git
5 years agoMerge branch 'ab/receive-pack-use-after-free-fix'
Junio C Hamano [Thu, 7 Mar 2019 00:59:57 +0000 (09:59 +0900)] 
Merge branch 'ab/receive-pack-use-after-free-fix'

Memfix.

* ab/receive-pack-use-after-free-fix:
  receive-pack: fix use-after-free bug

5 years agoMerge branch 'dl/doc-submodule-wo-subcommand'
Junio C Hamano [Thu, 7 Mar 2019 00:59:57 +0000 (09:59 +0900)] 
Merge branch 'dl/doc-submodule-wo-subcommand'

Doc update.

* dl/doc-submodule-wo-subcommand:
  submodule: document default behavior

5 years agoMerge branch 'jk/unused-params'
Junio C Hamano [Thu, 7 Mar 2019 00:59:56 +0000 (09:59 +0900)] 
Merge branch 'jk/unused-params'

Code clean-up.

* jk/unused-params:
  ref-filter: drop unused "sz" parameters
  ref-filter: drop unused "obj" parameters
  ref-filter: drop unused buf/sz pairs
  files-backend: drop refs parameter from split_symref_update()
  pack-objects: drop unused parameter from oe_map_new_pack()
  merge-recursive: drop several unused parameters
  diff: drop complete_rewrite parameter from run_external_diff()
  diff: drop unused emit data parameter from sane_truncate_line()
  diff: drop unused color reset parameters
  diff: drop options parameter from diffcore_fix_diff_index()

5 years agoMerge branch 'jk/prune-optim'
Junio C Hamano [Thu, 7 Mar 2019 00:59:56 +0000 (09:59 +0900)] 
Merge branch 'jk/prune-optim'

"git prune" has been taught to take advantage of reachability
bitmap when able.

* jk/prune-optim:
  t5304: rename "sha1" variables to "oid"
  prune: check SEEN flag for reachability
  prune: use bitmaps for reachability traversal
  prune: lazily perform reachability traversal

5 years agoMerge branch 'jh/trace2'
Junio C Hamano [Thu, 7 Mar 2019 00:59:56 +0000 (09:59 +0900)] 
Merge branch 'jh/trace2'

A more structured way to obtain execution trace has been added.

* jh/trace2:
  trace2: add for_each macros to clang-format
  trace2: t/helper/test-trace2, t0210.sh, t0211.sh, t0212.sh
  trace2:data: add subverb for rebase
  trace2:data: add subverb to reset command
  trace2:data: add subverb to checkout command
  trace2:data: pack-objects: add trace2 regions
  trace2:data: add trace2 instrumentation to index read/write
  trace2:data: add trace2 hook classification
  trace2:data: add trace2 transport child classification
  trace2:data: add trace2 sub-process classification
  trace2:data: add editor/pager child classification
  trace2:data: add trace2 regions to wt-status
  trace2: collect Windows-specific process information
  trace2: create new combined trace facility
  trace2: Documentation/technical/api-trace2.txt

5 years agoMerge branch 'js/doc-symref-in-proto-v1'
Junio C Hamano [Thu, 7 Mar 2019 00:59:56 +0000 (09:59 +0900)] 
Merge branch 'js/doc-symref-in-proto-v1'

Doc update.

* js/doc-symref-in-proto-v1:
  protocol-capabilities.txt: document symref

5 years agoMerge branch 'nd/split-index-null-base-fix'
Junio C Hamano [Thu, 7 Mar 2019 00:59:56 +0000 (09:59 +0900)] 
Merge branch 'nd/split-index-null-base-fix'

Split-index fix.

* nd/split-index-null-base-fix:
  read-cache.c: fix writing "link" index ext with null base oid

5 years agoMerge branch 'rj/prune-packed-excess-args'
Junio C Hamano [Thu, 7 Mar 2019 00:59:55 +0000 (09:59 +0900)] 
Merge branch 'rj/prune-packed-excess-args'

"git prune-packed" did not notice and complain against excess
arguments given from the command line, which now it does.

* rj/prune-packed-excess-args:
  prune-packed: check for too many arguments

5 years agoMerge branch 'jc/test-yes-doc'
Junio C Hamano [Thu, 7 Mar 2019 00:59:54 +0000 (09:59 +0900)] 
Merge branch 'jc/test-yes-doc'

Test doc update.

* jc/test-yes-doc:
  test: caution on our version of 'yes'

5 years agoMerge branch 'en/combined-all-paths'
Junio C Hamano [Thu, 7 Mar 2019 00:59:54 +0000 (09:59 +0900)] 
Merge branch 'en/combined-all-paths'

Output from "diff --cc" did not show the original paths when the
merge involved renames.  A new option adds the paths in the
original trees to the output.

* en/combined-all-paths:
  log,diff-tree: add --combined-all-paths option

5 years agoMerge branch 'sc/pack-redundant'
Junio C Hamano [Thu, 7 Mar 2019 00:59:54 +0000 (09:59 +0900)] 
Merge branch 'sc/pack-redundant'

Update the implementation of pack-redundant for performance in a
repository with many packfiles.

* sc/pack-redundant:
  pack-redundant: consistent sort method
  pack-redundant: rename pack_list.all_objects
  pack-redundant: new algorithm to find min packs
  pack-redundant: delete redundant code
  pack-redundant: delay creation of unique_objects
  t5323: test cases for git-pack-redundant

5 years agoMerge branch 'du/branch-show-current'
Junio C Hamano [Thu, 7 Mar 2019 00:59:53 +0000 (09:59 +0900)] 
Merge branch 'du/branch-show-current'

"git branch" learned a new subcommand "--show-current".

* du/branch-show-current:
  branch: introduce --show-current display option

5 years agoMerge branch 'dl/complete-submodule-absorbgitdirs'
Junio C Hamano [Thu, 7 Mar 2019 00:59:53 +0000 (09:59 +0900)] 
Merge branch 'dl/complete-submodule-absorbgitdirs'

Command-line completion (in contrib/) learned to tab-complete the
"git submodule absorbgitdirs" subcommand.

* dl/complete-submodule-absorbgitdirs:
  completion: complete git submodule absorbgitdirs

5 years agoMerge branch 'wh/author-committer-ident-config'
Junio C Hamano [Thu, 7 Mar 2019 00:59:53 +0000 (09:59 +0900)] 
Merge branch 'wh/author-committer-ident-config'

Four new configuration variables {author,committer}.{name,email}
have been introduced to override user.{name,email} in more specific
cases.

