Remove double check when decoding ExpnId to avoid races

[zetanumbers]

Fixes debug assertion failure as described in #141540 (comment)

Essentially failure happens during the race while decoding one ExpnId from different threads. This ICE doesn't happen with single threaded thread_pool due to early return within decode_expn_id with the same condition:

rust/compiler/rustc_middle/src/query/on_disk_cache.rs

Lines 605 to 607 in 8205e6b

	if let Some(expn_id) = ExpnId::from_hash(hash) {
	return expn_id;
	}

However I believe this race does not hurt because register_expn_id is pretty much idempotent:

rust/compiler/rustc_span/src/hygiene.rs

Lines 1397 to 1413 in 8205e6b

	pub fn register_expn_id(
	krate: CrateNum,
	local_id: ExpnIndex,
	data: ExpnData,
	hash: ExpnHash,
	) -> ExpnId {
	debug_assert!(data.parent == ExpnId::root() || krate == data.parent.krate);
	let expn_id = ExpnId { krate, local_id };
	HygieneData::with(|hygiene_data| {
	let _old_data = hygiene_data.foreign_expn_data.insert(expn_id, data);
	let _old_hash = hygiene_data.foreign_expn_hashes.insert(expn_id, hash);
	debug_assert!(_old_hash.is_none() || _old_hash == Some(hash));
	let _old_id = hygiene_data.expn_hash_to_expn_id.insert(hash, expn_id);
	debug_assert!(_old_id.is_none() || _old_id == Some(expn_id));
	});
	expn_id
	}

[rustbot]

r? @nnethercote

rustbot has assigned @nnethercote.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[Kivooeo]

cc @cjgillot

[nnethercote]

Fixes #141540 (comment)

I'm surprised you're seeing a debug assertion failure. Did you build your own compiler?

As for the fix, it seems plausible, but I'm not familiar with this code. @Zoxc or @Mark-Simulacrum, can you evaluate this change?

[zetanumbers]

I'm surprised you're seeing a debug assertion failure. Did you build your own compiler?

Yes, I did. I am trying to solve #141540, it is the most active A-parallel-compiler issue. Got this debug assertion failure while trying to reproduce that bug, so I've investigated it.

I usually keep rust.debug = true set in my bootstrap.toml file to make sure my code is good. But even if I turned it off, strangely that debug assertion kept failing even after ./x clean. Checked everything, still no clue. But anyway.

[nnethercote]

Ok, seems reasonable.

@bors r+ rollup

[bors]

📌 Commit 2a5d830 has been approved by nnethercote

It is now in the queue for this repository.

[Zoxc]

It indeed looks like expn_hash_to_expn_id is allowed to race here, so this removal looks correct.