Sean T Allen

Interesting so this might be aarch64 + linux issue.

@ergl can you try on m1 with the host.cc change removed? so that debug is doing no optimization?

@sylvanc says he is believes that yes, the current usage of catch all code is probably wrong for arm exceptions. but when i told him that aarch64 docs say that...

@jemc same boom

Interestingly, this also segfaults because of the message send that originates in the `else` of the `try`: ```pony actor Main let _out: OutStream new create(env: Env) => _out = env.out...

I discussed this with @sylvanc. He took a look at the pony_try llvm ir that we use and said it looked correct and the personality function looked correct as well....

I've found a difference in our personality function from the c++ abi. When we set registers, we do not do what the C++ does: ```c _Unwind_SetGR(context, __builtin_eh_return_data_regno(1), static_cast(results.ttypeIndex)); ``` It...

So we get the ttypeIndex from: ``` /// Read a sleb128 encoded value and advance pointer /// See Variable Length Data Appendix C in: /// @link http://dwarfstd.org/Dwarf4.pdf @unlink /// @param...

I changed this to "triggers release" as this is actually incorrect exception handling on aarch64.

I talked through this with Sylvan and we came to the conclusion that this is a probably a bug in LLVM like I initially thought. In particular a bug in...