Andrew Anderson

@jrp2014 There are a couple of different things going on here, but to me this looks like a bug in this Kaleidoscope implementation. What this Kaleidoscope does to implement anonymous...

@jrp2014 It might be worthwhile to have a go at updating the "official" llvm-hs kaleidoscope (https://github.com/llvm-hs/llvm-hs-kaleidoscope) to use the OrcJIT interface instead of the MCJIT interface? It would be great...

Ah, I see the root of the problem: `disposeMangledSymbol` disposes our handle to the mangled symbol. It does not *destroy* the mangled symbol; the mangled symbol still exists in the...

@jrp2014 thanks for updating the kaleidoscope! I've merged your changes onto the llvm-12 branch. It seems like all that remains for basic OrcJIT v2 support is to uncomment and fix...

@cartazio If you have some time, fixing up those tests would be very helpful! This is fundamentally a bit of a messy task in general, because upstream is still very...

@cartazio we should probably remove the old ExecutionEngine and MCJIT code as well, which will require some surgery. The tests for that code should be retargeted to test the new...

@cartazio I've had a look through the functionality exposed by upstream via the C API (https://llvm.org/doxygen/Orc_8h.html) and we definitely are not at 100% coverage, so it would be helpful to...

@jrp2014 this is a lot to discuss in the comments of a specific issue, so I'd suggest opening issues for these things so they can be tracked properly, but briefly:...

As of #340 we now have JIT, optimization of modules with the pass manager, and calls/resolution of external library function symbols operational and being tested, so I think we can...

Thanks for the heads up! Any idea if this will make it into a 12.X release? If not I will move it to the LLVM 13 milestone, but if it's...