OSDN Git Service

git-core/git.git
9 years agoMerge branch 'jc/merge'
Junio C Hamano [Tue, 19 May 2015 20:17:57 +0000 (13:17 -0700)]
Merge branch 'jc/merge'

"git merge FETCH_HEAD" learned that the previous "git fetch" could
be to create an Octopus merge, i.e. recording multiple branches
that are not marked as "not-for-merge"; this allows us to lose an
old style invocation "git merge <msg> HEAD $commits..." in the
implementation of "git pull" script; the old style syntax can now
be deprecated.

* jc/merge:
  merge: deprecate 'git merge <message> HEAD <commit>' syntax
  merge: handle FETCH_HEAD internally
  merge: decide if we auto-generate the message early in collect_parents()
  merge: make collect_parents() auto-generate the merge message
  merge: extract prepare_merge_message() logic out
  merge: narrow scope of merge_names
  merge: split reduce_parents() out of collect_parents()
  merge: clarify collect_parents() logic
  merge: small leakfix and code simplification
  merge: do not check argc to determine number of remote heads
  merge: clarify "pulling into void" special case
  t5520: test pulling an octopus into an unborn branch
  t5520: style fixes
  merge: simplify code flow
  merge: test the top-level merge driver

9 years agoMerge branch 'ph/rebase-i-redo'
Junio C Hamano [Tue, 19 May 2015 20:17:56 +0000 (13:17 -0700)]
Merge branch 'ph/rebase-i-redo'

"git rebase -i" moved the "current" command from "todo" to "done" a
bit too prematurely, losing a step when a "pick" did not even start.

* ph/rebase-i-redo:
  rebase -i: redo tasks that die during cherry-pick

9 years agoMerge branch 'jc/test-prereq-validate'
Junio C Hamano [Tue, 19 May 2015 20:17:55 +0000 (13:17 -0700)]
Merge branch 'jc/test-prereq-validate'

Help us to find broken test script that splits the body part of the
test by mistaken use of wrong kind of quotes.

* jc/test-prereq-validate:
  test: validate prerequistes syntax

9 years agoMerge branch 'bc/connect-plink'
Junio C Hamano [Tue, 19 May 2015 20:17:54 +0000 (13:17 -0700)]
Merge branch 'bc/connect-plink'

The connection initiation code for "ssh" transport tried to absorb
differences between the stock "ssh" and Putty-supplied "plink" and
its derivatives, but the logic to tell that we are using "plink"
variants were too loose and falsely triggered when "plink" appeared
anywhere in the path (e.g. "/home/me/bin/uplink/ssh").

* bc/connect-plink:
  connect: improve check for plink to reduce false positives
  t5601: fix quotation error leading to skipped tests
  connect: simplify SSH connection code path

9 years agoMerge branch 'jk/test-chain-lint'
Junio C Hamano [Tue, 19 May 2015 20:17:54 +0000 (13:17 -0700)]
Merge branch 'jk/test-chain-lint'

Developer support to automatically detect broken &&-chain in the
test scripts is now turned on by default.

* jk/test-chain-lint:
  test-lib: turn on GIT_TEST_CHAIN_LINT by default
  t7502-commit.sh: fix a broken and-chain

9 years agoMerge branch 'fg/document-commit-message-stripping'
Junio C Hamano [Tue, 19 May 2015 20:17:53 +0000 (13:17 -0700)]
Merge branch 'fg/document-commit-message-stripping'

* fg/document-commit-message-stripping:
  Documentation: clarify how "git commit" cleans up the edited log message

9 years agoMerge branch 'jk/stash-require-clean-index'
Junio C Hamano [Tue, 19 May 2015 20:17:52 +0000 (13:17 -0700)]
Merge branch 'jk/stash-require-clean-index'

"git stash pop/apply" forgot to make sure that not just the working
tree is clean but also the index is clean. The latter is important
as a stash application can conflict and the index will be used for
conflict resolution.

* jk/stash-require-clean-index:
  stash: require a clean index to apply
  t3903: avoid applying onto dirty index
  t3903: stop hard-coding commit sha1s

9 years agoMerge branch 'jk/git-no-more-argv0-path-munging'
Junio C Hamano [Tue, 19 May 2015 20:17:52 +0000 (13:17 -0700)]
Merge branch 'jk/git-no-more-argv0-path-munging'

We have prepended $GIT_EXEC_PATH and the path "git" is installed in
(typically "/usr/bin") to $PATH when invoking subprograms and hooks
for almost eternity, but the original use case the latter tried to
support was semi-bogus (i.e. install git to /opt/foo/git and run it
without having /opt/foo on $PATH), and more importantly it has
become less and less relevant as Git grew more mainstream (i.e. the
users would _want_ to have it on their $PATH).  Stop prepending the
path in which "git" is installed to users' $PATH, as that would
interfere the command search order people depend on (e.g. they may
not like versions of programs that are unrelated to Git in /usr/bin
and want to override them by having different ones in /usr/local/bin
and have the latter directory earlier in their $PATH).

* jk/git-no-more-argv0-path-munging:
  stop putting argv[0] dirname at front of PATH

9 years agoMerge branch 'jc/gitignore-precedence'
Junio C Hamano [Tue, 19 May 2015 20:17:50 +0000 (13:17 -0700)]
Merge branch 'jc/gitignore-precedence'

core.excludesfile (defaulting to $XDG_HOME/git/ignore) is supposed
to be overridden by repository-specific .git/info/exclude file, but
the order was swapped from the beginning. This belatedly fixes it.

* jc/gitignore-precedence:
  ignore: info/exclude should trump core.excludesfile

9 years agoMerge branch 'nd/diff-i-t-a'
Junio C Hamano [Tue, 19 May 2015 20:17:49 +0000 (13:17 -0700)]
Merge branch 'nd/diff-i-t-a'

After "git add -N", the path appeared in output of "git diff HEAD"
and "git diff --cached HEAD", leading "git status" to classify it
as "Changes to be committed".  Such a path, however, is not yet to
be scheduled to be committed.  "git diff" showed the change to the
path as modification, not as a "new file", in the header of its
output.

Treat such paths as "yet to be added to the index but Git already
know about them"; "git diff HEAD" and "git diff --cached HEAD"
should not talk about them, and "git diff" should show them as new
files yet to be added to the index.

