Re: [PATCH] attr: do not mark queried macros as unset

Jeff King [off-list ref] writes:

And here's a resurrection of the optimization that _seems_ to work, but
I'm not 100% confident in.

In particular, it does not care about macros at all. It simply asks: is
this queried attribute a thing which was ever mentioned in the
attributes files (either as a path match or as a possible macro
expansion). If not, then we know we do not need to look further for it.

So, if we are looking for 'diff' and we know no .gitattributes (or
$GIT_DIR/info/attributes) entry for 'diff' or any macro that expands
to touch 'diff' (e.g. 'binary') is in use, we know for any path
governed by the current attr-stack 'diff' attribute is unspecified.
But if we see an entry, say, "*.exe binary", then we do need to be
aware of the possibility that 'diff' may be unset for some paths.

Makes sense.

I guess maybe what I'm missing is that asking for "diff" means that we
need to care about:

  - whether "diff" was mentioned in the stack

  - whether "binary" was mentioned in the stack

But just "binary" mentioning "diff" is not interesting without somebody
actually mentioning "binary".

Yeah, that matches my understanding (which mostly comes from the
original design before even Duy's optimization).

I'm not sure how to do it robustly without being able to reverse-map all
of the macros after we've resolved them (i.e., to know that "diff" gets
mentioned by "binary", and then check if "binary" is actually
mentioned). I think that would be possible now, as we should know that
after determine_macros(). But I also wonder if we are hitting
diminishing returns (after all, determine_macros() is already walking
the attr stack).