]> git.ipfire.org Git - thirdparty/git.git/log
thirdparty/git.git
3 years agoMerge branch 'rc/fetch-refetch'
Junio C Hamano [Mon, 4 Apr 2022 17:56:23 +0000 (10:56 -0700)] 
Merge branch 'rc/fetch-refetch'

"git fetch --refetch" learned to fetch everything without telling
the other side what we already have, which is useful when you
cannot trust what you have in the local object store.

* rc/fetch-refetch:
  docs: mention --refetch fetch option
  fetch: after refetch, encourage auto gc repacking
  t5615-partial-clone: add test for fetch --refetch
  fetch: add --refetch option
  builtin/fetch-pack: add --refetch option
  fetch-pack: add refetch
  fetch-negotiator: add specific noop initializer

3 years agoMerge branch 'jc/mailsplit-warn-on-tty'
Junio C Hamano [Mon, 4 Apr 2022 17:56:23 +0000 (10:56 -0700)] 
Merge branch 'jc/mailsplit-warn-on-tty'

"git am" can read from the standard input when no mailbox is given
on the command line, but the end-user gets no indication when it
happens, making Git appear stuck.

* jc/mailsplit-warn-on-tty:
  am/apply: warn if we end up reading patches from terminal

3 years agoMerge branch 'ns/trace2-fsync-stat'
Junio C Hamano [Mon, 4 Apr 2022 17:56:23 +0000 (10:56 -0700)] 
Merge branch 'ns/trace2-fsync-stat'

Trace2 code has been taught to report stats for fsync operations.

* ns/trace2-fsync-stat:
  trace2: add stats for fsync operations

3 years agoMerge branch 'gc/branch-recurse-submodules-fix'
Junio C Hamano [Mon, 4 Apr 2022 17:56:22 +0000 (10:56 -0700)] 
Merge branch 'gc/branch-recurse-submodules-fix'

A handful of obvious clean-ups around a topic that is already in
'master'.

* gc/branch-recurse-submodules-fix:
  branch.c: simplify advice-and-die sequence
  branch: rework comments for future developers
  branch: remove negative exit code
  branch --set-upstream-to: be consistent when advising
  branch: give submodule updating advice before exit
  branch: support more tracking modes when recursing

3 years agoMerge branch 'ns/fsync-or-die-message-fix'
Junio C Hamano [Mon, 4 Apr 2022 17:56:22 +0000 (10:56 -0700)] 
Merge branch 'ns/fsync-or-die-message-fix'

When creating a loose object file, we didn't report the exact
filename of the file we failed to fsync, even though the
information was readily available, which has been corrected.

* ns/fsync-or-die-message-fix:
  object-file: pass filename to fsync_or_die

3 years agoMerge branch 'ns/core-fsyncmethod'
Junio C Hamano [Mon, 4 Apr 2022 17:56:22 +0000 (10:56 -0700)] 
Merge branch 'ns/core-fsyncmethod'

A couple of fix-up to a topic that is now in 'master'.

* ns/core-fsyncmethod:
  core.fsyncmethod: correctly camel-case warning message
  core.fsync: fix incorrect expression for default configuration

3 years agoMerge branch 'dp/worktree-repair-in-usage'
Junio C Hamano [Mon, 4 Apr 2022 17:56:22 +0000 (10:56 -0700)] 
Merge branch 'dp/worktree-repair-in-usage'

Usage string fix.

* dp/worktree-repair-in-usage:
  worktree: include repair cmd in usage

3 years agoMerge branch 'ab/reftable-aix-xlc-12'
Junio C Hamano [Mon, 4 Apr 2022 17:56:22 +0000 (10:56 -0700)] 
Merge branch 'ab/reftable-aix-xlc-12'

Work around AIX C compiler that does not seem to grok
initialization of a union member of a struct.

* ab/reftable-aix-xlc-12:
  reftable: make assignments portable to AIX xlc v12.01

3 years agoMerge branch 'gc/submodule-update-part2'
Junio C Hamano [Mon, 4 Apr 2022 17:56:21 +0000 (10:56 -0700)] 
Merge branch 'gc/submodule-update-part2'

Move more "git submodule update" to C.

* gc/submodule-update-part2:
  submodule--helper: remove forward declaration
  submodule: move core cmd_update() logic to C
  submodule--helper: reduce logic in run_update_procedure()
  submodule--helper: teach update_data more options
  builtin/submodule--helper.c: rename option struct to "opt"
  submodule update: use die_message()
  submodule--helper: run update using child process struct

3 years agoMerge branch 'ds/t7700-kept-pack-test'
Junio C Hamano [Mon, 4 Apr 2022 17:56:21 +0000 (10:56 -0700)] 
Merge branch 'ds/t7700-kept-pack-test'

Test clean-up.

* ds/t7700-kept-pack-test:
  test-lib-functions: remove test_subcommand_inexact
  t7700: check post-condition in kept-pack test

3 years agoMerge branch 'ds/partial-bundle-more'
Junio C Hamano [Mon, 4 Apr 2022 17:56:21 +0000 (10:56 -0700)] 
Merge branch 'ds/partial-bundle-more'

Code clean-up.

* ds/partial-bundle-more:
  pack-objects: lazily set up "struct rev_info", don't leak
  bundle: output hash information in 'verify'
  bundle: move capabilities to end of 'verify'
  pack-objects: parse --filter directly into revs.filter
  pack-objects: move revs out of get_object_list()
  list-objects-filter: remove CL_ARG__FILTER

3 years agoMerge branch 'tl/ls-tree-oid-only'
Junio C Hamano [Mon, 4 Apr 2022 17:56:21 +0000 (10:56 -0700)] 
Merge branch 'tl/ls-tree-oid-only'

"git ls-tree" learns "--oid-only" option, similar to "--name-only",
and more generalized "--format" option.

* tl/ls-tree-oid-only:
  ls-tree: split up "fast path" callbacks
  ls-tree: detect and error on --name-only --name-status
  ls-tree: support --object-only option for "git-ls-tree"
  ls-tree: introduce "--format" option
  cocci: allow padding with `strbuf_addf()`
  ls-tree: introduce struct "show_tree_data"
  ls-tree: slightly refactor `show_tree()`
  ls-tree: fix "--name-only" and "--long" combined use bug
  ls-tree: simplify nesting if/else logic in "show_tree()"
  ls-tree: rename "retval" to "recurse" in "show_tree()"
  ls-tree: use "size_t", not "int" for "struct strbuf"'s "len"
  ls-tree: use "enum object_type", not {blob,tree,commit}_type
  ls-tree: add missing braces to "else" arms
  ls-tree: remove commented-out code
  ls-tree tests: add tests for --name-status

3 years agoMerge branch 'ab/reflog-parse-options'
Junio C Hamano [Mon, 4 Apr 2022 17:56:21 +0000 (10:56 -0700)] 
Merge branch 'ab/reflog-parse-options'

"git reflog" command now uses parse-options API to parse its
command line options.

* ab/reflog-parse-options:
  reflog: fix 'show' subcommand's argv
  reflog [show]: display sensible -h output
  reflog: convert to parse_options() API
  reflog exists: use parse_options() API
  git reflog [expire|delete]: make -h output consistent with SYNOPSIS
  reflog: move "usage" variables and use macros
  reflog tests: add missing "git reflog exists" tests
  reflog: refactor cmd_reflog() to "if" branches
  reflog.c: indent argument lists

3 years agobranch.c: simplify advice-and-die sequence
Glen Choo [Thu, 31 Mar 2022 22:41:18 +0000 (15:41 -0700)] 
branch.c: simplify advice-and-die sequence

In the dwim_branch_start(), when we cannot find an appropriate
upstream, we will die with the same message anyway, whether we
issue an advice message.

Flip the code around a bit and simplify the flow using
advise_if_enabled() function.

Helped-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Glen Choo <chooglen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agobranch: rework comments for future developers
Glen Choo [Thu, 31 Mar 2022 22:41:17 +0000 (15:41 -0700)] 
branch: rework comments for future developers

For two cases in which we do not explicitly pass --track=<choice>
option down to the submodule--helper subprocess, we have comments
that say "we do not have to pass --track", but in fact we not just
do not have to, but it would be incorrect to pass any --track option
to the subprocess (instead, the correct behaviour is to let the
subprocess figure out what is the appropriate tracking mode to use).