* nd/diff-i-t-a:
  diff-lib.c: adjust position of i-t-a entries in diff

9 years agoSync with 2.4.1
Junio C Hamano [Wed, 13 May 2015 21:34:46 +0000 (14:34 -0700)]
Sync with 2.4.1

* maint:
  Git 2.4.1

9 years agoGit 2.4.1 v2.4.1
Junio C Hamano [Wed, 13 May 2015 21:11:43 +0000 (14:11 -0700)]
Git 2.4.1

Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoMerge branch 'sb/line-log-plug-pairdiff-leak' into maint
Junio C Hamano [Wed, 13 May 2015 21:05:56 +0000 (14:05 -0700)]
Merge branch 'sb/line-log-plug-pairdiff-leak' into maint

* sb/line-log-plug-pairdiff-leak:
  line-log.c: fix a memleak

9 years agoMerge branch 'sb/test-bitmap-free-at-end' into maint
Junio C Hamano [Wed, 13 May 2015 21:05:55 +0000 (14:05 -0700)]
Merge branch 'sb/test-bitmap-free-at-end' into maint

* sb/test-bitmap-free-at-end:
  pack-bitmap.c: fix a memleak

9 years agoMerge branch 'nd/t1509-chroot-test' into maint
Junio C Hamano [Wed, 13 May 2015 21:05:55 +0000 (14:05 -0700)]
Merge branch 'nd/t1509-chroot-test' into maint

Correct test bitrot.

* nd/t1509-chroot-test:
  t1509: update prepare script to be able to run t1509 in chroot again

9 years agoMerge branch 'jk/type-from-string-gently' into maint
Junio C Hamano [Wed, 13 May 2015 21:05:54 +0000 (14:05 -0700)]
Merge branch 'jk/type-from-string-gently' into maint

"git cat-file bl $blob" failed to barf even though there is no
object type that is "bl".

* jk/type-from-string-gently:
  type_from_string_gently: make sure length matches

9 years agoMerge branch 'ep/fix-test-lib-functions-report' into maint
Junio C Hamano [Wed, 13 May 2015 21:05:52 +0000 (14:05 -0700)]
Merge branch 'ep/fix-test-lib-functions-report' into maint

* ep/fix-test-lib-functions-report:
  test-lib-functions.sh: fix the second argument to some helper functions

9 years agoMerge branch 'cn/bom-in-gitignore' into maint
Junio C Hamano [Wed, 13 May 2015 21:05:51 +0000 (14:05 -0700)]
Merge branch 'cn/bom-in-gitignore' into maint

Teach the codepaths that read .gitignore and .gitattributes files
that these files encoded in UTF-8 may have UTF-8 BOM marker at the
beginning; this makes it in line with what we do for configuration
files already.

* cn/bom-in-gitignore:
  attr: skip UTF8 BOM at the beginning of the input file
  config: use utf8_bom[] from utf.[ch] in git_parse_source()
  utf8-bom: introduce skip_utf8_bom() helper
  add_excludes_from_file: clarify the bom skipping logic
  dir: allow a BOM at the beginning of exclude files

9 years agoMerge branch 'jk/prune-mtime' into maint
Junio C Hamano [Wed, 13 May 2015 21:05:50 +0000 (14:05 -0700)]
Merge branch 'jk/prune-mtime' into maint

Access to objects in repositories that borrow from another one on a
slow NFS server unnecessarily got more expensive due to recent code
becoming more cautious in a naive way not to lose objects to pruning.

* jk/prune-mtime:
  sha1_file: only freshen packs once per run
  sha1_file: freshen pack objects before loose
  reachable: only mark local objects as recent

9 years agoMerge branch 'jk/init-core-worktree-at-root' into maint
Junio C Hamano [Wed, 13 May 2015 21:05:49 +0000 (14:05 -0700)]
Merge branch 'jk/init-core-worktree-at-root' into maint

We avoid setting core.worktree when the repository location is the
".git" directory directly at the top level of the working tree, but
the code misdetected the case in which the working tree is at the
root level of the filesystem (which arguably is a silly thing to
do, but still valid).

* jk/init-core-worktree-at-root:
  init: don't set core.worktree when initializing /.git

9 years agoMerge branch 'maint'
Junio C Hamano [Mon, 11 May 2015 21:39:39 +0000 (14:39 -0700)]
Merge branch 'maint'

* maint:
  Git 2.3.8

9 years agoSync with 2.3.8
Junio C Hamano [Mon, 11 May 2015 21:39:28 +0000 (14:39 -0700)]
Sync with 2.3.8

Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoGit 2.3.8 v2.3.8
Junio C Hamano [Mon, 11 May 2015 21:36:31 +0000 (14:36 -0700)]
Git 2.3.8

Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoMerge branch 'mm/usage-log-l-can-take-regex' into maint-2.3
Junio C Hamano [Mon, 11 May 2015 21:34:01 +0000 (14:34 -0700)]
Merge branch 'mm/usage-log-l-can-take-regex' into maint-2.3

Documentation fix.

* mm/usage-log-l-can-take-regex:
  log -L: improve error message on malformed argument
  Documentation: change -L:<regex> to -L:<funcname>

9 years agoMerge branch 'jc/diff-no-index-d-f' into maint-2.3
Junio C Hamano [Mon, 11 May 2015 21:34:00 +0000 (14:34 -0700)]
Merge branch 'jc/diff-no-index-d-f' into maint-2.3

The usual "git diff" when seeing a file turning into a directory
showed a patchset to remove the file and create all files in the
directory, but "git diff --no-index" simply refused to work.  Also,
when asked to compare a file and a directory, imitate POSIX "diff"
and compare the file with the file with the same name in the
directory, instead of refusing to run.

* jc/diff-no-index-d-f:
  diff-no-index: align D/F handling with that of normal Git
  diff-no-index: DWIM "diff D F" into "diff D/F F"

9 years agoMerge branch 'oh/fix-config-default-user-name-section' into maint-2.3
Junio C Hamano [Mon, 11 May 2015 21:33:59 +0000 (14:33 -0700)]
Merge branch 'oh/fix-config-default-user-name-section' into maint-2.3

The default $HOME/.gitconfig file created upon "git config --global"
that edits it had incorrectly spelled user.name and user.email
entries in it.

* oh/fix-config-default-user-name-section:
  config: fix settings in default_user_config template

9 years agoMerge branch 'jc/epochtime-wo-tz' into maint-2.3
Junio C Hamano [Mon, 11 May 2015 21:33:58 +0000 (14:33 -0700)]
Merge branch 'jc/epochtime-wo-tz' into maint-2.3

"git commit --date=now" or anything that relies on approxidate lost
the daylight-saving-time offset.

* jc/epochtime-wo-tz:
  parse_date_basic(): let the system handle DST conversion
  parse_date_basic(): return early when given a bogus timestamp

9 years agoSecond batch for 2.5 cycle
Junio C Hamano [Mon, 11 May 2015 21:31:28 +0000 (14:31 -0700)]
Second batch for 2.5 cycle

Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoMerge branch 'pt/xdg-config-path'
Junio C Hamano [Mon, 11 May 2015 21:24:01 +0000 (14:24 -0700)]
Merge branch 'pt/xdg-config-path'

Code clean-up for xdg configuration path support.

