This is the mail archive of the
libstdc++@gcc.gnu.org
mailing list for the libstdc++ project.
Re: libstdc++ vs. -fshow-column
- From: Aldy Hernandez <aldyh at redhat dot com>
- To: Benjamin Kosnik <bkoz at redhat dot com>
- Cc: libstdc++ at gcc dot gnu dot org, janis187 at us dot ibm dot com
- Date: Thu, 11 Jun 2009 15:00:38 -0400
- Subject: Re: libstdc++ vs. -fshow-column
- References: <20090611111711.7fa26b49@mcgee.artheist.org>
On Thu, Jun 11, 2009 at 11:17:11AM -0700, Benjamin Kosnik wrote:
>
> Hey y'all:
>
> > The libstdc++ testsuite doesn't use the overrides for dg-error et al.
> > I believe Janis is looking into this.
>
> Thanks for the workaround on this Aldy.
>
> Is there any way I can get more detail on the "dg-error overrides"
> problem? Not quite sure what you mean here, or what it is that should be
> done. Any pointers appreciated.
Janis-- please correct me if I'm wrong.
The default dg-{error,warning,etc} procedures in dejagnu are currently
overloaded in the C and C++ dg.exp machinery to handle columns. I
believe Janis did this because it was easier to do this, than to require
GCC to use a newer version of dejagnu (and get our patches into
dejagnu).
GCC now defaults to -fshow-column, but stock dejagnu does not handle
columns gracefully. It gets confused if a column number matches a line
number that you want to match. There are thre possible solutions for
the libstdc++ testsuite:
a) Override the libstdc++ harness like we're currently doing for
C and C++ dg.exp.
b) Modify any tests affected to match the actual column.
c) Make the libstdc++ testsuite harness pass -fshow-column.
We need to do (a) sooner or later, and ideally (b) as well as we become
more comfortable with the C++ front-end emitting correct column
information. Approach (c) is just papering over the problem.
Aldy