git
8 years agobuiltin/am: use apply API in run_apply()
Christian Couder [Sun, 4 Sep 2016 20:18:33 +0000 (22:18 +0200)] 
builtin/am: use apply API in run_apply()

This replaces run_apply() implementation with a new one that
uses the apply API that has been previously prepared in
apply.c and apply.h.

This shoud improve performance a lot in certain cases.

As the previous implementation was creating a new `git apply`
process to apply each patch, it could be slow on systems like
Windows where it is costly to create new processes.

Also the new `git apply` process had to read the index from
disk, and when the process was done the calling process
discarded its own index and read back from disk the new
index that had been created by the `git apply` process.

This could be very inefficient with big repositories that
have big index files, especially when the system decided
that it was a good idea to run the `git apply` processes on
a different processor core.

Also eliminating index reads enables further performance
improvements by using:

`git update-index --split-index`

For example here is a benchmark of a multi hundred commit
rebase on the Linux kernel on a Debian laptop with SSD:

command: git rebase --onto 1993b17 52bef0c 29dde7c

Vanilla "next" without split index:                1m54.953s
Vanilla "next" with split index:                   1m22.476s
This series on top of "next" without split index:  1m12.034s
This series on top of "next" with split index:     0m15.678s

(using branch "next" from mid April 2016.)

Benchmarked-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoapply: learn to use a different index file
Christian Couder [Sun, 4 Sep 2016 20:18:32 +0000 (22:18 +0200)] 
apply: learn to use a different index file

Sometimes we want to apply in a different index file.

Before the apply functionality was libified it was possible to
use the GIT_INDEX_FILE environment variable, for this purpose.

But now, as the apply functionality has been libified, it should
be possible to do that in a libified way.

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoapply: pass apply state to build_fake_ancestor()
Christian Couder [Sun, 4 Sep 2016 20:18:31 +0000 (22:18 +0200)] 
apply: pass apply state to build_fake_ancestor()

To libify git apply functionality, we will need to read from a
different index file in get_current_sha1(). This index file will be
stored in "struct apply_state", so let's pass the state to
build_fake_ancestor() which will later pass it to get_current_sha1().

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoapply: refactor `git apply` option parsing
Christian Couder [Sun, 4 Sep 2016 20:18:30 +0000 (22:18 +0200)] 
apply: refactor `git apply` option parsing

Parsing `git apply` options can be useful to other commands that
want to call the libified apply functionality, because this way
they can easily pass some options from their own command line to
the libified apply functionality.

This will be used by `git am` in a following patch.

To make this possible, let's refactor the `git apply` option
parsing code into a new libified apply_parse_options() function.

Doing that makes it possible to remove some functions definitions
from "apply.h" and make them static in "apply.c".

Helped-by: Ramsay Jones <ramsay@ramsayjones.plus.com>
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoapply: change error_routine when silent
Christian Couder [Sun, 4 Sep 2016 20:18:29 +0000 (22:18 +0200)] 
apply: change error_routine when silent

To avoid printing anything when applying with
`state->apply_verbosity == verbosity_silent`, let's save the
existing warn and error routines before applying, and let's
replace them with a routine that does nothing.

Then after applying, let's restore the saved routines.

Note that, as we need to restore the saved routines in all
cases, we cannot return early any more in apply_all_patches().

Helped-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agousage: add get_error_routine() and get_warn_routine()
Christian Couder [Sun, 4 Sep 2016 20:18:28 +0000 (22:18 +0200)] 
usage: add get_error_routine() and get_warn_routine()

Let's make it possible to get the current error_routine and warn_routine,
so that we can store them before using set_error_routine() or
set_warn_routine() to use new ones.

This way we will be able put back the original routines, when we are done
with using new ones.

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agousage: add set_warn_routine()
Christian Couder [Sun, 4 Sep 2016 20:18:27 +0000 (22:18 +0200)] 
usage: add set_warn_routine()

There are already set_die_routine() and set_error_routine(),
so let's add set_warn_routine() as this will be needed in a
following commit.

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoapply: don't print on stdout in verbosity_silent mode
Christian Couder [Sun, 4 Sep 2016 20:18:26 +0000 (22:18 +0200)] 
apply: don't print on stdout in verbosity_silent mode

When apply_verbosity is set to verbosity_silent nothing should be
printed on both stderr and stdout.

To avoid printing on stdout, we can just skip calling the following
functions:

- stat_patch_list(),
- numstat_patch_list(),
- summary_patch_list().

It is safe to do that because the above functions have no side
effects other than printing:

- stat_patch_list() only computes some local values and then call
show_stats() and print_stat_summary(), those two functions only
compute local values and call printing functions,
- numstat_patch_list() also only computes local values and calls
printing functions,
- summary_patch_list() calls show_file_mode_name(), printf(),
show_rename_copy(), show_mode_change() that are only printing.

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoapply: make it possible to silently apply
Christian Couder [Sun, 4 Sep 2016 20:18:25 +0000 (22:18 +0200)] 
apply: make it possible to silently apply

This changes 'int apply_verbosely' into 'enum apply_verbosity', and
changes the possible values of the variable from a bool to
a tristate.

The previous 'false' state is changed into 'verbosity_normal'.
The previous 'true' state is changed into 'verbosity_verbose'.

The new added state is 'verbosity_silent'. It should prevent
anything to be printed on both stderr and stdout.

This is needed because `git am` wants to first call apply
functionality silently, if it can then fall back on 3-way merge
in case of error.

Printing on stdout, and calls to warning() or error() are not
taken care of in this patch, as that will be done in following
patches.

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoapply: use error_errno() where possible
Christian Couder [Sun, 4 Sep 2016 20:18:24 +0000 (22:18 +0200)] 
apply: use error_errno() where possible

To avoid possible mistakes and to uniformly show the errno
related messages, let's use error_errno() where possible.

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoapply: make some parsing functions static again
Christian Couder [Sun, 4 Sep 2016 20:18:23 +0000 (22:18 +0200)] 
apply: make some parsing functions static again

Some parsing functions that were used in both "apply.c" and
"builtin/apply.c" are now only used in the former, so they
can be made static to "apply.c".

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoapply: move libified code from builtin/apply.c to apply.{c,h}
Christian Couder [Fri, 22 Apr 2016 18:55:46 +0000 (20:55 +0200)] 
apply: move libified code from builtin/apply.c to apply.{c,h}

As most of the apply code in builtin/apply.c has been libified by a number of
previous commits, it can now be moved to apply.{c,h}, so that more code can
use it.