* pt/xdg-config-path:
  path.c: remove home_config_paths()
  git-config: replace use of home_config_paths()
  git-commit: replace use of home_config_paths()
  credential-store.c: replace home_config_paths() with xdg_config_home()
  dir.c: replace home_config_paths() with xdg_config_home()
  attr.c: replace home_config_paths() with xdg_config_home()
  path.c: implement xdg_config_home()

9 years agoMerge branch 'ep/do-not-feed-a-pointer-to-array-size'
Junio C Hamano [Mon, 11 May 2015 21:24:00 +0000 (14:24 -0700)]
Merge branch 'ep/do-not-feed-a-pointer-to-array-size'

Catch a programmer mistake to feed a pointer not an array to
ARRAY_SIZE() macro, by using a couple of GCC extensions.

* ep/do-not-feed-a-pointer-to-array-size:
  git-compat-util.h: implement a different ARRAY_SIZE macro for for safely deriving the size of array

9 years agoMerge branch 'jc/hash-object'
Junio C Hamano [Mon, 11 May 2015 21:23:59 +0000 (14:23 -0700)]
Merge branch 'jc/hash-object'

"hash-object --literally" introduced in v2.2 was not prepared to
take a really long object type name.

* jc/hash-object:
  write_sha1_file(): do not use a separate sha1[] array
  t1007: add hash-object --literally tests
  hash-object --literally: fix buffer overrun with extra-long object type
  git-hash-object.txt: document --literally option

9 years agoMerge branch 'sb/prefix-path-free-results'
Junio C Hamano [Mon, 11 May 2015 21:23:57 +0000 (14:23 -0700)]
Merge branch 'sb/prefix-path-free-results'

Code clean-up (not a leak-fix).

* sb/prefix-path-free-results:
  prefix_path(): unconditionally free results in the callers

9 years agoMerge branch 'sg/completion-no-redundant-all-command-list'
Junio C Hamano [Mon, 11 May 2015 21:23:57 +0000 (14:23 -0700)]
Merge branch 'sg/completion-no-redundant-all-command-list'

Code simplification.

* sg/completion-no-redundant-all-command-list:
  completion: remove redundant __git_compute_all_commands() call

9 years agoMerge branch 'sg/complete-decorate-full-not-long'
Junio C Hamano [Mon, 11 May 2015 21:23:55 +0000 (14:23 -0700)]
Merge branch 'sg/complete-decorate-full-not-long'

The completion for "log --decorate=" parameter value was incorrect.

* sg/complete-decorate-full-not-long:
  completion: fix and update 'git log --decorate=' options

9 years agoMerge branch 'jk/filter-branch-use-of-sed-on-incomplete-line'
Junio C Hamano [Mon, 11 May 2015 21:23:54 +0000 (14:23 -0700)]
Merge branch 'jk/filter-branch-use-of-sed-on-incomplete-line'

"filter-branch" corrupted commit log message that ends with an
incomplete line on platforms with some "sed" implementations that
munge such a line.  Work it around by avoiding to use "sed".

* jk/filter-branch-use-of-sed-on-incomplete-line:
  filter-branch: avoid passing commit message through sed

9 years agoMerge branch 'jc/daemon-no-ipv6-for-2.4.1'
Junio C Hamano [Mon, 11 May 2015 21:23:53 +0000 (14:23 -0700)]
Merge branch 'jc/daemon-no-ipv6-for-2.4.1'

"git daemon" fails to build from the source under NO_IPV6
configuration (regression in 2.4).

* jc/daemon-no-ipv6-for-2.4.1:
  daemon: unbreak NO_IPV6 build regression

9 years agoMerge branch 'jn/clean-use-error-not-fprintf-on-stderr'
Junio C Hamano [Mon, 11 May 2015 21:23:52 +0000 (14:23 -0700)]
Merge branch 'jn/clean-use-error-not-fprintf-on-stderr'

Some error messages in "git config" were emitted without calling
the usual error() facility.

* jn/clean-use-error-not-fprintf-on-stderr:
  config: use error() instead of fprintf(stderr, ...)

9 years agoMerge branch 'tb/blame-resurrect-convert-to-git'
Junio C Hamano [Mon, 11 May 2015 21:23:52 +0000 (14:23 -0700)]
Merge branch 'tb/blame-resurrect-convert-to-git'

Some time ago, "git blame" (incorrectly) lost the convert_to_git()
call when synthesizing a fake "tip" commit that represents the
state in the working tree, which broke folks who record the history
with LF line ending to make their project portabile across
platforms while terminating lines in their working tree files with
CRLF for their platform.

* tb/blame-resurrect-convert-to-git:
  blame: CRLF in the working tree and LF in the repo

9 years agoMerge branch 'va/fix-git-p4-tests'
Junio C Hamano [Mon, 11 May 2015 21:23:51 +0000 (14:23 -0700)]
Merge branch 'va/fix-git-p4-tests'

* va/fix-git-p4-tests:
  git-p4: t9814: prevent --chain-lint failure

9 years agoMerge branch 'ld/p4-case-fold'
Junio C Hamano [Mon, 11 May 2015 21:23:50 +0000 (14:23 -0700)]
Merge branch 'ld/p4-case-fold'

* ld/p4-case-fold:
  git-p4: add failing tests for case-folding p4d

9 years agoMerge branch 'jk/rebase-quiet-noop'
Junio C Hamano [Mon, 11 May 2015 21:23:49 +0000 (14:23 -0700)]
Merge branch 'jk/rebase-quiet-noop'

"git rebase --quiet" was not quite quiet when there is nothing to
do.

* jk/rebase-quiet-noop:
  rebase: silence "git checkout" for noop rebase

9 years agoMerge branch 'va/p4-client-path'
Junio C Hamano [Mon, 11 May 2015 21:23:48 +0000 (14:23 -0700)]
Merge branch 'va/p4-client-path'

git p4 attempts to better handle branches in Perforce.

* va/p4-client-path:
  git-p4: improve client path detection when branches are used
  t9801: check git-p4's branch detection with client spec enabled

9 years agoMerge branch 'mm/add-p-split-error'
Junio C Hamano [Mon, 11 May 2015 21:23:47 +0000 (14:23 -0700)]
Merge branch 'mm/add-p-split-error'

When "add--interactive" splits a hunk into two overlapping hunks
and then let the user choose only one, it sometimes feeds an
incorrect patch text to "git apply".  Add tests to demonstrate
this.

