[COMMIT] Eliminate some non-X11 build failures

[Stephen J. Turnbull]

On 10/15/07, Aidan Kehoe <kehoea at parhasard.net> wrote:
>
>  Ar an triú lá déag de mí Deireadh Fómhair, scríobh mike.kupfer at xemacs.org:

>  > That is, if the function isn't bound at compile time, don't bother
>  > checking at runtime.  Is that going to be a problem?
>
> The check should be at runtime, and only at runtime.

Yes.  MH-E has had problems with this a number of times in the past.  Keep an
eye out for such code.

>  > Also, will I need to do a test-build with a non-X11 XEmacs before
>  > committing?
>
> That would be polite, but no-one else is holding themselves to that. The
> package smoketests: http://labb.contactor.se/~matsl/smoketest/logs/
> use an X11 build, and they're what people look to.

I agree.  In fact, I don't see a point to doing multiple test builds
for different
builds of XEmacs of a given version at all.  I'm not an expert on the byte-
compiler, but AFAIK in the absence of compilation conditioned on features
there should be no differences between a package byte-compiled by an
X11 build and one byte-compiled by a non-X11 build of XEmacs.

In theory, if there are conditional compilations such that a macro needed by
the build gets optimized out of existence or a callable changes signature for
different features of XEmacs, you could get build problems with feature-
conditional compilation.  But that seems unlikely; macros are tricky enough as
it is, people probably don't do that kind of thing.

What would be nice would be to do functional tests in several different
builds of XEmacs for the package, but that's unreasonable without an
automated test.

I would say that the best policy is to do a build with 21.4 with Mule.  That is
what Norbert will distribute AFAIK, and it will catch the vast
majority of thinkos
and typos.  Then keep an eye on the smoketest for the other three principal
variations on the byte-compiler.