git
8 years agogrep/pcre: support utf-8
Nguyễn Thái Ngọc Duy [Sat, 25 Jun 2016 05:22:35 +0000 (07:22 +0200)] 
grep/pcre: support utf-8

In the previous change in this function, we add locale support for
single-byte encodings only. It looks like pcre only supports utf-* as
multibyte encodings, the others are left in the cold (which is
fine).

We need to enable PCRE_UTF8 so pcre can find character boundary
correctly. It's needed for case folding (when --ignore-case is used)
or '*', '+' or similar syntax is used.

The "has_non_ascii()" check is to be on the conservative side. If
there's non-ascii in the pattern, the searched content could still be
in utf-8, but we can treat it just like a byte stream and everything
should work. If we force utf-8 based on locale only and pcre validates
utf-8 and the file content is in non-utf8 encoding, things break.

Noticed-by: Plamen Totev <plamen.totev@abv.bg>
Helped-by: Plamen Totev <plamen.totev@abv.bg>
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agogettext: add is_utf8_locale()
Nguyễn Thái Ngọc Duy [Sat, 25 Jun 2016 05:22:34 +0000 (07:22 +0200)] 
gettext: add is_utf8_locale()

This function returns true if git is running under an UTF-8
locale. pcre in the next patch will need this.

is_encoding_utf8() is used instead of strcmp() to catch both "utf-8"
and "utf8" suffixes.

When built with no gettext support, we peek in several env variables
to detect UTF-8. pcre library might support utf-8 even if libc is
built without locale support.. The peeking code is a copy from
compat/regex/regcomp.c

Helped-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agogrep/pcre: prepare locale-dependent tables for icase matching
Nguyễn Thái Ngọc Duy [Sat, 25 Jun 2016 05:22:33 +0000 (07:22 +0200)] 
grep/pcre: prepare locale-dependent tables for icase matching

The default tables are usually built with C locale and only suitable
for LANG=C or similar.  This should make case insensitive search work
correctly for all single-byte charsets.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agogrep: rewrite an if/else condition to avoid duplicate expression
Nguyễn Thái Ngọc Duy [Sat, 25 Jun 2016 05:22:32 +0000 (07:22 +0200)] 
grep: rewrite an if/else condition to avoid duplicate expression

"!icase || ascii_only" is repeated twice in this if/else chain as this
series evolves. Rewrite it (and basically revert the first if
condition back to before the "grep: break down an "if" stmt..." commit).

Helped-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agogrep/icase: avoid kwsset when -F is specified
Nguyễn Thái Ngọc Duy [Sat, 25 Jun 2016 05:22:31 +0000 (07:22 +0200)] 
grep/icase: avoid kwsset when -F is specified

Similar to the previous commit, we can't use kws on icase search
outside ascii range. But we can't simply pass the pattern to
regcomp/pcre like the previous commit because it may contain regex
special characters, so we need to quote the regex first.

To avoid misquote traps that could lead to undefined behavior, we
always stick to basic regex engine in this case. We don't need fancy
features for grepping a literal string anyway.

basic_regex_quote_buf() assumes that if the pattern is in a multibyte
encoding, ascii chars must be unambiguously encoded as single
bytes. This is true at least for UTF-8. For others, let's wait until
people yell up. Chances are nobody uses multibyte, non utf-8 charsets
anymore.

Noticed-by: Plamen Totev <plamen.totev@abv.bg>
Helped-by: René Scharfe <l.s.r@web.de>
Helped-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agocommit -S: avoid invalid pointer with empty message
Johannes Schindelin [Wed, 29 Jun 2016 14:14:54 +0000 (16:14 +0200)] 
commit -S: avoid invalid pointer with empty message

While it is not recommended, fsck.c says:

Not having a body is not a crime [...]

... which means that we cannot assume that the commit buffer
contains an empty line to separate header from body.  A commit
object with only a header without any body, not even without
a blank line after the header, is valid.

So let's tread carefully here.  strstr("\n\n") may find nothing
and return NULL.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoreset --hard: skip blank lines when reporting the commit subject
Johannes Schindelin [Wed, 29 Jun 2016 14:14:50 +0000 (16:14 +0200)] 
reset --hard: skip blank lines when reporting the commit subject

When there are blank lines at the beginning of a commit message, the
pretty printing machinery already skips them when showing a commit
subject (or the complete commit message). We shall henceforth do the
same when reporting the commit subject after the user called

git reset --hard <commit>

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agosequencer: use skip_blank_lines() to find the commit subject
Johannes Schindelin [Wed, 29 Jun 2016 14:14:46 +0000 (16:14 +0200)] 
sequencer: use skip_blank_lines() to find the commit subject

Just like we already taught the find_commit_subject() function (to make
it consistent with the code in pretty.c), we now simply skip leading
blank lines of the commit message.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agocommit -C: skip blank lines at the beginning of the message
Johannes Schindelin [Wed, 29 Jun 2016 14:14:42 +0000 (16:14 +0200)] 
commit -C: skip blank lines at the beginning of the message

Consistent with the pretty-printing machinery, we skip leading blank
lines (if any) of existing commit messages.

While Git itself only produces commit objects with a single empty line
between commit header and commit message, it is legal to have more than
one blank line (i.e. lines containing only white space, or no
characters) at the beginning of the commit message, and the
pretty-printing code already handles that.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agot3404: fix a grammo (commands are ran -> commands are run)
Johannes Schindelin [Wed, 29 Jun 2016 14:31:03 +0000 (16:31 +0200)] 
t3404: fix a grammo (commands are ran -> commands are run)

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agorebase -i: restore autostash on abort
Patrick Steinhardt [Wed, 29 Jun 2016 06:21:27 +0000 (08:21 +0200)] 
rebase -i: restore autostash on abort

When we abort an interactive rebase we do so by calling
`die_abort`, which cleans up after us by removing the rebase
state directory. If the user has requested to use the autostash
feature, though, the state directory may also contain a reference
to the autostash, which will now be deleted.

Fix the issue by trying to re-apply the autostash in `die_abort`.
This will also handle the case where the autostash does not apply
cleanly anymore by recording it in a user-visible stash.

Reported-by: Daniel Hahler <git@thequod.de>
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agocontrib/subtree: Add a test for subtree rebase that loses commits
David A. Greene [Tue, 28 Jun 2016 10:54:02 +0000 (05:54 -0500)] 
contrib/subtree: Add a test for subtree rebase that loses commits

This test merges an external tree in as a subtree, makes some commits
on top of it and splits it back out.  In the process the added commits
are lost or the rebase aborts with an internal error.  The tests are
marked to expect failure so that we don't forget to fix it.

Signed-off-by: David A. Greene <greened@obbligato.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agocolor.h: remove obsolete comment about limitations on Windows
Johannes Schindelin [Tue, 28 Jun 2016 10:02:49 +0000 (12:02 +0200)] 
color.h: remove obsolete comment about limitations on Windows

Originally, ANSI color sequences were supported on Windows only by
overriding the printf() and fprintf() functions, as mentioned in e7821d7
(Add a notice that only certain functions can print color escape codes,
2009-11-27).

