Re: Bug #593838: AX_CFLAGS_GCC_OPTION misuses AS_VAR_PUSHDEF variable

[Eric Blake]

On 10/01/2010 06:51 AM, Paolo Bonzini wrote:
> On 09/27/2010 07:46 PM, Eric Blake wrote:
> > > I was able to reduce the problem
> > > to the following trivial case:
> > >
> > > --8<--------------------------cut here-------------------------->8--
> > > AC_DEFUN([MY_MACRO],
> > > [AS_VAR_PUSHDEF([VAR],[prefix_$1])
> > >
> > > dnl This is the version used by AX_CFLAGS_GCC_OPTION.
> > > dnl I believe that it is incorrect. It does not work on Autoconf>= 2.67.
> > > VAR=the_value
> >
> > You are correct that this will not work if VAR is an indirect variable
> > name. It all depends on whether $1 is determined to be an indirection.
>
> The testcase with MY_MACRO([a=b]), which was in Ben's email, does point
> out however a small efficiency regression. I think this is worth fixing
> in Autoconf:
>
> Up to 2.67, the AS_TR_SH would be evaluated at autoconf-time:
>
> prefix_a_b=value
>
> Now instead it is evaluated at configure-time:
>
> as_VAR=`$as_echo "prefix_a=b" | $as_tr_sh`
>
> eval "$as_VAR=value"

Either way, the name of the indirect variable being assigned is 
$prefix_a_b, whether you use pre- or post-2.67.  Why is 'a=b' being 
passed instead of 'a_b' in the first place?  I'd need to see more 
context of the real macro usage (rather than this trivial test case) to 
see whether the efficiency hit is a corner case of a=b is likely in real 
life and thus a performance regression, or just an extreme that is not 
worth worrying about.

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