Helped-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Helped-by: Ramsay Jones <ramsay@ramsayjones.plus.com>
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoapply: rename and move opt constants to apply.h
Christian Couder [Sun, 4 Sep 2016 20:18:21 +0000 (22:18 +0200)] 
apply: rename and move opt constants to apply.h

The constants for the "inaccurate-eof" and the "recount" options will
be used in both "apply.c" and "builtin/apply.c", so they need to go
into "apply.h", and therefore they need a name that is more specific
to the API they belong to.

Helped-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: rename option parsing functions
Christian Couder [Sun, 4 Sep 2016 20:18:20 +0000 (22:18 +0200)] 
builtin/apply: rename option parsing functions

As these functions are going to be part of the libified
apply API, let's give them a name that is more specific
to the apply API.

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: make create_one_file() return -1 on error
Christian Couder [Sun, 4 Sep 2016 20:18:19 +0000 (22:18 +0200)] 
builtin/apply: make create_one_file() return -1 on error

To libify `git apply` functionality we have to signal errors to the
caller instead of exit()ing.

To do that in a compatible manner with the rest of the error handling
in "builtin/apply.c", create_one_file() should return -1 instead of
calling exit().

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: make try_create_file() return -1 on error
Christian Couder [Sun, 4 Sep 2016 20:18:18 +0000 (22:18 +0200)] 
builtin/apply: make try_create_file() return -1 on error

To libify `git apply` functionality we have to signal errors to the
caller instead of die()ing.

To do that in a compatible manner with the rest of the error handling
in "builtin/apply.c", try_create_file() should return -1 in case of
error.

Unfortunately try_create_file() currently returns -1 to signal a
recoverable error. To fix that, let's make it return 1 in case of
a recoverable error and -1 in case of an unrecoverable error.

Helped-by: Eric Sunshine <sunshine@sunshineco.com>
Helped-by: Jeff King <peff@peff.net>
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: make write_out_results() return -1 on error
Christian Couder [Mon, 8 Aug 2016 21:03:22 +0000 (23:03 +0200)] 
builtin/apply: make write_out_results() return -1 on error

To libify `git apply` functionality we have to signal errors to the
caller instead of exit()ing.

To do that in a compatible manner with the rest of the error handling
in "builtin/apply.c", write_out_results() should return -1 instead of
calling exit().

Helped-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: make write_out_one_result() return -1 on error
Christian Couder [Mon, 8 Aug 2016 21:03:21 +0000 (23:03 +0200)] 
builtin/apply: make write_out_one_result() return -1 on error

To libify `git apply` functionality we have to signal errors to the
caller instead of exit()ing.

To do that in a compatible manner with the rest of the error handling
in "builtin/apply.c", write_out_one_result() should just return what
remove_file() and create_file() are returning instead of calling
exit().

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: make create_file() return -1 on error
Christian Couder [Mon, 8 Aug 2016 21:03:20 +0000 (23:03 +0200)] 
builtin/apply: make create_file() return -1 on error

To libify `git apply` functionality we have to signal errors to the
caller instead of exit()ing.

To do that in a compatible manner with the rest of the error handling
in "builtin/apply.c", create_file() should just return what
add_conflicted_stages_file() and add_index_file() are returning
instead of calling exit().

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: make add_index_file() return -1 on error
Christian Couder [Mon, 8 Aug 2016 21:03:19 +0000 (23:03 +0200)] 
builtin/apply: make add_index_file() return -1 on error

To libify `git apply` functionality we have to signal errors to the
caller instead of die()ing.

To do that in a compatible manner with the rest of the error handling
in "builtin/apply.c", add_index_file() should return -1 instead of
calling die().

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: make add_conflicted_stages_file() return -1 on error
Christian Couder [Mon, 8 Aug 2016 21:03:18 +0000 (23:03 +0200)] 
builtin/apply: make add_conflicted_stages_file() return -1 on error

To libify `git apply` functionality we have to signal errors to the
caller instead of die()ing.

To do that in a compatible manner with the rest of the error handling
in "builtin/apply.c", add_conflicted_stages_file() should return -1
instead of calling die().

Helped-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: make remove_file() return -1 on error
Christian Couder [Mon, 8 Aug 2016 21:03:17 +0000 (23:03 +0200)] 
builtin/apply: make remove_file() return -1 on error

To libify `git apply` functionality we have to signal errors to the
caller instead of die()ing.

To do that in a compatible manner with the rest of the error handling
in "builtin/apply.c", remove_file() should return -1 instead of
calling die().

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: make build_fake_ancestor() return -1 on error
Christian Couder [Mon, 8 Aug 2016 21:03:16 +0000 (23:03 +0200)] 
builtin/apply: make build_fake_ancestor() return -1 on error

To libify `git apply` functionality we have to signal errors to the
caller instead of die()ing.

To do that in a compatible manner with the rest of the error handling
in "builtin/apply.c", build_fake_ancestor() should return -1 instead
of calling die().

Helped-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: change die_on_unsafe_path() to check_unsafe_path()
Christian Couder [Mon, 8 Aug 2016 21:03:15 +0000 (23:03 +0200)] 
builtin/apply: change die_on_unsafe_path() to check_unsafe_path()

To libify `git apply` functionality we have to signal errors to the
caller instead of die()ing.

To do that in a compatible manner with the rest of the error handling
in "builtin/apply.c", die_on_unsafe_path() should return a negative
integer instead of calling die(), so while doing that let's change
its name to check_unsafe_path().

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: make gitdiff_*() return -1 on error
Christian Couder [Mon, 8 Aug 2016 21:03:14 +0000 (23:03 +0200)] 
builtin/apply: make gitdiff_*() return -1 on error

To libify `git apply` functionality we have to signal errors to the
caller instead of die()ing.

To do that in a compatible manner with the rest of the error handling
in "builtin/apply.c", gitdiff_*() functions should return -1 instead
of calling die().

A previous patch made it possible for gitdiff_*() functions to
return -1 in case of error. Let's take advantage of that to
make gitdiff_verify_name() return -1 on error, and to have
gitdiff_oldname() and gitdiff_newname() directly return
what gitdiff_verify_name() returns.

