Misc changes from my parallel rustc branch

[Zoxc]

r? @michaelwoerister

[michaelwoerister]

"Make Span and Symbol implement Send and Sync" (04e2088) is the only change that worries me a little. Since #48811 landed, it should be correct to share these between threads (with and without cfg(parallel_queries)). You would just get a runtime error when you try to look something up in an interner. So why do things differently for cfg(parallel_queries)?

[Zoxc]

@michaelwoerister It is always memory-safe to share them between threads. However it only actually works properly if you use them with the same interner and point the scoped thread local to the right interner on other threads. With cfg(not(parallel_queries)), we only use the interner on one thread, so using a Span or Symbol on other threads would result in a panic or a reference to the wrong thing, removing the Send and Sync impls gives us a compile time error instead.

[michaelwoerister]

Can you add a comment that explains why the #[cfg] is there, i.e. that it's just an additional convenience in the single-threaded case?

Ideally we'll have lifetimes on everything that's interned. That would make us not depend on runtime errors.

[Zoxc]

@michaelwoerister I added comment about that.

I also removed a change which caused a panic on Travis. I couldn't reproduce it locally so I just removed it for now.

[michaelwoerister]

Thanks, @Zoxc, I'll re-review asap.

[michaelwoerister]

Looks good to me, thanks, @Zoxc!

@bors r+

[bors]

📌 Commit e09c2ff has been approved by michaelwoerister