* wh/author-committer-ident-config:
  config: allow giving separate author and committer idents

5 years agoMerge branch 'aw/pretty-trailers'
Junio C Hamano [Thu, 7 Mar 2019 00:59:52 +0000 (09:59 +0900)] 
Merge branch 'aw/pretty-trailers'

The %(trailers) formatter in "git log --format=..."  now allows to
optionally pick trailers selectively by keyword, show only values,
etc.

* aw/pretty-trailers:
  pretty: add support for separator option in %(trailers)
  strbuf: separate callback for strbuf_expand:ing literals
  pretty: add support for "valueonly" option in %(trailers)
  pretty: allow showing specific trailers
  pretty: single return path in %(trailers) handling
  pretty: allow %(trailers) options with explicit value
  doc: group pretty-format.txt placeholders descriptions

5 years agoMerge branch 'nd/diff-parseopt'
Junio C Hamano [Thu, 7 Mar 2019 00:59:52 +0000 (09:59 +0900)] 
Merge branch 'nd/diff-parseopt'

The diff machinery, one of the oldest parts of the system, which
long predates the parse-options API, uses fairly long and complex
handcrafted option parser.  This is being rewritten to use the
parse-options API.

* nd/diff-parseopt:
  diff.c: convert --raw
  diff.c: convert -W|--[no-]function-context
  diff.c: convert -U|--unified
  diff.c: convert -u|-p|--patch
  diff.c: prepare to use parse_options() for parsing
  diff.h: avoid bit fields in struct diff_flags
  diff.h: keep forward struct declarations sorted
  parse-options: allow ll_callback with OPTION_CALLBACK
  parse-options: avoid magic return codes
  parse-options: stop abusing 'callback' for lowlevel callbacks
  parse-options: add OPT_BITOP()
  parse-options: disable option abbreviation with PARSE_OPT_KEEP_UNKNOWN
  parse-options: add one-shot mode
  parse-options.h: remove extern on function prototypes

5 years agoMerge branch 'tg/checkout-no-overlay'
Junio C Hamano [Thu, 7 Mar 2019 00:59:51 +0000 (09:59 +0900)] 
Merge branch 'tg/checkout-no-overlay'

"git checkout --no-overlay" can be used to trigger a new mode of
checking out paths out of the tree-ish, that allows paths that
match the pathspec that are in the current index and working tree
and are not in the tree-ish.

* tg/checkout-no-overlay:
  revert "checkout: introduce checkout.overlayMode config"
  checkout: introduce checkout.overlayMode config
  checkout: introduce --{,no-}overlay option
  checkout: factor out mark_cache_entry_for_checkout function
  checkout: clarify comment
  read-cache: add invalidate parameter to remove_marked_cache_entries
  entry: support CE_WT_REMOVE flag in checkout_entry
  entry: factor out unlink_entry function
  move worktree tests to t24*

5 years agoMerge tag 'l10n-2.21.0-rnd2.1' of git://github.com/git-l10n/git-po
Junio C Hamano [Tue, 5 Mar 2019 12:53:10 +0000 (21:53 +0900)] 
Merge tag 'l10n-2.21.0-rnd2.1' of git://github.com/git-l10n/git-po

L10n for Git 2.21.0 round 2.1

* tag 'l10n-2.21.0-rnd2.1' of git://github.com/git-l10n/git-po:
  l10n: Fixes to Catalan translation
  l10n: Updated Vietnamese translation for v2.21 rd2
  l10n: fr.po remove obsolete entries

5 years agol10n: Fixes to Catalan translation
Jordi Mas [Sat, 2 Mar 2019 18:12:58 +0000 (19:12 +0100)] 
l10n: Fixes to Catalan translation

Signed-off-by: Jordi Mas <jmas@softcatala.org>
5 years agol10n: Updated Vietnamese translation for v2.21 rd2
Tran Ngoc Quan [Tue, 26 Feb 2019 07:50:59 +0000 (14:50 +0700)] 
l10n: Updated Vietnamese translation for v2.21 rd2

Signed-off-by: Tran Ngoc Quan <vnwildman@gmail.com>
5 years agol10n: fr.po remove obsolete entries
Jean-Noël Avila [Mon, 25 Feb 2019 21:11:15 +0000 (22:11 +0100)] 
l10n: fr.po remove obsolete entries

On NetBSD, the version of msgfmt is still 0.14.4. There's no hope for
an upgrade due to some GPLv3 allergy of NetBSD's. This version chokes
on heavily decorated commented entries in po files. It's safer to get
rid of all these obsolete entries.

Signed-off-by: Jean-Noël Avila <jn.avila@free.fr>
5 years agoGit 2.21 v2.21.0
Junio C Hamano [Sun, 24 Feb 2019 15:55:19 +0000 (07:55 -0800)] 
Git 2.21

Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agoMerge branch 'yn/checkout-doc-fix'
Junio C Hamano [Sun, 24 Feb 2019 15:18:00 +0000 (07:18 -0800)] 
Merge branch 'yn/checkout-doc-fix'

Doc fix.

* yn/checkout-doc-fix:
  checkout doc: fix an unmatched double-quote pair

5 years agoMerge tag 'l10n-2.21.0-rnd2' of git://github.com/git-l10n/git-po
Junio C Hamano [Sun, 24 Feb 2019 15:03:39 +0000 (07:03 -0800)] 
Merge tag 'l10n-2.21.0-rnd2' of git://github.com/git-l10n/git-po

l10n-2.21.0-rnd2

* tag 'l10n-2.21.0-rnd2' of git://github.com/git-l10n/git-po:
  l10n: bg.po: Updated Bulgarian translation (4363t)
  l10n: update German translation
  l10n: zh_CN: Revision for git v2.21.0 l10n
  l10n: zh_CN: for git v2.21.0 l10n round 1~2
  l10n: bg.po: correct typo
  l10n: Update Swedish translation (4363t0f0u)
  l10n: de.po: fix grammar in message for tag.c
  l10n: de.po: fix a message for index-pack.c
  l10n: de.po: consistent translation of 'root commit'
  l10n: it: update the Italian translation
  l10n: es: 2.21.0 round 2
  l10n: el: add Greek l10n team and essential translations
  l10n: fr.po v2.21.0 rnd 2
  l10n: fr.po Fix some typos from round3
  l10n: fr.po Fix some typos
  l10n: Fixes to Catalan translation
  l10n: git.pot: v2.21.0 round 2 (3 new, 3 removed)
  l10n: git.pot: v2.21.0 round 1 (214 new, 38 removed)
  l10n: zh_CN: fix typo of submodule init message
  l10n: Update Catalan translation

5 years agoREADME: adjust for final Azure Pipeline ID
Johannes Schindelin [Sat, 23 Feb 2019 14:49:23 +0000 (06:49 -0800)] 
README: adjust for final Azure Pipeline ID

During the six months of development of the Azure Pipelines support, the
patches went through quite a few iterations of changes, and to test
those iterations, a temporary build definition was used.

