gc.m4 and hard failure

[Ralf Wildenhues]

It would be great if the gc.m4 module could be taught not to fail when
no random devices exist (as seen on hppa2.0-hp-hpux10.20).  Passing

  --disable-random-device --disable-pseudo-random-device \
  --disable-nonce-device

did not help.  libgcrypt was not installed either.

This would help automated testing of gnulib mega test; faking a cross
compile helps but is so different as to be an actually different test.


More generally, I think it would be good if all Gnulib modules allowed
at least some way of invocation that allowed configure to continue even
if the tested resources are not present; an extra RUN-IF-FAIL argument
(possibly defaulting to AC_MSG_ERROR or so) could serve well, for
example.

Cheers,
Ralf