gcc-bugs Digest 31 Jul 1999 20:05:06 -0000 Issue 383

[Casper Dik]

>From: Ben Elliston <bje@cygnus.com>
>To: Richard Henderson <rth@cygnus.com>
>cc: "H.J. Lu" <hjl@lucon.org>, gcc-bugs@egcs.cygnus.com

>> > Shouldn't the -DGOOD version be used?

>> Yes.
>
>> > Is that a bug in vfork test in autoconf?

>> Yes.

>Can I get as little more context here?  I'm not familiar with this `GOOD'
>define, and find it hard to believe that this is the name of a
>preprocessor macro that would feature in C library headers for altering
>the behaviour of vfork().

>Solaris is not my main target and I took another look at the code.
>It seems that my Solaris may have a vfork bug. I don't have the
>time for it.
>
>-- 
>H.J. Lu (hjl@gnu.org)


As I've said in:
	http://egcs.cygnus.com/ml/gcc-bugs/1999-07/msg00618.html
	http://egcs.cygnus.com/ml/gcc-bugs/1999-07/msg00559.html

There is a bug in the HAIFA scheduler for SPARC; one instruction is
moved to the delay slot of "call vfork", but code further on
depends on the output of that register. vfork() will clobber all
registers.

Without the "standard" scheduler, the vfork tests succeeds (at -O2)

Casper