As of eac14f8 (Win32: Thread-safe windows console output, 2012-01-14),
however, this is no longer the case, as the ANSI color sequence support
code needed to be replaced with a thread-safe version, one side effect
being that stdout and stderr handled no matter which function is used to
write to it.

So let's just remove the comment that is now obsolete.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agodoc: typeset HEAD and variants as literal
Matthieu Moy [Tue, 28 Jun 2016 11:40:15 +0000 (13:40 +0200)] 
doc: typeset HEAD and variants as literal

This is an application of the newly added CodingGuidelines to HEAD and
variants like FETCH_HEAD. It was obtained with:

  perl -pi -e "s/'([A-Z_]*HEAD)'/\`\$1\`/g" *.txt

Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoCodingGuidelines: formatting HEAD in documentation
Matthieu Moy [Tue, 28 Jun 2016 11:40:14 +0000 (13:40 +0200)] 
CodingGuidelines: formatting HEAD in documentation

The current practice is:

git/Documentation$ git grep "'HEAD'" | wc -l
24
git/Documentation$ git grep "\`HEAD\`" | wc -l
66

Let's adopt the majority as a guideline.

Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agodoc: typeset long options with argument as literal
Matthieu Moy [Tue, 28 Jun 2016 11:40:13 +0000 (13:40 +0200)] 
doc: typeset long options with argument as literal

We previously reformatted '--option' to `--option`. This patch reformats
'--option <arg>' to `--option <arg>`. Obtained with:

  perl -pi -e "s/'(--[a-z][a-z=<>-]* <[^>]*>)'/\`\$1\`/g" *.txt

Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agodoc: typeset '--' as literal
Matthieu Moy [Tue, 28 Jun 2016 11:40:12 +0000 (13:40 +0200)] 
doc: typeset '--' as literal

This was obtained with:

  perl -pi -e "s/'--'/\`--\`/g" *.txt

Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agodoc: typeset long command-line options as literal
Matthieu Moy [Tue, 28 Jun 2016 11:40:11 +0000 (13:40 +0200)] 
doc: typeset long command-line options as literal

Similarly to the previous commit, use backquotes instead of
forward-quotes, for long options.

This was obtained with:

  perl -pi -e "s/'(--[a-z][a-z=<>-]*)'/\`\$1\`/g" *.txt

and manual tweak to remove false positive in ascii-art (o'--o'--o' to
describe rewritten history).

Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agodoc: typeset short command-line options as literal
Matthieu Moy [Tue, 28 Jun 2016 11:40:10 +0000 (13:40 +0200)] 
doc: typeset short command-line options as literal

It was common in our documentation to surround short option names with
forward quotes, which renders as italic in HTML. Instead, use backquotes
which renders as monospace. This is one more step toward conformance to
Documentation/CodingGuidelines.

This was obtained with:

  perl -pi -e "s/'(-[a-z])'/\`\$1\`/g" *.txt

Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoDocumentation/git-mv.txt: fix whitespace indentation
Matthieu Moy [Tue, 28 Jun 2016 11:40:09 +0000 (13:40 +0200)] 
Documentation/git-mv.txt: fix whitespace indentation

Replace spaces with tabs to avoid a warning when further patches change
these lines.

Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoStart preparing for 2.9.1
Junio C Hamano [Mon, 27 Jun 2016 16:59:51 +0000 (09:59 -0700)] 
Start preparing for 2.9.1

Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoMerge branch 'rs/xdiff-hunk-with-func-line' into maint
Junio C Hamano [Mon, 27 Jun 2016 16:56:24 +0000 (09:56 -0700)] 
Merge branch 'rs/xdiff-hunk-with-func-line' into maint

"git show -W" (extend hunks to cover the entire function, delimited
by lines that match the "funcname" pattern) used to show the entire
file when a change added an entire function at the end of the file,
which has been fixed.

* rs/xdiff-hunk-with-func-line:
  xdiff: fix merging of appended hunk with -W
  grep: -W: don't extend context to trailing empty lines
  t7810: add test for grep -W and trailing empty context lines
  xdiff: don't trim common tail with -W
  xdiff: -W: don't include common trailing empty lines in context
  xdiff: ignore empty lines before added functions with -W
  xdiff: handle appended chunks better with -W
  xdiff: factor out match_func_rec()
  t4051: rewrite, add more tests

8 years agoMerge branch 'jk/rev-list-count-with-bitmap' into maint
Junio C Hamano [Mon, 27 Jun 2016 16:56:24 +0000 (09:56 -0700)] 
Merge branch 'jk/rev-list-count-with-bitmap' into maint

"git rev-list --count" whose walk-length is limited with "-n"
option did not work well with the counting optimized to look at the
bitmap index.

* jk/rev-list-count-with-bitmap:
  rev-list: disable bitmaps when "-n" is used with listing objects
  rev-list: "adjust" results of "--count --use-bitmap-index -n"

8 years agoMerge branch 'et/pretty-format-c-auto' into maint
Junio C Hamano [Mon, 27 Jun 2016 16:56:23 +0000 (09:56 -0700)] 
Merge branch 'et/pretty-format-c-auto' into maint

The commands in `git log` family take %C(auto) in a custom format
string.  This unconditionally turned the color on, ignoring
--no-color or with --color=auto when the output is not connected to
a tty; this was corrected to make the format truly behave as
"auto".

* et/pretty-format-c-auto:
  format_commit_message: honor `color=auto` for `%C(auto)`

8 years agoMerge branch 'ew/daemon-socket-keepalive' into maint
Junio C Hamano [Mon, 27 Jun 2016 16:56:22 +0000 (09:56 -0700)] 
Merge branch 'ew/daemon-socket-keepalive' into maint

When "git daemon" is run without --[init-]timeout specified, a
connection from a client that silently goes offline can hang around
for a long time, wasting resources.  The socket-level KEEPALIVE has
been enabled to allow the OS to notice such failed connections.

* ew/daemon-socket-keepalive:
  daemon: enable SO_KEEPALIVE for all sockets

8 years agoxwrite: poll on non-blocking FDs
Eric Wong [Sun, 26 Jun 2016 23:21:12 +0000 (23:21 +0000)] 
xwrite: poll on non-blocking FDs

write(2) can hit the same EAGAIN/EWOULDBLOCK errors as read(2),
so busy-looping on a non-blocking FD is a waste of resources.

Currently, I do not know of a way for this happen:

* the NonBlocking directive in systemd does not apply to stdin,
  stdout, or stderr.

* xinetd provides no way to set the non-blocking flag at all

But theoretically, it's possible a careless C10K HTTP server
could use pipe2(..., O_NONBLOCK) to setup a pipe for
git-http-backend with only the intent to use non-blocking reads;
but accidentally leave non-blocking set on the write end passed
as stdout to git-upload-pack.