In the meantime, Azure Pipelines support made it to `master`, and we now
have a regular Azure Pipeline, installed via the common GitHub App
workflow. This new pipeline has a different name (git.git instead of
test-git.git), and a new ID (11 instead of 2).

Let's adjust the badge in our README to reflect that final shape of the
Azure Pipeline.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agocheckout doc: fix an unmatched double-quote pair
Yoichi Nakayama [Sat, 23 Feb 2019 06:33:40 +0000 (15:33 +0900)] 
checkout doc: fix an unmatched double-quote pair

Signed-off-by: Yoichi Nakayama <yoichi.nakayama@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agol10n: bg.po: Updated Bulgarian translation (4363t)
Alexander Shopov [Sat, 23 Feb 2019 16:39:07 +0000 (18:39 +0200)] 
l10n: bg.po: Updated Bulgarian translation (4363t)

Signed-off-by: Alexander Shopov <ash@kambanaria.org>
5 years agoMerge branch 'ab/bsd-fixes'
Junio C Hamano [Sat, 23 Feb 2019 05:20:19 +0000 (21:20 -0800)] 
Merge branch 'ab/bsd-fixes'

Test portability fix.

* ab/bsd-fixes:
  commit-graph tests: fix unportable "dd" invocation
  tests: fix unportable "\?" and "\+" regex syntax

5 years agoMerge branch 'ab/workaround-dash-bug-in-test'
Junio C Hamano [Sat, 23 Feb 2019 05:20:19 +0000 (21:20 -0800)] 
Merge branch 'ab/workaround-dash-bug-in-test'

* ab/workaround-dash-bug-in-test:
  tests: avoid syntax triggering old dash bug

5 years agotrace2: add for_each macros to clang-format
Jeff Hostetler [Fri, 22 Feb 2019 22:25:11 +0000 (14:25 -0800)] 
trace2: add for_each macros to clang-format

Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agotrace2: t/helper/test-trace2, t0210.sh, t0211.sh, t0212.sh
Jeff Hostetler [Fri, 22 Feb 2019 22:25:10 +0000 (14:25 -0800)] 
trace2: t/helper/test-trace2, t0210.sh, t0211.sh, t0212.sh

Create unit tests for Trace2.

Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agotrace2:data: add subverb for rebase
Jeff Hostetler [Fri, 22 Feb 2019 22:25:10 +0000 (14:25 -0800)] 
trace2:data: add subverb for rebase

Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agotrace2:data: add subverb to reset command
Jeff Hostetler [Fri, 22 Feb 2019 22:25:09 +0000 (14:25 -0800)] 
trace2:data: add subverb to reset command

Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agotrace2:data: add subverb to checkout command
Jeff Hostetler [Fri, 22 Feb 2019 22:25:08 +0000 (14:25 -0800)] 
trace2:data: add subverb to checkout command

Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agotrace2:data: pack-objects: add trace2 regions
Derrick Stolee [Fri, 22 Feb 2019 22:25:07 +0000 (14:25 -0800)] 
trace2:data: pack-objects: add trace2 regions

When studying the performance of 'git push' we would like to know
how much time is spent at various parts of the command. One area
that could cause performance trouble is 'git pack-objects'.

Add trace2 regions around the three main actions taken in this
command:

1. Enumerate objects.
2. Prepare pack.
3. Write pack-file.

Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agotrace2:data: add trace2 instrumentation to index read/write
Jeff Hostetler [Fri, 22 Feb 2019 22:25:07 +0000 (14:25 -0800)] 
trace2:data: add trace2 instrumentation to index read/write

Add trace2 events to measure reading and writing the index.

Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agotrace2:data: add trace2 hook classification
Jeff Hostetler [Fri, 22 Feb 2019 22:25:06 +0000 (14:25 -0800)] 
trace2:data: add trace2 hook classification

Classify certain child processes as hooks.

Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agotrace2:data: add trace2 transport child classification
Jeff Hostetler [Fri, 22 Feb 2019 22:25:05 +0000 (14:25 -0800)] 
trace2:data: add trace2 transport child classification

Add trace2 child classification for transport processes.

Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agotrace2:data: add trace2 sub-process classification
Jeff Hostetler [Fri, 22 Feb 2019 22:25:05 +0000 (14:25 -0800)] 
trace2:data: add trace2 sub-process classification

Add trace2 classification for long-running processes
started in sub-process.c

Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agotrace2:data: add editor/pager child classification
Jeff Hostetler [Fri, 22 Feb 2019 22:25:04 +0000 (14:25 -0800)] 
trace2:data: add editor/pager child classification

Add trace2 process classification for editor and pager
child processes.

Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agotrace2:data: add trace2 regions to wt-status
Jeff Hostetler [Fri, 22 Feb 2019 22:25:03 +0000 (14:25 -0800)] 
trace2:data: add trace2 regions to wt-status

Add trace2_region_enter() and trace2_region_leave() calls around the
various phases of a status scan.  This gives elapsed time for each
phase in the GIT_TR2_PERF and GIT_TR2_EVENT trace target.

Also, these Trace2 calls now use s->repo rather than the_repository.

Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agotrace2: collect Windows-specific process information
Jeff Hostetler [Fri, 22 Feb 2019 22:25:02 +0000 (14:25 -0800)] 
trace2: collect Windows-specific process information

Add platform-specific interface to log information about the current
process.

On Windows, this interface is used to indicate whether the git process
is running under a debugger and list names of the process ancestors.

Information for other platforms is left for a future effort.

Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agotrace2: create new combined trace facility
Jeff Hostetler [Fri, 22 Feb 2019 22:25:01 +0000 (14:25 -0800)] 
trace2: create new combined trace facility

Create a new unified tracing facility for git.  The eventual intent is to
replace the current trace_printf* and trace_performance* routines with a
unified set of git_trace2* routines.

In addition to the usual printf-style API, trace2 provides higer-level
event verbs with fixed-fields allowing structured data to be written.
This makes post-processing and analysis easier for external tools.