Helped-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: make gitdiff_*() return 1 at end of header
Christian Couder [Mon, 8 Aug 2016 21:03:13 +0000 (23:03 +0200)] 
builtin/apply: make gitdiff_*() return 1 at end of header

The gitdiff_*() functions that are called as p->fn() in parse_git_header()
should return 1 instead of -1 in case of end of header or unrecognized
input, as these are not real errors. It just instructs the parser to break
out.

This makes it possible for gitdiff_*() functions to return -1 in case of a
real error. This will be done in a following patch.

Helped-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: make parse_traditional_patch() return -1 on error
Christian Couder [Mon, 8 Aug 2016 21:03:12 +0000 (23:03 +0200)] 
builtin/apply: make parse_traditional_patch() return -1 on error

To libify `git apply` functionality we have to signal errors to the
caller instead of die()ing.

To do that in a compatible manner with the rest of the error handling
in "builtin/apply.c", parse_traditional_patch() should return -1
instead of calling die().

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: make apply_all_patches() return 128 or 1 on error
Christian Couder [Mon, 8 Aug 2016 21:03:11 +0000 (23:03 +0200)] 
builtin/apply: make apply_all_patches() return 128 or 1 on error

To finish libifying the apply functionality, apply_all_patches() should not
die() or exit() in case of error, but return either 128 or 1, so that it
gives the same exit code as when die() or exit(1) is called. This way
scripts relying on the exit code don't need to be changed.

While doing that we must take care that file descriptors are properly closed
and, if needed, reset to a sensible value.

Also, according to the lockfile API, when finished with a lockfile, one
should either commit it or roll it back.

This is even more important now that the same lockfile can be passed
to init_apply_state() many times to be reused by series of calls to
the apply lib functions.

Helped-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Helped-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Helped-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: move check_apply_state() to apply.c
Christian Couder [Mon, 8 Aug 2016 21:03:10 +0000 (23:03 +0200)] 
builtin/apply: move check_apply_state() to apply.c

To libify `git apply` functionality we must make check_apply_state()
usable outside "builtin/apply.c".

Let's do that by moving it into "apply.c".

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: make check_apply_state() return -1 instead of die()ing
Christian Couder [Mon, 8 Aug 2016 21:03:09 +0000 (23:03 +0200)] 
builtin/apply: make check_apply_state() return -1 instead of die()ing

To libify `git apply` functionality we have to signal errors to the
caller instead of die()ing.

To do that in a compatible manner with the rest of the error handling
in "builtin/apply.c", check_apply_state() should return -1 instead of
calling die().

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoapply: make init_apply_state() return -1 instead of exit()ing
Christian Couder [Mon, 8 Aug 2016 21:03:08 +0000 (23:03 +0200)] 
apply: make init_apply_state() return -1 instead of exit()ing

To libify `git apply` functionality we have to signal errors to the
caller instead of exit()ing.

To do that in a compatible manner with the rest of the error handling
in "builtin/apply.c", init_apply_state() should return -1 instead of
calling exit().

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: move init_apply_state() to apply.c
Christian Couder [Mon, 8 Aug 2016 21:03:07 +0000 (23:03 +0200)] 
builtin/apply: move init_apply_state() to apply.c

To libify `git apply` functionality we must make init_apply_state()
usable outside "builtin/apply.c".

Let's do that by moving it into a new "apply.c".

Helped-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: make parse_ignorewhitespace_option() return -1 instead of die()ing
Christian Couder [Mon, 8 Aug 2016 21:03:06 +0000 (23:03 +0200)] 
builtin/apply: make parse_ignorewhitespace_option() return -1 instead of die()ing

To libify `git apply` functionality we have to signal errors to the
caller instead of die()ing.

To do that in a compatible manner with the rest of the error handling
in "builtin/apply.c", parse_ignorewhitespace_option() should return
-1 instead of calling die().

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: make parse_whitespace_option() return -1 instead of die()ing
Christian Couder [Mon, 8 Aug 2016 21:03:05 +0000 (23:03 +0200)] 
builtin/apply: make parse_whitespace_option() return -1 instead of die()ing

To libify `git apply` functionality we have to signal errors to the
caller instead of die()ing.

To do that in a compatible manner with the rest of the error handling
in builtin/apply.c, parse_whitespace_option() should return -1 instead
of calling die().

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: make parse_single_patch() return -1 on error
Christian Couder [Mon, 8 Aug 2016 21:03:04 +0000 (23:03 +0200)] 
builtin/apply: make parse_single_patch() return -1 on error

To libify `git apply` functionality we have to signal errors to the
caller instead of die()ing.

To do that in a compatible manner with the rest of the error handling
in builtin/apply.c, parse_single_patch() should return a negative
integer instead of calling die().

Let's do that by using error() and let's adjust the related test
cases accordingly.

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: make parse_chunk() return a negative integer on error
Christian Couder [Mon, 8 Aug 2016 21:03:03 +0000 (23:03 +0200)] 
builtin/apply: make parse_chunk() return a negative integer on error

To libify `git apply` functionality we have to signal errors to the
caller instead of die()ing or exit()ing.

To do that in a compatible manner with the rest of the error handling
in builtin/apply.c, parse_chunk() should return a negative integer
instead of calling die() or exit().

As parse_chunk() is called only by apply_patch() which already
returns either -1 or -128 when an error happened, let's make it also
return -1 or -128.

This makes it compatible with what find_header() and parse_binary()
already return.

Helped-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: make find_header() return -128 instead of die()ing
Christian Couder [Mon, 8 Aug 2016 21:03:02 +0000 (23:03 +0200)] 
builtin/apply: make find_header() return -128 instead of die()ing

To libify `git apply` functionality we have to signal errors to the
caller instead of die()ing.

To do that in a compatible manner with the rest of the error handling
in builtin/apply.c, let's make find_header() return -128 instead of
calling die().

We could make it return -1, unfortunately find_header() already
returns -1 when no header is found.

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: read_patch_file() return -1 instead of die()ing
Christian Couder [Mon, 8 Aug 2016 21:03:01 +0000 (23:03 +0200)] 
builtin/apply: read_patch_file() return -1 instead of die()ing

To libify `git apply` functionality we have to signal errors to the
caller instead of die()ing. Let's do that by returning -1 instead of
die()ing in read_patch_file().

Helped-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agobuiltin/apply: make apply_patch() return -1 or -128 instead of die()ing
Christian Couder [Mon, 8 Aug 2016 21:03:00 +0000 (23:03 +0200)] 
builtin/apply: make apply_patch() return -1 or -128 instead of die()ing

To libify `git apply` functionality we have to signal errors
to the caller instead of die()ing.

As a first step in this direction, let's make apply_patch() return
-1 or -128 in case of errors instead of dying. For now its only
caller apply_all_patches() will exit(128) when apply_patch()
return -128 and it will exit(1) when it returns -1.

We exit() with code 128 because that was what die() was doing
and we want to keep the distinction between exiting with code 1
and exiting with code 128.

Helped-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoapply: move 'struct apply_state' to apply.h
Christian Couder [Mon, 8 Aug 2016 21:02:59 +0000 (23:02 +0200)] 
apply: move 'struct apply_state' to apply.h

To libify `git apply` functionality we must make 'struct apply_state'
usable outside "builtin/apply.c".

Let's do that by creating a new "apply.h" and moving
'struct apply_state' there.

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoapply: make some names more specific
Christian Couder [Thu, 11 Aug 2016 08:52:29 +0000 (10:52 +0200)] 
apply: make some names more specific

To prepare for some structs and constants being moved from
builtin/apply.c to apply.h, we should give them some more
specific names to avoid possible name collisions in the global
namespace.

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoSync with maint
Junio C Hamano [Thu, 28 Jul 2016 21:21:18 +0000 (14:21 -0700)] 
Sync with maint

* maint:
  Some fixes for 2.9.3

8 years agoEighth batch of topics for 2.10
Junio C Hamano [Thu, 28 Jul 2016 20:14:53 +0000 (13:14 -0700)] 
Eighth batch of topics for 2.10

Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agot9100: portability fix
Junio C Hamano [Thu, 28 Jul 2016 21:20:13 +0000 (14:20 -0700)] 
t9100: portability fix

Do not say "export VAR=VAL"; "VAR=VAL && export VAR" is always more
portable.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoMerge branch 'master' of git://git.bogomips.org/git-svn
Junio C Hamano [Thu, 28 Jul 2016 20:13:53 +0000 (13:13 -0700)] 
Merge branch 'master' of git://git.bogomips.org/git-svn

* 'master' of git://git.bogomips.org/git-svn:
  git-svn: allow --version to work anywhere
  git-svn: document svn.authorsProg in config

8 years agoSome fixes for 2.9.3
Junio C Hamano [Thu, 28 Jul 2016 18:28:32 +0000 (11:28 -0700)] 
Some fixes for 2.9.3

Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoMerge branch 'ak/lazy-prereq-mktemp' into maint
Junio C Hamano [Thu, 28 Jul 2016 18:26:03 +0000 (11:26 -0700)] 
Merge branch 'ak/lazy-prereq-mktemp' into maint

A test that unconditionally used "mktemp" learned that the command
is not necessarily available everywhere.

* ak/lazy-prereq-mktemp:
  t7610: test for mktemp before test execution

8 years agoMerge branch 'nd/icase' into maint
Junio C Hamano [Thu, 28 Jul 2016 18:26:03 +0000 (11:26 -0700)] 
Merge branch 'nd/icase' into maint

"git grep -i" has been taught to fold case in non-ascii locales
correctly.

* nd/icase:
  grep.c: reuse "icase" variable
  diffcore-pickaxe: support case insensitive match on non-ascii
  diffcore-pickaxe: Add regcomp_or_die()
  grep/pcre: support utf-8
  gettext: add is_utf8_locale()
  grep/pcre: prepare locale-dependent tables for icase matching
  grep: rewrite an if/else condition to avoid duplicate expression
  grep/icase: avoid kwsset when -F is specified
  grep/icase: avoid kwsset on literal non-ascii strings
  test-regex: expose full regcomp() to the command line
  test-regex: isolate the bug test code
  grep: break down an "if" stmt in preparation for next changes

8 years agoMerge branch 'sb/submodule-parallel-fetch' into maint
Junio C Hamano [Thu, 28 Jul 2016 18:26:02 +0000 (11:26 -0700)] 
Merge branch 'sb/submodule-parallel-fetch' into maint

Fix recently introduced codepaths that are involved in parallel
submodule operations, which gave up on reading too early, and
could have wasted CPU while attempting to write under a corner
case condition.

* sb/submodule-parallel-fetch:
  hoist out handle_nonblock function for xread and xwrite
  xwrite: poll on non-blocking FDs
  xread: retry after poll on EAGAIN/EWOULDBLOCK

8 years agoMerge branch 'dk/blame-move-no-reason-for-1-line-context' into maint
Junio C Hamano [Thu, 28 Jul 2016 18:26:01 +0000 (11:26 -0700)] 
Merge branch 'dk/blame-move-no-reason-for-1-line-context' into maint

"git blame -M" missed a single line that was moved within the file.

* dk/blame-move-no-reason-for-1-line-context:
  blame: require 0 context lines while finding moved lines with -M

8 years agoMerge branch 'jk/test-match-signal' into maint
Junio C Hamano [Thu, 28 Jul 2016 18:26:00 +0000 (11:26 -0700)] 
Merge branch 'jk/test-match-signal' into maint

The test framework learned a new helper test_match_signal to
check an exit code from getting killed by an expected signal.

* jk/test-match-signal:
  t/lib-git-daemon: use test_match_signal
  test_must_fail: use test_match_signal
  t0005: use test_match_signal as appropriate
  tests: factor portable signal check out of t0005

8 years agoMerge branch 'js/am-call-theirs-theirs-in-fallback-3way' into maint
Junio C Hamano [Thu, 28 Jul 2016 18:25:59 +0000 (11:25 -0700)] 
Merge branch 'js/am-call-theirs-theirs-in-fallback-3way' into maint

One part of "git am" had an oddball helper function that called
stuff from outside "his" as opposed to calling what we have "ours",
which was not gender-neutral and also inconsistent with the rest of
the system where outside stuff is usuall called "theirs" in
contrast to "ours".

* js/am-call-theirs-theirs-in-fallback-3way:
  am: counteract gender bias

8 years agoMerge branch 'js/t3404-grammo-fix' into maint
Junio C Hamano [Thu, 28 Jul 2016 18:25:58 +0000 (11:25 -0700)] 
Merge branch 'js/t3404-grammo-fix' into maint

Grammofix.

* js/t3404-grammo-fix:
  t3404: fix a grammo (commands are ran -> commands are run)

8 years agoMerge branch 'nd/doc-new-command' into maint
Junio C Hamano [Thu, 28 Jul 2016 18:25:57 +0000 (11:25 -0700)] 
Merge branch 'nd/doc-new-command' into maint

Typofix in a doc.

