Re: Fix libstdc++ testing using upstream dejagnu

[Daniel Jacobowitz]

On Tue, May 18, 2004 at 08:28:59PM -0600, Rob Savoye wrote:
> On Tue, May 18, 2004 at 06:20:37PM -0400, Daniel Jacobowitz wrote:
> > This patch fixes an issue I found last week when trying to test libstdc++. 
> > -isystem on many targets implies an implicit extern "C" { }, which doesn't
> > work at all for the C++ headers.  The version of DejaGNU in
> > sources.redhat.com has been using -I for as far back as CVS history goes.
> 
>   I'll have to dig through old email, but I believe that change to -isystem
> was a patch somebody submitted for cross-testing. I'll take your word
> for it that there is a current problem, but I need to track down that
> patch from about a year ago. -I was the original behaviour for most of
> DejaGnu's history.  To be honest, shouldn't this behaviour in G++ be fixed ?
> Shouldn't it work the same for all targets ?

No.  Some targets the system headers (/usr/include) are C++ aware, and
some they aren't.  If they aren't, you need this.

Hmm, the patch was from H-P.  What do you remember about this?  Maybe
we need to -isystem the newlib/libgloss headers, but not the libstdc++
ones.

-- 
Daniel Jacobowitz