On 08/24/2010 11:37 PM, Ralf Wildenhues wrote:
>> The more aggressive approach is to assume that all shells that
>> support shell functions also support : in variable substitutions;
>
> I can't really judge this one well.  If Paul doesn't know, Sven
> Mascheck's pages are a good indicator.

I think I'll go with the more aggressive patch and assume that : works
based on [1]; hopefully enough people with alternate shells run the
testsuite and point out any regressions (although such are unlikely to
occur); I've already tested some older versions of ash, bash, zsh, posh,
dash, yash, ksh, and Solaris /bin/sh.

[1] http://www.in-ulm.de/~mascheck/bourne/
says that System III added ${a:=b} at the same time as [!..] negated
globbing, and that functions came later in SVR2.

>
>> Should we also update tests/m4sh.at to add a test next
>> to 'Negated classes in globbing' and 'Functions support'
>> to reiterate this spy in the autoconf testsuite?
>
> Yes, that would be good.  Thanks!

Some of the comments in the existing tests are out of date, too.  I
guess that means I need to respin and re-post this patch with all of
those fixes.

>> +_AS_RUN([_AS_SHELL_NULL_WORKS]) || {
>> +  echo No shell found that supports \${a:-b}.
>
> Please quote {a:-b} otherwise there is the slight chance that the { }
> get removed if there is a file named '$a:-b' in the current directory.
> Also, output to stderr?

Sure.  I was blindly copying from commit 734abde3, which was the last
time we had a spy in autoconf code, and used stdout instead of stderr.
But since I'm going to go with the more aggressive approach of assuming
: works, we won't need the spy after all.

--
Eric Blake   eblake@redh...    +1-801-349-2682
Libvirt virtualization library http://libvirt.org