Oh - I am bit disappointed I think. The hash tests didn’t fail on a second run: https://github.com/LinkiTools/racket/runs/336295671

I just noticed one of the tess failed in a documentation PR: https://github.com/racket/racket/pull/2952/checks?check_run_id=333170253#step:9:28

This is macos / cify / unsafe section

The other one is macos / nocify / hash section

argh - I know, lets drop support for macos. :exploding_head:

Maybe it was a butterfly glitch : https://xkcd.com/378/

These hash issues are on weak hashes, which means that the gc behaviour here is really important.

So this might not be a cosmic fluke… :disappointed:

The problems I posted on unsafe seem to be happening on the same set of calls but on unsafe-weak-hashes instead.

Going to open bugs for these so we can think what to do about it.

In most cases these are bugs in the tests rather than in the system, but it’s still good to fix them when possible