Trace2 defines 3 output targets.  These are set using the environment
variables "GIT_TR2", "GIT_TR2_PERF", and "GIT_TR2_EVENT".  These may be
set to "1" or to an absolute pathname (just like the current GIT_TRACE).

* GIT_TR2 is intended to be a replacement for GIT_TRACE and logs command
  summary data.

* GIT_TR2_PERF is intended as a replacement for GIT_TRACE_PERFORMANCE.
  It extends the output with columns for the command process, thread,
  repo, absolute and relative elapsed times.  It reports events for
  child process start/stop, thread start/stop, and per-thread function
  nesting.

* GIT_TR2_EVENT is a new structured format. It writes event data as a
  series of JSON records.

Calls to trace2 functions log to any of the 3 output targets enabled
without the need to call different trace_printf* or trace_performance*
routines.

Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agotrace2: Documentation/technical/api-trace2.txt
Jeff Hostetler [Fri, 22 Feb 2019 22:25:00 +0000 (14:25 -0800)] 
trace2: Documentation/technical/api-trace2.txt

Created design document for Trace2 feature.

Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agocommit-graph tests: fix unportable "dd" invocation
Ævar Arnfjörð Bjarmason [Thu, 21 Feb 2019 19:28:49 +0000 (20:28 +0100)] 
commit-graph tests: fix unportable "dd" invocation

Change an unportable invocation of "dd" with count=0, that wanted to
truncate the commit-graph file.  In POSIX it is unspecified what
happens when count=0 is provided[1]. The NetBSD "dd" behavior
differs from GNU (and seemingly other BSDs), which has left this test
broken since d2b86fbaa1 ("commit-graph: fix buffer read-overflow",
2019-01-15).

Copying from /dev/null would seek/truncate to seek=$zero_pos and
stop immediately after that (without being able to copy anything),
which is the right way to truncate the file.

1. http://pubs.opengroup.org/onlinepubs/9699919799/utilities/dd.html

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Helped-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agoMerge branch 'master' of https://github.com/ralfth/git-po-de
Jiang Xin [Fri, 22 Feb 2019 14:18:12 +0000 (22:18 +0800)] 
Merge branch 'master' of https://github.com/ralfth/git-po-de

5 years agol10n: update German translation
Ralf Thielow [Fri, 22 Feb 2019 07:45:40 +0000 (08:45 +0100)] 
l10n: update German translation

Signed-off-by: Ralf Thielow <ralf.thielow@gmail.com>
Reviewed-by: Matthias Rüster <matthias.ruester@gmail.com>
5 years agotests: fix unportable "\?" and "\+" regex syntax
Ævar Arnfjörð Bjarmason [Thu, 21 Feb 2019 19:28:48 +0000 (20:28 +0100)] 
tests: fix unportable "\?" and "\+" regex syntax

Fix widely supported but non-POSIX basic regex syntax introduced in
[1] and [2]. On GNU, NetBSD and FreeBSD the following works:

    $ echo xy >f
    $ grep 'xy\?' f; echo $?
    xy
    0

The same goes for "\+". The "?" and "+" syntax is not in the BRE
syntax, just in ERE, but on some implementations it can be invoked by
prefixing the meta-operator with "\", but not on OpenBSD:

    $ uname -a
    OpenBSD obsd.my.domain 6.2 GENERIC#132 amd64
    $ grep --version
    grep version 0.9
    $ grep 'xy\?' f; echo $?
    1

Let's fix this by moving to ERE syntax instead, where "?" and "+" are
universally supported:

    $ grep -E 'xy?' f; echo $?
    xy
    0

1. 2ed5c8e174 ("describe: setup working tree for --dirty", 2019-02-03)
2. c801170b0c ("t6120: test for describe with a bare repository",
   2019-02-03)

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agoprotocol-capabilities.txt: document symref
Josh Steadmon [Wed, 20 Feb 2019 00:32:26 +0000 (16:32 -0800)] 
protocol-capabilities.txt: document symref

In 7171d8c15f ("upload-pack: send symbolic ref information as
capability"), we added a symref capability to the pack protocol, but it
was never documented. Adapt the patch notes from that commit and add
them to the capabilities documentation.

While we're at it, add a disclaimer to the top of
protocol-capabilities.txt noting that the doc only applies to v0/v1 of
the wire protocol.

Signed-off-by: Josh Steadmon <steadmon@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agoreceive-pack: fix use-after-free bug
Ævar Arnfjörð Bjarmason [Wed, 20 Feb 2019 00:00:33 +0000 (01:00 +0100)] 
receive-pack: fix use-after-free bug

The resolve_ref_unsafe() function can, and sometimes will in the case
of this codepath, return the char * passed to it to the caller. In
this case we construct a strbuf, free it, and then continue using the
dst_name after that free().

The code being fixed dates back to da3efdb17b ("receive-pack: detect
aliased updates which can occur with symrefs", 2010-04-19). When it
was originally added it didn't have this bug, it was introduced when
it was subsequently modified to use strbuf in 6b01ecfe22 ("ref
namespaces: Support remote repositories via upload-pack and
receive-pack", 2011-07-08).

This is theoretically a security issue, the C standard makes no
guarantees that a value you use after free() hasn't been poked at or
changed by something else on the system, but in practice modern OSs
will have mapped the relevant page to this process, so nothing else
would have used it. We do no further allocations between the free()
and use-after-free, so we ourselves didn't corrupt or change the
value.

Jeff investigated that and found: "It probably would be an issue if
the allocation were larger. glibc at least will use mmap()/munmap()
after some cutoff[1], in which case we'd get a segfault from hitting
the unmapped page. But for small allocations, it just bumps brk() and
the memory is still available for further allocations after
free(). [...] If you had a sufficiently large refname you might be
able to trigger the bug [...]. I tried to push such a ref. I had to
manually make a packed-refs file with the long name to avoid
filesystem limits (though probably you could have a long a/b/c/ name
on ext4).  But the result can't actually be pushed, because it all has
to fit into a 64k pkt-line as part of the push protocol.".

An a alternative and more succinct way of implementing this would have
been to do the strbuf_release() at the end of check_aliased_update()
and use "goto out" instead of the early "return" statements. Hopefully
this approach of using a helper instead makes it easier to follow.

1. Jeff: "Weirdly, the mmap() cutoff on my glibc system is 135168
   bytes. Which is...2^17 + 2^12? 33 pages? I'm sure there's a good
   reason for that, but I didn't dig into it."

