Re: [libstdc++] Port-specific symbol exports, porting notes

[Andrew Pinski <pinskia at physics dot uc dot edu>]

On Sunday, Jul 6, 2003, at 17:12 US/Eastern, Gerald Pfeifer wrote:

> On Sun, 6 Jul 2003, Andreas Jaeger wrote:
> > But we need to support binutils 2.13 - and disabling symbol versioning
> > for older binutils is not an option!
>
> And _at least_ binutils 2.12.1!

No because 2.12.1 has an x86 bug which causes it to look like gcc 
produces bad code, see PR10877.

> On Sun, 6 Jul 2003, Phil Edwards wrote:
> > My opinion is that /by the time 3.4 is released/, binutils 2.14 will 
> > be
> > far more widespread than it is now.  The kind of people who install 
> > 3.4
> > the day it's released will already have installed 2.14.  The kind of
> > people who stay with 3.3 for years to come will also stay with 2.1[23]
> > for years to come, and symvers will continues to work for them as 
> > well.
>
> Negative.  I am bootstrapping GCC mainline on a daily base on FreeBSD
> 4.8-RELEASE and FreeBSD 4-STABLE production and SuSE 8.2 systems here,
> which are using binutils 2.12.1 and 2.13, and will probably keep them
> for quite some time.

I do not think we should support 2.12.1 or 2.13 but only 2.13.1 and 
above as pointed out, they both contain a bug which causes a seg fault, 
see again PR10877.
> > So I submit that nearly all the folks mixing development-stage 
> > compilers
> > with older released binutils will be us.
>
> ...and thousands of users of the BSD ports collections, and many users
> who need to use non-released GCC for various reasons, and so on.
>
> If we can work with vendor tools (as, ld) we really should also support
> older versions of the GNU tools.


Andrew Pinski