Followup-to: 1079c4be0b720 ("xread: poll on non blocking fds")

Signed-off-by: Eric Wong <e@80x24.org>
Reviewed-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoxread: retry after poll on EAGAIN/EWOULDBLOCK
Eric Wong [Mon, 27 Jun 2016 03:56:35 +0000 (03:56 +0000)] 
xread: retry after poll on EAGAIN/EWOULDBLOCK

We should continue to loop after EAGAIN/EWOULDBLOCK as the
intent of xread is to try until there is available data,
EOF, or an unrecoverable error.

Fixes: 1079c4be0b720 ("xread: poll on non blocking fds")

Signed-off-by: Eric Wong <e@80x24.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agogc: fix off-by-one error with gc.autoPackLimit
Eric Wong [Sat, 25 Jun 2016 06:46:47 +0000 (06:46 +0000)] 
gc: fix off-by-one error with gc.autoPackLimit

This matches the documentation and allows gc.autoPackLimit=1
to maintain a single pack without attempting a repack on every
"git gc --auto" invocation.

Signed-off-by: Eric Wong <e@80x24.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agogrep/icase: avoid kwsset on literal non-ascii strings
Nguyễn Thái Ngọc Duy [Sat, 25 Jun 2016 05:22:30 +0000 (07:22 +0200)] 
grep/icase: avoid kwsset on literal non-ascii strings

When we detect the pattern is just a literal string, we avoid heavy
regex engine and use fast substring search implemented in kwsset.c.
But kws uses git-ctype which is locale-independent so it does not know
how to fold case properly outside ascii range. Let regcomp or pcre
take care of this case instead. Slower, but accurate.

Noticed-by: Plamen Totev <plamen.totev@abv.bg>
Helped-by: René Scharfe <l.s.r@web.de>
Helped-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agotest-regex: expose full regcomp() to the command line
Nguyễn Thái Ngọc Duy [Sat, 25 Jun 2016 05:22:29 +0000 (07:22 +0200)] 
test-regex: expose full regcomp() to the command line

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agotest-regex: isolate the bug test code
Nguyễn Thái Ngọc Duy [Sat, 25 Jun 2016 05:22:28 +0000 (07:22 +0200)] 
test-regex: isolate the bug test code

This is in preparation to turn test-regex into some generic regex
testing command.

Helped-by: Eric Sunshine <sunshine@sunshineco.com>
Helped-by: Ramsay Jones <ramsay@ramsayjones.plus.com>
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agogrep: break down an "if" stmt in preparation for next changes
Nguyễn Thái Ngọc Duy [Sat, 25 Jun 2016 05:22:27 +0000 (07:22 +0200)] 
grep: break down an "if" stmt in preparation for next changes

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agonew-command.txt: correct the command description file
Nguyễn Thái Ngọc Duy [Sat, 25 Jun 2016 13:55:14 +0000 (15:55 +0200)] 
new-command.txt: correct the command description file

It has always been command-list.txt even at the time this
new-command.txt document is added.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agot2300: "git --exec-path" is not usable in $PATH on Windows as-is
Johannes Schindelin [Sat, 18 Jun 2016 10:49:11 +0000 (12:49 +0200)] 
t2300: "git --exec-path" is not usable in $PATH on Windows as-is

The "git" command prepends the exec-path to the PATH environment
variable for processes it spawns.  That is how ". git-sh-setup" in
our scripted Porcelains can find the dot-sourced file in the
exec-path location that is not usually on user's PATH.

When t2300 runs, because it is not spawned by the "git" command, the
scriptlet being tested did not run with a realistic setting of PATH
environment.  It lacked the exec-path on the PATH, and failed to
find the dot-sourced file.  A recent update to t2300 attempted to
fix this, with "PATH=$(git --exec-path):$PATH", which has been the
recommended way around v1.6.0 days (a script whose original was
written before that release that survives to this day is likely to
have such a line).

However, the "git --exec-path" command outputs C:\path\to\exec\dir
(not /c/path/to/exec/dir) on Windows; the recent update failed to
consider the problem that comes from it.

Even though Git itself, when doing the equivalent internally, does
so in a platform native way (i.e. on Windows, C:\path\to\exec\dir is
prepended to the existing value of %PATH% using ';' as a component
separator), the result is further massaged by bash and gets turned
into $PATH that uses /c/path/to/exec/dir with ':' separating the
components, which is the form understood by bash, so scripted
Porcelains find commands from PATH correctly.

An end user script written in shell, however, cannot prepend
"C:\path\to\exec\dir:" to the existing value of $PATH and expect
bash to magically turn it into the form it understands.  In other
words, "PATH=$(git --exec-path):$PATH" does not work as an emulation
of what "Git" internally does to the PATH on Windows.

To correctly emulate how exec-path is prepended to the PATH
environment internally on Windows, we'd need to convert
C:\git-sdk-64\usr\src\git to at least /c\git-sdk-64\usr\src\git
ourselves before prepending it to PATH.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agocommit.c: make find_commit_subject() more robust
Johannes Schindelin [Wed, 22 Jun 2016 20:20:20 +0000 (22:20 +0200)] 
commit.c: make find_commit_subject() more robust

Just like the pretty printing machinery, we should simply ignore
blank lines at the beginning of the commit messages.

This discrepancy was noticed when an early version of the
rebase--helper produced commit objects with more than one empty line
between the header and the commit message.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agopretty: make the skip_blank_lines() function public
Johannes Schindelin [Wed, 22 Jun 2016 20:20:16 +0000 (22:20 +0200)] 
pretty: make the skip_blank_lines() function public

This function will be used also in the find_commit_subject()
function.

While at it, rename the function to reflect that it skips not only
empty lines, but any lines consisting of only whitespace, too.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agodoc: git-htmldocs.googlecode.com is no more
Jonathan Nieder [Wed, 22 Jun 2016 17:38:25 +0000 (10:38 -0700)] 
doc: git-htmldocs.googlecode.com is no more

http://git-htmldocs.googlecode.com/git/git.html says

 There was no service found for the uri requested.

Link to the rendered documentation on Jekyll instead.

Reported-by: Andrea Stacchiotti <andreastacchiotti@gmail.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agogit-p4: correct hasBranchPrefix verbose output
Andrew Oakley [Wed, 22 Jun 2016 09:26:11 +0000 (10:26 +0100)] 
git-p4: correct hasBranchPrefix verbose output

The logic here was inverted, you got a message saying the file is
ignored for each file that is not ignored.

Signed-off-by: Andrew Oakley <aoakley@roku.com>
Acked-by: Lars Schneider <larsxschneider@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agot7810: fix duplicated test title
Charles Bailey [Tue, 21 Jun 2016 21:14:11 +0000 (14:14 -0700)] 
t7810: fix duplicated test title

Signed-off-by: Charles Bailey <charles@hashpling.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agot5614: don't use subshells
Stefan Beller [Mon, 20 Jun 2016 17:21:18 +0000 (10:21 -0700)] 
t5614: don't use subshells

Using a subshell for just one git command is both a waste in compute
overhead (create a new process) as well as in line count.

