Re: Segmentation fault building libg++ without named returns

[Manfred Hollstein <manfredh at redhat dot com>]

On Saturday, 9 September 2000, 18:06:51 -0400, dave@hiauly1.hia.nrc.ca wrote:

> > Thanks for the patch you sent with your previous email; I observed the
> > same failures when Mark installed his patch to  warn about using named
> > return values.  I took a differend approach  than you, i.e. instead of
> > removing  the code completely, I used  the opposite logic  with a flag
> > _G_USE_NRV  as  opposed to the  former   _G_NO_NRV; I've attached that
> > patch to this email.
> 
> I tend to think the code should be removed.  There were essentially two
> sets with the "same" code.

I agree; however, I'll keep the code for some time, at least until it's
working again with the current gcc versions.... and until I've checked
the _NO_NRV code works with older compilers, too.

> However, the unused branch which didn't use
> named returns had rotted a bit.  Apparently, `and', `or' and `xor' are
> c++ diagraphs and can't be used.

Yup, my fix included patches for this also.

> > Regarding the missing   weak symbols,  I'm pretty   much  at a   loss;
> > building Fix.cc and  Rational.cc  doesn't generate all  required  weak
> > symbols as  it  did  before; even worse,   using  a reduced test  case
> > doesn't fail to  generate those symbols...  I'll dig  further and keep
> > you informed.
> 
> I don't have any time this weekend to work on it but I think a first
> step would be to dump the assembly code for Fix.cc and see what is
> happening with respect to the week definitions.

I did already, they aren't even emitted...

> Then, looking at the
> rtl may help to pinpoint where things are going wrong.  It's a fairly
> recent change to the compiler that has caused the problem.
> 
> Dave
> -- 
> J. David Anglin                                  dave.anglin@nrc.ca
> National Research Council of Canada              (613) 990-0752 (FAX: 952-6605)

Cheers, manfred.