JS code can now catch Ruby exceptions

[bnoordhuis]

Before this commit exceptions from Ruby callbacks were translated to termination exceptions that cannot be caught by JS code. Turn them into regular exceptions that can be caught.

Fixes: #357

[tisba]

I think the truffelruby tests are not supposed to fail 🤔

[bnoordhuis]

I disabled the test for truffleruby. Ideally truffleruby and mruby should behave the same but I don't know enough about the former to say whether that's feasible.

[bnoordhuis]

That "corrupted double-linked list" test failure on the linux-ubuntu-24.04 - ruby-3.2 - gnu buildbot is worrying but...

I tried valgrind --trace-children=yes rake test but ruby itself is so valgrind-unclean, it's useless. Just loading an empty script emits thousands upon thousands of "Conditional jump or move depends on uninitialised value(s)" warnings coming from /usr/lib/x86_64-linux-gnu/libruby-3.3.so.3.3.4 🤦

[bnoordhuis]

The error went away after a rerun. I ran valgrind locally a couple of times but I've not been able to definitely pinpoint any problems in mini_racer itself.

As I mentioned earlier, there are plenty of warnings, and mini_racer.so shows up in some of the stack traces, but the top-most frame is always some internal ruby function operating on a VALUE. I've not been able to trace any of those back to mini_racer not initializing a VALUE properly.

I can't conclusively say it's not a mini_racer issue (the data is too tenuous) but there are no obvious smoking guns.

[tisba]

IIRC, @eregon took a look from time to time to mini_racer's truffleruby support, so I'm just pinging him for FYI.