Suggested-by: Jeff King <peff@peff.net>
Signed-off-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agot7800: readlink may not be available
Armin Kunaschik [Tue, 31 May 2016 00:26:12 +0000 (02:26 +0200)] 
t7800: readlink may not be available

The readlink(1) command is not available on all platforms (notably
not on AIX and HP-UX) and can be replaced in this test with the
"workaround"

ls -ld <name> | sed -e 's/.* -> //'

This is no universal readlink replacement but works in the
controlled test environment well enough.

Signed-off-by: Armin Kunaschik <megabreit@googlemail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoperf: accommodate for MacOSX
Johannes Schindelin [Tue, 21 Jun 2016 13:53:43 +0000 (15:53 +0200)] 
perf: accommodate for MacOSX

As this developer has no access to MacOSX developer setups anymore,
Travis becomes the best bet to run performance tests on that OS.

However, on MacOSX /usr/bin/time is that good old BSD executable that
no Linux user cares about, as demonstrated by the perf-lib.sh's use
of GNU-ish extensions. And by the hard-coded path.

Let's just work around this issue by using gtime on MacOSX, the
Homebrew-provided GNU implementation onto which pretty much every
MacOSX power user falls back anyway.

To help other developers use Travis to run performance tests on
MacOSX, the .travis.yml file now sports a commented-out line that
installs GNU time via Homebrew.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Reviewed-by: Lars Schneider <larsxschneider@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agolocal_tzoffset: detect errors from tm_to_time_t
Jeff King [Mon, 20 Jun 2016 21:14:14 +0000 (17:14 -0400)] 
local_tzoffset: detect errors from tm_to_time_t

When we want to know the local timezone offset at a given
timestamp, we compute it by asking for localtime() at the
given time, and comparing the offset to GMT at that time.
However, there's some juggling between time_t and "struct
tm" which happens, which involves calling our own
tm_to_time_t().

If that function returns an error (e.g., because it only
handles dates up to the year 2099), it returns "-1", which
we treat as a time_t, and is clearly bogus, leading to
bizarre timestamps (that seem to always adjust the time back
to (time_t)(uint32_t)-1, in the year 2106).

It's not a good idea for local_tzoffset() to simply die
here; it would make it hard to run "git log" on a repository
with funny timestamps. Instead, let's just treat such cases
as "zero offset".

Reported-by: Norbert Kiesel <nkiesel@gmail.com>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agot0006: test various date formats
Jeff King [Mon, 20 Jun 2016 21:11:59 +0000 (17:11 -0400)] 
t0006: test various date formats

We ended up testing some of these date formats throughout
the rest of the suite (e.g., via for-each-ref's
"$(authordate:...)" format), but we never did so
systematically. t0006 is the right place for unit-testing of
our date-handling code.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agot0006: rename test-date's "show" to "relative"
Jeff King [Mon, 20 Jun 2016 21:10:29 +0000 (17:10 -0400)] 
t0006: rename test-date's "show" to "relative"

The "show" tests are really only checking relative formats;
we should make that more clear.

This also frees up the "show" name to later check other
formats. We could later fold "relative" into a more generic
"show" command, but it's not worth it.  Relative times are a
special case already because we have to munge the concept of
"now" in our tests.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agomingw: let the build succeed with DEVELOPER=1
Johannes Schindelin [Sat, 18 Jun 2016 12:38:36 +0000 (14:38 +0200)] 
mingw: let the build succeed with DEVELOPER=1

The recently introduced developer flags identified a couple of
old-style function declarations in the Windows-specific code where
the parameter list was left empty instead of specifying "void"
explicitly. Let's just fix them.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoclone: do not let --depth imply --shallow-submodules
Junio C Hamano [Sun, 19 Jun 2016 20:51:56 +0000 (13:51 -0700)] 
clone: do not let --depth imply --shallow-submodules

In v2.9.0, we prematurely flipped the default to force cloning
submodules shallowly, when the superproject is getting cloned
shallowly.  This is likely to fail when the upstream repositories
submodules are cloned from a repository that is not prepared to
serve histories that ends at a commit that is not at the tip of a
branch, and we know the world is not yet ready.

Use a safer default to clone the submodules fully, unless the user
tells us that she knows that the upstream repository of the
submodules are willing to cooperate with "--shallow-submodules"
option.

Noticed-by: Vadim Eisenberg <VADIME@il.ibm.com>
Helped-by: Jeff King <peff@peff.net>
Helped-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agosh-setup: enclose setting of ${VAR=default} in double-quotes
LE Manh Cuong [Sat, 18 Jun 2016 20:26:03 +0000 (03:26 +0700)] 
sh-setup: enclose setting of ${VAR=default} in double-quotes

We often make sure an environment variable is set to
something, either set by the user (in which case we do not
molest it) or set it to our default value (otherwise), with

: ${VAR=default value}

i.e. running the no-op command ":" with ${VAR} as its
parameters (or the default value we supply), relying on that
":" is a no-op.

This pattern, even though it is no-op from correctness point
of view, still can be expensive if the existing value in VAR
has shell glob (because they will be expanded against
filesystem entities) and IFS whitespaces (because the value
need to be split into multiple parameters).  Our invocation
of ":" command does not care if the parameter given to it is
after the value in VAR goes through these processing.

Enclosing the whole thing in double-quote, i.e.

: "${VAR=default value}"

avoids paying the unnecessary cost, so let's do so.

Signed-off-by: LE Manh Cuong <cuong.manhle.vn@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoDocumentation/technical: signed merge tag format
Michael J Gruber [Fri, 17 Jun 2016 07:46:11 +0000 (09:46 +0200)] 
Documentation/technical: signed merge tag format

Signed-off-by: Michael J Gruber <git@drmicha.warpmail.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoDocumentation/technical: signed commit format
Michael J Gruber [Fri, 17 Jun 2016 07:46:10 +0000 (09:46 +0200)] 
Documentation/technical: signed commit format

Signed-off-by: Michael J Gruber <git@drmicha.warpmail.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoDocumentation/technical: signed tag format
Michael J Gruber [Fri, 17 Jun 2016 07:46:09 +0000 (09:46 +0200)] 
Documentation/technical: signed tag format

Signed-off-by: Michael J Gruber <git@drmicha.warpmail.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoDocumentation/technical: describe signature formats
Michael J Gruber [Fri, 17 Jun 2016 07:46:08 +0000 (09:46 +0200)] 
Documentation/technical: describe signature formats

We use different types of signature formats in different places.
Set up the infrastructure and overview to describe them systematically
in our technical documentation.

Signed-off-by: Michael J Gruber <git@drmicha.warpmail.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agorebase: update comment about FreeBSD /bin/sh
Ed Maste [Fri, 17 Jun 2016 15:33:29 +0000 (11:33 -0400)] 
rebase: update comment about FreeBSD /bin/sh

Commit 9f50d32 introduced a fix for FreeBSD /bin/sh misbehaviour
when dot-sourcing a file containing "return" statements outside of
any function, from a function in another shell script. That issue
affects FreeBSD 9.x, and is not present in the /bin/sh in FreeBSD
10.3 and later. Update the comment to clarify this.

