Re: linux-next: Fixes tag needs some work in the rdma-fixes tree

[Jason Gunthorpe]

On Mon, Oct 21, 2019 at 10:25:41AM -0700, Joe Perches wrote:
> On Mon, 2019-10-21 at 17:11 +0000, Jason Gunthorpe wrote:
> > On Mon, Oct 21, 2019 at 07:08:21PM +0200, Matteo Croce wrote:
> > > On Mon, Oct 21, 2019 at 7:01 PM Jason Gunthorpe <jgg@xxxxxxxxxxxx> wrote:
> > > > On Mon, Oct 21, 2019 at 05:39:06PM +0200, Matteo Croce wrote:
> > > > > > I thought I saw that checkpatch was checking this now?
> > > > > > 
> > > > > > commit a8dd86bf746256fbf68f82bc13356244c5ad8efa
> > > > > > Author: Matteo Croce <mcroce@xxxxxxxxxx>
> > > > > > Date:   Wed Sep 25 16:46:38 2019 -0700
> > > > > > 
> > > > > >     checkpatch.pl: warn on invalid commit id
> > > > > > 
> > > > > > Maybe that check should also check that enough hash is provided and
> > > > > > other details like the correct subject line?
> > > > > > 
> > > > > > I also use a check that builds the fixes line from the commit id and
> > > > > > requires it to be the same as the patch provided. This catches all
> > > > > > sorts of wrong fixes lines, and sometimes git even recommends 13 chars
> > > > > > :\
> > > > > > 
> > > > > > Jason
> > > > > 
> > > > > Hi,
> > > > > 
> > > > > actually I just call git_commit_info() which checks for validness.
> > > > > I could also check that the hash is at least 12 digits, would be very easy.
> > > > 
> > > > IMHO you should do
> > > > 
> > > >   git log --abbrev=12 -1 --format='Fixes: %h (\"%s\")'
> > > > 
> > > > And check that the provided fixes line matches the above output
> > > > exactly, or nearly exactly. People do lots of funny things to fixes
> > > > lines..
> > > > 
> > > 
> > > The point in using git_commit_info() instead of calling git directly
> > > is that the latter would generate an error if the working copy is not
> > > a git tree (e.g. a tar.xz downloaded from kernel.org).
> > 
> > Well, it does some checks and calls 'git log' so it seems like it
> > could learn to call git log with different arguments, right?
> 
> git commit SHA1's are not just 12 chars and could be any length.

--abbrev forces a minimum bound to the kernel recommendation if the
user has an old git or misconfigured git. git auto-detects if it needs
more digits beyond 12.

> And checkpatch already does use specific arguments
> 
> 	my $output = `${git_command} log --no-color --format='%H %s' -1 $commit 2>&1`;
> 
> and then parses that $output.

Maybe output format "%H %h %s" and then parse it to check the min
length and verify the subject?

Jason