* nd/doc-new-command:
  new-command.txt: correct the command description file

8 years agoMerge branch 'ew/gc-auto-pack-limit-fix' into maint
Junio C Hamano [Thu, 28 Jul 2016 18:25:56 +0000 (11:25 -0700)] 
Merge branch 'ew/gc-auto-pack-limit-fix' into maint

"gc.autoPackLimit" when set to 1 should not trigger a repacking
when there is only one pack, but the code counted poorly and did
so.

* ew/gc-auto-pack-limit-fix:
  gc: fix off-by-one error with gc.autoPackLimit

8 years agoMerge branch 'js/color-on-windows-comment' into maint
Junio C Hamano [Thu, 28 Jul 2016 18:25:55 +0000 (11:25 -0700)] 
Merge branch 'js/color-on-windows-comment' into maint

For a long time, we carried an in-code comment that said our
colored output would work only when we use fprintf/fputs on
Windows, which no longer is the case for the past few years.

* js/color-on-windows-comment:
  color.h: remove obsolete comment about limitations on Windows

8 years agoMerge branch 'mm/doc-tt' into maint
Junio C Hamano [Thu, 28 Jul 2016 18:25:54 +0000 (11:25 -0700)] 
Merge branch 'mm/doc-tt' into maint

More mark-up updates to typeset strings that are expected to
literally typed by the end user in fixed-width font.

* mm/doc-tt:
  doc: typeset HEAD and variants as literal
  CodingGuidelines: formatting HEAD in documentation
  doc: typeset long options with argument as literal
  doc: typeset '--' as literal
  doc: typeset long command-line options as literal
  doc: typeset short command-line options as literal
  Documentation/git-mv.txt: fix whitespace indentation

8 years agoMerge branch 'js/sign-empty-commit-fix' into maint
Junio C Hamano [Thu, 28 Jul 2016 18:25:53 +0000 (11:25 -0700)] 
Merge branch 'js/sign-empty-commit-fix' into maint

"git commit --amend --allow-empty-message -S" for a commit without
any message body could have misidentified where the header of the
commit object ends.

* js/sign-empty-commit-fix:
  commit -S: avoid invalid pointer with empty message

8 years agoMerge branch 'ps/rebase-i-auto-unstash-upon-abort' into maint
Junio C Hamano [Thu, 28 Jul 2016 18:25:52 +0000 (11:25 -0700)] 
Merge branch 'ps/rebase-i-auto-unstash-upon-abort' into maint

"git rebase -i --autostash" did not restore the auto-stashed change
when the operation was aborted.

* ps/rebase-i-auto-unstash-upon-abort:
  rebase -i: restore autostash on abort

8 years agoMerge branch 'nd/ita-cleanup' into maint
Junio C Hamano [Thu, 28 Jul 2016 18:25:51 +0000 (11:25 -0700)] 
Merge branch 'nd/ita-cleanup' into maint

Git does not know what the contents in the index should be for a
path added with "git add -N" yet, so "git grep --cached" should not
show hits (or show lack of hits, with -L) in such a path, but that
logic does not apply to "git grep", i.e. searching in the working
tree files.  But we did so by mistake, which has been corrected.

* nd/ita-cleanup:
  grep: fix grepping for "intent to add" files
  t7810-grep.sh: fix a whitespace inconsistency
  t7810-grep.sh: fix duplicated test name

8 years agoMerge branch 'js/find-commit-subject-ignore-leading-blanks' into maint
Junio C Hamano [Thu, 28 Jul 2016 18:25:50 +0000 (11:25 -0700)] 
Merge branch 'js/find-commit-subject-ignore-leading-blanks' into maint

A helper function that takes the contents of a commit object and
finds its subject line did not ignore leading blank lines, as is
commonly done by other codepaths.  Make it ignore leading blank
lines to match.

* js/find-commit-subject-ignore-leading-blanks:
  reset --hard: skip blank lines when reporting the commit subject
  sequencer: use skip_blank_lines() to find the commit subject
  commit -C: skip blank lines at the beginning of the message
  commit.c: make find_commit_subject() more robust
  pretty: make the skip_blank_lines() function public

8 years agoMerge branch 'dg/subtree-rebase-test' into maint
Junio C Hamano [Thu, 28 Jul 2016 18:25:49 +0000 (11:25 -0700)] 
Merge branch 'dg/subtree-rebase-test' into maint

Add a test to specify the desired behaviour that currently is not
available in "git rebase -Xsubtree=...".

* dg/subtree-rebase-test:
  contrib/subtree: Add a test for subtree rebase that loses commits

8 years agoMerge branch 'sb/submodule-deinit-all'
Junio C Hamano [Thu, 28 Jul 2016 17:34:45 +0000 (10:34 -0700)] 
Merge branch 'sb/submodule-deinit-all'

A comment update for a topic that was merged to Git v2.8.

* sb/submodule-deinit-all:
  submodule deinit: remove outdated comment

8 years agoMerge branch 'ew/find-perl-on-freebsd-in-local'
Junio C Hamano [Thu, 28 Jul 2016 17:34:44 +0000 (10:34 -0700)] 
Merge branch 'ew/find-perl-on-freebsd-in-local'

Recent FreeBSD stopped making perl available at /usr/bin/perl;
switch the default the built-in path to /usr/local/bin/perl on not
too ancient FreeBSD releases.

* ew/find-perl-on-freebsd-in-local:
  config.mak.uname: correct perl path on FreeBSD

8 years agoMerge branch 'ew/daemon-socket-keepalive'
Junio C Hamano [Thu, 28 Jul 2016 17:34:43 +0000 (10:34 -0700)] 
Merge branch 'ew/daemon-socket-keepalive'

Recent update to "git daemon" tries to enable the socket-level
KEEPALIVE, but when it is spawned via inetd, the standard input
file descriptor may not necessarily be connected to a socket.
Suppress an ENOTSOCK error from setsockopt().

* ew/daemon-socket-keepalive:
  Windows: add missing definition of ENOTSOCK
  daemon: ignore ENOTSOCK from setsockopt

8 years agoMerge branch 'nd/pack-ofs-4gb-limit'
Junio C Hamano [Thu, 28 Jul 2016 17:34:42 +0000 (10:34 -0700)] 
Merge branch 'nd/pack-ofs-4gb-limit'

"git pack-objects" and "git index-pack" mostly operate with off_t
when talking about the offset of objects in a packfile, but there
were a handful of places that used "unsigned long" to hold that
value, leading to an unintended truncation.

