0.22: "stg mail" generated cover letter is not correct about stgit HOT 8 CLOSED

Another approach for __shortlog would be to stick with "git log" and simply pass it the range of commit IDs. This approach also addresses the diffstat issue:

def __diffstat(stack, patches, diff_opts):
    rev1 = stack.patches.get(patches[0])
    rev2 = stack.patches.get(patches[-1])
    return stack.repository.diff_tree(
            rev1.commit.data.tree,
            rev2.commit.data.tree,
            diff_opts,
            stat=True,
    )

from stgit.

My Python is immensely rusty, but here's a patch that addresses the issues.

https://github.com/chucklever/stgit/commit/d120b78ab791d467a1c14ae894142de5c6971c70

from stgit.

My approach works fine for ranges, but providing "stg mail" with a list of individual patches will require __shortlog to continue to use a loop that processes each patch in the list. I will update my patch to address this.

Unfortunately, a list-based approach does not work for the diffstat: There's no way to make that output meaningful for a sparse list of patches. For the moment, I'm sticking with a range-based approach, which was what was in place before 5b99acc .

from stgit.

I forgot to update here last week. The fix has been split into two commits:

shortlog:
https://github.com/chucklever/stgit/commit/5ad3d67035132122986c9b5ae3ce0284c6f5f120

diffstat:
https://github.com/chucklever/stgit/commit/aad399762a4dc359d7e4846180dfd3e84de5b48e

from stgit.

Thanks @chucklever for this issue report and for these patches. I have not yet had time to dig into the details, but I appreciate this help and I will look into this deeper in my next StGit cycles.

I will admit that stg mail is a bit of a blind spot for me. The test suite does not cover stg mail particularly well and my personal workflows have not needed stg mail. A longer-term goal I have in mind for stg mail is to see if we can replace its guts with git send-email. My general belief is that stg mail could shed a lot of code and become more correct by leveraging git send-email.

So it's nice to know that there is at least one stg mail user out there!

from stgit.

The one open issue for me here is I would like to know how to generate a proper diffstat for a sparse list of patches! Perhaps git send-email would help with that.

from stgit.

Fwiw, the diffstat situation has not gone unnoticed within git itself:

https://lwn.net/Articles/802797/

from stgit.

I have pushed the two repairs you made, @chucklever. Thank you!

I updated a unit test to check for the shortlog problem.

And I made one change to not pass the user-provided diff options (from --diff-opt) to the diff stat in the shortlog. I think those options should only be applied to the diffs in the patch messages themselves. Instead, we now pass --stat-width=72 to for the shortlog's diffstat. This is similar to what git format-patch does.

Anyway, I think this issue can be put to rest. Thank you again for finding and fixing this problem!

from stgit.