Signed-off-by: Glen Choo <chooglen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agoMerge branch 'ab/usage-die-message' into gc/branch-recurse-submodules-fix
Junio C Hamano [Thu, 31 Mar 2022 22:32:48 +0000 (15:32 -0700)] 
Merge branch 'ab/usage-die-message' into gc/branch-recurse-submodules-fix

* ab/usage-die-message:
  config API: use get_error_routine(), not vreportf()
  usage.c + gc: add and use a die_message_errno()
  gc: return from cmd_gc(), don't call exit()
  usage.c API users: use die_message() for error() + exit 128
  usage.c API users: use die_message() for "fatal :" + exit 128
  usage.c: add a die_message() routine

3 years agoThe 17th batch
Junio C Hamano [Thu, 31 Mar 2022 00:43:19 +0000 (17:43 -0700)] 
The 17th batch

Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agoMerge branch 'ab/test-tap-fix-for-immediate'
Junio C Hamano [Thu, 31 Mar 2022 01:01:11 +0000 (18:01 -0700)] 
Merge branch 'ab/test-tap-fix-for-immediate'

Fix test framework a bit.

* ab/test-tap-fix-for-immediate:
  test-lib: have --immediate emit valid TAP on failure

3 years agoMerge branch 'ab/hook-tests-updates'
Junio C Hamano [Thu, 31 Mar 2022 01:01:11 +0000 (18:01 -0700)] 
Merge branch 'ab/hook-tests-updates'

Update tests around the use of hook scripts.

* ab/hook-tests-updates:
  http tests: use "test_hook" for "smart" and "dumb" http tests
  proc-receive hook tests: use "test_hook" instead of "write_script"
  tests: extend "test_hook" for "rm" and "chmod -x", convert "$HOOK"
  tests: use "test_hook" for misc "mkdir -p" and "chmod" cases
  tests: change "mkdir -p && write_script" to use "test_hook"
  tests: change "cat && chmod +x" to use "test_hook"
  gc + p4 tests: use "test_hook", remove sub-shells
  fetch+push tests: use "test_hook" and "test_when_finished" pattern
  bugreport tests: tighten up "git bugreport -s hooks" test
  tests: assume the hooks are disabled by default
  http tests: don't rely on "hook/post-update.sample"
  hook tests: turn exit code assertions into a loop
  test-lib-functions: add and use a "test_hook" wrapper

3 years agoMerge branch 'jd/prompt-upstream-mark'
Junio C Hamano [Thu, 31 Mar 2022 01:01:11 +0000 (18:01 -0700)] 
Merge branch 'jd/prompt-upstream-mark'

Tweaks in the command line prompt (in contrib/) code around its
GIT_PS1_SHOWUPSTREAM feature.

* jd/prompt-upstream-mark:
  git-prompt: put upstream comments together
  git-prompt: make long upstream state indicator consistent
  git-prompt: make upstream state indicator location consistent
  git-prompt: rename `upstream` to `upstream_type`

3 years agoMerge branch 'pw/add-p-single-key'
Junio C Hamano [Thu, 31 Mar 2022 01:01:11 +0000 (18:01 -0700)] 
Merge branch 'pw/add-p-single-key'

Finishing touches to C rewrite of "git add -i" in single-key
interactive mode.

* pw/add-p-single-key:
  terminal: restore settings on SIGTSTP
  terminal: work around macos poll() bug
  terminal: don't assume stdin is /dev/tty
  terminal: use flags for save_term()

3 years agoMerge branch 'ab/make-optim-noop'
Junio C Hamano [Thu, 31 Mar 2022 01:01:11 +0000 (18:01 -0700)] 
Merge branch 'ab/make-optim-noop'

A micro fix to a topic earlier merged to 'master'

* ab/make-optim-noop:
  Makefile: use ' ', not non-existing $(wspfx_SQ)

3 years agoMerge branch 'vd/stash-silence-reset'
Junio C Hamano [Thu, 31 Mar 2022 01:01:10 +0000 (18:01 -0700)] 
Merge branch 'vd/stash-silence-reset'

"git stash" does not allow subcommands it internally runs as its
implementation detail, except for "git reset", to emit messages;
now "git reset" part has also been squelched.

* vd/stash-silence-reset:
  reset: show --no-refresh in the short-help
  reset: remove 'reset.refresh' config option
  reset: remove 'reset.quiet' config option
  reset: do not make '--quiet' disable index refresh
  stash: make internal resets quiet and refresh index
  reset: suppress '--no-refresh' advice if logging is silenced
  reset: replace '--quiet' with '--no-refresh' in performance advice
  reset: introduce --[no-]refresh option to --mixed
  reset: revise index refresh advice

3 years agoMerge branch 'ab/racy-hooks'
Junio C Hamano [Thu, 31 Mar 2022 01:01:10 +0000 (18:01 -0700)] 
Merge branch 'ab/racy-hooks'

Regression fix.

* ab/racy-hooks:
  hooks: fix "invoked hook" regression in a8cc5943338

3 years agobranch: remove negative exit code
Glen Choo [Tue, 29 Mar 2022 20:01:19 +0000 (20:01 +0000)] 
branch: remove negative exit code

Replace an instance of "exit(-1)" with "exit(1)". We don't use negative
exit codes - they are misleading because Unix machines will coerce them
to 8-bit unsigned values, losing the sign.

Signed-off-by: Glen Choo <chooglen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agoobject-file: pass filename to fsync_or_die
Neeraj Singh [Wed, 30 Mar 2022 18:14:15 +0000 (18:14 +0000)] 
object-file: pass filename to fsync_or_die

If we die while trying to fsync a loose object file, pass the actual
filename we're trying to sync. This is likely to be more helpful for a
user trying to diagnose the cause of the failure than the former
'loose object file' string. It also sidesteps any concerns about
translating the die message differently for loose objects versus
something else that has a real path.

Reported-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Neeraj Singh <neerajsi@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agocore.fsyncmethod: correctly camel-case warning message
Neeraj Singh [Wed, 30 Mar 2022 18:08:36 +0000 (18:08 +0000)] 
core.fsyncmethod: correctly camel-case warning message

The warning for an unrecognized fsyncMethod was not
camel-cased.

Reported-by: Jiang Xin <worldhello.net@gmail.com>
Signed-off-by: Neeraj Singh <neerajsi@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agobranch --set-upstream-to: be consistent when advising
Glen Choo [Tue, 29 Mar 2022 20:01:18 +0000 (20:01 +0000)] 
branch --set-upstream-to: be consistent when advising

"git branch --set-upstream-to" behaves differently when advice is
enabled/disabled:

|                 | error prefix | exit code |
|-----------------+--------------+-----------|
| advice enabled  | error:       |         1 |
| advice disabled | fatal:       |       128 |

Make both cases consistent by using die_message() when advice is
enabled (this was first proposed in [1]).

[1] https://lore.kernel.org/git/211210.86ee6ldwlc.gmgdl@evledraar.gmail.com

Signed-off-by: Glen Choo <chooglen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agobranch: give submodule updating advice before exit
Glen Choo [Tue, 29 Mar 2022 20:01:17 +0000 (20:01 +0000)] 
branch: give submodule updating advice before exit

Fix a bug where "hint:" was printed _before_ "fatal:" (instead of the
other way around).

Signed-off-by: Glen Choo <chooglen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agobranch: support more tracking modes when recursing
Glen Choo [Tue, 29 Mar 2022 20:01:16 +0000 (20:01 +0000)] 
branch: support more tracking modes when recursing

"git branch --recurse-submodules" does not propagate "--track=inherit"
or "--no-track" to submodules, which causes submodule branches to use
the wrong tracking mode [1]. To fix this, pass the correct options to
the "submodule--helper create-branch" child process and test for it.

While we are refactoring the same code, replace "--track" with the
synonymous, but more consistent-looking "--track=direct" option
(introduced at the same time as "--track=inherit", d3115660b4 (branch:
add flags and config to inherit tracking, 2021-12-20)).

[1] This bug is partially a timing issue: "branch --recurse-submodules"
 was introduced around the same time as "--track=inherit", and even
 though I rebased "branch --recurse-submodules" on top of that, I had
 neglected to support the new tracking mode. Omitting "--no-track"
 was just a plain old mistake, though.

Signed-off-by: Glen Choo <chooglen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agotrace2: add stats for fsync operations
Neeraj Singh [Wed, 30 Mar 2022 05:06:40 +0000 (05:06 +0000)] 
trace2: add stats for fsync operations

Add some global trace2 statistics for the number of fsyncs performed
during the lifetime of a Git process.

These stats are printed as part of trace2_cmd_exit_fl, which is
presumably where we might want to print any other cross-cutting
statistics.

