Spurious assertion in rustc_codegen_ssa/src/back/write.rs

[michaelwoerister]

There seems to be a race condition that sometimes triggers this assertion in CI builds:

rust/compiler/rustc_codegen_ssa/src/back/write.rs

Line 1464 in e97c29b

assert_eq!(main_thread_worker_state, MainThreadWorkerState::Codegenning);

See #86965 (comment) for an example.

Let's start collecting instances of this issue here.

[lqd]

When I looked at this last time, it seemed:

• in the messages with/from the coordinator, codegen being aborted expects that codegen is not complete and the main worker thread is still codegenning (the failing assertion).
• but it seems the main thread worker can start LLVMing in different situations, e.g. before codegen is complete and depending on the queue which would appear to break the previous state transition precondition ?

There are other such assertions when other messages are received, maybe the state machine could contain other unexpected state transitions with respect to the main thread worker state ? (For example -- and I don't know if it's possible for this to happen -- while the main thread worker is LLVMing, receiving a Message::Token() would transition the main thread worker to idle, which could then trigger similar assertions if Message::Codegen{Done,Aborted,Complete} messages are then received)

[Enselic]

Triage: The assert was removed in b00d0fa so let's close this as obsolete.