[PLUG] 2.0.36 with Slack 3.5 warning

[Mark Crichton]

On Tue, 17 Nov 1998, James Lewis wrote:

> 
> Even though 2.0.34 and 2.0.35 seemed to complie fine with egcs, 2.0.36 broke.  This
> is not a problem with egcs, but a problem with
> the kernel.  The 2.1 series is suppose to complie fine.  I figured since 2.0.36 is

Does egcs give any indication as to where the problem is?  Linus has been
having some real screaming matches on the Linux kernel development list,
since it seems that the kernel code is causing all the bugs in egcs to pop
up.  He's a little fed up on people blaming the kernel code, when it's
obviously the compiler's optimizer gumming things up.

What are your CFLAGS during the compile, btw?

> suppose to complie fine with gcc 2.8, it'd work fine
> with egcs, but I guess that's what I get for thinking.

>From what I've grepped from the mailing list, is that egcs has problems
with determining which registers are in use.  Granted, last bug report
I've seen on this matter is for egcs 1.1.1, but I'd suspect the bug to be
there in 1.0.3.  GCC works since IIRC, it's optimizer isn't so hell-bent
on trying to get fast code, but fast enough.

Anyways, the 2 workarounds for the egcs bugs are:

IIRC, if X doesn't run, iopl is foobared.  I think you need to take the
arch/i386/io.c file and copy it to your 2.0 tree, and it should work.

Compile with -O2 -fno-inline-functions.  It seems inlined asm is where
egcs goes ballistic most of the time. :|

Give those a shot, I'm curious as to some quickie workarounds for this.
(The correct thing is to fix egcs...but I'm no compiler internals
wizard... ;) )

M


--
This message is from the Purdue Linux Users' Group (PLUG) mailing list.
Send mail server commands to <plug-request@csociety.purdue.edu>.