Bruno Haible <bruno <at> clisp.org> writes:
> So, how to resolve this contradicting requirements?
>
> Possibility 1: Put the SIGNATURE_CHECK into a file tests/signature.h, and put
> ASSERT, SIZEOF and a few others into tests/macros.h. A test looks like this:
>
> #include <config.h>
>
> #include <specification.h>
>
> #include "signature.h"
> SIGNATURE_CHECK (...);
>
> #include <other.h>
> #include <other2.h>
> #include "macros.h"
>
> Possibility 2: Keep the SIGNATURE_CHECK definition in config.h, on behalf of
> m4/gnulib-common.m4, and put ASSERT, SIZEOF and a few others into
tests/macros.h.
> A test looks like this:
>
> #include <config.h>
>
> #include <specification.h>
>
> SIGNATURE_CHECK (...);
>
> #include <other.h>
> #include <other2.h>
> #include "macros.h"
>
> I've explained why possibility 1 is preferable, but now I am actually
undecided.
> What do you think?
It's less work for me if we go with possibility 2, but I have no qualms with
the work to go with possibility 1. Either way we create macros.h and put
ASSERT there, and the creation of macros.h is now decoupled from my signature
check macro. The difference all boils down to whether SIGNATURE_CHECK is in
its own header ("signature.h", affecting only the unit tests that include it),
or whether it is in a common header (<config.h>, and probably should be named
_GL_SIGNATURE_CHECK to avoid pollution, since config.h is also included by the
actual project). I'm 55-45 for going with option 1. Anyone else want to speak
up?
--
Eric Blake
