Re: [PATCH] am: Allow passing exclude and include args to apply

maximilian attems <max@stro.at> writes:

You need to do that with test suite, not with the log message. Otherwise you are adding undue burden on the Maintainer to download klibc and dash just to run regression testing whenever somebody else makes changes to "am/apply" callchain down the road.

While I love patches that are backed by a strong "here is a real-world problem we needed to solve, and this change made our life much easier by doing so-and-so" statement, I also tend to think twice before considering a change that could potentially encourage a bad version control discipline.

Your use case description in the log message however lacks crucial information to be useful when judging that aspect. You said that the directory structure is "different", but didn't say they are different in what way. In order to skip one mail exchange turnaround, I'd speculate.

If dash repository keeps (perhaps slightly stale version of) the same files as klibc repository in its libc/ subdirectory, a patch to dash that fixes its libc part may all have its pathnames prefixed with libc/. In order to apply such a patch to the klibc tree, you would need to give -p2 to strip one extra level (if you are going the other way, you would instead give --directory=libc/ to deepen it). But then I do not see a need for --exclude to remove parts from the patch that touch outside of libc/ tree.

If the dash patch you needed to deal with touched both inside libc/ and outside, and if you are taking only libc/ part and discarding everything else, I see two issues with respect to promoting pottentially bad version control disciplines.

 - Should you be reusing the information in the commit without editing?  I
   am not worried about Signed-off-by which is about asserting the origin,
   and origin of the libc/ part is the same as the origin of the whole.
   But what about reviewers' and tester's assertion at the end?  Also the
   description of the change itself may need to be adjusted to the new
   context you are reusing the change for.

 - Why does the patch touch two unrelated parts in the first place, if its
   libc/ part can stand on its own?  This is not about the discipline of
   the user of "am", but of the originating project.

Another thing that came to my mind around the vague "different directory structure" is this question: what if directory A/ in "dash" corresponded to directory B/ in "klibc" and you saw a patch to A/ (and some others) for "dash" that you wanted to reuse in "klibc"? Do we need more changes to make it work, or do we already have enough support for this combination?

I would imagine that "git am --directory=B/ -p2 --exclude=\* --include=A/" or something like that should work, but I didn't think it through nor I didn't check the command line syntax, either.