Signed-off-by: Neeraj Singh <neerajsi@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agocore.fsync: fix incorrect expression for default configuration
Neeraj Singh [Tue, 29 Mar 2022 21:41:52 +0000 (21:41 +0000)] 
core.fsync: fix incorrect expression for default configuration

Commit b9f5d035 (core.fsync: documentation and user-friendly
aggregate options, 2022-03-15) introduced an incorrect value for
FSYNC_COMPONENTS_DEFAULT. We need an AND-NOT rather than OR-NOT.

Signed-off-by: Neeraj Singh <neerajsi@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agoThe 16th batch
Junio C Hamano [Tue, 29 Mar 2022 19:14:56 +0000 (12:14 -0700)] 
The 16th batch

Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agoMerge branch 'jc/rebase-detach-fix'
Junio C Hamano [Tue, 29 Mar 2022 19:22:03 +0000 (12:22 -0700)] 
Merge branch 'jc/rebase-detach-fix'

"git rebase $base $non_branch_commit", when $base is an ancestor or
the $non_branch_commit, modified the current branch, which has been
corrected.

* jc/rebase-detach-fix:
  rebase: set REF_HEAD_DETACH in checkout_up_to_date()
  rebase: use test_commit helper in setup

3 years agoMerge branch 'jt/reset-grafts-when-resetting-shallow'
Junio C Hamano [Tue, 29 Mar 2022 19:22:03 +0000 (12:22 -0700)] 
Merge branch 'jt/reset-grafts-when-resetting-shallow'

When "shallow" information is updated, we forgot to update the
in-core equivalent, which has been corrected.

* jt/reset-grafts-when-resetting-shallow:
  shallow: reset commit grafts when shallow is reset

3 years agoMerge branch 'vd/cache-bottom-fix'
Junio C Hamano [Tue, 29 Mar 2022 19:22:02 +0000 (12:22 -0700)] 
Merge branch 'vd/cache-bottom-fix'

Correct a bug in unpack-trees introduced earlier.

* vd/cache-bottom-fix:
  Revert "unpack-trees: improve performance of next_cache_entry"
  unpack-trees: increment cache_bottom for sparse directories
  t1092: add sparse directory before cone in test repo

3 years agoMerge branch 'ab/refs-various-fixes'
Junio C Hamano [Tue, 29 Mar 2022 19:22:02 +0000 (12:22 -0700)] 
Merge branch 'ab/refs-various-fixes'

Code clean-up.

* ab/refs-various-fixes:
  refs debug: add a wrapper for "read_symbolic_ref"
  packed-backend: remove stub BUG(...) functions
  misc *.c: use designated initializers for struct assignments
  refs: use designated initializers for "struct ref_iterator_vtable"
  refs: use designated initializers for "struct ref_storage_be"

3 years agoworktree: include repair cmd in usage
Des Preston [Mon, 28 Mar 2022 20:47:05 +0000 (20:47 +0000)] 
worktree: include repair cmd in usage

The worktree repair command was not added to the usage menu for the
worktree command. This commit adds the usage of 'worktree repair'
according to the existing docs.

Signed-off-by: Des Preston <despreston@gmail.com>
Acked-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agoreflog: fix 'show' subcommand's argv
SZEDER Gábor [Mon, 28 Mar 2022 21:21:52 +0000 (23:21 +0200)] 
reflog: fix 'show' subcommand's argv

cmd_reflog() invokes parse_options() with PARSE_OPT_KEEP_ARGV0, but it
doesn't account for the retained argv[0] before invoking
cmd_reflog_show() to handle the 'git reflog show' subcommand.
Consequently, cmd_reflog_show() always gets an 'argv' array starting
with elements argv[0]="reflog" and argv[1]="show".

Strip the name of the git command from the 'argv' array before passing
it to the function handling the 'show' subcommand.

There is no user-visible bug here, because cmd_reflog_show() doesn't
have any options or parameters of its own.

Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agoreftable: make assignments portable to AIX xlc v12.01
Ævar Arnfjörð Bjarmason [Mon, 28 Mar 2022 19:10:04 +0000 (21:10 +0200)] 
reftable: make assignments portable to AIX xlc v12.01

Change the assignment syntax introduced in 66c0dabab5e (reftable: make
reftable_record a tagged union, 2022-01-20) to be portable to AIX xlc
v12.1:

    avar@gcc111:[/home/avar]xlc -qversion
    IBM XL C/C++ for AIX, V12.1 (5765-J02, 5725-C72)
    Version: 12.01.0000.0000

The error emitted before this was e.g.:

    "reftable/generic.c", line 133.26: 1506-196 (S) Initialization
    between types "char*" and "struct reftable_ref_record" is not
    allowed.

The syntax in the pre-image is supported by e.g. xlc 13.01 on a newer
AIX version:

    avar@gcc119:[/home/avar]xlc -qversion
    IBM XL C/C++ for AIX, V13.1.3 (5725-C72, 5765-J07)
    Version: 13.01.0003.0006

But as we've otherwise supported this compiler let's not break it
entirely if it's easy to work around it.

Suggested-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agodocs: mention --refetch fetch option
Robert Coup [Mon, 28 Mar 2022 14:02:11 +0000 (14:02 +0000)] 
docs: mention --refetch fetch option

Document it for partial clones as a means to apply a new filter, and
reference it from the remote.<name>.partialclonefilter config parameter.

Signed-off-by: Robert Coup <robert@coup.net.nz>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agofetch: after refetch, encourage auto gc repacking
Robert Coup [Mon, 28 Mar 2022 14:02:10 +0000 (14:02 +0000)] 
fetch: after refetch, encourage auto gc repacking

After invoking `fetch --refetch`, the object db will likely contain many
duplicate objects. If auto-maintenance is enabled, invoke it with
appropriate settings to encourage repacking/consolidation.

* gc.autoPackLimit: unless this is set to 0 (disabled), override the
  value to 1 to force pack consolidation.
* maintenance.incremental-repack.auto: unless this is set to 0, override
  the value to -1 to force incremental repacking.

Signed-off-by: Robert Coup <robert@coup.net.nz>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agot5615-partial-clone: add test for fetch --refetch
Robert Coup [Mon, 28 Mar 2022 14:02:09 +0000 (14:02 +0000)] 
t5615-partial-clone: add test for fetch --refetch

Add a test for doing a refetch to apply a changed partial clone filter
under protocol v0 and v2.

Signed-off-by: Robert Coup <robert@coup.net.nz>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agofetch: add --refetch option
Robert Coup [Mon, 28 Mar 2022 14:02:08 +0000 (14:02 +0000)] 
fetch: add --refetch option

Teach fetch and transports the --refetch option to force a full fetch
without negotiating common commits with the remote. Use when applying a
new partial clone filter to refetch all matching objects.

Signed-off-by: Robert Coup <robert@coup.net.nz>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agobuiltin/fetch-pack: add --refetch option
Robert Coup [Mon, 28 Mar 2022 14:02:07 +0000 (14:02 +0000)] 
builtin/fetch-pack: add --refetch option

Add a refetch option to fetch-pack to force a full fetch. Use when
applying a new partial clone filter to refetch all matching objects.

Signed-off-by: Robert Coup <robert@coup.net.nz>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agofetch-pack: add refetch
Robert Coup [Mon, 28 Mar 2022 14:02:06 +0000 (14:02 +0000)] 
fetch-pack: add refetch

Allow a "refetch" where the contents of the local object store are
ignored and a full fetch is performed, not attempting to find or
negotiate common commits with the remote.

A key use case is to apply a new partial clone blob/tree filter and
refetch all the associated matching content, which would otherwise not
be transferred when the commit objects are already present locally.

Signed-off-by: Robert Coup <robert@coup.net.nz>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agofetch-negotiator: add specific noop initializer
Robert Coup [Mon, 28 Mar 2022 14:02:05 +0000 (14:02 +0000)] 
fetch-negotiator: add specific noop initializer

Add a specific initializer for the noop fetch negotiator. This is
introduced to support allowing partial clones to skip commit negotiation
when performing a "refetch".

Signed-off-by: Robert Coup <robert@coup.net.nz>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agopack-objects: lazily set up "struct rev_info", don't leak
Ævar Arnfjörð Bjarmason [Mon, 28 Mar 2022 15:43:18 +0000 (17:43 +0200)] 
pack-objects: lazily set up "struct rev_info", don't leak

In the preceding [1] (pack-objects: move revs out of
get_object_list(), 2022-03-22) the "repo_init_revisions()" was moved
to cmd_pack_objects() so that it unconditionally took place for all
invocations of "git pack-objects".

We'd thus start leaking memory, which is easily reproduced in
e.g. git.git by feeding e83c5163316 (Initial revision of "git", the
information manager from hell, 2005-04-07) to "git pack-objects";

    $ echo e83c5163316f89bfbde7d9ab23ca2e25604af290 | ./git pack-objects initial
    [...]
==19130==ERROR: LeakSanitizer: detected memory leaks

Direct leak of 7120 byte(s) in 1 object(s) allocated from:
    #0 0x455308 in __interceptor_malloc (/home/avar/g/git/git+0x455308)
    #1 0x75b399 in do_xmalloc /home/avar/g/git/wrapper.c:41:8
    #2 0x75b356 in xmalloc /home/avar/g/git/wrapper.c:62:9
    #3 0x5d7609 in prep_parse_options /home/avar/g/git/diff.c:5647:2
    #4 0x5d415a in repo_diff_setup /home/avar/g/git/diff.c:4621:2
    #5 0x6dffbb in repo_init_revisions /home/avar/g/git/revision.c:1853:2
    #6 0x4f599d in cmd_pack_objects /home/avar/g/git/builtin/pack-objects.c:3980:2
    #7 0x4592ca in run_builtin /home/avar/g/git/git.c:465:11
    #8 0x457d81 in handle_builtin /home/avar/g/git/git.c:718:3
    #9 0x458ca5 in run_argv /home/avar/g/git/git.c:785:4
    #10 0x457b40 in cmd_main /home/avar/g/git/git.c:916:19
    #11 0x562259 in main /home/avar/g/git/common-main.c:56:11
    #12 0x7fce792ac7ec in __libc_start_main csu/../csu/libc-start.c:332:16
    #13 0x4300f9 in _start (/home/avar/g/git/git+0x4300f9)

SUMMARY: LeakSanitizer: 7120 byte(s) leaked in 1 allocation(s).
Aborted

Narrowly fixing that commit would have been easy, just add call
repo_init_revisions() right before get_object_list(), which is
effectively what was done before that commit.

But an unstated constraint when setting it up early is that it was
needed for the subsequent [2] (pack-objects: parse --filter directly
into revs.filter, 2022-03-22), i.e. we might have a --filter
command-line option, and need to either have the "struct rev_info"
setup when we encounter that option, or later.

Let's just change the control flow so that we'll instead set up the
"struct rev_info" only when we need it. Doing so leads to a bit more
verbosity, but it's a lot clearer what we're doing and why.

An earlier version of this commit[3] went behind
opt_parse_list_objects_filter()'s back by faking up a "struct option"
before calling it. Let's avoid that and instead create a blessed API
for this pattern.

We could furthermore combine the two get_object_list() invocations
here by having repo_init_revisions() invoked on &pfd.revs, but I think
clearly separating the two makes the flow clearer. Likewise
redundantly but explicitly (i.e. redundant v.s. a "{ 0 }") "0" to
"have_revs" early in cmd_pack_objects().

While we're at it add parentheses around the arguments to the OPT_*
macros in in list-objects-filter-options.h, as we need to change those
lines anyway. It doesn't matter in this case, but is good general
practice.

1. https://lore.kernel.org/git/619b757d98465dbc4995bdc11a5282fbfcbd3daa.1647970119.git.gitgitgadget@gmail.com
2. https://lore.kernel.org/git/97de926904988b89b5663bd4c59c011a1723a8f5.1647970119.git.gitgitgadget@gmail.com
3. https://lore.kernel.org/git/patch-1.1-193534b0f07-20220325T121715Z-avarab@gmail.com/

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agoThe 15th batch
Junio C Hamano [Fri, 25 Mar 2022 23:35:36 +0000 (16:35 -0700)] 
The 15th batch

Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agoMerge branch 'gc/recursive-fetch-with-unused-submodules'
Junio C Hamano [Fri, 25 Mar 2022 23:38:25 +0000 (16:38 -0700)] 
Merge branch 'gc/recursive-fetch-with-unused-submodules'

When "git fetch --recurse-submodules" grabbed submodule commits
that would be needed to recursively check out newly fetched commits
in the superproject, it only paid attention to submodules that are
in the current checkout of the superproject.  We now do so for all
submodules that have been run "git submodule init" on.

* gc/recursive-fetch-with-unused-submodules:
  submodule: fix latent check_has_commit() bug
  fetch: fetch unpopulated, changed submodules
  submodule: move logic into fetch_task_create()
  submodule: extract get_fetch_task()
  submodule: store new submodule commits oid_array in a struct
  submodule: inline submodule_commits() into caller
  submodule: make static functions read submodules from commits
  t5526: create superproject commits with test helper
  t5526: stop asserting on stderr literally
  t5526: introduce test helper to assert on fetches

3 years agoMerge branch 'ps/fsync-refs'
Junio C Hamano [Fri, 25 Mar 2022 23:38:24 +0000 (16:38 -0700)] 
Merge branch 'ps/fsync-refs'

Updates to refs traditionally weren't fsync'ed, but we can
configure using core.fsync variable to do so.

* ps/fsync-refs:
  core.fsync: new option to harden references

3 years agoMerge branch 'ns/core-fsyncmethod'
Junio C Hamano [Fri, 25 Mar 2022 23:38:24 +0000 (16:38 -0700)] 
Merge branch 'ns/core-fsyncmethod'

Replace core.fsyncObjectFiles with two new configuration variables,
core.fsync and core.fsyncMethod.

* ns/core-fsyncmethod:
  core.fsync: documentation and user-friendly aggregate options
  core.fsync: new option to harden the index
  core.fsync: add configuration parsing
  core.fsync: introduce granular fsync control infrastructure
  core.fsyncmethod: add writeout-only mode
  wrapper: make inclusion of Windows csprng header tightly scoped

3 years agotest-lib-functions: remove test_subcommand_inexact
Derrick Stolee [Fri, 25 Mar 2022 19:02:47 +0000 (19:02 +0000)] 
test-lib-functions: remove test_subcommand_inexact

The implementation of test_subcommand_inexact() was originally
introduced in e4d0c11c0 (repack: respect kept objects with '--write-midx
-b', 2021-12-20) with the intention to allow finding a subcommand based
on an initial set of arguments. The inexactness was intended as a way to
allow flexible options beyond that initial set, as opposed to
test_subcommand() which requires that the full list of options is
provided in its entirety.

The implementation began by copying test_subcommand() and replaced the
repeated argument 'printf' statement to append ".*" instead of "," to
each argument. This caused it to be more flexible than initially
intended.

The previous change deleted the only use of test_subcommand_inexact, so
instead of editing the helper, delete it.

Helped-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agot7700: check post-condition in kept-pack test
Derrick Stolee [Fri, 25 Mar 2022 19:02:46 +0000 (19:02 +0000)] 
t7700: check post-condition in kept-pack test

The '--write-midx -b packs non-kept objects' test in t7700-repack.sh
uses test_subcommand_inexact to check that 'git repack' properly adds
the '--honor-pack-keep' flag to the 'git pack-objects' subcommand.
However, the test_subcommand_inexact helper is more flexible than
initially designed, and this instance is the only one that makes use of
it: there are additional arguments between 'git pack-objects' and the
'--honor-pack-keep' flag. In order to make test_subcommand_inexact more
strict, we need to fix this instance.

This test checks that 'git repack --write-midx -a -b -d' will create a
new pack-file that does not contain the objects within the kept pack.
This behavior is possible because of the multi-pack-index bitmap that
will bitmap objects against multiple packs. Without --write-midx, the
objects in the kept pack would be duplicated so the resulting pack is
closed under reachability and bitmaps can be created against it. This is
discussed in more detail in e4d0c11c0 (repack: respect kept objects with
'--write-midx -b', 2021-12-20) which also introduced this instance of
test_subcommand_inexact.

To better verify the intended post-conditions while also removing this
instance of test_subcommand_inexact, rewrite the test to check the list
of packed objects in the kept pack and the list of the objects in the
newly-repacked pack-file _other_ than the kept pack. These lists should
be disjoint.

Be sure to include a non-kept pack-file and loose objects to be extra
careful that this is properly behaving with kept packs and not just
avoiding repacking all pack-files.

