I've just uploaded InlineX-CPP2XS-0.21, which now has the capability of
writing a portable Makefile.PL and xs file - by specifying WRITE_MAKEFILE_PL
=> 'p' (or --write_makefile_pl=p if you use the cpp2xs utility).

The test script (t_portability.t) uses the Math::Primes::FastSieve demo to
do some checking. It just checks that the expected files are generated, and
does some sanity checks on them.

One thing I noticed in CPP.pm's validate subroutine is that it does:

#################################
$o->{ILSM}{AUTO_INCLUDE} ||= $auto_include;
$o->{ILSM}{AUTO_INCLUDE} =   $flavor_defs  .  $o->{ILSM}{AUTO_INCLUDE};
#################################

(You've probably always done it that way - I haven't checked.)

For the purposes of InlineX::CPP2XS (and perhaps other purposes, too), it
would make things easier if it did:

#################################
$o->{ILSM}{AUTO_INCLUDE} ||= $flavor_defs . $auto_include;
#################################

But I don't know whether that fixes a bug in validate(), or creates one, or
does neither.
As regards the tests in the I::CPP test suite it seems we can do it either
way, and it makes no difference.

In case you haven't spotted the relevance, validate() will currently always
prepend auto_include with $flavor_defs. But you can avoid that if you use
the latter.
And, for portability, we don't want the contents of $flavor_defs to appear
anywhere in the shipped C++ source code - hence the relevance to
InlineX::CPP2XS.

(Btw, don't stress about this .... I already have "workarounds" in place,
anyway :-)

Cheers,
Rob