[PATCH (!) 0/2] Fix serious regressions in latest master

Hi,

It turns out that status.short and status.branch introduce two very
serious regressions.  This series fixes them.

Thanks.

Ramkumar Ramachandra (2):
  status: really ignore config with --porcelain
  commit: make it work with status.short

 builtin/commit.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

-- 
1.8.3.1.550.gd96f26e.dirty

1a22bd3 (Merge branch 'jg/status-config', 2013-06-23) introduced a
serious regression in --porcelain by introducing the configuration
variables status.short and status.branch.  Contrary to its description,
the output of

  $ git status --porcelain

now depends on the configuration variables status.short and
status.branch.  As a result, callers that expect parsable output to be
returned are broken.  For instance, in a repository with submodules with
status.branch and status.short set,

  $ git status

always reports all submodules as containing modified content, even if
they are clean.

One solution to the problem is to turn off s->show_branch in
wt_porcelain_print() just like we turn off other s->* variables, but
that would break callers of --porcelain --branch (in fact, there is such
a caller in t/t7508-status.sh).  Besides, we never said that --porcelain
cannot be combined with other options.  The larger problem is that the
config parser and command-line option parser set the same variables,
making it impossible to determine who set them.

The correct solution is therefore to skip the config parser completely
when --porcelain is given.  Unfortunately, to determine that --porcelain
is given, we have to run the command-line option parser.  Running the
command-line option parser before the config parser is undesirable, as
configuration variables would override options on the command-line.  As
a fair compromise, check that argv[1] is equal to the string
"--porcelain" and skip the config parser in this case.  It is a
compromise, because we expect --porcelain to be specified as the first
argument to status.

On a related note, the command-line parser is not very robust.

  $ git status --short --long
  $ git status --long --long
  $ git status --porcelain --long
  $ git status --long --porcelain
  $ git status --porcelain --short
  $ git status --short --porcelain

all return different outputs.  This bug is left as an exercise for
future contributors to fix.

Signed-off-by: Ramkumar Ramachandra <redacted>
---
 builtin/commit.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/builtin/commit.c b/builtin/commit.c
index b589ce0..896f002 100644
--- a/builtin/commit.c
+++ b/builtin/commit.c

@@ -1193,7 +1193,12 @@ int cmd_status(int argc, const char **argv, const char *prefix)
 
 	wt_status_prepare(&s);
 	gitmodules_config();
-	git_config(git_status_config, &s);
+
+	if (argc > 1 && !strcmp(argv[1], "--porcelain"))
+		; /* Do not read user configuration */
+	else
+		git_config(git_status_config, &s);
+
 	determine_whence(&s);
 	argc = parse_options(argc, argv, prefix,
 			     builtin_status_options,

-- 
1.8.3.1.550.gd96f26e.dirty

50e4f75 (status: introduce status.short to enable --short by default,
2013-06-11) introduced a regression in git commit; it is now impossible
to commit with status.short set.

This happens because commit internally runs run_status() to set
s->commitable and determine whether or not there is something to
commit.  The problem arises from the fact that only STATUS_FORMAT_NONE
(or STATUS_FORMAT_LONG) is equipped to set s->commitable.
7c9f7038 (commit: support alternate status formats, 2009-09-05) clearly
states that --short and --porcelain imply --dry-run and are therefore
only intended for display purposes.

The bigger problem is that it is impossible to differentiate between a
status_format set by the command-line option parser versus that set by
the config parser.  So these two are exactly equivalent:

  $ git -c status.short=true commit
  $ git commit --short

To alleviate this problem, clear status_format as soon as the config
parser has finished its work to nullify the effect of parsing
status.short, and get commit to work again.

Signed-off-by: Ramkumar Ramachandra <redacted>
---
 builtin/commit.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/builtin/commit.c b/builtin/commit.c
index 896f002..dc5ed7d 100644
--- a/builtin/commit.c
+++ b/builtin/commit.c

@@ -1448,6 +1448,7 @@ int cmd_commit(int argc, const char **argv, const char *prefix)
 	wt_status_prepare(&s);
 	gitmodules_config();
 	git_config(git_commit_config, &s);
+	status_format = STATUS_FORMAT_NONE; /* Ignore status.short */
 	determine_whence(&s);
 	s.colopts = 0;
 

-- 
1.8.3.1.550.gd96f26e.dirty