Re: gcc 2.8.1 (Re: Uploaded 2.0.34pre9)

Mikael Pettersson (Mikael.Pettersson@sophia.inria.fr)
Tue, 21 Apr 1998 13:44:04 +0200 (MET DST)


On Mon, 20 Apr 1998, Dean Gaudet wrote:
> > If you force people to stick to old gcc versions just to compile
> > a linux kernel, you will never discover the bugs in gcc-2.8.1,
> > and you will never discover broken hacks in the kernel source.
>
> I think it can be put more like this: the folks working on 2.0.3x
> probably have their hands full enough as it is just dealing with the bugs
> in linux-2.0.3x. Changing the compiler adds a whole slew of new bugs
> which they don't have to find; folks don't have to use gcc-2.8.x.

People here seem to assume that just because gcc 2.7.x compiles
the 2.0.x linux kernels ok, while gcc 2.8.x sometimes don't, that
gcc 2.7.x therefore is the better/less buggy compiler. 'Tis not so.

Let me explain. I've written compilers for >10 years, and on several
occasions I've constructed compilers that compile via ANSI-C.
Since 1995, I've must have submitted at least 10 bug reports for gcc
2.6.3/2.7.x, where gcc would blatantly miscompile fairly simple code.
In fact, one of these gcc bugs was a complete show-stopper for
the x86 target. No workaround, bogus code, core dumps.
This particular bug wasn't fixed officially (*) until the 2.8.x
series. Now you tell me: "folks don't have to use gcc-2.8.x" ?!
(Ok, so forget 2.8.0. But 2.8.1?) _I_ have _zero_ faith in unpatched
gcc 2.7.x for x86 machines.

[(*) I had an unofficial fix, but that doesn't help third
parties much.]

I wish people would give gcc 2.8.1 a try. If it doesn't work,
_report_ the problem so that compiler hackers like myself can
investigate it and (hopefully) identify the cause and correct it.

As for myself, I've compiled every linux 2.0.3x/2.1.x kernel since
February with 2.8.0/2.8.1. Apart from the ioport.c fix, no problems.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu