Re: Bytecode-to-native regression tests

[Tom Tromey <tromey at redhat dot com>]

>>>>> "Bryce" == Bryce McKinlay <mckinlay@redhat.com> writes:

Bryce> I propose adding a libjava.bytecode test suite which will contain
Bryce> binary .class files for such tests. Obviously for licensing reasons we
Bryce> (in general) can't just go adding bytecode from random failing .jar
Bryce> files in there, but in cases where a reproducable test case exists, we
Bryce> could place its bytecode there. The source code would be included too,
Bryce> but only for reference purposes - only the bytecode version would
Bryce> actually be used/compiled.

Bryce> Anyone have any concerns about this?

What if instead of source code we checked in the jasmin-syntax
assembler source (as well as the class files)?  Then we could
conceivably make minor tweaks and regenerate if needed.  This would
also make it simpler to test variations on a theme.  I'm just
thinking of a situation where we need to make some change (dunno,
change a method name or something) but we can't because no existing
compiler will generate the same weird bytecode.

Tom