Regression in Truffleruby-19.3.0 at memcpy() in C-ext

[larskanis]

Truffleruby-19.3.0 fails with TypeError in two test cases of the pg gem, whereas Truffleruby-19.2.1 works:

TypeError:
  TruffleRuby doesn't have a case for the org.truffleruby.cext.CExtNodesFactory$StringPointerWriteNodeFactory$StringPointerWriteNodeGen node
  with values of type  String(com.oracle.truffle.object.DynamicObjectBasic) java.lang.Integer=0 java.lang.Long=8751735898823355977

It fails in this line of the C code. Full stack trace is here: https://travis-ci.org/larskanis/ruby-pg/jobs/615474667#L628

[eregon]

Thank you for the report.
@aardvark179 Could you look at this?

[aardvark179]

This is another problem with the new memcpy implementation in Sulong. Even if we ensure that the RSTRING_PTR is typed as a polyglot i8 array ``memcpy` still attempts to copy into it with i64s.

[aardvark179]

I have had a discussion with @rschatz and we have a plan to resolve this and #1772. This requires a small fix in TruffleRuby to declare the correct polyglot type for RSTRING_PTR, and some work in our LLVM interpreter to convert managed objects to native in certain cases. This will cause slightly more conversions to native objects than happens currently, but is relatively simple fix to resolve these issues until a more complete solution can be built.

[eregon]

We should try to solve this soon as it fails in pg's CI, for instance:
https://travis-ci.org/ged/ruby-pg/jobs/625409241

[eregon]

This is fixed with 9a22abd and 1bf2f6a, by always returning a native pointer for RSTRING_PTR. It will be in the 20.0 release. Thank you for the bug report!

[eregon]

Ref: ged/ruby-pg#331