Reported-by: 王健强 <jianqiang.wang@securitygossip.com>
Helped-by: Jeff King <peff@peff.net>
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agoMerge branch 'bg-submodule-helper-typo' of github.com:pclouds/git-po
Jiang Xin [Wed, 20 Feb 2019 13:40:54 +0000 (21:40 +0800)] 
Merge branch 'bg-submodule-helper-typo' of github.com:pclouds/git-po

5 years agol10n: zh_CN: Revision for git v2.21.0 l10n
Fangyi Zhou [Mon, 11 Feb 2019 11:28:55 +0000 (11:28 +0000)] 
l10n: zh_CN: Revision for git v2.21.0 l10n

Signed-off-by: Fangyi Zhou <fangyi.zhou@yuriko.moe>
5 years agol10n: zh_CN: for git v2.21.0 l10n round 1~2
Jiang Xin [Tue, 20 Nov 2018 02:18:06 +0000 (10:18 +0800)] 
l10n: zh_CN: for git v2.21.0 l10n round 1~2

Translate 214 new messages (4363t0f0u) for git 2.21.0.

Signed-off-by: Jiang Xin <worldhello.net@gmail.com>
5 years agol10n: bg.po: correct typo
Nguyễn Thái Ngọc Duy [Wed, 20 Feb 2019 09:59:24 +0000 (16:59 +0700)] 
l10n: bg.po: correct typo

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
5 years agol10n: Update Swedish translation (4363t0f0u)
Peter Krefting [Mon, 11 Feb 2019 21:34:05 +0000 (22:34 +0100)] 
l10n: Update Swedish translation (4363t0f0u)

Signed-off-by: Peter Krefting <peter@softwolves.pp.se>
5 years agoGit 2.21-rc2 v2.21.0-rc2
Junio C Hamano [Tue, 19 Feb 2019 21:20:23 +0000 (13:20 -0800)] 
Git 2.21-rc2

Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agoMerge branch 'js/test-tool-gen-nuls'
Junio C Hamano [Tue, 19 Feb 2019 21:18:08 +0000 (13:18 -0800)] 
Merge branch 'js/test-tool-gen-nuls'

* js/test-tool-gen-nuls:
  tests: teach the test-tool to generate NUL bytes and use it

5 years agoMerge branch 'mk/t5562-no-input-to-too-large-an-input-test'
Junio C Hamano [Tue, 19 Feb 2019 21:18:08 +0000 (13:18 -0800)] 
Merge branch 'mk/t5562-no-input-to-too-large-an-input-test'

* mk/t5562-no-input-to-too-large-an-input-test:
  t5562: do not depend on /dev/zero
  Revert "t5562: replace /dev/zero with a pipe from generate_zero_bytes"

5 years agoMerge branch 'mk/t5562-do-not-reuse-output-files'
Junio C Hamano [Tue, 19 Feb 2019 21:18:08 +0000 (13:18 -0800)] 
Merge branch 'mk/t5562-do-not-reuse-output-files'

* mk/t5562-do-not-reuse-output-files:
  t5562: do not reuse output files

5 years agot5562: do not reuse output files
Max Kirillov [Sat, 24 Nov 2018 09:37:19 +0000 (11:37 +0200)] 
t5562: do not reuse output files

Some expected failures of git-http-backend leaves running its children
(receive-pack or upload-pack) which still hold opened descriptors
to act.err and with some probability they live long enough to write
there their failure messages after next test has already truncated
the files. This causes occasional failures of the test script.

Avoid the issue by using separated output and error file for each test,
apprending the test number to their name.

Reported-by: Carlo Arenas <carenas@gmail.com>
Helped-by: Carlo Arenas <carenas@gmail.com>
Helped-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Max Kirillov <max@max630.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agotests: teach the test-tool to generate NUL bytes and use it
Johannes Schindelin [Thu, 14 Feb 2019 21:33:12 +0000 (13:33 -0800)] 
tests: teach the test-tool to generate NUL bytes and use it

In cc95bc2025 (t5562: replace /dev/zero with a pipe from
generate_zero_bytes, 2019-02-09), we replaced usage of /dev/zero (which
is not available on NonStop, apparently) by a Perl script snippet to
generate NUL bytes.

Sadly, it does not seem to work on NonStop, as t5562 reportedly hangs.

Worse, this also hangs in the Ubuntu 16.04 agents of the CI builds on
Azure Pipelines: for some reason, the Perl script snippet that is run
via `generate_zero_bytes` in t5562's 'CONTENT_LENGTH overflow ssite_t'
test case tries to write out an infinite amount of NUL bytes unless a
broken pipe is encountered, that snippet never encounters the broken
pipe, and keeps going until the build times out.

Oddly enough, this does not reproduce on the Windows and macOS agents,
nor in a local Ubuntu 18.04.

This developer tried for a day to figure out the exact circumstances
under which this hang happens, to no avail, the details remain a
mystery.

In the end, though, what counts is that this here change incidentally
fixes that hang (maybe also on NonStop?). Even more positively, it gets
rid of yet another unnecessary Perl invocation.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agot5562: do not depend on /dev/zero
Max Kirillov [Fri, 15 Feb 2019 16:42:37 +0000 (18:42 +0200)] 
t5562: do not depend on /dev/zero

It was reported [1] that NonStop platform does not have /dev/zero.

The test uses /dev/zero as a dummy input. Passing case (http-backed
failed because of too big input size) should not be reading anything
from it. If http-backend would erroneously try to read any data
returning EOF probably would be even safer than providing some
meaningless data.

Replace /dev/zero with /dev/null to avoid issues with platforms which do
not have /dev/zero.

[1] https://public-inbox.org/git/20190209185930.5256-4-randall.s.becker@rogers.com/

Reported-by: Randall S. Becker <rsbecker@nexbridge.com>
Signed-off-by: Max Kirillov <max@max630.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agoRevert "t5562: replace /dev/zero with a pipe from generate_zero_bytes"
Junio C Hamano [Tue, 19 Feb 2019 18:18:15 +0000 (10:18 -0800)] 
Revert "t5562: replace /dev/zero with a pipe from generate_zero_bytes"

Revert cc95bc20 ("t5562: replace /dev/zero with a pipe from
generate_zero_bytes", 2019-02-09), as not feeding anything to the
command is a better way to test it.

