Re: [PATCH 0/2] First steps in making autoconf ensure a POSIX shell

[Stefano Lattarini]

Hi Eric.

On 11/07/2012 06:18 PM, Eric Blake wrote:
> On 11/07/2012 09:56 AM, Adrian Bunk wrote:
>> On Wed, Nov 07, 2012 at 05:33:54PM +0100, Stefano Lattarini wrote:
>>> On 11/07/2012 04:41 PM, Paul Eggert wrote:
>>>> On 11/07/2012 04:37 AM, Stefano Lattarini wrote:
>>>>> I want to be able to assume the make recipes
>>>>> are run by a POSIX shell.
>>>>
>>>> Thanks.
>>>>
>>>> Are all the features you're testing for specified by
>>>> POSIX 1003.2-1992? (That is, are they all suitably *old* POSIX?)
>>>>
>>> I'm not really sure: I only looked at the last standard available
>>> online on the Austing Group site:
>>>
>>> <http://pubs.opengroup.org/onlinepubs/009695399/utilities/xcu_chap02.html>
>>>
>>> In any case, the features required seem common and simple enough, and
>>> I'd rather start aiming a little high, and then lower our expectations
>>> if we hit a real-world shell that doesn't support all the features
>>> we are testing
>>> ...
>>
>> That sounds like a recipe for getting emergency 2.71, 2.72 and 2.73 
>> releases, plus bad press as a bonus.
> 
> What we've done in the past is to do a release that probes for the new
> features but does not mandate them, where the probe is quite verbal to
> tell people to report the situation if the probe fails; then the next
> release can actually start relying on the features.  (Remember how we
> did it when we started requiring shell functions, for example).  That
> is, autoconf 2.70 must provide the mechanisms for checking for a POSIX
> shell, and even allow clients like automake to use those mechanisms to
> require a POSIX shell, but autoconf itself won't require a POSIX shell
> until 2.71 at the earliest, even though 2.70 is probing for one.
> 
Actually, my patch make the Autoconf-generated configure *require* a
POSIX shell...   I mean, what system worth supporting today doesn't
have such a shell *somewhere*?  What we can't do (thank you, Solaris)
is assume that /bin/sh is a POSIX shell; and my patch doesn't change
that assumption: it just uses the existing infrastructure to look for
a proper POSIX shell at runtime, assuming only old Bourne features
until that point.

Of course, if you want to proceed with your slower & safer path,
that's OK with me (albeit I believe that's overkill); feel free to
adapt my patch accordingly, or to write a follow-up on it.

Regards,
  Stefano