Co-authored-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agotest-lib: have --immediate emit valid TAP on failure
Ævar Arnfjörð Bjarmason [Wed, 23 Mar 2022 20:51:31 +0000 (21:51 +0100)] 
test-lib: have --immediate emit valid TAP on failure

Change the "--immediate" option so that it emits valid TAP on
failure. Before this it would omit the required plan at the end,
e.g. under SANITIZE=leak we'd show a "No plan found in TAP output"
error from "prove":

    $ prove t0006-date.sh ::  --immediate
    t0006-date.sh .. Dubious, test returned 1 (wstat 256, 0x100)
    Failed 1/22 subtests

    Test Summary Report
    -------------------
    t0006-date.sh (Wstat: 256 Tests: 22 Failed: 1)
      Failed test:  22
      Non-zero exit status: 1
      Parse errors: No plan found in TAP output
    Files=1, Tests=22,  0 wallclock secs ( 0.02 usr  0.01 sys +  0.18 cusr  0.06 csys =  0.27 CPU)
    Result: FAIL

Now we'll emit output that doesn't result in TAP parsing failures:

    $ prove t0006-date.sh ::  --immediate
    t0006-date.sh .. Dubious, test returned 1 (wstat 256, 0x100)
    Failed 1/22 subtests

    Test Summary Report
    -------------------
    t0006-date.sh (Wstat: 256 Tests: 22 Failed: 1)
      Failed test:  22
      Non-zero exit status: 1
    Files=1, Tests=22,  0 wallclock secs ( 0.02 usr  0.00 sys +  0.19 cusr  0.05 csys =  0.26 CPU)
    Result: FAIL

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agoreset: show --no-refresh in the short-help
Junio C Hamano [Thu, 24 Mar 2022 17:33:10 +0000 (10:33 -0700)] 
reset: show --no-refresh in the short-help

In the short help output from "git reset -h", the recently added
"--[no-]refresh" option is shown like so:

        --refresh             skip refreshing the index after reset

which explains what happens when the option is given in the negative
form, i.e. "--no-refresh".  We could rephrase the explanation to
read "refresh the index after reset (default)" to hint that the user
can say "--no-refresh" to override the default, but listing the
"--no-refresh" form in the list of options would be more helpful.

Helped-by: Phillip Wood <phillip.wood@dunelm.org.uk>
Acked-by: Victoria Dye <vdye@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agoreflog [show]: display sensible -h output
Ævar Arnfjörð Bjarmason [Thu, 17 Mar 2022 18:08:40 +0000 (19:08 +0100)] 
reflog [show]: display sensible -h output

Change the "git reflog show -h" output to show the usage summary
relevant to it, rather than displaying the same output that "git log
-h" would show.

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agoreflog: convert to parse_options() API
Ævar Arnfjörð Bjarmason [Thu, 17 Mar 2022 18:08:39 +0000 (19:08 +0100)] 
reflog: convert to parse_options() API

Continue the work started in 33d7bdd6459 (builtin/reflog.c: use
parse-options api for expire, delete subcommands, 2022-01-06) and
convert the cmd_reflog() function itself to use the parse_options()
API.

Let's also add a test which would fail if we forgot
PARSE_OPT_NO_INTERNAL_HELP here, as well as making sure that we'll
still pass through "--" by supplying PARSE_OPT_KEEP_DASHDASH. For that
test we need to change "test_commit()" to accept files starting with
"--".

The "git reflog -h" usage will now show the usage for all of the
sub-commands, rather than a terse summary which wasn't
correct (e.g. "git reflog exists" is not a valid command). See my
8757b35d443 (commit-graph: define common usage with a macro,
2021-08-23) for prior art.

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agoreset: remove 'reset.refresh' config option
Victoria Dye [Wed, 23 Mar 2022 18:18:00 +0000 (18:18 +0000)] 
reset: remove 'reset.refresh' config option

Remove the 'reset.refresh' option, requiring that users explicitly specify
'--no-refresh' if they want to skip refreshing the index.

The 'reset.refresh' option was introduced in 101cee42dd (reset: introduce
--[no-]refresh option to --mixed, 2022-03-11) as a replacement for the
refresh-skipping behavior originally controlled by 'reset.quiet'.

Although 'reset.refresh=false' functionally served the same purpose as
'reset.quiet=true', it exposed [1] the fact that the existence of a global
"skip refresh" option could potentially cause problems for users. Allowing a
global config option to avoid refreshing the index forces scripts using 'git
reset --mixed' to defensively use '--refresh' if index refresh is expected;
if that option is missing, behavior of a script could vary from user-to-user
without explanation.

Furthermore, globally disabling index refresh in 'reset --mixed' was
initially devised as a passive performance improvement; since the
introduction of the option, other changes have been made to Git (e.g., the
sparse index) with a greater potential performance impact without
sacrificing index correctness. Therefore, we can more aggressively err on
the side of correctness and limit the cases of skipping index refresh to
only when a user specifies the '--no-refresh' option.

[1] https://lore.kernel.org/git/xmqqy2179o3c.fsf@gitster.g/

Signed-off-by: Victoria Dye <vdye@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agoreset: remove 'reset.quiet' config option
Victoria Dye [Wed, 23 Mar 2022 18:17:59 +0000 (18:17 +0000)] 
reset: remove 'reset.quiet' config option

Remove the 'reset.quiet' config option, remove '--no-quiet' documentation in
'Documentation/git-reset.txt'. In 4c3abd0551 (reset: add new reset.quiet
config setting, 2018-10-23), 'reset.quiet' was introduced as a way to
globally change the default behavior of 'git reset --mixed' to skip index
refresh.

However, now that '--quiet' does not affect index refresh, 'reset.quiet'
would only serve to globally silence logging. This was not the original
intention of the config setting, and there's no precedent for such a setting
in other commands with a '--quiet' option, so it appears to be obsolete.

In addition to the options & its documentation, remove 'reset.quiet' from
the recommended config for 'scalar'.

Signed-off-by: Victoria Dye <vdye@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agoreset: do not make '--quiet' disable index refresh
Victoria Dye [Wed, 23 Mar 2022 18:17:58 +0000 (18:17 +0000)] 
reset: do not make '--quiet' disable index refresh

Update '--quiet' to no longer implicitly skip refreshing the index in a
mixed reset. Users now have the ability to explicitly disable refreshing the
index with the '--no-refresh' option, so they no longer need to use
'--quiet' to do so. Moreover, we explicitly remove the refresh-skipping
behavior from '--quiet' because it is completely unrelated to the stated
purpose of the option: "Be quiet, only report errors."

Helped-by: Phillip Wood <phillip.wood@dunelm.org.uk>
Signed-off-by: Victoria Dye <vdye@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agoThe 14th batch
Junio C Hamano [Wed, 23 Mar 2022 21:04:45 +0000 (14:04 -0700)] 
The 14th batch

Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agoMerge branch 'ab/plug-random-leaks'
Junio C Hamano [Wed, 23 Mar 2022 21:09:31 +0000 (14:09 -0700)] 
Merge branch 'ab/plug-random-leaks'

Double-free fix for a recently merged topic.

* ab/plug-random-leaks:
  diff.c: fix a double-free regression in a18d66cefb
  tests: demonstrate "show --word-diff --color-moved" regression

3 years agoMerge branch 'dc/complete-restore'
Junio C Hamano [Wed, 23 Mar 2022 21:09:31 +0000 (14:09 -0700)] 
Merge branch 'dc/complete-restore'

The command line completion support (in contrib/) learns to give
modified paths to the "git restore" command.

* dc/complete-restore:
  completion: tab completion of filenames for 'git restore'

3 years agoMerge branch 'jc/cat-file-batch-default-format-optim'
Junio C Hamano [Wed, 23 Mar 2022 21:09:31 +0000 (14:09 -0700)] 
Merge branch 'jc/cat-file-batch-default-format-optim'

Optimize away strbuf_expand() call with a hardcoded formatting logic
specific for the default format in the --batch and --batch-check
options of "git cat-file".

* jc/cat-file-batch-default-format-optim:
  cat-file: skip expanding default format

3 years agoMerge branch 'js/in-place-reverse-in-sequencer'
Junio C Hamano [Wed, 23 Mar 2022 21:09:31 +0000 (14:09 -0700)] 
Merge branch 'js/in-place-reverse-in-sequencer'

Code clean-up.

* js/in-place-reverse-in-sequencer:
  sequencer: use reverse_commit_list() helper

3 years agoMerge branch 'ac/test-lazy-fetch'
Junio C Hamano [Wed, 23 Mar 2022 21:09:30 +0000 (14:09 -0700)] 
Merge branch 'ac/test-lazy-fetch'

A new test to ensure a lazy fetching is not triggered when it
should not be.

* ac/test-lazy-fetch:
  partial-clone: add a partial-clone test case

3 years agoMerge branch 'ps/repack-with-server-info'
Junio C Hamano [Wed, 23 Mar 2022 21:09:30 +0000 (14:09 -0700)] 
Merge branch 'ps/repack-with-server-info'

"git repack" learned a new configuration to disable triggering of
age-old "update-server-info" command, which is rarely useful these
days.