5 years agol10n: de.po: fix grammar in message for tag.c
Sebastian Staudt [Sun, 3 Feb 2019 10:33:39 +0000 (11:33 +0100)] 
l10n: de.po: fix grammar in message for tag.c

Signed-off-by: Sebastian Staudt <koraktor@gmail.com>
5 years agol10n: de.po: fix a message for index-pack.c
Sebastian Staudt [Sun, 3 Feb 2019 10:32:20 +0000 (11:32 +0100)] 
l10n: de.po: fix a message for index-pack.c

Signed-off-by: Sebastian Staudt <koraktor@gmail.com>
5 years agol10n: de.po: consistent translation of 'root commit'
Sebastian Staudt [Sun, 3 Feb 2019 10:21:14 +0000 (11:21 +0100)] 
l10n: de.po: consistent translation of 'root commit'

'root commit' is usually translated as 'Root-Commit'. But in one
occasion it‘s translated as 'Basis-Commit' which is the translation
for 'base commit'.

Signed-off-by: Sebastian Staudt <koraktor@gmail.com>
5 years agol10n: it: update the Italian translation
Alessandro Menti [Sun, 17 Feb 2019 17:04:51 +0000 (18:04 +0100)] 
l10n: it: update the Italian translation

Signed-off-by: Alessandro Menti <alessandro.menti@alessandromenti.it>
5 years agoMerge branch 'master' of https://github.com/Softcatala/git-po
Jiang Xin [Sun, 17 Feb 2019 07:28:15 +0000 (15:28 +0800)] 
Merge branch 'master' of https://github.com/Softcatala/git-po

5 years agol10n: es: 2.21.0 round 2
Christopher Diaz Riveros [Wed, 6 Feb 2019 13:36:53 +0000 (08:36 -0500)] 
l10n: es: 2.21.0 round 2

Signed-off-by: Christopher Diaz Riveros <chrisadr@gentoo.org>
5 years agoMerge branch 'fr_2.21.0_rnd2' of git://github.com/jnavila/git
Jiang Xin [Sat, 16 Feb 2019 01:54:53 +0000 (09:54 +0800)] 
Merge branch 'fr_2.21.0_rnd2' of git://github.com/jnavila/git

5 years agol10n: el: add Greek l10n team and essential translations
Jimmy Angelakos [Fri, 15 Feb 2019 02:28:19 +0000 (02:28 +0000)] 
l10n: el: add Greek l10n team and essential translations

Signed-off-by: Jimmy Angelakos <vyruss@hellug.gr>
5 years agol10n: fr.po v2.21.0 rnd 2
Jean-Noël Avila [Wed, 13 Feb 2019 19:35:13 +0000 (20:35 +0100)] 
l10n: fr.po v2.21.0 rnd 2

Signed-off-by: Jean-Noël Avila <jn.avila@free.fr>
5 years agol10n: fr.po Fix some typos from round3
Fabien Villepinte [Thu, 20 Dec 2018 08:59:45 +0000 (09:59 +0100)] 
l10n: fr.po Fix some typos from round3

Signed-off-by: Fabien Villepinte <fabien.villepinte@gmail.com>
5 years agol10n: fr.po Fix some typos
Fabien Villepinte [Mon, 17 Dec 2018 11:51:58 +0000 (12:51 +0100)] 
l10n: fr.po Fix some typos

Signed-off-by: Fabien Villepinte <fabien.villepinte@gmail.com>
5 years agomingw: safe-guard a bit more against getenv() problems
Johannes Schindelin [Fri, 15 Feb 2019 15:17:45 +0000 (07:17 -0800)] 
mingw: safe-guard a bit more against getenv() problems

Running up to v2.21.0, we fixed two bugs that were made prominent by the
Windows-specific change to retain copies of only the 30 latest getenv()
calls' returned strings, invalidating any copies of previous getenv()
calls' return values.

While this really shines a light onto bugs of the form where we hold
onto getenv()'s return values without copying them, it is also a real
problem for users.

And even if Jeff King's patches merged via 773e408881 (Merge branch
'jk/save-getenv-result', 2019-01-29) provide further work on that front,
we are far from done. Just one example: on Windows, we unset environment
variables when spawning new processes, which potentially invalidates
strings that were previously obtained via getenv(), and therefore we
have to duplicate environment values that are somehow involved in
spawning new processes (e.g. GIT_MAN_VIEWER in show_man_page()).

We do not have a chance to investigate, let address, all of those issues
in time for v2.21.0, so let's at least help Windows users by increasing
the number of getenv() calls' return values that are kept valid. The
number 64 was determined by looking at the average number of getenv()
calls per process in the entire test suite run on Windows (which is
around 40) and then adding a bit for good measure. And it is a power of
two (which would have hit yesterday's theme perfectly).

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agosubmodule: document default behavior
Denton Liu [Fri, 15 Feb 2019 09:26:41 +0000 (01:26 -0800)] 
submodule: document default behavior

submodule's default behavior wasn't documented in both git-submodule.txt
and in the usage text of git-submodule. Document the default behavior
similar to how git-remote does it.

Signed-off-by: Denton Liu <liu.denton@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agol10n: Fixes to Catalan translation
Jordi Mas [Fri, 15 Feb 2019 15:43:59 +0000 (16:43 +0100)] 
l10n: Fixes to Catalan translation

Signed-off-by: Jordi Mas <jmas@softcatala.org>
5 years agol10n: git.pot: v2.21.0 round 2 (3 new, 3 removed)
Jiang Xin [Fri, 15 Feb 2019 02:12:14 +0000 (10:12 +0800)] 
l10n: git.pot: v2.21.0 round 2 (3 new, 3 removed)

Introduce 3 update messages for v2.21.0 l10n round 2 from commit
32ceace39f (Fix typos in translatable strings for v2.21.0, 2019-02-11).

Signed-off-by: Jiang Xin <worldhello.net@gmail.com>
5 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/git/git
Jiang Xin [Fri, 15 Feb 2019 02:09:07 +0000 (10:09 +0800)] 
Merge branch 'master' of git://git./git/git

5 years agoref-filter: drop unused "sz" parameters
Jeff King [Thu, 14 Feb 2019 05:51:03 +0000 (00:51 -0500)] 
ref-filter: drop unused "sz" parameters

Many of our grab_* functions, which parse the object content, take a
buf/sz pair of the object bytes. However, the functions which actually
parse the buffers (like find_wholine() and find_subpos()) never look at
"sz", and instead use functions like strchr() and strchrnul() that
assume the result is NUL-terminated.

This is OK in practice (and common for Git's parsing code), since we
always allocate an extra NUL when loading an object into memory (and
likewise, we are OK with stopping parsing if a commit or tag contains an
embedded NUL).

Let's drop these extra "sz" parameters, as they are misleading about how
the functions intend to access the buffer. We can drop from both the
functions mentioned above, which in turn lets us drop from their
callers, cascading all the way up to the top-level grab_values().

