reset --no-mixed" behave identically to
"git reset --separate" and vice-versa, but because we have many
types of reset, let's just forbid "--no-mixed" and negated form of
other types.
Junio C Hamano [Wed, 19 Jul 2023 16:32:36 +0000 (09:32 -0700)]
show-branch: reject --[no-](topo|date)-order
"git show-branch --no-topo-order" behaved exactly the same way as
"git show-branch --topo-order" did, which was nonsense. This was
because we choose between topo- and date- by setting a variable to
either REV_SORT_IN_GRAPH_ORDER or REV_SORT_BY_COMMIT_DATE with
OPT_SET_INT() and REV_SORT_IN_GRAPH_ORDER happens to be 0. The
OPT_SET_INT() macro assigns 0 to the target variable in respose to
the negated form of its option.
"--no-date-order" by luck behaves identically to "--topo-order"
exactly for the same reason, and it sort-of makes sense right now,
but the "sort-of makes sense" will quickly break down once we add a
third way to sort. Not-A may be B when there are only two choices
between A and B, but once your choices become among A, B, and C,
not-A does not mean B.
Just mark these two ordering options to reject negation, and add a
test, which was missing. "git show-branch --no-reflog" is also
unnegatable, so throw in a test for that while we are at it.
When the trace2 counter mechanism was added in 81071626ba (trace2: add
global counter mechanism, 2022-10-24), the name of the file where new
counters are added was misspelled in a comment.
Use the correct file name.
Signed-off-by: Beat Bolli <dev+git@drbeat.li> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Junio C Hamano [Tue, 18 Jul 2023 22:58:00 +0000 (15:58 -0700)]
short help: allow a gap smaller than USAGE_GAP
The parse-options API responds to "git cmd -h" by listing the option
flag (padded to the USAGE_OPTS_WIDTH column), followed by USAGE_GAP
(set to 2) whitespaces, followed by the help text. If the flags
part does not fit within the USAGE_OPTS_WIDTH, the help text is given
on its own line. Imagine that "@" below depicts the USAGE_OPTS_WIDTH'th
column, and "#" are for the usage help text, the output may look
like this:
@@@@@@@@@@@@@ ########################################
-f description of the flag '-f' comes here
--short=<num> description of the flag '--short'
--very-long-option=<number>
description of the flag '--very-long-option'
This is all good and nice in principle, but it becomes awkward when
the flags part is just one column over the limit and forces a line
break. See the description of the "--almost" option below:
@@@@@@@@@@@@@ ########################################
-f description of the flag '-f' comes here
--short=<num> description of the flag '--short'
--almost=<num>
description of the flag '--almost'
--very-long-option=<number>
description of the flag '--very-long-option'
If we allow shrinking the gap to a single whitespace only in such a
case, we would instead get:
@@@@@@@@@@@@@ ########################################
-f description of the flag '-f' comes here
--short=<num> description of the flag '--short'
--almost=<num> description of the flag '--almost'
--very-long-option=<number>
description of the flag '--very-long-option'
and the boundary between the flags and their descriptions does not
become any harder to see, while saving precious vertical screen real
estate.
Junio C Hamano [Tue, 18 Jul 2023 22:31:03 +0000 (15:31 -0700)]
remote: simplify "remote add --tags" help text
The help text for the --tags option was split into two option[]
entries, which was a hacky way to give two lines of help text (the
second entry did not have either short or long help, and there was
no way to invoke its entry---it was there only for the help text).
As we now support multi-line text in the option help, let's make
the second line of the help a proper second line and remove the
hacky second entry.
Junio C Hamano [Tue, 18 Jul 2023 22:54:04 +0000 (15:54 -0700)]
short help: allow multi-line opthelp
When "-h" triggers the short-help in a command that implements its
option parsing using the parse-options API, the option help text is
shown with a single fprintf() as a long line. When the text is
multi-line, the second and subsequent lines are not left padded,
that breaks the alignment across options.
Borrowing the idea from the advice API where its hint strings are
shown with (localized) "hint:" prefix, let's internally split the
(localized) help text into lines, and showing the first line, pad
the remaining lines to align.
