@bors r+

// The "thread pool" is a single spawned thread in the non-parallel

// compiler. We run on a spawned thread instead of the main thread (a) to
// provide control over the stack size, and (b) to increase similarity with
// the parallel compiler, in particular to ensure there is no accidental
// sharing of data between the main thread and the compilation thread
// (which might cause problems for the parallel compiler).

This hurts performance, right, because we have to start a thread even for very small programs? Does it make sense to avoid using a thread pool for the non-parallel compiler? We can use stacker to run things on an executable heap if we need more stack size; if we want to test that parallel_compiler still works (which imo is not necessary worth the effort) we can run in a separate thread only when debug assertions are enabled.

📌 Commit 5d716fd has been approved by jyn514

It is now in the queue for this repository.

I haven't measured, but I'd be surprised (horrified) if the cost of spawning one thread is significant even on the smallest programs.

@bors rollup=always