Re: [RFC] Could autoconf-generated configure scripts start requiring a POSIX shell?

[Stefano Lattarini]

On 03/04/2012 04:48 AM, Eric Blake wrote:
> On 03/03/2012 07:46 PM, Paul Eggert wrote:
>> On 03/03/2012 11:01 AM, Stefano Lattarini wrote:
>>> would you think about the possibility of making autoconf-generated
>>> configure scripts *require* a POSIX shell in order to run
>>
>> Doesn't it already do that?  'configure' scripts already look for
>> a better shell, one that presumably conforms to POSIX better.
>>
>> I don't think 'configure' needs to check for strict conformance
>> to POSIX (hardly any shell does that), only stuff that's useful
>> and practical -- which is what it does now, no?  So perhaps I am
>> not understanding your proposal.
> 
> I think the idea is to find a shell that supports $(), ${foo#bar}, and
> other POSIX constructs that Solaris /bin/sh lacks, but which can be
> found on other shells installed on Solaris.
>
Exactly.  And punt if it doesn't find such a shell.  Sorry for not having
been clearer.

> To some extent, Jim
> Meyering has already insisted on finding a POSIX shell to run the
> coreutils testsuite, borrowing ideas from autoconf for finding a better
> shell, but insisting that the shell that is found has more than the bare
> minimum required by current autoconf.  I'm actually in favor of the
> idea,
>
I'm very glad to hear that.  And I believe that, in the long run, this move
will help improving automake as well, since the recipes it generates (which
are run with the configure-determined $SHELL) will be able to assume more
POSIX features.

> post-2.69, because we haven't had any complaints about the
> inability to run the coreutils testsuite, and therefore we can assume
> that most systems these days have a shell with a bare minimum of $().
>
It might also be worth noting that Git too assumes a POSIX-ish shell
(i.e., on Solaris, its build system and its installed scripts work with
/usr/xpg4/bin/sh, but not with /bin/sh); to my knowledge, nobody has
ever complained.

> It may be worth injecting a probe even into 2.69 (remember, for several
> releases, we probed whether shell functions were available, before
> requiring them).
> 

Thanks,
  Stefano