This is in order to test against compilers other than GCC.

This fixes <#13>.

…PERL_MM_OPT

`PERL_MM_OPT` does not work with Makefile.PLs in subdirectories. Running

    perl Makefile.PL CC=clang; find -name Makefile -exec grep -H '^CC =' {} \;

shows that `CC = clang` in the toplevel Makefile, but `CC = cc`
elsewhere.

All of these came from one line in Basic/Gen/PP.pm that used an
assignment in a macro argument. But in the Perl macro cascade
there is #defined HvAUX(hv), which uses its argument twice.
This caused the assignment to happen twice. It was a harmless
unintended consequence, but at least the warnings have gone away.

It used to be that you could call printf(buf), but now for security
purposes it needs to be printf("%s",buf). Since Perl_croak just uses
printf (I think), I just added "%s" to a single line in Dev.pm.

This is a bit nanny-ish, just asking for an extra set of parenthesis
to make sure we understand the C operator precedence rules. We do.

This warning happens when you do an assignment in a conditional:
if (a=something()).  Compiler checks to make sure you didn't mean
if (a==something()). Get around that by adding an extra set of
parentheses around the assignment.

clang was complaining because the generated code turned out to
be somthing like 'PDL_short foo = 0.25;' which is a type mismatch.
0.25 got rounded to 0 before assignment to PDL_short foo.
This is actually what the cheesy test for floating-pointiness
intended, so I suppressed that warning in the Makefile.PL.

An alternative would be to have a type() look for integers and
one for floating-point datatypes.

It's been a few decades since void was introduced...