* ps/repack-with-server-info:
  repack: add config to skip updating server info
  repack: refactor to avoid double-negation of update-server-info

3 years agoMerge branch 'ds/doc-maintenance-synopsis-fix'
Junio C Hamano [Wed, 23 Mar 2022 21:09:30 +0000 (14:09 -0700)] 
Merge branch 'ds/doc-maintenance-synopsis-fix'

Doc update.

* ds/doc-maintenance-synopsis-fix:
  maintenance: fix synopsis in documentation

3 years agoMerge branch 'ab/reflog-prep-fix'
Junio C Hamano [Wed, 23 Mar 2022 21:09:30 +0000 (14:09 -0700)] 
Merge branch 'ab/reflog-prep-fix'

Regression fix.

* ab/reflog-prep-fix:
  reflog: don't be noisy on empty reflogs

3 years agoMerge branch 'ep/remove-duplicated-includes'
Junio C Hamano [Wed, 23 Mar 2022 21:09:30 +0000 (14:09 -0700)] 
Merge branch 'ep/remove-duplicated-includes'

Code clean-up.

* ep/remove-duplicated-includes:
  attr.h: remove duplicate struct definition
  t/helper/test-run-command.c: delete duplicate include
  builtin/stash.c: delete duplicate include
  builtin/sparse-checkout.c: delete duplicate include
  builtin/gc.c: delete duplicate include
  attr.c: delete duplicate include

3 years agoMerge branch 'ep/t6423-modernize'
Junio C Hamano [Wed, 23 Mar 2022 21:09:29 +0000 (14:09 -0700)] 
Merge branch 'ep/t6423-modernize'

Code clean-up.

* ep/t6423-modernize:
  t6423-merge-rename-directories.sh: use the $(...) construct

3 years agoMerge branch 'jk/name-rev-w-genno'
Junio C Hamano [Wed, 23 Mar 2022 21:09:29 +0000 (14:09 -0700)] 
Merge branch 'jk/name-rev-w-genno'

"git name-rev" learned to use the generation numbers when setting
the lower bound of searching commits used to explain the revision,
when available, instead of committer time.

* jk/name-rev-w-genno:
  name-rev: use generation numbers if available

3 years agoMerge branch 'jd/userdiff-kotlin'
Junio C Hamano [Wed, 23 Mar 2022 21:09:29 +0000 (14:09 -0700)] 
Merge branch 'jd/userdiff-kotlin'

A new built-in userdiff driver for kotlin.

* jd/userdiff-kotlin:
  userdiff: add builtin diff driver for kotlin language.

3 years agoMerge branch 'bc/block-sha1-without-gcc-asm-extension'
Junio C Hamano [Wed, 23 Mar 2022 21:09:29 +0000 (14:09 -0700)] 
Merge branch 'bc/block-sha1-without-gcc-asm-extension'

Get rid of one use of __asm__() GCC extension that does not help us
much these days, which has an added advantage of not having to
worry about -pedantic complaining.

* bc/block-sha1-without-gcc-asm-extension:
  block-sha1: remove use of obsolete x86 assembly

3 years agoMerge branch 'gc/submodule-update-part1'
Junio C Hamano [Wed, 23 Mar 2022 21:09:28 +0000 (14:09 -0700)] 
Merge branch 'gc/submodule-update-part1'

Rewrite of "git submodule update" in C (early part).

* gc/submodule-update-part1:
  submodule--helper update-clone: check for --filter and --init
  submodule update: add tests for --filter
  submodule--helper: remove ensure-core-worktree
  submodule--helper update-clone: learn --init
  submodule--helper: allow setting superprefix for init_submodule()
  submodule--helper: refactor get_submodule_displaypath()
  submodule--helper run-update-procedure: learn --remote
  submodule--helper: don't use bitfield indirection for parse_options()
  submodule--helper: get remote names from any repository
  submodule--helper run-update-procedure: remove --suboid
  submodule--helper: reorganize code for sh to C conversion
  submodule--helper: remove update-module-mode
  submodule tests: test for init and update failure output

3 years agobundle: output hash information in 'verify'
Derrick Stolee [Tue, 22 Mar 2022 17:28:39 +0000 (17:28 +0000)] 
bundle: output hash information in 'verify'

The previous change moved the 'filter' capability to the end of the 'git
bundle verify' output. Now, add the 'object-format' capability to the
output, when it exists.

This change makes 'git bundle verify' output the hash used in all cases,
even if the capability is not in the bundle. This means that v2 bundles
will always output that they use "sha1". This might look noisy to some
users, but it does simplify the implementation and the test strategy for
this feature.

Since 'verify' ends early when a prerequisite commit is missing, we need
to insert this hash message carefully into our expected test output
throughout t6020.

Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agobundle: move capabilities to end of 'verify'
Derrick Stolee [Tue, 22 Mar 2022 17:28:38 +0000 (17:28 +0000)] 
bundle: move capabilities to end of 'verify'

The 'filter' capability was added in 105c6f14a (bundle: parse filter
capability, 2022-03-09), but was added in a strange place in the 'git
bundle verify' output.

The tests for this show output like the following:

The bundle contains these 2 refs:
<COMMIT1> <REF1>
<COMMIT2> <REF2>
The bundle uses this filter: blob:none
The bundle records a complete history.

This looks very odd if we have a thin bundle that contains boundary
commits instead of a complete history:

The bundle contains these 2 refs:
<COMMIT1> <REF1>
<COMMIT2> <REF2>
The bundle uses this filter: blob:none
The bundle requires these 2 refs:
<COMMIT3>
<COMMIT4>

This separation between tip refs and boundary refs is unfortunate. Move
the filter capability output to the end of the output. Update the
documentation to match.

Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agopack-objects: parse --filter directly into revs.filter
Derrick Stolee [Tue, 22 Mar 2022 17:28:37 +0000 (17:28 +0000)] 
pack-objects: parse --filter directly into revs.filter

The previous change moved the 'revs' variable into cmd_pack_objects()
and now we can remove the global filter_options in favor of revs.filter.

Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agopack-objects: move revs out of get_object_list()
Derrick Stolee [Tue, 22 Mar 2022 17:28:36 +0000 (17:28 +0000)] 
pack-objects: move revs out of get_object_list()

We intend to parse the --filter option directly into revs.filter, but we
first need to move the 'revs' variable out of get_object_list() and pass
it as a pointer instead. This change only deals with the issues of
making that work.

Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agolist-objects-filter: remove CL_ARG__FILTER
Derrick Stolee [Tue, 22 Mar 2022 17:28:35 +0000 (17:28 +0000)] 
list-objects-filter: remove CL_ARG__FILTER

We have established the command-line interface for the --[no-]filter
options for a while now, so we do not need a helper to make this
editable in the future.

Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agogit-prompt: put upstream comments together
Justin Donnelly [Sun, 27 Feb 2022 19:57:12 +0000 (19:57 +0000)] 
git-prompt: put upstream comments together

Commit 6d158cba28 (bash completion: Support "divergence from upstream"
messages in __git_ps1, 2010-06-17) introduced support for indicating
divergence from upstream in the PS1 prompt. The comments at the top of
git-prompt.sh that were introduced with that commit are several
paragraphs long. Over the years, other comments have been inserted in
between the paragraphs relating to divergence from upstream.

This commit puts the comments relating to divergence from upstream back
together.

Signed-off-by: Justin Donnelly <justinrdonnelly@gmail.com>
Reviewed-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agogit-prompt: make long upstream state indicator consistent
Justin Donnelly [Sun, 27 Feb 2022 19:57:11 +0000 (19:57 +0000)] 
git-prompt: make long upstream state indicator consistent

Use a pipe as a separator before long upstream state indicator. This is
consistent with long state indicators for sparse and in-progress
operations (e.g. merge).

For comparison, `__git_ps1` examples without upstream state indicator:
(main)
(main %)
(main *%)
(main|SPARSE)
(main %|SPARSE)
(main *%|SPARSE)
(main|SPARSE|REBASE 1/2)
(main %|SPARSE|REBASE 1/2)

Note that if there are long state indicators, they appear after short
state indicators if there are any, or after the branch name if there are
no short state indicators. Each long state indicator begins with a pipe
(`|`) as a separator.

