Re: Error with vtable thunks in EGCS under Linux

[Bill Kelly <billk at cts dot com>]

> Date: Thu, 11 Feb 1999 11:24:27 -0800
> From: mrs@wrs.com (Mike Stump)
>   To: hjl@lucon.org, hoffman@albirio.crd.ge.com
>
> > From: hjl@lucon.org (H.J. Lu)
> > To: hoffman@albirio.crd.ge.com (William A. Hoffman)
> > Date: Thu, 11 Feb 1999 08:14:15 -0800 (PST)
>
> >           if (flag_vtable_thunks)
> >             {
> >               /* We don't have dynamic thunks yet!
> >                  So for now, just fail silently.  */
> >             }
>
> > We chose to generate bad code instead of abort (). I don't
> > agree with it. But .......
>
> While we could do this, g++ would then cease being a C++ compiler,
> what it would become is a tool for creating core files.

But it *isn't* a C++ compiler when it's generating bad code.  Or do you
mean the if (flag_vtable_thunks) test would trap legitimate cases, as
well as those where bad code is generated?

I know programmers place different priorities on issues like code
correctness versus, in the case of vtable thunks, execution speed.  But
before now I would have assumed that given a scenario where the compiler
could reliably detect it was about to generate bad code, everyone
concerned would prefer it abort rather than fail silently.

For the record, I think vtable-thunks should be disabled by default
until it works correctly.  When enabled, it should--at least--complain
loudly before generating bad code.

Bill Kelly
<billk@cts.com>