I have a slight suspicion that this may be $gmane/87202 coming back
and biting us (I seem to have said "let's run with this and see
what happens" back then).

* mm/add-p-split-error:
  stash -p: demonstrate failure of split with mixed y/n
  t3904-stash-patch: factor PERL prereq at the top of the file
  t3904-stash-patch: fix test description
  add -p: demonstrate failure when running 'edit' after a split
  t3701-add-interactive: simplify code

9 years agoMerge branch 'tb/t0027-crlf'
Junio C Hamano [Mon, 11 May 2015 21:23:47 +0000 (14:23 -0700)]
Merge branch 'tb/t0027-crlf'

More line-ending tests.

* tb/t0027-crlf:
  t0027: Add repoMIX and LF_nul
  t0027: support NATIVE_CRLF platforms
  t0027: cleanup: rename functions; avoid non-leading TABs

9 years agoMerge branch 'ls/p4-changes-block-size'
Junio C Hamano [Mon, 11 May 2015 21:23:46 +0000 (14:23 -0700)]
Merge branch 'ls/p4-changes-block-size'

"git p4" learned "--changes-block-size <n>" to read the changes in
chunks from Perforce, instead of making one call to "p4 changes"
that may trigger "too many rows scanned" error from Perforce.

* ls/p4-changes-block-size:
  git-p4: use -m when running p4 changes

9 years agoMerge branch 'jc/plug-fmt-merge-msg-leak'
Junio C Hamano [Mon, 11 May 2015 21:23:45 +0000 (14:23 -0700)]
Merge branch 'jc/plug-fmt-merge-msg-leak'

* jc/plug-fmt-merge-msg-leak:
  fmt-merge-msg: plug small leak of commit buffer

9 years agoMerge branch 'nd/slim-index-pack-memory-usage'
Junio C Hamano [Mon, 11 May 2015 21:23:44 +0000 (14:23 -0700)]
Merge branch 'nd/slim-index-pack-memory-usage'

Memory usage of "git index-pack" has been trimmed by tens of
per-cent.

* nd/slim-index-pack-memory-usage:
  index-pack: kill union delta_base to save memory
  index-pack: reduce object_entry size to save memory

9 years agoMerge branch 'jk/still-interesting'
Junio C Hamano [Mon, 11 May 2015 21:23:43 +0000 (14:23 -0700)]
Merge branch 'jk/still-interesting'

"git rev-list --objects $old --not --all" to see if everything that
is reachable from $old is already connected to the existing refs
was very inefficient.

* jk/still-interesting:
  limit_list: avoid quadratic behavior from still_interesting

9 years agoMerge branch 'jk/reading-packed-refs'
Junio C Hamano [Mon, 11 May 2015 21:23:42 +0000 (14:23 -0700)]
Merge branch 'jk/reading-packed-refs'

An earlier rewrite to use strbuf_getwholeline() instead of fgets(3)
to read packed-refs file revealed that the former is unacceptably
inefficient.

* jk/reading-packed-refs:
  t1430: add another refs-escape test
  read_packed_refs: avoid double-checking sane refs
  strbuf_getwholeline: use getdelim if it is available
  strbuf_getwholeline: avoid calling strbuf_grow
  strbuf_addch: avoid calling strbuf_grow
  config: use getc_unlocked when reading from file
  strbuf_getwholeline: use getc_unlocked
  git-compat-util: add fallbacks for unlocked stdio
  strbuf_getwholeline: use getc macro

9 years agoMerge branch 'lm/squelch-bg-progress'
Junio C Hamano [Mon, 11 May 2015 21:23:42 +0000 (14:23 -0700)]
Merge branch 'lm/squelch-bg-progress'

Many long-running operations show progress eye-candy, even when
they are later backgrounded.  Hide the eye-candy when the process
is sent to the background instead.

* lm/squelch-bg-progress:
  compat/mingw: stubs for getpgid() and tcgetpgrp()
  progress: no progress in background

9 years agoMerge branch 'jk/sha1-file-reduce-useless-warnings'
Junio C Hamano [Mon, 11 May 2015 21:23:40 +0000 (14:23 -0700)]
Merge branch 'jk/sha1-file-reduce-useless-warnings'

* jk/sha1-file-reduce-useless-warnings:
  sha1_file: squelch "packfile cannot be accessed" warnings

9 years agoMerge branch 'nd/multiple-work-trees'
Junio C Hamano [Mon, 11 May 2015 21:23:39 +0000 (14:23 -0700)]
Merge branch 'nd/multiple-work-trees'

A replacement for contrib/workdir/git-new-workdir that does not
rely on symbolic links and make sharing of objects and refs safer
by making the borrowee and borrowers aware of each other.

* nd/multiple-work-trees: (41 commits)
  prune --worktrees: fix expire vs worktree existence condition
  t1501: fix test with split index
  t2026: fix broken &&-chain
  t2026 needs procondition SANITY
  git-checkout.txt: a note about multiple checkout support for submodules
  checkout: add --ignore-other-wortrees
  checkout: pass whole struct to parse_branchname_arg instead of individual flags
  git-common-dir: make "modules/" per-working-directory directory
  checkout: do not fail if target is an empty directory
  t2025: add a test to make sure grafts is working from a linked checkout
  checkout: don't require a work tree when checking out into a new one
  git_path(): keep "info/sparse-checkout" per work-tree
  count-objects: report unused files in $GIT_DIR/worktrees/...
  gc: support prune --worktrees
  gc: factor out gc.pruneexpire parsing code
  gc: style change -- no SP before closing parenthesis
  checkout: clean up half-prepared directories in --to mode
  checkout: reject if the branch is already checked out elsewhere
  prune: strategies for linked checkouts
  checkout: support checking out into a new working directory
  ...

9 years agoMerge branch 'pt/credential-xdg'
Junio C Hamano [Mon, 11 May 2015 21:23:38 +0000 (14:23 -0700)]
Merge branch 'pt/credential-xdg'

Tweak the sample "store" backend of the credential helper to honor
XDG configuration file locations when specified.

* pt/credential-xdg:
  t0302: "unreadable" test needs POSIXPERM
  t0302: test credential-store support for XDG_CONFIG_HOME
  git-credential-store: support XDG_CONFIG_HOME
  git-credential-store: support multiple credential files

9 years agopath.c: remove home_config_paths()
Paul Tan [Wed, 6 May 2015 08:01:04 +0000 (16:01 +0800)]
path.c: remove home_config_paths()

home_config_paths() combines distinct functionality already implemented
by expand_user_path() and xdg_config_home(), and it also hard-codes the
path ~/.gitconfig, which makes it unsuitable to use for other home
config file paths. Since its use will just add unnecessary complexity to
the code, remove it.

