Refactor lookupPath to avoid recursion. NFC

[hoodmane]

This removes the recursion from lookupPath and in my opinion makes the control flow much more explicit and comprehensible.

[sbc100]

Thanks for working on this!

[sbc100]

Wow, labeled continue targets!

[sbc100]

Can you rebase the codesize tests?

[hoodmane]

Well it's going to conflict with #22998, I suppose you can pick which one to merge first.

[sbc100]

Can you confirm those codesize changes are all due to this change? i.e. can you run the rebase on main and confirm you don't see any changes?

[hoodmane]

There are a lot of similar changes when I run it on main too.

[sbc100]

There are a lot of similar changes when I run it on main too.

Maybe try with emsdk install tot (or otherwise make sure you have the latest llvm and binaryen). (Also try ./emcc --clear-cache)

[hoodmane]

Even after ./emsdk install tot, and clearing cache and getting the most up to date main branch I get a bunch of changes from the rebaseline.

[sbc100]

Even after ./emsdk install tot, and clearing cache and getting the most up to date main branch I get a bunch of changes from the rebaseline.

Hmm.. that is strange/annoying. Maybe just revert those files then, and I can do a rebaseline after this lands.

[sbc100]

OK, I just pushed another rebaseline commit, perhaps this one will match what you are generated on main now?

[hoodmane]

Nope, even if I check out your rebaseline commit and do a clean install of tot from emsdk I get a bunch of changes in code size by one to ten bytes.

[hoodmane]

I think this PR fixes the case when you symlink a directory to itself symlink(".", "dir") and then try to glob("**") it. The previous lookupPath doesn't seem to detect this loop.

[hoodmane]

@sbc100 can this be merged?

[hoodmane]

Thanks!