Re: RFC: two minor tweaks to check-ignore to help git-annex assistant

Adam Spiers [off-list ref] writes:

I already have a rough fix for the second issue, but I wanted to
solicit feedback on the appropriate UI changes before proceeding much
further.  Does something like the below patch seem reasonable, modulo
the lack of tests?  In case the UI changes I am proposing are not
clear from the patch, here's some example output from running it
inside a clone of the git source tree:

    $ git check-ignore -v -n foo.tar.{gz,bz2}
    .gitignore:203:*.tar.gz foo.tar.gz
    ::      foo.tar.bz2

So the number of output fields does not change depending on whether
the pattern matches or not, and any caller can determine whether it
does simply by checking whether the first field is non-empty.

Haven't looked at the proposed patch very carefully, but the design
looks sound.  The above output screams "empty! nothing!", and I do
not think there is any other way :: will show up in that position.

Also, does it make sense to write a new test to accompany the fix to
the first (streaming) issue?

Would it be tricky to write safely not to get stuck?  You feed one
line, stop feeding, while checking that the output has arrived, and
then kill the whole thing?  Feels somewhat yucky, but sounds doable.