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agoref-filter: drop unused "obj" parameters
Jeff King [Thu, 14 Feb 2019 05:50:58 +0000 (00:50 -0500)] 
ref-filter: drop unused "obj" parameters

The grab_person() and grab_sub_body_contents() functions take both an
object struct and a buf/sz pair of the object bytes. However, they use
only the latter, since "struct object" does not contain the parsed ident
(nor the whole commit message, of course).

Let's get rid of these misleading "struct object" parameters. It's
possible we may want them in the future (e.g., to generate error
messages that mention the object id), but since these are static
functions, we can easily add them back in later (and if we do want that
information, it's likely we'd pass it through a more generalized
"parsing context" struct anyway).

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agoref-filter: drop unused buf/sz pairs
Jeff King [Thu, 14 Feb 2019 05:50:54 +0000 (00:50 -0500)] 
ref-filter: drop unused buf/sz pairs

The grab_tag_values() and grab_commit_values() functions take both the
"struct object" as well as the buf/sz pair for the actual object bytes.
However, neither function uses the latter, as they pull the data
directly from the parsed object struct.

Let's get rid of these misleading parameters.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agofiles-backend: drop refs parameter from split_symref_update()
Jeff King [Thu, 14 Feb 2019 05:50:46 +0000 (00:50 -0500)] 
files-backend: drop refs parameter from split_symref_update()

This parameter was added in fcc42ea0c9 (split_symref_update(): add a
files_ref_store argument, 2016-09-04) without comment, but never used.
The splitting is purely mechanical, and doesn't depend on the particular
ref-store. Let's drop this parameter in the name of simplicity.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agopack-objects: drop unused parameter from oe_map_new_pack()
Jeff King [Thu, 14 Feb 2019 05:50:32 +0000 (00:50 -0500)] 
pack-objects: drop unused parameter from oe_map_new_pack()

Since 43fa44fa3b (pack-objects: move in_pack out of struct object_entry,
2018-04-14), we store the source pack for each object as a small index
rather than as a pointer. When we see a new pack that has no allocated
index, we fall back to generating an array of pointers by calling
oe_map_new_pack().

Perhaps counter-intuitively, that function does not need to actually see
our new index-less pack. It only allocates and populates the array with
the existing packs, after which oe_set_in_pack() actually adds the new
pack to the array.

Let's drop the unused "struct packed_git" argument to oe_map_new_pack()
to avoid confusion.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agomerge-recursive: drop several unused parameters
Jeff King [Thu, 14 Feb 2019 05:50:02 +0000 (00:50 -0500)] 
merge-recursive: drop several unused parameters

There are a few functions related to directory renames that have unused
parameters. After consulting with the author in [1], these seem to be
leftover cruft from the development process, and not signs of any bug.
Let's drop them.

[1] https://public-inbox.org/git/CABPp-BHobf8wbBsXF97scNQCzkxQukziODRXq6JOOWq61cAd9g@mail.gmail.com/

Helped-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agodiff: drop complete_rewrite parameter from run_external_diff()
Jeff King [Thu, 14 Feb 2019 05:49:42 +0000 (00:49 -0500)] 
diff: drop complete_rewrite parameter from run_external_diff()

Our builtin_diff() wants to know whether break-detection found a
complete rewrite, because it changes how the diff is shown. However,
when calling out to an external diff, we don't pass this information
along (and doing so would require designing a new interface to the
user-provided program).

Let's drop the unused parameter to make this fact more clear.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agodiff: drop unused emit data parameter from sane_truncate_line()
Jeff King [Thu, 14 Feb 2019 05:48:27 +0000 (00:48 -0500)] 
diff: drop unused emit data parameter from sane_truncate_line()

We pass the "struct emit_callback" (which contains all of the context
for our diff) into sane_truncate_line(), but that function doesn't
actually use it. In theory we might eventually develop a diff option
that impacts this, but in the meantime let's not mislead anybody reading
the code. Since the function is static, it would be easy to pass it
again if it should ever become useful.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agodiff: drop unused color reset parameters
Jeff King [Thu, 14 Feb 2019 05:48:13 +0000 (00:48 -0500)] 
diff: drop unused color reset parameters

Several of the emit_* functions take a "reset" color parameter, but
never actually look at it (instead, they call into emit_diff_symbol,
which handles the colors itself). Let's drop these unused parameters.

Note that emit_line() does still take a color/reset pair, and actually
uses it. It cannot be refactored to match these other functions because
it's the thing that emit_diff_symbol eventually calls into (i.e., it
does not by itself know which colors to use, and must be told by the
caller).

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agodiff: drop options parameter from diffcore_fix_diff_index()
Jeff King [Thu, 14 Feb 2019 05:48:03 +0000 (00:48 -0500)] 
diff: drop options parameter from diffcore_fix_diff_index()

The sole purpose of this function is to fix the sorting order of the
queued diff entries. It doesn't need to know about any diff options, so
we can drop the unused parameter.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agot5304: rename "sha1" variables to "oid"
Jeff King [Thu, 14 Feb 2019 11:07:36 +0000 (06:07 -0500)] 
t5304: rename "sha1" variables to "oid"

Let's make the script less jarring to read in a post-sha1 world by
using more hash-agnostic variable names.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agoprune: check SEEN flag for reachability
Jeff King [Thu, 14 Feb 2019 04:38:21 +0000 (23:38 -0500)] 
prune: check SEEN flag for reachability

The git-prune command checks reachability by doing a traversal, and then
checking whether a given object exists in the global object hash. This
can yield false positives if any other part of the code had to create an
object struct for some reason. It's not clear whether this is even
possible, but it's more robust to rely on something a little more
concrete: the SEEN flag set by our traversal.

Note that there is a slight possibility of regression here, as we're
relying on mark_reachable_objects() to consistently set the flag.
However, it has always done so, and we're already relying on that fact
in prune_shallow(), which is called as part of git-prune. So this is
making these two parts of the prune operation more consistent.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agoprune: use bitmaps for reachability traversal
Jeff King [Thu, 14 Feb 2019 04:37:43 +0000 (23:37 -0500)] 
prune: use bitmaps for reachability traversal

Pruning generally has to traverse the whole commit graph in order to
see which objects are reachable. This is the exact problem that
reachability bitmaps were meant to solve, so let's use them (if they're
available, of course).