Andreas Herrmann [Wed, 19 Jul 2023 14:29:56 +0000 (16:29 +0200)]
configure.ac: don't overwrite NO_CURL option
Even if 'configure --with-curl=no' was run, curl support is used,
because library detection overwrites it. Avoid this overwrite.
Configure should obey what the user has specified.
Signed-off-by: Andreas Herrmann <aherrmann@suse.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Andreas Herrmann [Wed, 19 Jul 2023 14:29:54 +0000 (16:29 +0200)]
configure.ac: don't overwrite NO_EXPAT option
Even if 'configure --with-expat=no' was run, expat support is used,
because library detection overwrites it. Avoid this overwrite.
Configure should obey what the user has specified.
Signed-off-by: Andreas Herrmann <aherrmann@suse.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Junio C Hamano [Wed, 19 Jul 2023 11:37:39 +0000 (04:37 -0700)]
show-branch: --no-sparse should give dense output
"git show-branch --no-sparse" behaved exactly the same way as "git
show-branch --sparse", which did not make any sense. This was
because it used a variable "dense" initialized to 1 by default to
give "non sparse" behaviour, and OPT_SET_INT() to set the varilable
to 0 in response to the "--sparse" option. Unfortunately,
OPT_SET_INT() sets 0 to the given variable when the option is
negated.
Flip the polarity of the variable "dense" by renaming it to "sparse"
and initializing it to 0, and have OPT_SET_INT() set the variable to
1 when "--sparse" is given. This way, "--no-sparse" would set 0 to
the variable and would give us the "dense" behaviour.
Junio C Hamano [Tue, 18 Jul 2023 21:45:59 +0000 (14:45 -0700)]
fetch: reject --no-ipv[46]
Now we have introduced OPT_IPVERSION(), tweak its implementation so
that "git clone", "git fetch", and "git push" reject the negated
form of "Use only IP version N" options.
Junio C Hamano [Tue, 18 Jul 2023 21:34:33 +0000 (14:34 -0700)]
parse-options: introduce OPT_IPVERSION()
The command line option parsing for "git clone", "git fetch", and
"git push" have duplicated implementations of parsing "--ipv4" and
"--ipv6" options, by having two OPT_SET_INT() for "ipv4" and "ipv6".
Introduce a new OPT_IPVERSION() macro and use it in these three
commands.
Junio C Hamano [Tue, 18 Jul 2023 18:27:49 +0000 (11:27 -0700)]
branch: reject "--no-all" and "--no-remotes" early
As the command line parser for "git branch --all" forgets to use
PARSE_OPT_NONEG, it accepted "git branch --no-all", and then passed
a nonsense value to the underlying machinery, leading to a fatal
error "filter_refs: invalid type". The "--remotes" option had
exactly the same issue.
Catch the unsupported options early in the option parser.
Junio C Hamano [Tue, 18 Jul 2023 18:22:40 +0000 (11:22 -0700)]
am: simplify parsing of "--[no-]keep-cr"
Command line options "--keep-cr" and its negation trigger
OPT_SET_INT_F(PARSE_OPT_NONEG) to set a variable to 1 and 0
respectively. Using OPT_SET_INT() to implement the positive variant
that sets the variable to 1 without specifying PARSE_OPT_NONEG gives
us the negative variant to set it to 0 for free.
Junio C Hamano [Tue, 18 Jul 2023 17:47:39 +0000 (10:47 -0700)]
gitignore.txt: mark up explanation of patterns consistently
In the "PATTERN FORMAT" section, all the other pattern elements are
shown as `monospace` literals inside "double quoted" strings. Do
the same for the explanation of a slash to make it consistent.
René Scharfe [Tue, 18 Jul 2023 15:44:13 +0000 (17:44 +0200)]
ls-tree: fix --no-full-name
Since 61fdbcf98b (ls-tree: migrate to parse-options, 2009-11-13) git
ls-tree has accepted the option --no-full-name, but it does the same
as --full-name, contrary to convention. That's because it's defined
using OPT_SET_INT with a value of 0, where the negative variant sets
0 as well.
