scoder
Does the chained attribute access mean that you can no longer traverse object attributes without owning the GIL?
Interesting. Don't do that. The second declaration simply replaces the first one (as always in Python). We should reject duplicate definitions also for `__cinit__`. It's probably just letting it pass...
Thanks for the report, that looks surprising. It's not constant folding, though. Your type cast prevents that.
Were you using Cython 3.0.x? The stack trace for the current master looks different.
The `constant_result` that I get is `'((double)10)'`. So, it might still be a problem in the constant folding if arbitrary type casts end up with a string as result.
I think the problem in your sagemath example is an inline function calling another inline function. We keep track of whether an inline function is actually used and discard it...
> The problem with that is that the module state lookup should be different depending on where you're requesting it from (and not the scope that the entry lives in)....
Merging current master in order to update the CI config.
I think the ambiguity behind this is a striking argument against it. It's unclear whether it means "returns `void`" or "only propagates exceptions" or "returns `None`". I'm sure many users...
> So the best short-term fix would probably be to accept and ignore the annotation (which practically will become "returning an unchecked Python object") OTOH, given that annotations didn't work...