Improve inlining of return_call*

[vouillon]

Use the previous implementation when no return_call is in a try block. This avoids moving code around (as a sibling of the caller body or the inlined body), so that should allow more local optimizations after inlining.

[tlively]

Thanks, this LGTM! Have you had a chance to run the fuzzer on this? Usually running it for about 10k iterations is enough to give us confidence for a change like this.

[vouillon]

It seems the fuzzer does not like the last commit. I think that's because try_table is not fully supported yet. For instance, fuzz_opt.py 9274061891157013786 yields the following error:

unimplemented DCE control flow structure
UNREACHABLE executed at /home/jerome/sources/binaryen/src/passes/DeadCodeElimination.cpp:189!

Without this commit, I have been able to run it for about 10k iterations, with only a few failures due to try_table, but maybe I'm no longer testing tail calls within try blocks?

[tlively]

Ahh, right. I was hoping to use try_table since that's the new standard instruction, but we should change the tests to use try instead because that's better supported by all the optimization passes. The fuzzer shouldn't have any issues with try.

[vouillon]

I did not found any issue with the fuzzer.

[tlively]

Great, thanks!