* nd/pack-ofs-4gb-limit:
  fsck: use streaming interface for large blobs in pack
  pack-objects: do not truncate result in-pack object size on 32-bit systems
  index-pack: correct "offset" type in unpack_entry_data()
  index-pack: report correct bad object offsets even if they are large
  index-pack: correct "len" type in unpack_data()
  sha1_file.c: use type off_t* for object_info->disk_sizep
  pack-objects: pass length to check_pack_crc() without truncation

8 years agoMerge branch 'nd/worktree-lock'
Junio C Hamano [Thu, 28 Jul 2016 17:34:41 +0000 (10:34 -0700)] 
Merge branch 'nd/worktree-lock'

"git worktree prune" protected worktrees that are marked as
"locked" by creating a file in a known location.  "git worktree"
command learned a dedicated command pair to create and remove such
a file, so that the users do not have to do this with editor.

* nd/worktree-lock:
  worktree.c: find_worktree() search by path suffix
  worktree: add "unlock" command
  worktree: add "lock" command
  worktree.c: add is_worktree_locked()
  worktree.c: add is_main_worktree()
  worktree.c: add find_worktree()

8 years agoMerge branch 'rs/notes-merge-no-toctou'
Junio C Hamano [Thu, 28 Jul 2016 17:34:41 +0000 (10:34 -0700)] 
Merge branch 'rs/notes-merge-no-toctou'

"git notes merge" had a code to see if a path exists (and fails if
it does) and then open the path for writing (when it doesn't).
Replace it with open with O_EXCL.

* rs/notes-merge-no-toctou:
  notes-merge: use O_EXCL to avoid overwriting existing files

8 years agoMerge branch 'js/rebase-i-tests'
Junio C Hamano [Thu, 28 Jul 2016 17:34:40 +0000 (10:34 -0700)] 
Merge branch 'js/rebase-i-tests'

A few tests that specifically target "git rebase -i" have been
added.

* js/rebase-i-tests:
  rebase -i: we allow extra spaces after fixup!/squash!
  rebase -i: demonstrate a bug with --autosquash
  t3404: add a test for the --gpg-sign option

8 years agosubmodule deinit: remove outdated comment
Stefan Beller [Tue, 26 Jul 2016 00:35:38 +0000 (17:35 -0700)] 
submodule deinit: remove outdated comment

Signed-off-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoconfig.mak.uname: correct perl path on FreeBSD
Nguyễn Thái Ngọc Duy [Mon, 25 Jul 2016 16:21:25 +0000 (18:21 +0200)] 
config.mak.uname: correct perl path on FreeBSD

It looks the the symlink /usr/bin/perl (to /usr/local/bin/perl) has
been removed at least on FreeBSD 10.3. See [1] for more information.

[1] https://svnweb.freebsd.org/ports/head/UPDATING?r1=386270&r2=386269&pathrev=386270&diff_format=c

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Helped-by: Eric Wong <e@80x24.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoSeventh batch of topics for 2.10
Junio C Hamano [Mon, 25 Jul 2016 21:17:28 +0000 (14:17 -0700)] 
Seventh batch of topics for 2.10

Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoMerge branch 'rs/use-strbuf-addbuf'
Junio C Hamano [Mon, 25 Jul 2016 21:13:47 +0000 (14:13 -0700)] 
Merge branch 'rs/use-strbuf-addbuf'

Code cleanup.

* rs/use-strbuf-addbuf:
  strbuf: avoid calling strbuf_grow() twice in strbuf_addbuf()
  use strbuf_addbuf() for appending a strbuf to another

8 years agoMerge branch 'ew/autoconf-pthread'
Junio C Hamano [Mon, 25 Jul 2016 21:13:46 +0000 (14:13 -0700)] 
Merge branch 'ew/autoconf-pthread'

Existing autoconf generated test for the need to link with pthread
library did not check all the functions from pthread libraries;
recent FreeBSD has some functions in libc but not others, and we
mistakenly thought linking with libc is enough when it is not.

* ew/autoconf-pthread:
  configure.ac: stronger test for pthread linkage

8 years agoMerge branch 'mh/blame-worktree'
Junio C Hamano [Mon, 25 Jul 2016 21:13:45 +0000 (14:13 -0700)] 
Merge branch 'mh/blame-worktree'

"git blame file" allowed the lineage of lines in the uncommitted,
unadded contents of "file" to be inspected, but it refused when
"file" did not appear in the current commit.  When "file" was
created by renaming an existing file (but the change has not been
committed), this restriction was unnecessarily tight.

* mh/blame-worktree:
  t/t8003-blame-corner-cases.sh: Use here documents
  blame: allow to blame paths freshly added to the index

8 years agoMerge branch 'js/fsck-name-object'
Junio C Hamano [Mon, 25 Jul 2016 21:13:44 +0000 (14:13 -0700)] 
Merge branch 'js/fsck-name-object'

When "git fsck" reports a broken link (e.g. a tree object contains
a blob that does not exist), both containing object and the object
that is referred to were reported with their 40-hex object names.
The command learned the "--name-objects" option to show the path to
the containing object from existing refs (e.g. "HEAD~24^2:file.txt").

* js/fsck-name-object:
  fsck: optionally show more helpful info for broken links
  fsck: give the error function a chance to see the fsck_options
  fsck_walk(): optionally name objects on the go
  fsck: refactor how to describe objects

8 years agoMerge branch 'nd/cache-tree-ita'
Junio C Hamano [Mon, 25 Jul 2016 21:13:44 +0000 (14:13 -0700)] 
Merge branch 'nd/cache-tree-ita'

"git add -N dir/file && git write-tree" produced an incorrect tree
when there are other paths in the same directory that sorts after
"file".

* nd/cache-tree-ita:
  cache-tree: do not generate empty trees as a result of all i-t-a subentries
  cache-tree.c: fix i-t-a entry skipping directory updates sometimes
  test-lib.sh: introduce and use $EMPTY_BLOB
  test-lib.sh: introduce and use $EMPTY_TREE

8 years agoMerge branch 'jk/push-scrub-url'
Junio C Hamano [Mon, 25 Jul 2016 21:13:43 +0000 (14:13 -0700)] 
Merge branch 'jk/push-scrub-url'

"git fetch http://user:pass@host/repo..." scrubbed the userinfo
part, but "git push" didn't.

