Re: less mixing of object and function pointers

[Ralf Wildenhues]

Hello Eric,

* Eric Blake wrote on Thu, Aug 31, 2006 at 05:47:05AM CEST:
> According to Ralf Wildenhues on 8/30/2006 9:28 PM:
> > 
> > Here's a patch to fix the issue without changing the public interface:

> And nicer than my idea of a union between a data and function pointer.  Is
> there any way to move the testsuite addition away from mdemo and into the
> autotest arena?  But as it is, this is an improvement over no test at all,
> so I would go ahead and check it in if trying to put it in autotest proves
> too challenging.

Well, tests/mdemo isn't going away any time soon, this additional test
comes at virtually zero additional test time cost, and my Libtool work
time is too limited to invest work in this move ATM.

See, I view moving the old testsuite to the new style as a desirable
goal, but as Autotest stands, its use in Libtool has several drawbacks:
- If the granularity of the normal testsuite feedback is limited to
  one line of output per set of tests, there will be a lot less
  information than provided currently by the old test suite
  (imagine mdemo-conf, -make, -exec, -inst, -unst, and all mdemo2-*
  tests executed in one AT_SETUP),
- mdemo in the new testsuite can't be used on systems without Autoconf
  and Automake, unless the generated files could be stored in some way.
- The test exposure of whole packages like mdemo, f77demo, is much too
  valuable to do away with (should you think of not using a generated
  package).
- rerunning the autotools for each testsuite run for each of several
  tests is just too expensive.

I've thus applied the patch as shown.  Thanks for the review!

Cheers,
Ralf