Turn --no-full-name into the opposite of --full-name by using OPT_BOOL
instead and storing the option's status directly in a variable named
"full_name" instead of in negated form in "chomp_prefix".
Signed-off-by: René Scharfe <l.s.r@web.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Junio C Hamano [Tue, 18 Jul 2023 14:28:53 +0000 (07:28 -0700)]
Merge branch 'tb/repack-cleanup'
The recent change to "git repack" made it react less nicely when a
leftover .idx file that no longer has the corresponding .pack file
in the repository, which has been corrected.
* tb/repack-cleanup:
builtin/repack.c: avoid dir traversal in `collect_pack_filenames()`
builtin/repack.c: only repack `.pack`s that exist
Jeff King [Sun, 2 Jul 2023 22:38:29 +0000 (18:38 -0400)]
ref-filter: simplify return type of match_points_at
We return the oid that matched, but the sole caller only cares whether
we matched anything at all. This is mostly academic, since there's only
one caller, but the lifetime of the returned pointer is not immediately
clear. Sometimes it points to an oid in a tag struct, which should live
forever. And sometimes to the oid passed in, which only lives as long as
the each_ref_fn callback we're called from.
Simplify this to a boolean return which is more direct and obvious. As a
bonus, this lets us avoid the weird pattern of overwriting our "oid"
parameter in the loop (since we now only refer to the tagged oid one
time, and can just inline the call to get it).
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Jeff King [Sun, 2 Jul 2023 22:37:47 +0000 (18:37 -0400)]
ref-filter: avoid parsing non-tags in match_points_at()
When handling --points-at, we have to try to peel each ref to see if
it's a tag that points at a requested oid. We start this process by
calling parse_object() on the oid pointed to by each ref.
The cost of parsing each object adds up, especially in an output that
doesn't otherwise need to open the objects at all. Ideally we'd use
peel_iterated_oid() here, which uses the cached information in the
packed-refs file. But we can't, because our --points-at must match not
only the fully peeled value, but any interim values (so if tag A points
to tag B which points to commit C, we should match --points-at=B, but
peel_iterated_oid() will only tell us about C).
So the best we can do (absent changes to the packed-refs peel traits) is
to avoid parsing non-tags. The obvious way to do that is to call
oid_object_info() to check the type before parsing. But there are a few
gotchas there, like checking if the object has already been parsed.
Instead we can just tell parse_object() that we are OK skipping the hash
check, which lets it turn on several optimizations. Commits can be
loaded via the commit graph (so it's both fast and we have the benefit
of the parsed data if we need it later at the output stage). Blobs are
not loaded at all. Trees are still loaded, but it's rather rare to have
a ref point directly to a tree (and since this is just an optimization,
kicking in 99% of the time is OK).
Even though we're paying for an extra lookup, the cost to avoid parsing
the non-tags is a net benefit. In my git.git repository with 941 tags
and 1440 other refs pointing to commits, this significantly cuts the
runtime:
Benchmark 1: ./git.old for-each-ref --points-at=HEAD
Time (mean ± σ): 26.8 ms ± 0.5 ms [User: 24.5 ms, System: 2.2 ms]
Range (min … max): 25.9 ms … 29.2 ms 107 runs
Benchmark 2: ./git.new for-each-ref --points-at=HEAD
Time (mean ± σ): 9.1 ms ± 0.3 ms [User: 6.8 ms, System: 2.2 ms]
Range (min … max): 8.6 ms … 10.2 ms 308 runs
Summary
'./git.new for-each-ref --points-at=HEAD' ran
2.96 ± 0.10 times faster than './git.old for-each-ref --points-at=HEAD'
In a repository that is mostly annotated tags, we'd expect less
improvement (we might still skip a few object loads, but that's balanced
by the extra lookups). In my clone of linux.git, which has 782 tags and
3 branches, the run-time is about the same (it's actually ~1% faster on
average after this patch, but that's within the run-to-run noise).
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Jeff King [Sun, 2 Jul 2023 22:35:40 +0000 (18:35 -0400)]
ref-filter: avoid parsing tagged objects in match_points_at()
When we peel tags to check if they match a --points-at oid, we
recursively parse the tagged object to see if it is also a tag. But
since the tag itself tells us the type of the object it points to (and
even gives us the appropriate object struct via its "tagged" member), we
can use that directly.
