Re: Fix for PR java/1895

[Tom Tromey <tromey at redhat dot com>]

>>>>> "Jeff" == Jeff Sturm <jeff.sturm@commerceone.com> writes:

Jeff> In the long haul, will you preserve the existing testsuite and
Jeff> Mauve, or replace the testsuite with Mauve?  Is there any good
Jeff> reason for both?  What guidelines do we have?

Mauve is really just for library tests.  At least, that's true right
now.  There's been talk of making it do more but no one has taken
action.

With the arrival of Jacks I think we can separate tests into 3
categories:

* Library tests, which should go into Mauve.  Since we have a fair
  amount of control over Mauve (eg, if a libgcj hacker wants write
  access they should just ask me...) we can commit any library tests,
  including regressions, there.

* Front end tests.  These should go to Jacks.  However, I suspect
  Jacks won't be interested in accepting mere regressions for gcj.

* Everything else, which we can put into the libgcj test suite.  This
  covers non-library runtime tests (eg, JNI tests), regressions
  unsuitable for other test suites, etc.

Jeff> I think it's important to have something in the GCC repo that
Jeff> runs when we do "make check"; otherwise it's not so easy to find
Jeff> regressions.  Especially for those who aren't maintaining Java;
Jeff> they may not even know of the existence of Mauve.

Yeah.

As of yesterday we have a web page documenting the other test suites
we use.  I'm skeptical that we'll ever get the non-Java gcc
maintainers to do any Java testing whatsoever.  It hasn't worked so
far...

Tom