Consider an API to reset thread-local Span data

[dtolnay]

Basically truncate this thing:

proc-macro2/src/fallback.rs

Lines 323 to 325 in fecb02d

	#[cfg(all(span_locations, not(fuzzing)))]
	thread_local! {
	static SOURCE_MAP: RefCell<SourceMap> = RefCell::new(SourceMap {

Resetting would be a claim that all currently existing non-call-site spans on the current thread are no longer in use, or it's okay if methods like line/column and source_text crash or produce wrong results.

A workload that involves parsing all versions of all crates on a thread pool of 64 threads would need this, because otherwise our 32-bit positions would overflow. See #413 (comment).

[buchgr]

Adding such an API would be much appreciated. My workload is similar to what you described in your comment. David, did you already have an API in mind? I'd be happy to file a PR.

Also for my understanding: The idea behind the thread-local source map is to avoid a ref on Span to safe bytes?

[dtolnay]

Something like proc_macro2::invalidate_current_thread_spans() with a realistic code example.

The idea behind the thread-local source map is to avoid a ref on Span to safe bytes?

Span needs to be Copy and 'static. I don't know any other way to implement that, regardless of how many bytes big Span is.