Open Quuxplusone opened 4 years ago
MadCoder, please advise whether you are aware of this issue. Thanks. I am going to have a debug version and dig this a bit today as well.
Did you mean to attach something to this bug?
If you can reduce a self-contained test case that crashes, that would be very helpful. Looks like it's in the constructor for SCTestMainEntityBuilder.
Attached reduced.mm
(314 bytes, text/x-objcsrc): Reduced test case
Thanks, that's perfect.
Attached direct.patch
(943 bytes, text/plain): patch to fix getSelfType
We can move discussions there. There are some other bugs I need to fix along the way. It appears there are additional mangling going on in CodeGen part that makes GetNameForMethod won't return correct linkage name for a method. Do you have any pointers on this?
(In reply to Liu Liu from comment #7)
> We can move discussions there. There are some other bugs I need to fix along
> the way. It appears there are additional mangling going on in CodeGen part
> that makes GetNameForMethod won't return correct linkage name for a method.
> Do you have any pointers on this?
Email would probably be a more productive venue for that. I don't know why
there'd be additional mangling; ObjCMethodDecls shouldn't be able to
accidentally
fall into the C++ name-mangling paths.
reduced.mm
(314 bytes, text/x-objcsrc)direct.patch
(943 bytes, text/plain)I am experimenting with this attribute from commit: https://github.com/llvm/llvm-project/commit/d4e1ba3fa9dfec2613bdcc7db0b58dea490c56b1
I am on trunk 705a6aef350246c790ff8e73864dd27a640c59c8
Adding direct in property attribute to ordinary objc class with callsites from C++ side seems cause issues. Attached project exhibit this. Also attached crash stacktraces and clang diagnostic package.