The example from 9f50d32's commit message produces the expected output
on FreeBSD 10.3 and -CURRENT (the upcoming 11.0):

% sh script1.sh
only this line should show
%

Signed-off-by: Ed Maste <emaste@freebsd.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoDocumentation: GPG capitalization
Dave Nicolson [Thu, 16 Jun 2016 22:15:44 +0000 (22:15 +0000)] 
Documentation: GPG capitalization

When "GPG" is used in a sentence it is now consistently capitalized.
When referring to the binary it is left as "gpg".

Signed-off-by: David Nicolson <david.nicolson@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobisect: always call setup_revisions after init_revisions
Jeff King [Thu, 16 Jun 2016 23:37:20 +0000 (19:37 -0400)] 
bisect: always call setup_revisions after init_revisions

init_revisions() initializes the rev_info struct to default
values, and setup_revisions() parses any command-line
arguments and finalizes the struct.

In e22278c (bisect: display first bad commit without forking
a new process, 2009-05-28), a show_diff_tree() was added
that calls the former but not the latter. It doesn't have
any arguments to parse, but it still should do the
finalizing step.

This may have caused other minor bugs over the years, but it
became much more prominent after fe37a9c (pretty: allow
tweaking tabwidth in --expand-tabs, 2016-03-29). That leaves
the expected tab width as "-1", rather than the true default
of "8". When we see a commit with tabs to be expanded, we
end up trying to add (size_t)-1 spaces to a strbuf, which
complains about the integer overflow.

The fix is easy: just call setup_revisions() with no
arguments.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agopretty.c: support <direction>|(<negative number>) forms
Nguyễn Thái Ngọc Duy [Thu, 16 Jun 2016 13:18:38 +0000 (20:18 +0700)] 
pretty.c: support <direction>|(<negative number>) forms

%>|(num), %><|(num) and %<|(num), where num is a positive number, sets a
fixed column from the screen's left border. There is no way for us to
specifiy a column relative to the right border, which is useful when you
want to make use of all terminal space (on big screens). Use negative
num for that. Inspired by Go's array syntax (*).

(*) I know Python has this first (or before Go, at least) but the idea
didn't occur to me until I learned Go.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agopretty: pass graph width to pretty formatting for use in '%>|(N)'
Josef Kufner [Thu, 16 Jun 2016 13:18:37 +0000 (20:18 +0700)] 
pretty: pass graph width to pretty formatting for use in '%>|(N)'

Pass graph width to pretty formatting, to make N in '%>|(N)'
include columns consumed by graph rendered when --graph option
is in use.

For example, in the output of

  git log --all --graph --pretty='format: [%>|(20)%h] %ar%d'

this change will make all commit hashes align at 20th column from
the edge of the terminal, not from the edge of the graph.

Signed-off-by: Josef Kufner <josef@kufner.cz>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoupload-pack.c: make send_client_data() return void
Lukas Fleischer [Tue, 14 Jun 2016 14:49:17 +0000 (16:49 +0200)] 
upload-pack.c: make send_client_data() return void

The send_client_data() function uses write_or_die() for writing data
which immediately terminates the process on errors. If no such error
occurred, send_client_data() always returned the value that was passed
as third parameter prior to this commit. This value is already known to
the caller in any case, so let's turn send_client_data() into a void
function instead.

Signed-off-by: Lukas Fleischer <lfleischer@lfos.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agosideband.c: make send_sideband() return void
Lukas Fleischer [Tue, 14 Jun 2016 14:49:16 +0000 (16:49 +0200)] 
sideband.c: make send_sideband() return void

The send_sideband() function uses write_or_die() for writing data which
immediately terminates the process on errors. If no such error occurred,
send_sideband() always returned the value that was passed as fourth
parameter prior to this commit. This value is already known to the
caller in any case, so let's turn send_sideband() into a void function
instead.

Signed-off-by: Lukas Fleischer <lfleischer@lfos.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoadd--interactive: respect diff.compactionHeuristic
Jeff King [Thu, 16 Jun 2016 12:27:29 +0000 (08:27 -0400)] 
add--interactive: respect diff.compactionHeuristic

We use plumbing to generate the diff, so it doesn't
automatically pick up UI config like compactionHeuristic.
Let's forward it on, since interactive adding is porcelain.

Note that we only need to handle the "true" case. There's no
point in passing --no-compaction-heuristic when the variable
is false, since nothing else could have turned it on.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agogit-svn: document the 'git svn propset' command
Alfred Perlstein [Wed, 15 Jun 2016 05:19:50 +0000 (22:19 -0700)] 
git-svn: document the 'git svn propset' command

Add example usage to the git-svn documentation.

Reported-by: Joseph Pecoraro <pecoraro@apple.com>
Signed-off-by: Alfred Perlstein <alfred@freebsd.org>
Reviewed-by: Eric Wong <e@80x24.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoblame, line-log: do not loop around deref_tag()
Junio C Hamano [Tue, 14 Jun 2016 20:38:14 +0000 (13:38 -0700)] 
blame, line-log: do not loop around deref_tag()

These callers appear to expect that deref_tag() is to peel one layer
of a tag, but the function does not work that way; it has its own
loop to unwrap tags until an object that is not a tag appears.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agognome-keyring: Don't hard-code pkg-config executable
Heiko Becker [Tue, 14 Jun 2016 11:27:05 +0000 (13:27 +0200)] 
gnome-keyring: Don't hard-code pkg-config executable

Helpful if your pkg-config executable has a prefix based on the
architecture, for example.

Signed-off-by: Heiko Becker <heirecka@exherbo.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/fetch.c: don't free remote->name after fetch
Keith McGuigan [Tue, 14 Jun 2016 18:28:56 +0000 (14:28 -0400)] 
builtin/fetch.c: don't free remote->name after fetch

Make fetch's string_list of remote names own all of its string items
(strdup'ing when necessary) so that it can deallocate them safely
when clearing.