We do still have to make sure to call parse_tag() before looking at each
tag. This is redundant for the outermost tag (since we did call
parse_object() to find its type), but that's OK; parse_tag() is smart
enough to make this a noop when the tag has already been parsed.
In my clone of linux.git, with 782 tags (and only 3 non-tags), this
yields a significant speedup (bringing us back where we were before the
commit before this one started recursively dereferencing tags):
Benchmark 1: ./git.old for-each-ref --points-at=HEAD --format="%(refname)"
Time (mean ± σ): 20.3 ms ± 0.5 ms [User: 11.1 ms, System: 9.1 ms]
Range (min … max): 19.6 ms … 21.5 ms 141 runs
Benchmark 2: ./git.new for-each-ref --points-at=HEAD --format="%(refname)"
Time (mean ± σ): 11.4 ms ± 0.2 ms [User: 6.3 ms, System: 5.0 ms]
Range (min … max): 11.0 ms … 12.2 ms 250 runs
Summary
'./git.new for-each-ref --points-at=HEAD --format="%(refname)"' ran
1.79 ± 0.05 times faster than './git.old for-each-ref --points-at=HEAD --format="%(refname)"'
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Jan Klötzke [Sat, 1 Jul 2023 20:57:02 +0000 (22:57 +0200)]
ref-filter: handle nested tags in --points-at option
Tags are dereferenced until reaching a different object type to handle
nested tags, e.g. on checkout. In contrast, "git tag --points-at=..."
fails to list such nested tags because only one level of indirection is
obtained in filter_refs(). Implement the recursive dereferencing for the
"--points-at" option when filtering refs to unify the behaviour.
Signed-off-by: Jan Klötzke <jan@kloetzke.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Junio C Hamano [Mon, 17 Jul 2023 18:30:42 +0000 (11:30 -0700)]
Merge branch 'cw/compat-util-header-cleanup'
Further shuffling of declarations across header files to streamline
file dependencies.
* cw/compat-util-header-cleanup:
git-compat-util: move alloc macros to git-compat-util.h
treewide: remove unnecessary includes for wrapper.h
kwset: move translation table from ctype
sane-ctype.h: create header for sane-ctype macros
git-compat-util: move wrapper.c funcs to its header
git-compat-util: move strbuf.c funcs to its header
Junio C Hamano [Mon, 17 Jul 2023 18:30:41 +0000 (11:30 -0700)]
Merge branch 'pw/diff-no-index-from-named-pipes'
"git diff --no-index" learned to read from named pipes as if they
were regular files, to allow "git diff <(process) <(substitution)"
some shells support.
* pw/diff-no-index-from-named-pipes:
diff --no-index: support reading from named pipes
t4054: test diff --no-index with stdin
diff --no-index: die on error reading stdin
diff --no-index: refuse to compare stdin to a directory
René Scharfe [Sun, 16 Jul 2023 08:52:33 +0000 (10:52 +0200)]
strbuf: use skip_prefix() in strbuf_addftime()
Use the now common skip_prefix() cascade instead of a case statement to
parse the strftime(3) format in strbuf_addftime(). skip_prefix() parses
the "fmt" pointer and advances it appropriately, making additional
pointer arithmetic unnecessary. The resulting code is more compact and
consistent with most other strbuf_expand_step() loops.
Signed-off-by: René Scharfe <l.s.r@web.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
René Scharfe [Sun, 16 Jul 2023 08:17:35 +0000 (10:17 +0200)]
t6300: fix setup with GPGSSH but without GPG
In a test introduced by 26c9c03f0a (ref-filter: add new "signature"
atom, 2023-06-04) the file named "file" is added by a setup step that
requires GPG and modified by a second setup step that requires GPGSSH.
Systems lacking the first prerequisite skip the initial setup step and
then "git commit -a" in the second one doesn't find the modified file.
Add it explicitly.
Signed-off-by: René Scharfe <l.s.r@web.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
projects / thirdparty / git.git / log
