On 11/07/2012 01:08 PM, Nick Bowler wrote: > > I really think this test needs to be done at runtime. Two reasons: > > (1) A user may first ugprade Automake, then upgrade Autoconf. They > will then get the spurious warnings even though they have > sufficiently recent versions of both Automake and Autoconf. >
Good argument. Also, a user may first upgrade autoconf and then automake. We want to make sure both upgrade paths are sane, if possible, rather than requiring a lockstep upgrade. > (2) A user may have more than one version of Autoconf installed, one > without the warning category and one with. The value hardcoded > into aclocal at build time is therefore guaranteed to be wrong > for at least one installed version. > > Furthermore, the test itself can be simplified: Just run autom4te > -Werror -Wwhatever on empty input (/dev/null will work). For example: > > autom4te -Werror -Wno-m4require-without-m4defun /dev/null Hmm, this goes back to my question of whether autoconf should expose the ability to silence the message by means of including an extra file which does an m4_define, rather than via a new -W command-line argument. Allowing file inclusion as the witness of whether to warn would work even for older autoconf (where including the file has no effect), rather than the current situation of warning about an unknown option. -- Eric Blake ebl...@redhat.com +1-919-301-3266 Libvirt virtualization library http://libvirt.org
signature.asc
Description: OpenPGP digital signature