Here are timings on git.git:

  Test                            HEAD^             HEAD
  ------------------------------------------------------------------------
  5304.6: prune with bitmaps      3.65(3.56+0.09)   1.01(0.92+0.08) -72.3%

And on linux.git:

  Test                            HEAD^               HEAD
  --------------------------------------------------------------------------
  5304.6: prune with bitmaps      35.05(34.79+0.23)   3.00(2.78+0.21) -91.4%

The tests show a pretty optimal case, as we'll have just repacked and
should have pretty good coverage of all refs with our bitmaps. But
that's actually pretty realistic: normally prune is run via "gc" right
after repacking.

A few notes on the implementation:

  - the change is actually in reachable.c, so it would improve
    reachability traversals by "reflog expire --stale-fix", as well.
    Those aren't performed regularly, though (a normal "git gc" doesn't
    use --stale-fix), so they're not really worth measuring. There's a
    low chance of regressing that caller, since the use of bitmaps is
    totally transparent from the caller's perspective.

  - The bitmap case could actually get away without creating a "struct
    object", and instead the caller could just look up each object id in
    the bitmap result. However, this would be a marginal improvement in
    runtime, and it would make the callers much more complicated. They'd
    have to handle both the bitmap and non-bitmap cases separately, and
    in the case of git-prune, we'd also have to tweak prune_shallow(),
    which relies on our SEEN flags.

  - Because we do create real object structs, we go through a few
    contortions to create ones of the right type. This isn't strictly
    necessary (lookup_unknown_object() would suffice), but it's more
    memory efficient to use the correct types, since we already know
    them.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agoprune: lazily perform reachability traversal
Jeff King [Thu, 14 Feb 2019 04:35:22 +0000 (23:35 -0500)] 
prune: lazily perform reachability traversal

The general strategy of "git prune" is to do a full reachability walk,
then for each loose object see if we found it in our walk. But if we
don't have any loose objects, we don't need to do the expensive walk in
the first place.

This patch postpones that walk until the first time we need to see its
results.

Note that this is really a specific case of a more general optimization,
which is that we could traverse only far enough to find the object under
consideration (i.e., stop the traversal when we find it, then pick up
again when asked about the next object, etc). That could save us in some
instances from having to do a full walk. But it's actually a bit tricky
to do with our traversal code, and you'd need to do a full walk anyway
if you have even a single unreachable object (which you generally do, if
any objects are actually left after running git-repack).

So in practice this lazy-load of the full walk catches one easy but
common case (i.e., you've just repacked via git-gc, and there's nothing
unreachable).

The perf script is fairly contrived, but it does show off the
improvement:

  Test                            HEAD^             HEAD
  -------------------------------------------------------------------------
  5304.4: prune with no objects   3.66(3.60+0.05)   0.00(0.00+0.00) -100.0%

and would let us know if we accidentally regress this optimization.

Note also that we need to take special care with prune_shallow(), which
relies on us having performed the traversal. So this optimization can
only kick in for a non-shallow repository. Since this is easy to get
wrong and is not covered by existing tests, let's add an extra test to
t5304 that covers this case explicitly.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agoMerge branch 'ea/rebase-compat-doc-fix'
Junio C Hamano [Thu, 14 Feb 2019 22:28:22 +0000 (14:28 -0800)] 
Merge branch 'ea/rebase-compat-doc-fix'

* ea/rebase-compat-doc-fix:
  docs/git-rebase: remove redundant entry in incompatible options list

5 years agoMerge branch 'jc/no-grepping-for-strerror-in-tests'
Junio C Hamano [Thu, 14 Feb 2019 22:28:21 +0000 (14:28 -0800)] 
Merge branch 'jc/no-grepping-for-strerror-in-tests'

* jc/no-grepping-for-strerror-in-tests:
  t1404: do not rely on the exact phrasing of strerror()

5 years agoMerge branch 'jt/fetch-v2-sideband'
Junio C Hamano [Thu, 14 Feb 2019 22:28:20 +0000 (14:28 -0800)] 
Merge branch 'jt/fetch-v2-sideband'

"git fetch" and "git upload-pack" learned to send all exchange over
the sideband channel while talking the v2 protocol.

* jt/fetch-v2-sideband:
  t/lib-httpd: pass GIT_TEST_SIDEBAND_ALL through Apache

5 years agoMerge branch 'en/rebase-merge-on-sequencer'
Junio C Hamano [Thu, 14 Feb 2019 22:28:20 +0000 (14:28 -0800)] 
Merge branch 'en/rebase-merge-on-sequencer'

"git rebase --merge" as been reimplemented by reusing the internal
machinery used for "git rebase -i".

* en/rebase-merge-on-sequencer:
  git-rebase.txt: update to reflect merge now implemented on sequencer

5 years agogit-rebase.txt: update to reflect merge now implemented on sequencer
Elijah Newren [Thu, 14 Feb 2019 20:25:41 +0000 (12:25 -0800)] 
git-rebase.txt: update to reflect merge now implemented on sequencer

Since commit 8fe9c3f21dff (Merge branch 'en/rebase-merge-on-sequencer',
2019-02-06), --merge now uses the interactive backend (and matches its
behavior) so there is no separate merge backend anymore.  Fix an
oversight in the docs that should have been updated with the previous
change.

Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agot/lib-httpd: pass GIT_TEST_SIDEBAND_ALL through Apache
Todd Zullinger [Thu, 14 Feb 2019 06:35:13 +0000 (01:35 -0500)] 
t/lib-httpd: pass GIT_TEST_SIDEBAND_ALL through Apache

07c3c2aa16 ("tests: define GIT_TEST_SIDEBAND_ALL", 2019-01-16) added
GIT_TEST_SIDEBAND_ALL to the apache.conf PassEnv list.  Avoid warnings
from Apache when the variable is unset, as we do for GIT_VALGRIND* and
GIT_TRACE, from f628825481 ("t/lib-httpd: handle running under
--valgrind", 2012-07-24) and 89c57ab3f0 ("t: pass GIT_TRACE through
Apache", 2015-03-13), respectively.

Signed-off-by: Todd Zullinger <tmz@pobox.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5 years agot1404: do not rely on the exact phrasing of strerror()
Junio C Hamano [Thu, 14 Feb 2019 20:16:20 +0000 (12:16 -0800)] 
t1404: do not rely on the exact phrasing of strerror()

Not even in C locale, it is wrong to expect that the exact phrasing
"File exists" is used to show EEXIST.

Reported-by: Randall S. Becker <rsbecker@nexbridge.com>
Helped-by: Duy Nguyen <pclouds@gmail.com>
Helped-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>