Signed-off-by: Paul Tan <pyokagan@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agogit-config: replace use of home_config_paths()
Paul Tan [Wed, 6 May 2015 08:01:03 +0000 (16:01 +0800)]
git-config: replace use of home_config_paths()

Since home_config_paths() combines distinct functionality already
implemented by expand_user_path() and xdg_config_home(), and hides the
home config file path ~/.gitconfig. Make the code more explicit by
replacing the use of home_config_paths() with expand_user_path() and
xdg_config_home().

Signed-off-by: Paul Tan <pyokagan@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agogit-commit: replace use of home_config_paths()
Paul Tan [Wed, 6 May 2015 08:01:02 +0000 (16:01 +0800)]
git-commit: replace use of home_config_paths()

Since home_config_paths() combines two distinct functionality already
implemented by expand_user_path() and xdg_config_home(), and hides the
home config file path ~/.gitconfig. Make the code more explicit by
replacing the use of home_config_paths() with expand_user_path() and
xdg_config_home().

Signed-off-by: Paul Tan <pyokagan@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agocredential-store.c: replace home_config_paths() with xdg_config_home()
Paul Tan [Wed, 6 May 2015 08:01:01 +0000 (16:01 +0800)]
credential-store.c: replace home_config_paths() with xdg_config_home()

Since only the xdg credentials file path is required, and
home_config_paths() is unable to construct the path ~/.git-credentials,
simplify the code by replacing home_config_paths() with
xdg_config_home().

Signed-off-by: Paul Tan <pyokagan@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agodir.c: replace home_config_paths() with xdg_config_home()
Paul Tan [Wed, 6 May 2015 08:01:00 +0000 (16:01 +0800)]
dir.c: replace home_config_paths() with xdg_config_home()

Since only the xdg excludes file path is required, simplify the code by
replacing use of home_config_paths() with xdg_config_home().

Signed-off-by: Paul Tan <pyokagan@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoattr.c: replace home_config_paths() with xdg_config_home()
Paul Tan [Wed, 6 May 2015 08:00:59 +0000 (16:00 +0800)]
attr.c: replace home_config_paths() with xdg_config_home()

Since only the xdg attributes file path is required, simplify the code
by using xdg_config_home() instead of home_config_paths().

Signed-off-by: Paul Tan <pyokagan@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agopath.c: implement xdg_config_home()
Paul Tan [Tue, 21 Apr 2015 04:06:27 +0000 (12:06 +0800)]
path.c: implement xdg_config_home()

The XDG base dir spec[1] specifies that configuration files be stored in
a subdirectory in $XDG_CONFIG_HOME. To construct such a configuration
file path, home_config_paths() can be used. However, home_config_paths()
combines distinct functionality:

1. Retrieve the home git config file path ~/.gitconfig

2. Construct the XDG config path of the file specified by `file`.

This function was introduced in commit 21cf3227 ("read (but not write)
from $XDG_CONFIG_HOME/git/config file").  While the intention of the
function was to allow the home directory configuration file path and the
xdg directory configuration file path to be retrieved with one function
call, the hard-coding of the path ~/.gitconfig prevents it from being
used for other configuration files. Furthermore, retrieving a file path
relative to the user's home directory can be done with
expand_user_path(). Hence, it can be seen that home_config_paths()
introduces unnecessary complexity, especially if a user just wants to
retrieve the xdg config file path.

As such, implement a simpler function xdg_config_home() for constructing
the XDG base dir spec configuration file path. This function, together
with expand_user_path(), can replace all uses of home_config_paths().

[1] http://standards.freedesktop.org/basedir-spec/basedir-spec-0.7.html

Helped-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Paul Tan <pyokagan@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoFirst batch for 2.5 cycle
Junio C Hamano [Wed, 6 May 2015 04:13:30 +0000 (21:13 -0700)]
First batch for 2.5 cycle

Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoMerge branch 'jk/prune-mtime'
Junio C Hamano [Wed, 6 May 2015 04:00:37 +0000 (21:00 -0700)]
Merge branch 'jk/prune-mtime'

Access to objects in repositories that borrow from another one on a
slow NFS server unnecessarily got more expensive due to recent code
becoming more cautious in a naive way not to lose objects to pruning.

* jk/prune-mtime:
  sha1_file: only freshen packs once per run
  sha1_file: freshen pack objects before loose
  reachable: only mark local objects as recent

9 years agoMerge branch 'mm/usage-log-l-can-take-regex'
Junio C Hamano [Wed, 6 May 2015 04:00:36 +0000 (21:00 -0700)]
Merge branch 'mm/usage-log-l-can-take-regex'

Documentation fix.

* mm/usage-log-l-can-take-regex:
  log -L: improve error message on malformed argument
  Documentation: change -L:<regex> to -L:<funcname>

9 years agoMerge branch 'ep/fix-test-lib-functions-report'
Junio C Hamano [Wed, 6 May 2015 04:00:35 +0000 (21:00 -0700)]
Merge branch 'ep/fix-test-lib-functions-report'

* ep/fix-test-lib-functions-report:
  test-lib-functions.sh: fix the second argument to some helper functions

9 years agoMerge branch 'cn/bom-in-gitignore'
Junio C Hamano [Wed, 6 May 2015 04:00:34 +0000 (21:00 -0700)]
Merge branch 'cn/bom-in-gitignore'

Teach the codepaths that read .gitignore and .gitattributes files
that these files encoded in UTF-8 may have UTF-8 BOM marker at the
beginning; this makes it in line with what we do for configuration
files already.

* cn/bom-in-gitignore:
  attr: skip UTF8 BOM at the beginning of the input file
  config: use utf8_bom[] from utf.[ch] in git_parse_source()
  utf8-bom: introduce skip_utf8_bom() helper
  add_excludes_from_file: clarify the bom skipping logic
  dir: allow a BOM at the beginning of exclude files

9 years agoMerge branch 'jc/epochtime-wo-tz'
Junio C Hamano [Wed, 6 May 2015 04:00:33 +0000 (21:00 -0700)]
Merge branch 'jc/epochtime-wo-tz'

"git commit --date=now" or anything that relies on approxidate lost
the daylight-saving-time offset.

* jc/epochtime-wo-tz:
  parse_date_basic(): let the system handle DST conversion
  parse_date_basic(): return early when given a bogus timestamp

9 years agoMerge branch 'nd/t1509-chroot-test'
Junio C Hamano [Wed, 6 May 2015 04:00:31 +0000 (21:00 -0700)]
Merge branch 'nd/t1509-chroot-test'

Correct test bitrot.

