Re: Problem with Sun Studio 12 C++

[Ralf Wildenhues]

* Eric Blake wrote on Sun, Mar 08, 2009 at 03:48:40PM CET:
> According to Ralf Wildenhues on 3/8/2009 4:21 AM:
> > I think the Autoconf testsuite deserves to detect this issue,
> > lest it happen again.  How about this patch?
> 
> I like it.  Fix the nits, then apply.

Done; thanks.

> > The cpp-works stuff is necessary for systems without a working C++
> > compiler.  AC_PROG_CXX sets CXX to g++ in that case.  We should find
> > out why that was done, and fix it.
> 
> Didn't libtool already have to hack around that fact?  Yes, it would be
> nice to make autoconf more honest when CXX cannot be found, but as a
> separate patch.

yes and yes and yes.

> > +int foo (int * restrict ip)
> > +{
> > +  return ip[0];
> > +}
> 
> If I understand restrict correctly (although I'm not entirely sure I do),
> it does not make sense to use restrict unless there are at least two
> pointers in play, because you can't have aliasing without multiple
> references.  While I don't think the language forbids a single parameter
> with restrict, I also think that it may be better to write this as:
> 
> int foo (int * restrict i1, int * restrict i2)
> {
>   return i1[0] + i2[0];
> }

Sure, why not.  I simply took the test from AC_C_RESTRICT itself, which
uses only one argument, too.  It doesn't really matter here, as the
compiler will typically barf with a syntax error if the restrict keyword
(whatever it's #define'd to) is not understood.

Cheers,
Ralf