These changes are in the same spirit as the six patches that
precede them, but they haven't been split into individually
justifiable patches yet.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>

git's --stat output is intended for humans and since v1.7.9.2~13
(2012-02-01) varies by locale.  The tests in this script using "apply
--stat" are meant to check two things:

 - how binary file changes are accounted for and printed in
   git's diffstat format

 - that "git apply" can parse the various forms of binary diff

Split these two kinds of check into separate tests, and use --numstat
instead of --stat in the latter.  This way, we lose less test coverage
when git is being run without writing its output in the C locale (for
example because GETTEXT_POISON is enabled) and there are fewer tests
to change if the --stat output needs to be tweaked again.

While at it, use commands separated by && that read and write to
temporary files in place of pipelines so segfaults and other failures
in the upstream of the processing pipeline don't get hidden.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>

git's diff --stat output is intended for human consumption and
since v1.7.9.2~13 (2012-02-01) varies by locale.  Add a test checking
that git stash show defaults to --stat and tweak the rest of the
"stash show" tests that showed a diffstat to use numstat.

This way, there are fewer tests to tweak if the diffstat format
changes again.  This also improves test coverage when running tests
with git configured not to write its output in the C locale (e.g.,
via GETTEXT_POISON=Yes).

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>

Since v1.7.3-rc0~26^2~9 (revert: report success when using option
--strategy, 2010-07-14), the cherry-pick-many-commits test checks the
format of output written to the terminal during a cherry-pick sequence
in addition to the functionality.  There is no reason those have to
be checked in the same test, though, and it has some downsides:

 - when progress output is broken, the test result does not convey
   whether the functionality was also broken or not

 - it is not immediately obvious when reading that these checks are
   meant to prevent regressions in details of the output format and
   are not just a roundabout way to check functional details like the
   number of commits produced

 - there is a temptation to include the same kind of output checking
   for every new cherry-pick test, which would make future changes
   to the output unnecessarily difficult

Put the tests from v1.7.3-rc0~26^2~9 in separate assertions, following
the principle "test one feature at a time".

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>

This is one of the early tests, so it uses a style that by modern
standards can be hard to read.  Tweak it to:

 - clearly declare what assertion each test is designed to check

 - mark tests that create state later tests will depend on with the
   word "setup" so people writing or running tests know the others
   can be skipped or reordered safely

 - put commands that populate a file with expected output inside
   the corresponding test stanza, so it is easier to see by eye
   where each test begins and ends

 - instead of pipelines, use commands that read and write a
   temporary file, so bugs causing commands to segfault or produce
   the wrong exit status can be caught.

More cosmetic changes:

 - put the opening quote starting each test on the same line as the
   test_expect_* invocation, and indent the commands in each test
   with a single tab

 - end the test early if the underlying filesystem cannot
   accomodate the filenames we use, instead of marking all tests
   with the same TABS_IN_FILENAMES prerequisite.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>

This test script checks that git's plumbing commands quote filenames
with special characters like space, tab, and double-quote
appropriately in their input and output.

Since commit v1.7.9.2~13 (Use correct grammar in diffstat summary
line, 2012-02-01), the final "1 file changed, 1 insertion(+)" line
from diffstats is translatable, meaning tests that rely on exact "git
apply --stat" output have to be skipped when git is not configured to
produce output in the C locale (for example, when GETTEXT_POISON is
enabled).  So:

 - Tweak the three "git apply --stat" tests that check "git apply"'s
   input parsing to use --numstat instead.

   --numstat output is more reliable, does not vary with locale, and
   is itself easier to parse.  These tests are mainly about how "git
   apply" parses its input so this should not result in much loss of
   coverage.

 - Add a new "apply --stat" test to check the quoting in --stat output
   format.

This wins back a little of the test coverage lost with the patch
"test: use test_i18ncmp to check --stat output" when GETTEXT_POISON is
enabled.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>

Ever since v1.7.9.2~13 (2012-02-01), git's diffstat-style summary line
produced by "git apply --stat", "git diff --stat", and "git commit"
varies by locale, producing test failures when GETTEXT_POISON is set.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>

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

"git tag -s" honored "gpg.program" configuration variable since
1.7.9, but "git tag -v" and "git verify-tag" didn't.

By Alex Zepeda
* az/verify-tag-use-gpg-config:
  verify-tag: Parse GPG configuration options.