A setter type is the argument's type #7711
Conversation
|
What about |
|
Oh, some obscure codegen bug. Maybe later... |
| bind_to block.break if block | ||
| # For a setter like `foo.x = y` we'd like to use `y`'s type | ||
| if self.setter? | ||
| bind_to args.first |
There was a problem hiding this comment.
Note for later: it should probably be bind_to args.last instead to handle []=
|
I've updated this PR and fixed the original codegen failure (it seems to be caused by the One spec causes a codegen bug. I've deactivated it for now, to check other things (seems it didn't get far, unfortunately). That spec was introduced by #9972. There's no obvious connection between both changes. The compiler change from #9972 itself doesn't seem to have an influence on any other code as far as I can see, it's just that single spec that breaks. But there's also an ICE not covered by specs. |
|
@straight-shoota Thanks! I'll take a look at this in about two weeks |
|
@straight-shoota I'm not sure what I'm supposed to do here. Which spec is failing with an ICE? I see CI just gives an ICE but I don't know where. |
|
I don't know anything more, really. The deactivated spec I talked about is https://github.com/asterite/crystal-1/blob/d864a13b1c6df229b520a9e98b1982aea646b1b9/spec/compiler/codegen/proc_spec.cr#L885 As for the ICE, I didn't dig any further since you said you would take a look at it. |
|
Oh, I thought there was a single spec that ICEd. Hmm... I might not have time to look at this these days, not sure. |
|
I think we won't go with this because it's considered a breaking change, so closing. |
|
For the record: Breaking change doesn't mean this feature will never come. We can have a feature flag for that and make it the default in 2.0. |
Fixes #7707
The way it's done is probably not very efficient, but it probably doesn't matter.