David Chisnall

The changes to be blocks trampolines look like they'll make the trampolines larger than two pointers and so require some changes to the size typedefs. I'm somewhat unconvinced by CFI...

I think this is the right thing (implicitly linking a non-default unwinded will break things for people), but I’d like CI to be green…

All of those errors look like they're from the CRT. If you're using a recent CMake (3.20 or later?), it provides a cleaner way of specifying the target than depending...

The libobjc2 CMake contains numerous hacks because it requires a version of CMake that predates CMake's official support for Objective-C. Riccardo objected to moving it forward. At some point I...

Can you provide a reduced test case? To be able to debug this I need a preprocessed source file and the minimal compiler command to reproduce it.

I am not actively using GNUstep or anything Objective-C related, and have almost no time to work on it at the moment. If there is a *reduced* test case, then...

Minimal test case: ``` $ cat crash.m int main() { @try { } @catch (id e) { } @finally { }; } $ clang -cc1 -triple x86_64-pc-windows-msvc -emit-obj -fobjc-runtime=gnustep-2.0 -fobjc-exceptions...

It looks as if this is a bug in the Windows EH code for Objective-C with all runtimes. Clang is trying to pop something off the EH stack and expecting...

> There may be other uses for read-only capabilities to writable capabilities but I'm not aware of them. Read-only pointers to mutable data is quite a common idiom. For example,...

Sounds good to me, thank you!