Signed-off-by: Keith McGuigan <kmcguigan@twopensource.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agostrbuf: describe the return value of strbuf_read_file
Pranit Bauva [Tue, 14 Jun 2016 06:14:11 +0000 (11:44 +0530)] 
strbuf: describe the return value of strbuf_read_file

Mentored-by: Lars Schneider <larsxschneider@gmail.com>
Mentored-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Pranit Bauva <pranit.bauva@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agofetch: document that pruning happens before fetching
Jeff King [Mon, 13 Jun 2016 23:58:51 +0000 (19:58 -0400)] 
fetch: document that pruning happens before fetching

This was changed in 10a6cc8 (fetch --prune: Run prune before
fetching, 2014-01-02), but it seems that nobody in that
discussion realized we were advertising the "after"
explicitly.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoGit 2.9 v2.9.0
Junio C Hamano [Mon, 13 Jun 2016 17:42:13 +0000 (10:42 -0700)] 
Git 2.9

Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoMerge tag 'l10n-2.9.0-rc0' of git://github.com/git-l10n/git-po
Junio C Hamano [Mon, 13 Jun 2016 01:00:57 +0000 (18:00 -0700)] 
Merge tag 'l10n-2.9.0-rc0' of git://github.com/git-l10n/git-po

l10n-2.9.0-rc0

* tag 'l10n-2.9.0-rc0' of git://github.com/git-l10n/git-po:
  l10n: ko.po: Update Korean translation
  l10n: ru.po: update Russian translation
  l10n: de.po: translate 104 new messages
  l10n: zh_CN: review for git v2.9.0 l10n round 1
  l10n: zh_CN: for git v2.9.0 l10n round 1
  l10n: pt_PT: update Portuguese translation
  l10n: pt_PT: update according to git-gui glossary
  l10n: pt_PT: merge git.pot file
  l10n: Updated Bulgarian translation of git (2597t,0f,0u)
  l10n: sv.po: Update Swedish translation (2597t0f0u)
  l10n: fr.po v2.9.0rnd1
  l10n: Updated Vietnamese translation (2597t)
  l10n: git.pot: v2.9.0 round 1 (104 new, 37 removed)
  l10n: fr.po Fixed grammar mistake

8 years agol10n: ko.po: Update Korean translation
Changwoo Ryu [Sat, 11 Jun 2016 16:25:58 +0000 (01:25 +0900)] 
l10n: ko.po: Update Korean translation

8 years agoMerge branch 'russian-l10n' of https://github.com/DJm00n/git-po-ru
Jiang Xin [Sat, 11 Jun 2016 12:21:52 +0000 (20:21 +0800)] 
Merge branch 'russian-l10n' of https://github.com/DJm00n/git-po-ru

* 'russian-l10n' of https://github.com/DJm00n/git-po-ru:
  l10n: ru.po: update Russian translation

8 years agol10n: ru.po: update Russian translation
Dimitriy Ryazantcev [Sat, 11 Jun 2016 09:53:43 +0000 (12:53 +0300)] 
l10n: ru.po: update Russian translation

Signed-off-by: Dimitriy Ryazantcev <dimitriy.ryazantcev@gmail.com>
8 years agoHopefully the final last-minute update before 2.9 final
Junio C Hamano [Fri, 10 Jun 2016 22:30:19 +0000 (15:30 -0700)] 
Hopefully the final last-minute update before 2.9 final

Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoMerge branch 'jk/diff-compact-heuristic'
Junio C Hamano [Fri, 10 Jun 2016 22:26:06 +0000 (15:26 -0700)] 
Merge branch 'jk/diff-compact-heuristic'

It turns out that the earlier effort to update the heuristics may
want to use a bit more time to mature.  Turn it off by default.

* jk/diff-compact-heuristic:
  diff: disable compaction heuristic for now

8 years agoMerge branch 'jk/shell-portability'
Junio C Hamano [Fri, 10 Jun 2016 22:26:04 +0000 (15:26 -0700)] 
Merge branch 'jk/shell-portability'

test fixes.

* jk/shell-portability:
  t5500 & t7403: lose bash-ism "local"
  test-lib: add in-shell "env" replacement

8 years agoMerge branch 'jc/t2300-setup'
Junio C Hamano [Fri, 10 Jun 2016 22:26:04 +0000 (15:26 -0700)] 
Merge branch 'jc/t2300-setup'

A test fix.

* jc/t2300-setup:
  t2300: run git-sh-setup in an environment that better mimics the real life

8 years agoconfig.c: fix misspelt "occurred" in an error message
Peter Colberg [Fri, 10 Jun 2016 19:05:26 +0000 (15:05 -0400)] 
config.c: fix misspelt "occurred" in an error message

Signed-off-by: Peter Colberg <peter@colberg.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agorefs.h: fix misspelt "occurred" in a comment
Peter Colberg [Fri, 10 Jun 2016 19:05:26 +0000 (15:05 -0400)] 
refs.h: fix misspelt "occurred" in a comment

Signed-off-by: Peter Colberg <peter@colberg.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agodiff: disable compaction heuristic for now
Junio C Hamano [Fri, 10 Jun 2016 17:58:55 +0000 (10:58 -0700)] 
diff: disable compaction heuristic for now

http://lkml.kernel.org/g/20160610075043.GA13411@sigill.intra.peff.net
reports that a change to add a new "function" with common ending
with the existing one at the end of the file is shown like this:

    def foo
      do_foo_stuff()

   +  common_ending()
   +end
   +
   +def bar
   +  do_bar_stuff()
   +
      common_ending()
    end

when the new heuristic is in use.  In reality, the change is to add
the blank line before "def bar" and everything below, which is what
the code without the new heuristic shows.

Disable the heuristics by default, and resurrect the documentation
for the option and the configuration variables, while clearly
marking the feature as still experimental.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agowrite_or_die: remove the unused write_or_whine() function
Ramsay Jones [Thu, 9 Jun 2016 22:52:22 +0000 (23:52 +0100)] 
write_or_die: remove the unused write_or_whine() function

Now the last caller of this function is gone, and new ones are
unlikely to appear, because this function is doing very little that
a regular if() does not besides obfuscating the error message (and
if we ever did want something like it, we would probably prefer the
function to come back with more "normal" return value semantics).

Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agol10n: de.po: translate 104 new messages
Ralf Thielow [Fri, 10 Jun 2016 16:00:46 +0000 (18:00 +0200)] 
l10n: de.po: translate 104 new messages

Translate 104 new messages came from git.pot update in f517e50
(l10n: git.pot: v2.9.0 round 1 (104 new, 37 removed)).

Signed-off-by: Ralf Thielow <ralf.thielow@gmail.com>
8 years agoxdiff: fix merging of appended hunk with -W
René Scharfe [Thu, 9 Jun 2016 21:54:48 +0000 (23:54 +0200)] 
xdiff: fix merging of appended hunk with -W

When -W is given we search the lines between the end of the current
context and the next change for a function line.  If there is none then
we merge those two hunks as they must be part of the same function.

If the next change is an appended chunk we abort the search early in
get_func_line(), however, because its line number is out of range.  Fix
that by searching from the end of the pre-image in that case instead.

Reported-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Rene Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoUse "working tree" instead of "working directory" for git status
Lars Vogel [Thu, 9 Jun 2016 18:19:30 +0000 (20:19 +0200)] 
Use "working tree" instead of "working directory" for git status

Working directory can be easily confused with the current directory.
In one of my patches I already updated the usage of working directory
with working tree for the man page but I noticed that git status also
uses this incorrect term.

