Ralf Wildenhues wrote, quoting me:
>> you could just inline the assignment:
>>
>> mandir=`echo $mandir | sed s?^${prefix}?/usr/local?`
>>
>> in your configure.ac.
>
> Which will make your package blatantly incompatible with the old
> version of the GCS;
Eh? The OP's intent may not be strictly GCS compliant in any case,
but I fail to see how this would make it any less so.
> and will stop working once your package uses Autoconf-2.60 which
> supports the newer version.
Sure. I did say the the suggestion was `off-the-cuff'; that means
it's quick and dirty, and by no means likely to be future proof.
It's compatible with autoconf <= 2.59, (at least >= 2.50, and maybe
even some earlie versions); looks like it will break in 2.60, but
this is always a peril when hacking below the surface, to achieve
something out of the ordinary.
> Why not *just* *use* --mandir?
Yep; that's what I would be inclined to do myself, but the OP
*explicitly* said that he wanted to avoid this. Perhaps the issues
this sort of hacking raises will help to convince him that just
using `--mandir=...' is a significantly more robust option.
> Why not a script that invokes `configure --mandir=... $@'
Because it doesn't conform to the standard `./configure && make ...'
build paradigm?
> If that's too much, why not add a setting to your and your preferred
> users' config.site file(s), so they don't have to remember this?
How the heck do you, as a developer, possibly hope to control what
your end users put in *their* config.site files?
Cheers,
Keith.
_______________________________________________
Autoconf mailing list
Autoconf@xxxxxxx
http://lists.gnu.org/mailman/listinfo/autoconf