* nd/t1509-chroot-test:
  t1509: update prepare script to be able to run t1509 in chroot again

9 years agoMerge branch 'oh/fix-config-default-user-name-section'
Junio C Hamano [Wed, 6 May 2015 04:00:30 +0000 (21:00 -0700)]
Merge branch 'oh/fix-config-default-user-name-section'

The default $HOME/.gitconfig file created upon "git config --global"
that edits it had incorrectly spelled user.name and user.email
entries in it.

* oh/fix-config-default-user-name-section:
  config: fix settings in default_user_config template

9 years agoMerge branch 'jk/type-from-string-gently'
Junio C Hamano [Wed, 6 May 2015 04:00:29 +0000 (21:00 -0700)]
Merge branch 'jk/type-from-string-gently'

"git cat-file bl $blob" failed to barf even though there is no
object type that is "bl".

* jk/type-from-string-gently:
  type_from_string_gently: make sure length matches

9 years agoMerge branch 'sb/test-bitmap-free-at-end'
Junio C Hamano [Wed, 6 May 2015 04:00:28 +0000 (21:00 -0700)]
Merge branch 'sb/test-bitmap-free-at-end'

* sb/test-bitmap-free-at-end:
  pack-bitmap.c: fix a memleak

9 years agoMerge branch 'ld/p4-filetype-detection'
Junio C Hamano [Wed, 6 May 2015 04:00:27 +0000 (21:00 -0700)]
Merge branch 'ld/p4-filetype-detection'

* ld/p4-filetype-detection:
  git-p4: fix filetype detection on files opened exclusively
  git-p4: small fix for locked-file-move-test
  git-p4: fix small bug in locked test scripts

9 years agoMerge branch 'ts/checkout-advice-plural'
Junio C Hamano [Wed, 6 May 2015 04:00:27 +0000 (21:00 -0700)]
Merge branch 'ts/checkout-advice-plural'

* ts/checkout-advice-plural:
  checkout: call a single commit "it" intead of "them"

9 years agoMerge branch 'jk/init-core-worktree-at-root'
Junio C Hamano [Wed, 6 May 2015 04:00:26 +0000 (21:00 -0700)]
Merge branch 'jk/init-core-worktree-at-root'

We avoid setting core.worktree when the repository location is the
".git" directory directly at the top level of the working tree, but
the code misdetected the case in which the working tree is at the
root level of the filesystem (which arguably is a silly thing to
do, but still valid).

* jk/init-core-worktree-at-root:
  init: don't set core.worktree when initializing /.git

9 years agoMerge branch 'mh/show-branch-topic'
Junio C Hamano [Wed, 6 May 2015 04:00:25 +0000 (21:00 -0700)]
Merge branch 'mh/show-branch-topic'

"git show-branch --topics HEAD" (with no other arguments) did not
do anything interesting.  Instead, contrast the given revision
against all the local branches by default.

* mh/show-branch-topic:
  show-branch: show all local heads when only giving one rev along --topics

9 years agoMerge branch 'sb/line-log-plug-pairdiff-leak'
Junio C Hamano [Wed, 6 May 2015 04:00:25 +0000 (21:00 -0700)]
Merge branch 'sb/line-log-plug-pairdiff-leak'

* sb/line-log-plug-pairdiff-leak:
  line-log.c: fix a memleak

9 years agoMerge branch 'jc/diff-no-index-d-f'
Junio C Hamano [Wed, 6 May 2015 04:00:24 +0000 (21:00 -0700)]
Merge branch 'jc/diff-no-index-d-f'

The usual "git diff" when seeing a file turning into a directory
showed a patchset to remove the file and create all files in the
directory, but "git diff --no-index" simply refused to work.  Also,
when asked to compare a file and a directory, imitate POSIX "diff"
and compare the file with the file with the same name in the
directory, instead of refusing to run.

* jc/diff-no-index-d-f:
  diff-no-index: align D/F handling with that of normal Git
  diff-no-index: DWIM "diff D F" into "diff D/F F"

9 years agoMerge branch 'bc/object-id'
Junio C Hamano [Wed, 6 May 2015 04:00:23 +0000 (21:00 -0700)]
Merge branch 'bc/object-id'

Identify parts of the code that knows that we use SHA-1 hash to
name our objects too much, and use (1) symbolic constants instead
of hardcoded 20 as byte count and/or (2) use struct object_id
instead of unsigned char [20] for object names.

* bc/object-id:
  apply: convert threeway_stage to object_id
  patch-id: convert to use struct object_id
  commit: convert parts to struct object_id
  diff: convert struct combine_diff_path to object_id
  bulk-checkin.c: convert to use struct object_id
  zip: use GIT_SHA1_HEXSZ for trailers
  archive.c: convert to use struct object_id
  bisect.c: convert leaf functions to use struct object_id
  define utility functions for object IDs
  define a structure for object IDs

9 years agogit-compat-util.h: implement a different ARRAY_SIZE macro for for safely deriving...
Elia Pinto [Thu, 30 Apr 2015 12:44:14 +0000 (14:44 +0200)]
git-compat-util.h: implement a different ARRAY_SIZE macro for for safely deriving the size of array

To get number of elements in an array git use the ARRAY_SIZE macro
defined as:

       #define ARRAY_SIZE(x) (sizeof(x)/sizeof((x)[0]))

The problem with it is a possibility of mistakenly passing to it a
pointer instead an array. The ARRAY_SIZE macro as conventionally
defined does not provide good type-safety and the open-coded
approach is more fragile, more verbose and provides no improvement in
type-safety.

Use instead a different but compatible ARRAY_SIZE() macro,
which will also break compile if you try to
use it on a pointer. This implemention revert to the original code
if the compiler doesn't know the typeof and __builtin_types_compatible_p
GCC extensions.

This can ensure our code is robust to changes, without
needing a gratuitous macro or constant. A similar
ARRAY_SIZE implementation also exists in the linux kernel.

Credits to Rusty Russell and his ccan library.

Signed-off-by: Elia Pinto <gitter.spiros@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agodaemon: unbreak NO_IPV6 build regression
Junio C Hamano [Tue, 5 May 2015 18:03:24 +0000 (11:03 -0700)]
daemon: unbreak NO_IPV6 build regression

When 01cec54e (daemon: deglobalize hostname information, 2015-03-07)
wrapped the global variables such as hostname inside a struct, it
forgot to convert one location that spelled "hostname" that needs to
be updated to "hi->hostname".

This was inside NO_IPV6 block, and was not caught by anybody.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoprefix_path(): unconditionally free results in the callers
Stefan Beller [Mon, 4 May 2015 19:11:54 +0000 (12:11 -0700)]
prefix_path(): unconditionally free results in the callers

As of d089ebaa (setup: sanitize absolute and funny paths in
get_pathspec(), 2008-01-28), prefix_path() always returns a
newly allocated string, so callers should free its result.

