git tag -h fatal error with global tag.sort config

Hi,

I stumbled on Debian bug report [1], and I can reproduce the bug.

In global config (~/.gitconfig), I set:

[tag]
	sort = creatordate

When I run `git tag -h` inside a Git repository (such as git.git), 
instead of usage summary, I got:

fatal: not a git repository, but the field 'creatordate' requires access 
to object data

But plain `git tag` works fine.

This bug occurs on v2.33.0 as well as version in the original bug report 
(v2.26.0-rc2) and v2.25.1.

[1]: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=954811

-- 
An old man doll... just what I always wanted! - Clara

On Sun, Sep 12, 2021 at 06:28:54PM +0700, Bagas Sanjaya wrote:

I stumbled on Debian bug report [1], and I can reproduce the bug.

In global config (~/.gitconfig), I set:

[tag]
	sort = creatordate

When I run `git tag -h` inside a Git repository (such as git.git), instead
of usage summary, I got:

fatal: not a git repository, but the field 'creatordate' requires access to
object data

But plain `git tag` works fine.

This bug occurs on v2.33.0 as well as version in the original bug report
(v2.26.0-rc2) and v2.25.1.

[1]: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=954811

Interesting.  It bisects to 47bd3d0c14 (ref-filter: don't look for
objects when outside of a repository, 2018-11-14), which, based on the
error message, kind of makes sense, because 'git tag' uses the general
ref-filter sorting facility.  Now, even if 'git tag -h' is executed in
a repository, since 99caeed05d (Let 'git <command> -h' show usage
without a git dir, 2009-11-09) run_builtin() special-cases the '-h'
option and does not call setup_git_directory(), so cmd_tag() and
everything invoked from within will mistakenly think that there is no
repository.  And cmd_tag() parses the config before parsing the
options (of course, otherwise command line options couldn't override
the config), so it hits this die() before parse_options would get a
change to act on the '-h' option.

Now, 'git branch' uses the same ref-filter sorting, but the equivalent
'git -c branch.sort=creatordate branch -h' command does show the usage
as expected.  The relevant difference between cmd_branch() and
cmd_tag() is that the former special-cases the '-h' option as well
just before it would call git_config().  Doing the same in cmd_tag()
like in the patch below seems to fix this issue, but I'm not sure that
this is the right fix.


  ---  >8  ---

diff --git a/builtin/tag.c b/builtin/tag.c
index 065b6bf093..31b8cc4600 100644
--- a/builtin/tag.c
+++ b/builtin/tag.c

@@ -485,6 +485,9 @@ int cmd_tag(int argc, const char **argv, const char *prefix)
 
 	setup_ref_filter_porcelain_msg();
 
+	if (argc == 2 && !strcmp(argv[1], "-h"))
+		usage_with_options(git_tag_usage, options);
+
 	git_config(git_tag_config, sorting_tail);
 
 	memset(&opt, 0, sizeof(opt));

On Sun, Sep 12, 2021 at 03:27:57PM +0200, SZEDER Gábor wrote:

I think part of the problem is that git_tag_config() is pretty eager to
parse the ref format. You can similarly see:

  $ git -c tag.sort=foobar tag -h
  fatal: unknown field name: foobar

If git_tag_config() just kept strings, and then we later fed them to
a parser (when we knew they were needed), that would be an appropriate
time to bail.

We do something similar with verify_ref_format(); it is just a string
until we know we are ready to use it. But here, the format that is being
used for sorting gets fed early to parse_ref_sorting(). I guess it is a
bit more complicated, because it is generating a linked list. But maybe
it could generate a list of to-be-parsed entries, with a function like
verify_sort_format() to validate them.

-Peff

On 12/09/21 20.27, SZEDER Gábor wrote:

Now, 'git branch' uses the same ref-filter sorting, but the equivalent
'git -c branch.sort=creatordate branch -h' command does show the usage
as expected.  The relevant difference between cmd_branch() and
cmd_tag() is that the former special-cases the '-h' option as well
just before it would call git_config().  Doing the same in cmd_tag()
like in the patch below seems to fix this issue, but I'm not sure that
this is the right fix.

Also note: specifying the config above as local config (in .git/config) 
doesn't trigger the error.

-- 
An old man doll... just what I always wanted! - Clara

On 12/09/21 20.27, SZEDER Gábor wrote:

Sorry for long reply.

Patch applied and usage help appears.

But when I do `./bin-wrappers/git tag -h`, I don't see `usage:` header, 
and instead show the help starting from `-F`.

-- 
An old man doll... just what I always wanted! - Clara