Before/after examples with long upstream state indicator:
| Before                          | After                           |
| ------------------------------- | ------------------------------- |
| (main u=)                       | (main|u=)                       |
| (main u= origin/main)           | (main|u= origin/main)           |
| (main u+1)                      | (main|u+1)                      |
| (main u+1 origin/main)          | (main|u+1 origin/main)          |
| (main % u=)                     | (main %|u=)                     |
| (main % u= origin/main)         | (main %|u= origin/main)         |
| (main % u+1)                    | (main %|u+1)                    |
| (main % u+1 origin/main)        | (main %|u+1 origin/main)        |
| (main|SPARSE u=)                | (main|SPARSE|u=)                |
| (main|SPARSE u= origin/main)    | (main|SPARSE|u= origin/main)    |
| (main|SPARSE u+1)               | (main|SPARSE|u+1)               |
| (main|SPARSE u+1 origin/main)   | (main|SPARSE|u+1 origin/main)   |
| (main %|SPARSE u=)              | (main %|SPARSE|u=)              |
| (main %|SPARSE u= origin/main)  | (main %|SPARSE|u= origin/main)  |
| (main %|SPARSE u+1)             | (main %|SPARSE|u+1)             |
| (main %|SPARSE u+1 origin/main) | (main %|SPARSE|u+1 origin/main) |

Signed-off-by: Justin Donnelly <justinrdonnelly@gmail.com>
Reviewed-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agogit-prompt: make upstream state indicator location consistent
Justin Donnelly [Sun, 27 Feb 2022 19:57:10 +0000 (19:57 +0000)] 
git-prompt: make upstream state indicator location consistent

Make upstream state indicator location more consistent with similar
state indicators (e.g. sparse). Group the short upstream state indicator
(`=`, `<`, `>`, or `<>`) with other short state indicators immediately
after the branch name. Previously short and long upstream state
indicators appeared after all other state indicators.

Use a separator (`SP` or `GIT_PS1_STATESEPARATOR`) between branch name
and short upstream state indicator. Previously the short upstream state
indicator would sometimes appear directly adjacent to the branch name
instead of being separated.

For comparison, `__git_ps1` examples without upstream state indicator:
(main)
(main %)
(main *%)
(main|SPARSE)
(main %|SPARSE)
(main *%|SPARSE)
(main|SPARSE|REBASE 1/2)
(main %|SPARSE|REBASE 1/2)

Note that if there are short state indicators, they appear together
after the branch name and separated from it by `SP` or
`GIT_PS1_STATESEPARATOR`.

Before/after examples with short upstream state indicator:
| Before           | After            |
| ---------------- | ---------------- |
| (main=)          | (main =)         |
| (main|SPARSE=)   | (main =|SPARSE)  |
| (main %|SPARSE=) | (main %=|SPARSE) |

Signed-off-by: Justin Donnelly <justinrdonnelly@gmail.com>
Reviewed-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agogit-prompt: rename `upstream` to `upstream_type`
Justin Donnelly [Sun, 27 Feb 2022 19:57:09 +0000 (19:57 +0000)] 
git-prompt: rename `upstream` to `upstream_type`

In `__git_ps1_show_upstream` rename the variable `upstream` to
`upstream_type`. This allows `__git_ps1_show_upstream` to reference a
variable named `upstream` that is declared `local` in `__git_ps1`, which
calls `__git_ps1_show_upstream`.

Signed-off-by: Justin Donnelly <justinrdonnelly@gmail.com>
Reviewed-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agohooks: fix "invoked hook" regression in a8cc5943338
Ævar Arnfjörð Bjarmason [Mon, 21 Mar 2022 23:15:13 +0000 (00:15 +0100)] 
hooks: fix "invoked hook" regression in a8cc5943338

Fix a regression in a8cc5943338 (hooks: fix an obscure TOCTOU "did we
just run a hook?" race, 2022-03-07): The "invoked_hook" variable
passed to run_commit_hook() wasn't passed forward to run_hooks_opt(),
as push_to_checkout() in that commit correctly did.

Whether we ran the code contingent on having run the hook or not was
thus undefined, but in practice on most (all?) modern platforms we'd
have run it (almost?) all the time, since stack variables will get
initialized to some random value, which most of the time isn't "0".

This bug was revealed by running e.g. "t5537-fetch-shallow.sh" with
the --valgrind option. Unfortunately running the whole test suite with
--valgrind is really slow, so we didn't have a CI job that spotted
this. The --valgrind output was:

    ==31275== Conditional jump or move depends on uninitialised value(s)
    ==31275==    at 0x43C63F: prepare_to_commit (commit.c:1058)
    ==31275==    by 0x4396A5: cmd_commit (commit.c:1722)
    ==31275==    by 0x407C8A: run_builtin (git.c:465)
    ==31275==    by 0x406741: handle_builtin (git.c:718)
    ==31275==    by 0x407665: run_argv (git.c:785)
    ==31275==    by 0x406500: cmd_main (git.c:916)
    ==31275==    by 0x510839: main (common-main.c:56)
    ==31275==  Uninitialised value was created by a stack allocation
    ==31275==    at 0x43B344: prepare_to_commit (commit.c:719)

Reported-by: Jonathan Tan <jonathantanmy@google.com>
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agols-tree: split up "fast path" callbacks
Ævar Arnfjörð Bjarmason [Wed, 23 Mar 2022 09:13:15 +0000 (17:13 +0800)] 
ls-tree: split up "fast path" callbacks

Make the various if/else in the callbacks for the "fast path" a lot
easier to read by just using common functions for the parts that are
common, and have per-format callbacks for those parts that are
different.

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Teng Long <dyroneteng@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agols-tree: detect and error on --name-only --name-status
Ævar Arnfjörð Bjarmason [Wed, 23 Mar 2022 09:13:14 +0000 (17:13 +0800)] 
ls-tree: detect and error on --name-only --name-status

The --name-only and --name-status options are synonyms, but let's
detect and error if both are provided.

In addition let's add explicit --format tests for the combination of
these various options.

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Teng Long <dyroneteng@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agols-tree: support --object-only option for "git-ls-tree"
Teng Long [Wed, 23 Mar 2022 09:13:13 +0000 (17:13 +0800)] 
ls-tree: support --object-only option for "git-ls-tree"

'--object-only' is an alias for '--format=%(objectname)'. It cannot
be used together other format-altering options like '--name-only',
'--long' or '--format', they are mutually exclusive.

The "--name-only" option outputs <filepath> only. Likewise, <objectName>
is another high frequency used field, so implement '--object-only' option
will bring intuitive and clear semantics for this scenario. Using
'--format=%(objectname)' we can achieve a similar effect, but the former
is with a lower learning cost(without knowing the format requirement
of '--format' option).

Even so, if a user is prefer to use "--format=%(objectname)", this is entirely
welcome because they are not only equivalent in function, but also have almost
identical performance. The reason is this commit also add the specific of
"--format=%(objectname)" to the current fast-pathes (builtin formats) to
avoid running unnecessary parsing mechanisms.

The following performance benchmarks are based on torvalds/linux.git:

  When hit the fast-path:

      Benchmark 1: /opt/git/ls-tree-oid-only/bin/git ls-tree -r --object-only HEAD
        Time (mean ± σ):      83.6 ms ±   2.0 ms    [User: 59.4 ms, System: 24.1 ms]
        Range (min … max):    80.4 ms …  87.2 ms    35 runs

      Benchmark 1: /opt/git/ls-tree-oid-only/bin/git ls-tree -r --format='%(objectname)' HEAD
        Time (mean ± σ):      84.1 ms ±   1.8 ms    [User: 61.7 ms, System: 22.3 ms]
        Range (min … max):    80.9 ms …  87.5 ms    35 runs

  But for a customized format, it will be slower:

       Benchmark 1: /opt/git/ls-tree-oid-only/bin/git ls-tree -r --format='oid: %(objectname)' HEAD
         Time (mean ± σ):      96.5 ms ±   2.5 ms    [User: 72.9 ms, System: 23.5 ms]
    Range (min … max):    93.1 ms … 104.1 ms    31 runs

Helped-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Teng Long <dyroneteng@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agols-tree: introduce "--format" option
Ævar Arnfjörð Bjarmason [Wed, 23 Mar 2022 09:13:12 +0000 (17:13 +0800)] 
ls-tree: introduce "--format" option

Add a --format option to ls-tree. It has an existing default output,
and then --long and --name-only options to emit the default output
along with the objectsize and, or to only emit object paths.

Rather than add --type-only, --object-only etc. we can just support a
--format using a strbuf_expand() similar to "for-each-ref
--format". We might still add such options in the future for
convenience.

The --format implementation is slower than the existing code, but this
change does not cause any performance regressions. We'll leave the
existing show_tree() unchanged, and only run show_tree_fmt() in if
a --format different than the hardcoded built-in ones corresponding to
the existing modes is provided.

