08:03 <_vertig0> The more I look at the gctransformer and the exception transformer the more I'm convinced they're explicitly designed for code being expressed in C, interestingly enough. No progress otherwise though, learning from the C backend is proving to be exceptionally difficult. The curse of not being one of RPython's original designers, I suppose :)

08:08 <cfbolz> _vertig0: which parts are specifically designed for the C backend?

08:21 <_vertig0> Not sure what you mean? At least both of them seem to lower graphs in a way C can specifically express I think (Such as a global exception object and return values for exceptions), if I understood you correctly

08:27 <cfbolz> _vertig0: sure, but you can also express a global exception value in LLVM bitcode, obviously?

08:31 <_vertig0> I was curious if I could use the explicit support that LLVM bitcode has for exceptions and a GC, but that seems to be a bit out of my paygrade for now. I do think if I go that route I can't use the transformers though since the information needed would probably be lost though

08:31 <_vertig0> (Did not mean to say "though" twice, oops)

08:45 <cfbolz> right, interesting plan

08:45 <cfbolz> we looked into LLVM GC support some (many?) years ago, but concluded that it wasn't stable enough/complete enough for what pypy needed at the time

08:45 <cfbolz> maybe that changed

20:21 <korvo> Aw, None and tuples can't be unified. I don't know why I thought that they could.

20:27 <cfbolz> I'm always annoyed by that

20:27 <korvo> ...Or is there something about dicts such that I can't unify None values and tuple values?