Re: testcase for "absolute source" patch

[Michael Chastain <mec.gnu@mindspring.com>]

Baurjan Ismagulov <ibr@ata.cs.hun.edu.tr> wrote:
> I don't remember any decision from the recent thread on this issue.

That's true.  The last time I posted about it, I was leaning towards
requiring that build != host work, partly because Dan Kegel's
documentation on testing cross-gcc explicitly describes how to
set up build != host, and partly because Felix Lee stepped up and
said that he had used it in the past.

But I failed to announce a decision.  "leaning towards" is not
really a decision.

> As far as I could understand, there are also other tests that do not
> work well in build != host scenario.

And that's also true.  I posted a list of 11 test scripts that don't
work with build != host because they neglect to download their .h files.
There are probably others that fail for other reasons as well.

> While I agree that fixing the infrastructure is also important, may I
> suggest that we have at least local checking in the trunk?

I'm going to reject that argument.  I don't want the test suite
to add more stuff that runs in some conditions but not others.
That becomes a source of unending maintenance work.

To me, a flawed test *is* sometimes worse than no test at all, because
then I have to maintain it.  I would rather have more time (and more of
other people's time) actually running the test suite, reporting the
results, analyzing the failures, and creating good PR's, rather than
adding test scripts that don't work in some supported environments.

As long as we have a "build != host" requirement, I won't accept any
scripts that obviously won't run in that environment.

I'm still open to killing that requirement.  You've exhibited a test
script that (a) is useful and (b) has a hard time meeting that
requirement.  That balances against the usefulness of keeping
build != host.

Michael