t7815-grep-binary.sh subtest 12 inexplicably passes

[me-and]

The t7008-grep-binary.sh test "git grep .fi a" passes. It's marked as an expected failure, and it does fail on my v2.0.4 Git build on CentOS.

[me-and]

Huh, that's interesting. If you skip the make configure && ./configure stages of the build process and just run make, the test continues to fail.

[me-and]

I thought it might be the --with-libpcre argument to ./configure which was causing the difference, but running ./configure without it produces the same result...

[me-and]

Just checked for comparison on my CentOS VM: whether or not you include the configure step, that test always fails.

Next step is probably to report this upstream.

[me-and]

Now this is odd: building using Cygport has the test unexpectedly passing, but building directly from a Git source tree including the Cygwin patches has the test failing!

[me-and]

Ah, here we go: the Cygport compilation runs make configure and ./configure --htmldir=<dir> --mandir=<dir> --with-libpcre before make. Presumably at least part of that is what's making the difference here.

[me-and]

… which is something I'd already discovered in my notes above. OH WELL.

[me-and]

Reported this upstream: http://lists-archives.com/git/864215-binary-grep-t7008-known-breakage-vanished-on-cygwin.html.

[me-and]

Test is now t7815, still passing at 811da4d (v2.26.1) when marked as an expected failure.