Remove metadata::decoder::each_path and adjust users to do something else

[catamorphism]

metadata::decoder::each_path takes a predicate that determines whether a path_entry is broken. FIXMEs (formerly XXXes) say "this is awful", but I don't understand the problem well enough to elaborate.

[catamorphism]

Nominating for milestone 5 (production-ready)

[pcwalton]

This function just needs to be nuked, as it's impossible to make fast. It's the source of most of the performance problems when compiling small crates.

[graydon]

Also a production-readiness issue. But we'd like to do it soon! This is a major slowdown in small-file compilation.

[thomaslee]

Performance issues aside, I have a feeling the "This is awful" comment may relate to the way the original implementation was using a boolean flag as a sort of half-arsed "break" (this must have been written before the "new" for loop syntax). This forces each_path to parse & iterate over every item in an imported crate during decoding even when the predicate returns false. I don't think that's necessary, but my gut says it's unlikely the typical case will be made any faster by cleaning that up.

I'm cleaning up each_path a little while I dig into another, unrelated metadata issue but I'm curious: if we were to nuke this, what would we replace it with?

[emberian]

Triage bump.

[pnkfelix]

this has been done. closing.