Implement proper symbol lookup in a collections of versions

[marxin]

Apparently, the matching logic for symbols in versions in more complex than we've got:
https://maskray.me/blog/2020-11-26-all-about-symbol-versioning#version-script

[marxin]

Thanks to the nice integration done by @lapla-cogito, I can say the PR fixes 10 linker script related tests in Mold.

[lapla-cogito]

Then this would allow us to remove those tests from skip_tests.toml, right? (I'm still figuring out how best to maintain the list of skipped tests—I plan to gather feedback on this in Zulip later.)

[mati865]

Sorry for being late to the party.
In overall, it looks quite good to me. How does it stand on performance side?

Not related to this PR, but maybe we should utf8-parse whole file at once with SIMD? For example, using https://github.com/rusticstuff/simdutf8
I imagine parsing short strings separately isn't the world's fastest thing.

[davidlattimore]

From a performance perspective, what's most important to me is that we only pay costs if features are used. Validating that all symbols are valid UTF-8 in a single function call could be an interesting thing to experiment, however, we'd only really want to do that for links where we need UTF-8 validity, which I think is only when we have a version script with extern "C++". From some very limited performance testing that I just did, this PR fixes #978 - which makes sense, since we're now skipping demangling and UTF-8 validation unless we have match rules that need them. I'm happy for this to be merged, but will leave it to you.

That's great to have a few more tests passing and to have the builds that originally motivated this change working as well!

[mati865]

There is also https://github.com/marxin/wild/blob/5d1b3303fc8500a82cea9be51cde7bb76c6116ee/libwild/src/version_script.rs#L474
So the only cases where UTF-8 is not parsed is when names are not mangled and are either an asterisk or plain text - adding globs requires parsing. Naturally, this would be a pessimisation for Rust case.

[davidlattimore]

Good point, I forgot about the globbing. I'm somewhat tempted to make a create that matches glob patterns against bytes