Signed-off-by: Lars Vogel <Lars.Vogel@vogella.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/commit.c: memoize git-path for COMMIT_EDITMSG
Pranit Bauva [Tue, 24 May 2016 19:19:50 +0000 (00:49 +0530)] 
builtin/commit.c: memoize git-path for COMMIT_EDITMSG

This is a follow up commit for f932729c (memoize common git-path
"constant" files, 10-Aug-2015).

The many function calls to git_path() are replaced by
git_path_commit_editmsg() and which thus eliminates the need to repeatedly
compute the location of "COMMIT_EDITMSG".

Mentored-by: Lars Schneider <larsxschneider@gmail.com>
Mentored-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Pranit Bauva <pranit.bauva@gmail.com>
Reviewed-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agol10n: zh_CN: review for git v2.9.0 l10n round 1
Ray Chen [Sun, 5 Jun 2016 16:06:17 +0000 (00:06 +0800)] 
l10n: zh_CN: review for git v2.9.0 l10n round 1

Signed-off-by: Ray Chen <oldsharp@gmail.com>
Signed-off-by: Jiang Xin <worldhello.net@gmail.com>
8 years agosend-pack: use buffered I/O to talk to pack-objects
Jeff King [Wed, 8 Jun 2016 19:42:16 +0000 (15:42 -0400)] 
send-pack: use buffered I/O to talk to pack-objects

We start a pack-objects process and then write all of the
positive and negative sha1s to it over a pipe. We do so by
formatting each item into a fixed-size buffer and then
writing each individually. This has two drawbacks:

  1. There's some manual computation of the buffer size,
     which is not immediately obvious is correct (though it
     is).

  2. We write() once per sha1, which means a lot more system
     calls than are necessary.

We can solve both by wrapping the pipe descriptor in a stdio
handle; this is the same technique used by upload-pack when
serving fetches.

Note that we can also simplify and improve the error
handling here. The original detected a single write error
and broke out of the loop (presumably to avoid writing the
error message over and over), but never actually acted on
seeing an error; we just fed truncated input and took
whatever pack-objects returned.

In practice, this probably didn't matter, as the likely
errors would be caused by pack-objects dying (and we'd
probably just die with SIGPIPE anyway). But we can easily
make this simpler and more robust; the stdio handle keeps an
error flag, which we can check at the end.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agodoc: change configuration variables format
Tom Russello [Wed, 8 Jun 2016 17:23:16 +0000 (19:23 +0200)] 
doc: change configuration variables format

This change configuration variables that where in italic style
to monospace font according to the guideline. It was obtained with

grep '[[:alpha:]]*\.[[:alpha:]]*::$' config.txt | \
sed -e 's/::$//' -e 's/\./\\\\./' | \
xargs -iP perl -pi -e "s/\'P\'/\`P\`/g" ./*.txt

Signed-off-by: Tom Russello <tom.russello@grenoble-inp.org>
Signed-off-by: Erwan Mathoniere <erwan.mathoniere@grenoble-inp.org>
Signed-off-by: Samuel Groot <samuel.groot@grenoble-inp.org>
Signed-off-by: Matthieu Moy <matthieu.moy@grenoble-inp.fr>
Reviewed-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agodoc: more consistency in environment variables format
Tom Russello [Tue, 7 Jun 2016 22:35:07 +0000 (00:35 +0200)] 
doc: more consistency in environment variables format

Wrap with backticks (monospaced font) unwrapped or single-quotes wrapped
(italic type) environment variables which are followed by the word
"environment". It was obtained with:

perl -pi -e "s/\'?(\\\$?[0-9A-Z\_]+)\'?(?= environment ?)/\`\1\`/g" *.txt

One of the main purposes is to stick to the CodingGuidelines as possible so
that people writting new documentation by mimicking the existing are more likely
to have it right (even if they didn't read the CodingGuidelines).

Signed-off-by: Tom Russello <tom.russello@grenoble-inp.org>
Signed-off-by: Erwan Mathoniere <erwan.mathoniere@grenoble-inp.org>
Signed-off-by: Samuel Groot <samuel.groot@grenoble-inp.org>
Signed-off-by: Matthieu Moy <matthieu.moy@grenoble-inp.fr>
Reviewed-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agodoc: change environment variables format
Tom Russello [Tue, 7 Jun 2016 22:35:06 +0000 (00:35 +0200)] 
doc: change environment variables format

This change GIT_* variables that where in italic style to monospaced font
according to the guideline. It was obtained with

perl -pi -e "s/\'(GIT_.*?)\'/\`\1\`/g" *.txt

One of the main purposes is to stick to the CodingGuidelines as possible so
that people writting new documentation by mimicking the existing are more likely
to have it right (even if they didn't read the CodingGuidelines).

Signed-off-by: Tom Russello <tom.russello@grenoble-inp.org>
Signed-off-by: Erwan Mathoniere <erwan.mathoniere@grenoble-inp.org>
Signed-off-by: Samuel Groot <samuel.groot@grenoble-inp.org>
Signed-off-by: Matthieu Moy <matthieu.moy@grenoble-inp.fr>
Reviewed-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agodoc: clearer rule about formatting literals
Tom Russello [Tue, 7 Jun 2016 22:35:05 +0000 (00:35 +0200)] 
doc: clearer rule about formatting literals

Make the guideline text that we want for our documentation clearer.

Signed-off-by: Tom Russello <tom.russello@grenoble-inp.org>
Signed-off-by: Erwan Mathoniere <erwan.mathoniere@grenoble-inp.org>
Signed-off-by: Samuel Groot <samuel.groot@grenoble-inp.org>
Signed-off-by: Matthieu Moy <matthieu.moy@grenoble-inp.fr>
Reviewed-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agotree-diff: avoid alloca for large allocations
Jeff King [Tue, 7 Jun 2016 22:53:00 +0000 (18:53 -0400)] 
tree-diff: avoid alloca for large allocations

Commit 72441af (tree-diff: rework diff_tree() to generate
diffs for multiparent cases as well, 2014-04-07) introduced
the use of alloca so that the common cases of commits with 1
or 2 parents would not be adversely affected by going
through the multi-parent code.

However, our xalloca is not ideal when the number of parents
grows very large:

  1. If the requested size is too large for our stack,
     alloca() has no way to tell us, and we simply segfault
     while trying to access the memory.

  2. It does not use our usual memory_limit_check() logic.

I measured, and alloca is indeed buying us a very small
speedup over xmalloc()/free(). So we'd want to keep
something like it.

This patch simply puts a conditional in place at each
callsite: we use alloca for common known-small numbers of
parents, and otherwise use the heap. We are technically
still vulnerable to (1), but no more so than if we simply
put a few dozen bytes on the stack, which we must do all the
time anyway. And likewise, we technically miss a memory
limit check if it is tiny, but such a limit is pointless.

An alternative to this would be implement something like:

  struct tree *tp, tp_fallback[2];
  if (nparent <= ARRAY_SIZE(tp_fallback))
          tp = tp_fallback;
  else
  ALLOC_ARRAY(tp, nparent);
  ...
  if (tp != tp_fallback)
  free(tp);

That would let us drop our xalloca() portability code
entirely. But in my measurements, this seemed to perform
slightly worse than the xalloca solution.