* jk/push-scrub-url:
  t5541: fix url scrubbing test when GPG is not set
  push: anonymize URL in status output

8 years agoMerge branch 'nd/test-helpers'
Junio C Hamano [Mon, 25 Jul 2016 21:13:42 +0000 (14:13 -0700)] 
Merge branch 'nd/test-helpers'

Build clean-up.

* nd/test-helpers:
  t/test-lib.sh: fix running tests with --valgrind
  Makefile: use VCSSVN_LIB to refer to svn library
  Makefile: drop extra dependencies for test helpers

8 years agoMerge branch 'jc/doc-diff-filter-exclude'
Junio C Hamano [Mon, 25 Jul 2016 21:13:41 +0000 (14:13 -0700)] 
Merge branch 'jc/doc-diff-filter-exclude'

Belated doc update for a feature added in v1.8.5.

* jc/doc-diff-filter-exclude:
  diff: document diff-filter exclusion

8 years agoMerge branch 'ls/travis-enable-httpd-tests'
Junio C Hamano [Mon, 25 Jul 2016 21:13:39 +0000 (14:13 -0700)] 
Merge branch 'ls/travis-enable-httpd-tests'

Allow http daemon tests in Travis CI tests.

* ls/travis-enable-httpd-tests:
  travis-ci: enable web server tests t55xx on Linux

8 years agoMerge branch 'jc/renormalize-merge-kill-safer-crlf'
Junio C Hamano [Mon, 25 Jul 2016 21:13:38 +0000 (14:13 -0700)] 
Merge branch 'jc/renormalize-merge-kill-safer-crlf'

"git merge" with renormalization did not work well with
merge-recursive, due to "safer crlf" conversion kicking in when it
shouldn't.

* jc/renormalize-merge-kill-safer-crlf:
  merge: avoid "safer crlf" during recording of merge results
  convert: unify the "auto" handling of CRLF

8 years agoMerge branch 'rs/worktree-use-strbuf-absolute-path'
Junio C Hamano [Mon, 25 Jul 2016 21:13:37 +0000 (14:13 -0700)] 
Merge branch 'rs/worktree-use-strbuf-absolute-path'

Code simplification.

* rs/worktree-use-strbuf-absolute-path:
  worktree: use strbuf_add_absolute_path() directly

8 years agoMerge branch 'rs/rm-strbuf-optim'
Junio C Hamano [Mon, 25 Jul 2016 21:13:36 +0000 (14:13 -0700)] 
Merge branch 'rs/rm-strbuf-optim'

The use of strbuf in "git rm" to build filename to remove was a bit
suboptimal, which has been fixed.

* rs/rm-strbuf-optim:
  rm: reuse strbuf for all remove_dir_recursively() calls

8 years agoMerge branch 'rw/make-needs-librt'
Junio C Hamano [Mon, 25 Jul 2016 21:13:35 +0000 (14:13 -0700)] 
Merge branch 'rw/make-needs-librt'

Makefile assumed that -lrt is always available on platforms that
want to use clock_gettime() and CLOCK_MONOTONIC, which is not a
case for recent Mac OS X.  The necessary symbols are often found in
libc on many modern systems and having -lrt on the command line, as
long as the library exists, had no effect, but when the platform
removes librt.a that is a different matter--having -lrt will break
the linkage.

This change could be seen as a regression for those who do need to
specify -lrt, as they now specifically ask for NEEDS_LIBRT when
building. Hopefully they are in the minority these days.

* rw/make-needs-librt:
  config.mak.uname: define NEEDS_LIBRT under Linux, for now
  Makefile: add NEEDS_LIBRT to optionally link with librt

8 years agoMerge branch 'js/ignore-space-at-eol'
Junio C Hamano [Mon, 25 Jul 2016 21:13:35 +0000 (14:13 -0700)] 
Merge branch 'js/ignore-space-at-eol'

An age old bug that caused "git diff --ignore-space-at-eol"
misbehave has been fixed.

* js/ignore-space-at-eol:
  diff: fix a double off-by-one with --ignore-space-at-eol
  diff: demonstrate a bug with --patience and --ignore-space-at-eol

8 years agoMerge branch 'mh/ref-iterators'
Junio C Hamano [Mon, 25 Jul 2016 21:13:33 +0000 (14:13 -0700)] 
Merge branch 'mh/ref-iterators'

The API to iterate over all the refs (i.e. for_each_ref(), etc.)
has been revamped.

* mh/ref-iterators:
  for_each_reflog(): reimplement using iterators
  dir_iterator: new API for iterating over a directory tree
  for_each_reflog(): don't abort for bad references
  do_for_each_ref(): reimplement using reference iteration
  refs: introduce an iterator interface
  ref_resolves_to_object(): new function
  entry_resolves_to_object(): rename function from ref_resolves_to_object()
  get_ref_cache(): only create an instance if there is a submodule
  remote rm: handle symbolic refs correctly
  delete_refs(): add a flags argument
  refs: use name "prefix" consistently
  do_for_each_ref(): move docstring to the header file
  refs: remove unnecessary "extern" keywords

8 years agoMerge branch 'mh/update-ref-errors'
Junio C Hamano [Mon, 25 Jul 2016 21:13:32 +0000 (14:13 -0700)] 
Merge branch 'mh/update-ref-errors'

Error handling in the codepaths that updates refs has been
improved.

* mh/update-ref-errors:
  lock_ref_for_update(): avoid a symref resolution
  lock_ref_for_update(): make error handling more uniform
  t1404: add more tests of update-ref error handling
  t1404: document function test_update_rejected
  t1404: remove "prefix" argument to test_update_rejected
  t1404: rename file to t1404-update-ref-errors.sh

8 years agoMerge branch 'mh/split-under-lock'
Junio C Hamano [Mon, 25 Jul 2016 21:13:32 +0000 (14:13 -0700)] 
Merge branch 'mh/split-under-lock'

Further preparatory work on the refs API before the pluggable
backend series can land.