I.e. something like the "--long" output would be much slower with
this, mainly due to how we need to allocate various things to do with
quote.c instead of spewing the output directly to stdout.

The new option of '--format' comes from Ævar Arnfjörð Bjarmasonn's
idea and suggestion, this commit makes modifications in terms of the
original discussion on community [1].

In [1] there was a "GIT_TEST_LS_TREE_FORMAT_BACKEND" variable to
ensure that we had test coverage for passing tests that would
otherwise use show_tree() through show_tree_fmt(), and thus that the
formatting mechanism could handle all the same cases as the
non-formatting options.

Somewhere in subsequent re-rolls of that we seem to have drifted away
from what the goal of these tests should be. We're trying to ensure
correctness of show_tree_fmt(). We can't tell if we "hit [the]
fast-path" here, and instead of having an explicit test for that, we
can just add it to something our "test_ls_tree_format" tests for.

Here is the statistics about performance tests:

1. Default format (hitten the builtin formats):

    "git ls-tree <tree-ish>" vs "--format='%(mode) %(type) %(object)%x09%(file)'"

    $hyperfine --warmup=10 "/opt/git/master/bin/git ls-tree -r HEAD"
    Benchmark 1: /opt/git/master/bin/git ls-tree -r HEAD
    Time (mean ± σ):     105.2 ms ±   3.3 ms    [User: 84.3 ms, System: 20.8 ms]
    Range (min … max):    99.2 ms … 113.2 ms    28 runs

    $hyperfine --warmup=10 "/opt/git/ls-tree-oid-only/bin/git ls-tree -r --format='%(mode) %(type) %(object)%x09%(file)'  HEAD"
    Benchmark 1: /opt/git/ls-tree-oid-only/bin/git ls-tree -r --format='%(mode) %(type) %(object)%x09%(file)'  HEAD
    Time (mean ± σ):     106.4 ms ±   2.7 ms    [User: 86.1 ms, System: 20.2 ms]
    Range (min … max):   100.2 ms … 110.5 ms    29 runs

2. Default format includes object size (hitten the builtin formats):

    "git ls-tree -l <tree-ish>" vs "--format='%(mode) %(type) %(object) %(size:padded)%x09%(file)'"

    $hyperfine --warmup=10 "/opt/git/master/bin/git ls-tree -r -l HEAD"
    Benchmark 1: /opt/git/master/bin/git ls-tree -r -l HEAD
    Time (mean ± σ):     335.1 ms ±   6.5 ms    [User: 304.6 ms, System: 30.4 ms]
    Range (min … max):   327.5 ms … 348.4 ms    10 runs

    $hyperfine --warmup=10 "/opt/git/ls-tree-oid-only/bin/git ls-tree -r --format='%(mode) %(type) %(object) %(size:padded)%x09%(file)'  HEAD"
    Benchmark 1: /opt/git/ls-tree-oid-only/bin/git ls-tree -r --format='%(mode) %(type) %(object) %(size:padded)%x09%(file)'  HEAD
    Time (mean ± σ):     337.2 ms ±   8.2 ms    [User: 309.2 ms, System: 27.9 ms]
    Range (min … max):   328.8 ms … 349.4 ms    10 runs

Links:
[1] https://public-inbox.org/git/RFC-patch-6.7-eac299f06ff-20211217T131635Z-avarab@gmail.com/
[2] https://lore.kernel.org/git/cb717d08be87e3239117c6c667cb32caabaad33d.1646390152.git.dyroneteng@gmail.com/

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Teng Long <dyroneteng@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agococci: allow padding with `strbuf_addf()`
Johannes Schindelin [Wed, 23 Mar 2022 09:13:11 +0000 (17:13 +0800)] 
cocci: allow padding with `strbuf_addf()`

A convenient way to pad strings is to use something like
`strbuf_addf(&buf, "%20s", "Hello, world!")`.

However, the Coccinelle rule that forbids a format `"%s"` with a
constant string argument cast too wide a net, and also forbade such
padding.

The original rule was introduced by commit:

    28c23cd4c39 (strbuf.cocci: suggest strbuf_addbuf() to add one strbuf to an other, 2019-01-25)

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Teng Long <dyroneteng@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agols-tree: introduce struct "show_tree_data"
Ævar Arnfjörð Bjarmason [Wed, 23 Mar 2022 09:13:10 +0000 (17:13 +0800)] 
ls-tree: introduce struct "show_tree_data"

"show_tree_data" is a struct that packages the necessary fields for
"show_tree()". This commit is a pre-prepared commit for supporting
"--format" option and it does not affect any existing functionality.

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Teng Long <dyroneteng@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agols-tree: slightly refactor `show_tree()`
Teng Long [Wed, 23 Mar 2022 09:13:09 +0000 (17:13 +0800)] 
ls-tree: slightly refactor `show_tree()`

This is a non-functional change, we introduce an enum "ls_tree_cmdmode"
then use it to mark which columns to output.

This has the advantage of making the show_tree logic simpler and more
readable, as well as making it easier to extend new options (for example,
if we want to add a "--object-only" option, we just need to add a similar
"short-circuit logic in "show_tree()").

Helped-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Teng Long <dyroneteng@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agols-tree: fix "--name-only" and "--long" combined use bug
Teng Long [Wed, 23 Mar 2022 09:13:08 +0000 (17:13 +0800)] 
ls-tree: fix "--name-only" and "--long" combined use bug

If we execute "git ls-tree" with combined "--name-only" and "--long"
, only the pathname will be printed, the size is omitted (the original
discoverer was Peff in [1]).

This commit fix this issue by using `OPT_CMDMODE()` instead to make both
of them mutually exclusive.

[1] https://public-inbox.org/git/YZK0MKCYAJmG+pSU@coredump.intra.peff.net/

Helped-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Teng Long <dyroneteng@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agols-tree: simplify nesting if/else logic in "show_tree()"
Teng Long [Wed, 23 Mar 2022 09:13:07 +0000 (17:13 +0800)] 
ls-tree: simplify nesting if/else logic in "show_tree()"

Use the object_type() function to determine the object type from the
"mode" passed to us by read_tree(), instead of doing so with the S_*()
macros.

Helped-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Teng Long <dyronetengb@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agols-tree: rename "retval" to "recurse" in "show_tree()"
Teng Long [Wed, 23 Mar 2022 09:13:06 +0000 (17:13 +0800)] 
ls-tree: rename "retval" to "recurse" in "show_tree()"

The variable which "show_tree()" return is named "retval", a name that's
a little hard to understand. The commit rename "retval" to "recurse"
which is a more meaningful name than before in the context. We do not
need to take a look at "read_tree_at()" in "tree.c" to make sure what
does "retval" mean.

Signed-off-by: Teng Long <dyroneteng@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agols-tree: use "size_t", not "int" for "struct strbuf"'s "len"
Ævar Arnfjörð Bjarmason [Wed, 23 Mar 2022 09:13:05 +0000 (17:13 +0800)] 
ls-tree: use "size_t", not "int" for "struct strbuf"'s "len"

The "struct strbuf"'s "len" member is a "size_t", not an "int", so
let's change our corresponding types accordingly. This also changes
the "len" and "speclen" variables, which are likewise used to store
the return value of strlen(), which returns "size_t", not "int".

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agols-tree: use "enum object_type", not {blob,tree,commit}_type
Ævar Arnfjörð Bjarmason [Wed, 23 Mar 2022 09:13:04 +0000 (17:13 +0800)] 
ls-tree: use "enum object_type", not {blob,tree,commit}_type

Change the ls-tree.c code to use type_name() on the enum instead of
using the string constants. This doesn't matter either way for
performance, but makes this a bit easier to read as we'll no longer
need a strcmp() here.

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agols-tree: add missing braces to "else" arms
Ævar Arnfjörð Bjarmason [Wed, 23 Mar 2022 09:13:03 +0000 (17:13 +0800)] 
ls-tree: add missing braces to "else" arms

Add missing {} to the "else" arms in show_tree() per the
CodingGuidelines.

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
3 years agols-tree: remove commented-out code
Ævar Arnfjörð Bjarmason [Wed, 23 Mar 2022 09:13:02 +0000 (17:13 +0800)] 
ls-tree: remove commented-out code

Remove code added in f35a6d3bce7 (Teach core object handling functions
about gitlinks, 2007-04-09), later patched in 7d0b18a4da1 (Add output
flushing before fork(), 2008-08-04), and then finally ending up in its
current form in d3bee161fef (tree.c: allow read_tree_recursive() to
traverse gitlink entries, 2009-01-25). All while being commented-out!

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>