Re: [Kissme-general] Re: Building kissme with classpath 0.04

[Tom Tromey]

>>>>> "Alex" == Alex Lau <address@hidden> writes:

Alex> That kind of give me a bad feeling how solid the test base are (
Alex> not the test case ) e.g. gnu/testlet/SimpleTestHarness.java
Alex> Unless the test case is bad also ( I read a couple test cast,
Alex> some shouldn't fail in jdk, but it did )

Mauve works fine for me with gcj.  I mean, libgcj has a few existing
failures, and some of these may well be Mauve bugs (I fixed a Mauve
bug today in fact).  Mauve also works fine when I compile it to
bytecode and run it with gij.  So I doubt that the harness is
significantly buggy.

>> 2)  the Mauve tests looked pretty thin on the ground, and

Alex> ah...I think they started it and havn't put much in lately....

Contributions to it are low.  I try to add tests as I find bugs.  I
encourage everybody else hacking on Classpath to do the same.

Alex> Is there anything better out-there that we can try? Or less
Alex> complex and closer to what we are doing? (in nature ) like a
Alex> vm checker ( debuger? ) and inner class tester

I don't know what you want from an inner class tester.  If you find
something, I'd be interested in it as well.

Mauve has a few VM-level tests that I wrote for testing the libgcj
bytecode verifier.  There aren't many.

I think there are other VM-level tests in kaffe.  I haven't looked at
those.  There may be some in Acunia's VM as well.

For Java compilers, there is the Jacks test suite.

I'm unaware of any other organized Java test suites.  (Or more
precisely, I can't remember any others right now.)

>> 3)  they seemed to focus on complex stuff, not on the low level
>> things we were doing.

The bulk of Mauve is aimed at library-level testing.

If you're interested in writing VM tests, I'm for it.  I can give you
write access; I suggest putting something like this into its own cvs
module.  Various people have suggested doing this before, but nobody
has actually done any work on it.

Tom