If there's alternate solutions let's explore them in a followup, for now let's just make main go green again

Thanks for fixing this.

• should ID(...) constructed strings really be immortal? They've been possibly incorrectly used in some pmgen FPGA backends

Yes, I intended ID() strings to be immortal. There's a limited number of them so that should be OK.

• is there a better way to GC at shutdown than in yosys_atexit with the same code but without marking anything live?

At shutdown, I would expect there to be no live designs, so tracing shouldn't trace anything, so it's not clear to me why you need to make tracing conditional there.

I did not think about shutdown leaks triggering sanitizer warnings, sorry. I actually designed the GC scheduling with the intent that we not GC on shutdown --- in production, if the process is about to exit, doing cleanup work is a waste of time. But I understand why you'd want to catch bugs with the sanitizers. You might want to consider doing cleanup-before-shutdown work only when sanitizers are enabled.