* mh/split-under-lock: (33 commits)
  lock_ref_sha1_basic(): only handle REF_NODEREF mode
  commit_ref_update(): remove the flags parameter
  lock_ref_for_update(): don't resolve symrefs
  lock_ref_for_update(): don't re-read non-symbolic references
  refs: resolve symbolic refs first
  ref_transaction_update(): check refname_is_safe() at a minimum
  unlock_ref(): move definition higher in the file
  lock_ref_for_update(): new function
  add_update(): initialize the whole ref_update
  verify_refname_available(): adjust constness in declaration
  refs: don't dereference on rename
  refs: allow log-only updates
  delete_branches(): use resolve_refdup()
  ref_transaction_commit(): correctly report close_ref() failure
  ref_transaction_create(): disallow recursive pruning
  refs: make error messages more consistent
  lock_ref_sha1_basic(): remove unneeded local variable
  read_raw_ref(): move docstring to header file
  read_raw_ref(): improve docstring
  read_raw_ref(): rename symref argument to referent
  ...

8 years agogit-svn: allow --version to work anywhere
Eric Wong [Fri, 22 Jul 2016 20:17:31 +0000 (20:17 +0000)] 
git-svn: allow --version to work anywhere

Checking the version of the installed SVN libraries should not
require a git repository at all.  This matches the behavior of
"git --version".

Add a test for "git svn help" for the same behavior while we're
at it, too.

Signed-off-by: Eric Wong <e@80x24.org>
8 years agoWindows: add missing definition of ENOTSOCK
Johannes Sixt [Thu, 21 Jul 2016 20:59:06 +0000 (22:59 +0200)] 
Windows: add missing definition of ENOTSOCK

The previous commit introduced the first use of ENOTSOCK. This macro is
not available on Windows. Define it as WSAENOTSOCK because that is the
corresponding error value reported by the Windows versions of socket
functions.

Signed-off-by: Johannes Sixt <j6t@kdbg.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agostrbuf: avoid calling strbuf_grow() twice in strbuf_addbuf()
René Scharfe [Thu, 21 Jul 2016 16:46:44 +0000 (18:46 +0200)] 
strbuf: avoid calling strbuf_grow() twice in strbuf_addbuf()

Implement strbuf_addbuf() as a normal function in order to avoid calling
strbuf_grow() twice, with the second callinside strbud_add() being a
no-op.  This is slightly faster and also reduces the text size a bit.

Signed-off-by: Rene Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agot5541: fix url scrubbing test when GPG is not set
Jeff King [Wed, 20 Jul 2016 11:32:26 +0000 (05:32 -0600)] 
t5541: fix url scrubbing test when GPG is not set

When the GPG prereq is not set, we do not run test 34. That
test changes the directory of the test script as a side
effect (something we usually frown on, but which matches the
style of the rest of this script). When test 35 (the
url-scrubbing test) runs, it expects to be in the directory
from test 34. If it's not, the test fails; we are in a
different sub-repo, our test-commit is built on a different
history, and the push becomes a non-fast-forward.

We can fix this by unconditionally moving to the directory
we expect (again, against our usual style but matching how
the rest of the script operates).

As an additional protection, let's also switch from "make a
new commit and push to master" to just "push to a new
branch". We don't care about the branch name; we just want
_some_ ref update to trigger the status output. Pushing to a
new branch is less likely to run into problems with
force-updates, changing the checked-out branch, etc.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoSixth batch of topics for 2.10
Junio C Hamano [Tue, 19 Jul 2016 20:26:16 +0000 (13:26 -0700)] 
Sixth batch of topics for 2.10

Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoMerge branch 'ls/p4-tmp-refs'
Junio C Hamano [Tue, 19 Jul 2016 20:22:24 +0000 (13:22 -0700)] 
Merge branch 'ls/p4-tmp-refs'

"git p4" used a location outside $GIT_DIR/refs/ to place its
temporary branches, which has been moved to refs/git-p4-tmp/.

* ls/p4-tmp-refs:
  git-p4: place temporary refs used for branch import under refs/git-p4-tmp

8 years agoMerge branch 'js/am-call-theirs-theirs-in-fallback-3way'
Junio C Hamano [Tue, 19 Jul 2016 20:22:23 +0000 (13:22 -0700)] 
Merge branch 'js/am-call-theirs-theirs-in-fallback-3way'

One part of "git am" had an oddball helper function that called
stuff from outside "his" as opposed to calling what we have "ours",
which was not gender-neutral and also inconsistent with the rest of
the system where outside stuff is usuall called "theirs" in
contrast to "ours".

* js/am-call-theirs-theirs-in-fallback-3way:
  am: counteract gender bias

8 years agoMerge branch 'jk/write-file'
Junio C Hamano [Tue, 19 Jul 2016 20:22:23 +0000 (13:22 -0700)] 
Merge branch 'jk/write-file'

General code clean-up around a helper function to write a
single-liner to a file.

* jk/write-file:
  branch: use write_file_buf instead of write_file
  use write_file_buf where applicable
  write_file: add format attribute
  write_file: add pointer+len variant
  write_file: use xopen
  write_file: drop "gently" form
  branch: use non-gentle write_file for branch description
  am: ignore return value of write_file()
  config: fix bogus fd check when setting up default config

8 years agoMerge branch 'jk/printf-format'
Junio C Hamano [Tue, 19 Jul 2016 20:22:22 +0000 (13:22 -0700)] 
Merge branch 'jk/printf-format'

Code clean-up to avoid using a variable string that compilers may
feel untrustable as printf-style format given to write_file()
helper function.

* jk/printf-format:
  commit.c: remove print_commit_list()
  avoid using sha1_to_hex output as printf format
  walker: let walker_say take arbitrary formats

8 years agoMerge branch 'rs/help-c-source-with-gitattributes'
Junio C Hamano [Tue, 19 Jul 2016 20:22:21 +0000 (13:22 -0700)] 
Merge branch 'rs/help-c-source-with-gitattributes'

The .c/.h sources are marked as such in our .gitattributes file so
that "git diff -W" and friends would work better.

* rs/help-c-source-with-gitattributes:
  .gitattributes: set file type for C files

8 years agoMerge branch 'nd/fetch-ref-summary'
Junio C Hamano [Tue, 19 Jul 2016 20:22:21 +0000 (13:22 -0700)] 
Merge branch 'nd/fetch-ref-summary'

Improve the look of the way "git fetch" reports what happened to
each ref that was fetched.

* nd/fetch-ref-summary:
  fetch: reduce duplicate in ref update status lines with placeholder
  fetch: align all "remote -> local" output
  fetch: change flag code for displaying tag update and deleted ref
  fetch: refactor ref update status formatting code
  git-fetch.txt: document fetch output