Re: [Sbcl-devel] defun-cached/csubtypep and memory reordering problems

[Douglas K. <do...@go...>]

Inserting '(sb!thread:barrier (:write))' at line# 668 of 'early-extensions'
(just before the cond) makes this problem go away.

I'm very surprised this is a recently-introduced flake though, but it's
certainly not out of the question.
I'm going to back out my last change to 'invoke-type-method' and see if the
problem is reproducible.
If not, then I have a theory: maybe better code for invoke-type-method
changes just enough about the program that as you speculated (intentional
pun), there is a memory-order thing that behaves differently. But I don't
think the code generated from the macro change is actually better, I just
think it *looks* better. But what do I know?