Additionally, drop the const from variables to which the result of
the prefix_path() is assigned, so they can be free()'d without
having to cast-away the constness.

Signed-off-by: Stefan Beller <sbeller@google.com>
Reviewed-by: Eric Sunshine <sunshine@sunshineco.com>
Helped-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agowrite_sha1_file(): do not use a separate sha1[] array
Junio C Hamano [Mon, 4 May 2015 18:08:10 +0000 (11:08 -0700)]
write_sha1_file(): do not use a separate sha1[] array

In the beginning, write_sha1_file() did not have a way to tell the
caller the name of the object it wrote to the caller.  This was
changed in d6d3f9d0 (This implements the new "recursive tree"
write-tree., 2005-04-09) by adding the "returnsha1" parameter to the
function so that the callers who are interested in the value can
optionally pass a pointer to receive it.

It turns out that all callers do want to know the name of the object
it just has written.  Nobody passes a NULL to this parameter, hence
it is not necessary to use a separate sha1[] array to receive the
result from  write_sha1_file_prepare(), and copy the result to the
returnsha1 supplied by the caller.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agot1007: add hash-object --literally tests
Eric Sunshine [Mon, 4 May 2015 07:25:14 +0000 (03:25 -0400)]
t1007: add hash-object --literally tests

git-hash-object learned a --literally option in 5ba9a93
(hash-object: add --literally option, 2014-09-11). Check that
--literally allows object creation with a bogus type, with two
type strings whose length is reasonably short and very long.

Signed-off-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agohash-object --literally: fix buffer overrun with extra-long object type
Eric Sunshine [Mon, 4 May 2015 07:25:15 +0000 (03:25 -0400)]
hash-object --literally: fix buffer overrun with extra-long object type

"hash-object" learned in 5ba9a93 (hash-object: add --literally
option, 2014-09-11) to allow crafting a corrupt/broken object of
unknown type.

When the user-provided type is particularly long, however, it can
overflow the relatively small stack-based character array handed to
write_sha1_file_prepare() by hash_sha1_file() and write_sha1_file(),
leading to stack corruption (and crash).  Introduce a custom helper
to allow arbitrarily long typenames just for "hash-object --literally".

[jc: Eric's original used a strbuf in the more common codepaths, and
I rewrote it to avoid penalizing the non-literally code. Bugs are mine]

Signed-off-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoconfig: use error() instead of fprintf(stderr, ...)
Jonathan Nieder [Mon, 4 May 2015 21:18:39 +0000 (14:18 -0700)]
config: use error() instead of fprintf(stderr, ...)

The die() / error() / warning() helpers put a fatal: / error: /
warning: prefix in front of the error message they print describing
the message's severity, which users are likely to be accustomed to
seeing these days.

This change will also be useful when marking the message for
translation: the argument to error() includes no newline at the end,
so it is less fussy for translators to translate without lines running
together in the translated output.

While we're here, start the error messages with a lowercase letter to
match the usual typography of error messages.

A quick web search and a code search at codesearch.debian.net finds no
scripts trying to parse these error messages, so this change should be
safe.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agogit-hash-object.txt: document --literally option
Eric Sunshine [Mon, 4 May 2015 07:25:13 +0000 (03:25 -0400)]
git-hash-object.txt: document --literally option

Document the git-hash-object --literally option added by 5ba9a93
(hash-object: add --literally option, 2014-09-11).

While here, also correct a minor typesetting oversight.

Signed-off-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agocompletion: fix and update 'git log --decorate=' options
SZEDER Gábor [Fri, 1 May 2015 17:21:00 +0000 (19:21 +0200)]
completion: fix and update 'git log --decorate=' options

'git log --decorate=' understands the 'full', 'short' and 'no' options.
From these the completion script only offered 'short' and it offered
'long' instead of 'full'.

Signed-off-by: SZEDER Gábor <szeder@ira.uka.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agocompletion: remove redundant __git_compute_all_commands() call
SZEDER Gábor [Fri, 1 May 2015 15:48:56 +0000 (17:48 +0200)]
completion: remove redundant __git_compute_all_commands() call

During lazy-initialization of the lists of all commands and porcelain
commands the function __git_compute_all_commands() is called twice.  The
relevant part of the call sequence looks like this:

  __git_compute_porcelain_commands()
     __git_compute_all_commands()
        <finds list of all commands uninitialized>
        __git_list_all_commands()
        <initializes list of all commands>
     __git_list_porcelain_commands()
        __git_compute_all_commands()
           <finds list of all commands already initialized, does nothing>
        <filters porcelains from list of all commands>

Either one of the two calls could be removed and the initialization of
both command lists would still work as a whole, but let's remove the call
from __git_compute_porcelain_commands(), because this way
__git_list_porcelain_commands() will keep working in itself.

Signed-off-by: SZEDER Gábor <szeder@ira.uka.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoblame: CRLF in the working tree and LF in the repo
Torsten Bögershausen [Sun, 3 May 2015 16:38:01 +0000 (18:38 +0200)]
blame: CRLF in the working tree and LF in the repo

A typical setup under Windows is to set core.eol to CRLF, and text
files are marked as "text" in .gitattributes, or core.autocrlf is
set to true.

After 4d4813a5 "git blame" no longer works as expected for such a
set-up.  Every line is annotated as "Not Committed Yet", even though
the working directory is clean.  This is because the commit removed
the conversion in blame.c for all files, with or without CRLF in the
repo.

Having files with CRLF in the repo and core.autocrlf=input is a
temporary situation, and the files, if committed as is, will be
normalized in the repo, which _will_ be a notable change.  Blaming
them with "Not Committed Yet" is the right result.  Revert commit
4d4813a5 which was a misguided attempt to "solve" a non-problem.

Add two test cases in t8003 to verify the correct CRLF conversion.

Suggested-By: Stepan Kasal <kasal@ucw.cz>
Signed-off-by: Torsten Bögershausen <tboegi@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoGit 2.4 v2.4.0
Junio C Hamano [Thu, 30 Apr 2015 18:25:06 +0000 (11:25 -0700)]
Git 2.4

Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agomerge: deprecate 'git merge <message> HEAD <commit>' syntax
Junio C Hamano [Thu, 26 Mar 2015 04:58:45 +0000 (21:58 -0700)]
merge: deprecate 'git merge <message> HEAD <commit>' syntax

We had this in "git merge" manual for eternity:

    'git merge' <msg> HEAD <commit>...

    [This] syntax (<msg> `HEAD` <commit>...) is supported for
    historical reasons.  Do not use it from the command line or in
    new scripts.  It is the same as `git merge -m <msg> <commit>...`.

