[PATCH] Add ALL to FileCheck

[Stephen Lin]

> I agree with Renato here that multi-pass is a different feature. It's
> arguable whether you would prefer multi-pass with separate prefixes or
> one pass with multiple prefixes matching; you can come up with valid
> use cases for either behavior. However, I think the later is more
> intuitive as the default implementation of multiple prefixes, and is
> also considerably simpler in implementation from what I remember of
> FileCheck internals.

I take back that last part; it seems like the entire source file is
being copied into an in-memory buffer already so it wouldn't be hard
to do either implementation. I still think the default implementation
being "one pass, recognize multiple prefixes" makes more sense,
though, since usually you want the shared code to recognize common
features like function definitions and the unshared code to detect
features in between those; if it's "multiple passes, each on a
separate prefix" then there is no guarantee that the specific output
is located in the right location relative to the generic output unless
you duplicate the generic output for each prefix, which seems to
defeat the purpose.

Stephen