Note in the example above, and in the patch below, I've used
ALLOC_ARRAY() to replace the manual xmalloc(nr * sizeof(*x)).
Besides being shorter, this has the bonus that one cannot
accidentally overflow a size_t during that computation.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoadd: add --chmod=+x / --chmod=-x options
Edward Thomson [Tue, 31 May 2016 22:08:18 +0000 (17:08 -0500)] 
add: add --chmod=+x / --chmod=-x options

The executable bit will not be detected (and therefore will not be
set) for paths in a repository with `core.filemode` set to false,
though the users may still wish to add files as executable for
compatibility with other users who _do_ have `core.filemode`
functionality.  For example, Windows users adding shell scripts may
wish to add them as executable for compatibility with users on
non-Windows.

Although this can be done with a plumbing command
(`git update-index --add --chmod=+x foo`), teaching the `git-add`
command allows users to set a file executable with a command that
they're already familiar with.

Signed-off-by: Edward Thomson <ethomson@edwardthomson.com>
Helped-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoregex: fix a SIZE_MAX macro redefinition warning
Ramsay Jones [Tue, 7 Jun 2016 00:35:33 +0000 (01:35 +0100)] 
regex: fix a SIZE_MAX macro redefinition warning

Since commit 56a1a3ab ("Silence GCC's \"cast of pointer to integer of a
different size\" warning", 26-10-2015), sparse has been issuing a macro
redefinition warning for the SIZE_MAX macro. However, gcc did not issue
any such warning.

After commit 56a1a3ab, in terms of the order of #includes and #defines,
the code looked something like:

  $ cat -n junk.c
       1 #include <stddef.h>
       2
       3 #define SIZE_MAX ((size_t) -1)
       4
       5 #include <stdint.h>
       6
       7 int main(int argc, char *argv[])
       8 {
       9 return 0;
      10 }
  $
  $ gcc junk.c
  $

However, if you compile that file with -Wsystem-headers, then it will
also issue a warning. Having set -Wsystem-headers in CFLAGS, using the
config.mak file, then (on cygwin):

  $ make compat/regex/regex.o
      CC compat/regex/regex.o
  In file included from /usr/lib/gcc/x86_64-pc-cygwin/4.9.3/include/stdint.h:9:0,
                   from compat/regex/regcomp.c:21,
                   from compat/regex/regex.c:77:
  /usr/include/stdint.h:362:0: warning: "SIZE_MAX" redefined
   #define SIZE_MAX (__SIZE_MAX__)
   ^
  In file included from compat/regex/regex.c:69:0:
  compat/regex/regex_internal.h:108:0: note: this is the location of the previous definition
   # define SIZE_MAX ((size_t) -1)
   ^
  $

The compilation of the compat/regex code is somewhat unusual in that the
regex.c file directly #includes the other c files (regcomp.c, regexec.c
and regex_internal.c). Commit 56a1a3ab added an #include of <stdint.h>
to the regcomp.c file, which results in the redefinition, since this is
included after the regex_internal.h header. This header file contains a
'fallback' definition for SIZE_MAX, in order to support systems which do
not have the <stdint.h> header (the HAVE_STDINT_H macro is not defined).

In order to suppress the warning, we move the #include of <stdint.h>
from regcomp.c to the start of the compilation unit, close to the top
of regex.c, prior to the #include of the regex_internal.h header.

Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoreflog: continue walking the reflog past root commits
SZEDER Gábor [Fri, 3 Jun 2016 20:42:35 +0000 (22:42 +0200)] 
reflog: continue walking the reflog past root commits

If a repository contains more than one root commit, then its HEAD
reflog may contain multiple "creation events", i.e. entries whose
"from" value is the null sha1.  Listing such a reflog currently stops
prematurely at the first such entry, even when the reflog still
contains older entries.  This can scare users into thinking that their
reflog got truncated after 'git checkout --orphan'.

Continue walking the reflog past such creation events based on the
preceeding reflog entry's "new" value.

The test 'symbolic-ref writes reflog entry' in t1401-symbolic-ref
implicitly relies on the current behavior of the reflog walker to stop
at a root commit and thus to list only the reflog entries that are
relevant for that test.  Adjust the test to explicitly specify the
number of relevant reflog entries to be listed.

Reported-by: Patrik Gustafsson <pvn@textalk.se>
Signed-off-by: SZEDER Gábor <szeder@ira.uka.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoGit 2.9-rc2 v2.9.0-rc2
Junio C Hamano [Mon, 6 Jun 2016 21:19:45 +0000 (14:19 -0700)] 
Git 2.9-rc2

Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoSync with 2.8.4
Junio C Hamano [Mon, 6 Jun 2016 21:30:49 +0000 (14:30 -0700)] 
Sync with 2.8.4

* maint:
  Git 2.8.4

8 years agoGit 2.8.4 v2.8.4
Junio C Hamano [Mon, 6 Jun 2016 21:29:32 +0000 (14:29 -0700)] 
Git 2.8.4

Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoMerge branch 'kb/msys2-tty' into maint
Junio C Hamano [Mon, 6 Jun 2016 21:27:38 +0000 (14:27 -0700)] 
Merge branch 'kb/msys2-tty' into maint

The "are we talking with TTY, doing an interactive session?"
detection has been updated to work better for "Git for Windows".

* kb/msys2-tty:
  mingw: make isatty() recognize MSYS2's pseudo terminals (/dev/pty*)

8 years agoMerge branch 'da/difftool' into maint
Junio C Hamano [Mon, 6 Jun 2016 21:27:37 +0000 (14:27 -0700)] 
Merge branch 'da/difftool' into maint

"git difftool" learned to handle unmerged paths correctly in
dir-diff mode.

* da/difftool:
  difftool: handle unmerged files in dir-diff mode
  difftool: initialize variables for readability

8 years agoMerge branch 'tb/core-eol-fix' into maint
Junio C Hamano [Mon, 6 Jun 2016 21:27:36 +0000 (14:27 -0700)] 
Merge branch 'tb/core-eol-fix' into maint

A couple of bugs around core.autocrlf have been fixed.

* tb/core-eol-fix:
  convert.c: ident + core.autocrlf didn't work
  t0027: test cases for combined attributes
  convert: allow core.autocrlf=input and core.eol=crlf
  t0027: make commit_chk_wrnNNO() reliable

8 years agoMerge branch 'ar/diff-args-osx-precompose' into maint
Junio C Hamano [Mon, 6 Jun 2016 21:27:35 +0000 (14:27 -0700)] 
Merge branch 'ar/diff-args-osx-precompose' into maint

Many commands normalize command line arguments from NFD to NFC
variant of UTF-8 on OSX, but commands in the "diff" family did
not, causing "git diff $path" to complain that no such path is
known to Git.  They have been taught to do the normalization.

* ar/diff-args-osx-precompose:
  diff: run arguments through precompose_argv