With the update to "git merge" to make it understand what is
recorded in FETCH_HEAD directly, including Octopus merge cases, we
now can rewrite the use of this syntax in "git pull" with a simple
"git merge FETCH_HEAD".

Also there are quite a few fallouts in the test scripts, and it
turns out that "git cvsimport" also uses this old syntax to record
a merge.

Judging from this result, I would not be surprised if dropping the
support of the old syntax broke scripts people have written and been
relying on for the past ten years.  But at least we can start the
deprecation process by throwing a warning message when the syntax is
used.

With luck, we might be able to drop the support in a few years.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agomerge: handle FETCH_HEAD internally
Junio C Hamano [Sun, 26 Apr 2015 01:47:21 +0000 (18:47 -0700)]
merge: handle FETCH_HEAD internally

The collect_parents() function now is responsible for

 1. parsing the commits given on the command line into a list of
    commits to be merged;

 2. filtering these parents into independent ones; and

 3. optionally calling fmt_merge_msg() via prepare_merge_message()
    to prepare an auto-generated merge log message, using fake
    contents that FETCH_HEAD would have had if these commits were
    fetched from the current repository with "git pull . $args..."

Make "git merge FETCH_HEAD" to be the same as the traditional

    git merge "$(git fmt-merge-msg <.git/FETCH_HEAD)" $commits

invocation of the command in "git pull", where $commits are the ones
that appear in FETCH_HEAD that are not marked as not-for-merge, by
making it do a bit more, specifically:

 - noticing "FETCH_HEAD" is the only "commit" on the command line
   and picking the commits that are not marked as not-for-merge as
   the list of commits to be merged (substitute for step #1 above);

 - letting the resulting list fed to step #2 above;

 - doing the step #3 above, using the contents of the FETCH_HEAD
   instead of fake contents crafted from the list of commits parsed
   in the step #1 above.

Note that this changes the semantics.  "git merge FETCH_HEAD" has
always behaved as if the first commit in the FETCH_HEAD file were
directly specified on the command line, creating a two-way merge
whose auto-generated merge log said "merge commit xyz".  With this
change, if the previous fetch was to grab multiple branches (e.g.
"git fetch $there topic-a topic-b"), the new world order is to
create an octopus, behaving as if "git pull $there topic-a topic-b"
were run.  This is a deliberate change to make that happen, and
can be seen in the changes to t3033 tests.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agomerge: decide if we auto-generate the message early in collect_parents()
Junio C Hamano [Sun, 26 Apr 2015 01:39:43 +0000 (18:39 -0700)]
merge: decide if we auto-generate the message early in collect_parents()

Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agomerge: make collect_parents() auto-generate the merge message
Junio C Hamano [Sun, 26 Apr 2015 01:34:22 +0000 (18:34 -0700)]
merge: make collect_parents() auto-generate the merge message

Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agomerge: extract prepare_merge_message() logic out
Junio C Hamano [Sun, 26 Apr 2015 01:29:44 +0000 (18:29 -0700)]
merge: extract prepare_merge_message() logic out

Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agomerge: narrow scope of merge_names
Junio C Hamano [Sat, 25 Apr 2015 19:31:57 +0000 (12:31 -0700)]
merge: narrow scope of merge_names

In order to pass the list of parents to fmt_merge_msg(), cmd_merge()
uses this strbuf to create something that look like FETCH_HEAD that
describes commits that are being merged.  This is necessary only
when we are creating the merge commit message ourselves, but was
done unconditionally.

Move the variable and the logic to populate it to confine them in a
block that needs them.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agomerge: split reduce_parents() out of collect_parents()
Junio C Hamano [Sat, 25 Apr 2015 19:00:14 +0000 (12:00 -0700)]
merge: split reduce_parents() out of collect_parents()

The latter does two separate things:

 - Parse the list of commits on the command line, and formulate the
   list of commits to be merged (including the current HEAD);

 - Compute the list of parents to be recorded in the resulting merge
   commit.

Split the latter into a separate helper function, so that we can
later supply the list commits to be merged from a different source
(namely, FETCH_HEAD).

Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agomerge: clarify collect_parents() logic
Junio C Hamano [Sat, 25 Apr 2015 17:25:43 +0000 (10:25 -0700)]
merge: clarify collect_parents() logic

Clarify this small function in three ways.

 - The function initially collects all commits to be merged into a
   commit_list "remoteheads"; the "remotes" pointer always points at
   the tail of this list (either the remoteheads variable itself, or
   the ->next slot of the element at the end of the list) to help
   elongate the list by repeated calls to commit_list_insert().
   Because the new element appended by commit_list_insert() will
   always have its ->next slot NULLed out, there is no need for us
   to assign NULL to *remotes to terminate the list at the end.

 - The variable "head_subsumed" always confused me every time I read
   this code.  What is happening here is that we inspect what the
   caller told us to merge (including the current HEAD) and come up
   with the list of parents to be recorded for the resulting merge
   commit, omitting commits that are ancestor of other commits.
   This filtering may remove the current HEAD from the resulting
   parent list---and we signal that fact with this variable, so that
   we can later record it as the first parent when "--no-ff" is in
   effect.

 - The "parents" list is created for this function by reduce_heads()
   and was not deallocated after its use, even though the loop
   control was written in such a way to allow us to do so by taking
   the "next" element in a separate variable so that it can be used
   in the next-step part of the loop control.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agomerge: small leakfix and code simplification
Junio C Hamano [Thu, 23 Apr 2015 21:37:13 +0000 (14:37 -0700)]
merge: small leakfix and code simplification

When parsing a merged object name like "foo~20" to formulate a merge
summary "Merge branch foo (early part)", a temporary strbuf is used,
but we forgot to deallocate it when we failed to find the named
branch.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agomerge: do not check argc to determine number of remote heads
Junio C Hamano [Thu, 23 Apr 2015 20:56:34 +0000 (13:56 -0700)]
merge: do not check argc to determine number of remote heads

To reject merging multiple commits into an unborn branch, we check
argc, thinking that collect_parents() that reads the remaining
command line arguments from <argc, argv> will give us the same
number of commits as its input, i.e. argc.

Because what we really care about is the number of commits, let the
function run and then make sure it returns only one commit instead.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agomerge: clarify "pulling into void" special case
Junio C Hamano [Thu, 23 Apr 2015 20:46:44 +0000 (13:46 -0700)]
merge: clarify "pulling into void" special case

Instead of having it as one of the three if/elseif/.. case arms,
test the condition and handle this special case upfront.  This makes
it easier to follow the flow of logic.

Signed-off-by: Junio C Hamano <gitster@pobox.com>