Re: git format-patch produces invalid patch if the commit adds an empty file?

On Thu, 2021-08-19 at 14:09 -0700, Junio C Hamano wrote:

Adam Williamson [off-list ref] writes:

quoted

Hi folks! So I ran into an odd issue with git today. I'm kinda
surprised I can't find any prior discussion of it, but oh well. The
situation is this: I ran git format-patch on a commit that adds three
empty files to a repository - this commit:
https://github.com/mesonbuild/meson/commit/5c87167a34c6ed703444af180fffd8a45a7928ee
the relevant lines from the patch file it produced look like this:

===

diff --git a/test cases/common/56 array methods/a.txt b/test cases/common/56 array methods/a.txt
new file mode 100644
index 000000000..e69de29bb
diff --git a/test cases/common/56 array methods/b.txt b/test cases/common/56 array methods/b.txt
new file mode 100644
index 000000000..e69de29bb
diff --git a/test cases/common/56 array methods/c.txt b/test cases/common/56 array methods/c.txt
new file mode 100644
index 000000000..e69de29bb

I do not have very ancient build of Git handy, but I know Git as old
as v1.3.0 (which I consider is one of the two versions of historical
importance, the other being v1.5.3) behaved this way and we haven't
changed it ever since, so I am surprised too to learn that "GNU
patch" cannot grok it.  Even though you didn't mention it, am I
correct to assume that "patch" has a similar issue with a change
that removes an empty file?

Hi Junio!

I didn't test that. It does seem likely, though.

I do not think our patch injestion machinery in "git apply" minds if
we added the "--- /dev/null" + "+++ b/<path>" headers (and the
reverse for removal of an empty file) to the current output, and I
am not fundamentally opposed to such a change.

But because it is such a rare event (and a discouraged practice) to
record a completely empty file, I wouldn't place a high priority on
doing so myself.

Thanks.
-- 
Adam Williamson
Fedora QA
IRC: adamw | Twitter: adamw_ha
https://www.happyassassin.net