4.6.0pre3 available

[Ralf Corsepius]

Am Fre, 2003-05-02 um 02.27 schrieb Angelo Fraietta:
> Joel Sherrill wrote:
> 
> >I need to see if they are happy with the code on the 3.3 branch and 
> >make it apply to 3.2.3.  I will likely just go ahead and put out the
> >first 3.2.3 RPMs when that happens.
> >
> >Sorry for it not being a clean patch.
> >
> I'll continue using Ralf's patch.
The gcc's libstdc++'s maintainers refuse to accept my patch, for 2
reasons:

1. The bug in the original code is caused by using i486
asm-instructions. All I did was to apply their generic (cpu-independent)
code for the i386. According to them, their generic code broken and
therefore the patch can't work in general.

2. There is a general disagreement on the general design of a fix to
this bug. I used #ifdef in a C++-system-header, something they refuse to
do.

IMO, they are probably right wrt. #1. (I already had wondered before,
but ..), but I disagree wrt. #2.

=> Very likely, my patch fixes the xadd issue, it probably introduces a
subtle general functional bug into libstdc++ for the i386. 
IMHO, you don't see this problem (libstdc++ atomic locks) due to certain
characteristics of RTEMS.or due to certain characteristics of your
application (i.e. due to random luck).

> I have noticed, however, that the current tools on the ftp still has gcc-rtems-20030326.diff
> and not gcc-3.2.2-rtems-20030425.diff.
> 
> Why is that?
The patch I sent to you is my personal, private testing version. The
version on ftp.rtems.com are Joel's.

Ralf