Re: [PATCH 1/1] checkpatch: check for subject uniqueness in git repository.

[Joe Perches <joe@perches.com>]

On Sat, 2014-09-20 at 12:31 +0200, Fabian Frederick wrote:
> > On 16 September 2014 at 18:31 Joe Perches <joe@perches.com> wrote:
> > On Tue, 2014-09-16 at 18:15 +0200, Fabian Frederick wrote:
> > > > On 16 September 2014 at 05:22 Joe Perches <joe@perches.com> wrote:
> > > > On Mon, 2014-09-15 at 20:43 +0200, Fabian Frederick wrote:
> > > > > Adding patch subject uniqueness check in checkpatch --strict mode.
> > > > > See Documentation/SubmittingPatches/globally-unique identifier.
> > > >
> > > > Perhaps something like this?
> > []
> > > > diff --git a/scripts/checkpatch.pl b/scripts/checkpatch.pl
> > []
> > > > +     if ($check && $#git_commits < 1 && which("git") && -e ".git") {
> > > > +             my $output = `git log --no-color --format='%H %s' 2>&1`;
> > > > +             $output =~ s/^\s*//gm;
> > > > +             @git_commits = split("\n", $output);
> > > > +     }
> >
> > > Perfect ! :)
> >
> > Except for that _really, really_ long time to do the
> > git log of all commits...
> >
> > Maybe the git log should be enabled only with another
> > command-line option.
> >
> Hello Joe,

Dag Fabian:

> checkpatck does already execute git log in seed_camelcase_includes and
> git_commit_info when .git exists

which takes almost no time

$ time git log --no-merges --pretty=format:"%h%n" -1 -- include
40182b1

real	0m0.013s
user	0m0.008s
sys	0m0.000s

and

$ time git log --no-color --format='%H %s' -1 ^HEAD 2>&1

real	0m0.013s
user	0m0.004s
sys	0m0.004s

vs

$ time git log --no-merges --pretty=format:"%H %s" > /dev/null 2>&1

real	0m19.386s
user	0m17.264s
sys	0m1.008s

> so we could add some specific option instead like --complete --verbose
> --checkduplicates or something for this test ?

Maybe.

I'm still dubious about both the git use because of
the runtime and the checkpatch use as it's after the
fact.

The concept seems more suited to a git commit-msg hook
to me.  The runtime there is still a consideration too.

In a bit more detail:

If this modified checkpatch was run on a branch where
the patch was committed, in this sort of sequence:

	git checkout -b [some_branch]
	[make changes]
	git commit [etc]
	git format-patch -1
	./scripts/checkpatch.pl <patch from git format-patch>

then this new warning would trigger which seems senseless.

And in the case where multiple patches are done and then
scanned with checkpatch in a single block, the damage has
already been done and the defect was not found when it
occurred.

Andrew? Do you have an opinion?