Hi Philip
Thanks for CC'ing me Peff.
On 10/08/18 19:27, Jeff King wrote:
On Thu, Aug 09, 2018 at 08:17:36PM -0700, Philip White wrote:
I’d like to report what I suspect is a bug in Git, tested in 2.18 and
2.14. (I’d be delighted to be corrected if it is my own
misunderstanding.) I’m reporting it here based on guidance from
https://git-scm.com/community.
I created a minimal testcase with a detailed README here:
https://github.com/philipmw/git-bugreport-2018-hunk-fail
Overview of the bug:
When interactively selecting hunks to apply, using `git checkout -p
<tree> <file>`, git will correctly apply an unmodified hunk, but will
refuse to apply a hunk that went through a text editor ("e" command),
even when I made no changes in the text editor.
Thanks for any advice or attention you can give to this matter.
This sounds like the bug discussed in:
https://public-inbox.org/git/be321106-2f10-e678-8237-449d2dd30fee@xxxxxxxxxxxx/
Looking at the bug report Philip linked to I think this is a long
standing bug where splitting a patch and then editing one of the
subhunks can fail because git-add--interactive.perl doesn't know how to
stitch the subhunks back together when one of them has been edited.
There's a failing test in t3701-add-interactive.sh to document this. It
is on my radar to fix this (at least in simple cases) but I haven't got
round to it yet.
Thanks Philip for the detailed report
Best Wishes
Phillip
(that message and the surrounding thread).
The fix is in f4d35a6b49 (add -p: fix counting empty context lines in
edited patches, 2018-06-11), which is in the current tip of 'master',
but not yet any released version.
-Peff
