Re: FYI: Adding RuntimeException checks

[Mark Wielaard <mark at klomp dot org>]

Hi,

On Tue, 2002-04-02 at 19:56, Tom Tromey wrote:
> >>>>> "Mark" == Mark Wielaard <mark@klomp.org> writes:
> 
> Mark> Also sometimes libgcj throws NullPointer exceptions were it
> Mark> shouldn't which aborts the rest of the test completely.
> 
> Please submit gcj bug reports for these.

I already submitted a patch for the java.lang.Long failures. And I will
go through all libgcj/mauve test results in the next days/weeks to
analyse them. I try to post daily summaries to gcc-testresults at
<http://gcc.gnu.org/ml/gcc-testresults/>. (But the site seems down at
the moment.)

> Mark> It is a workaround for bugs in libgcj. But some testlets have
> Mark> hundreds of tests which just don't run if one of those throws an
> Mark> unexpected RuntimeException or Error. The real solution is to
> Mark> split those testlets into smaller ones that only test one
> Mark> particular method.
> 
> I appreciate what you're doing here.  However even though I care a lot
> about gcj I want to emphasize that I think we shouldn't let these
> changes impact the maintainability of the tests themselves.

To be honest some of the tests are just big code blobs and it would take
real effort to make them less maintainable :)

What I am currently doing is based on results with gcj but should
benefit the robustness of those tests on every platform. Some of these
things should be done for all the tests or might actually be done by
improving the framework. I will keep this in mind when making changes. 

Please let me know if you would like to see me ask for permission to
commit patches on the list. Having people review code changes is a good
thing, if people have time to actually do it.

Hopefully when mauve can be used to run daily automatic tests against
gcj I can also set this up for GNU Classpath+Orp. So maintainability is
a high priority for me.

Cheers,

Mark

P.S. I cannot reach sources.redhat.com at the moment since AboveNet
seems to be down or blocks all my traffic. So this message might take a
while to reach the list.