Mail Archives: pgcc/1999/03/14/20:08:34
On Sun, Mar 14, 1999 at 02:15:17PM -0800, David Ford wrote:
>
> I do follow it and my statement is in regards to your implications of how
> broke the kernel is and etc.
>
> Often I tend to agree with Linus after a little thought has been put to the
This doesn't allow him to flame people(!) in public that told him again and
again that forgot to read the documentation (RTFM).
> Your previous emails gave the impression that you don't intend to listen to
> Linus wrt the kernel and compiler related bugs.
This impression is wrong. I don't listen to Linus wrt to linux specific
bugs that he claims to be in the compiler because he is too lazy to check
and read the documentation.
Compare this to how he reacted to the many asm bugs in the kernel (that have
been fixed by Bernd Schmidt in tedious work). He claimed until last that egcs
is broken (and gcc-2.8, too).
The course he takes is pretty clear (and he also stated this in public).
Linus will only support gcc-2.7.2.3, he does not support egcs, pgcc, or
gcc.
This is what he said, and I comply. I just don't want to receive bug
reports or claims that pgcc is broken in cases where the kernel is.
> On March 9th, you stated that the linux kernels, including 2.2.3 are broken.
>
> I am still asking you what specifically is broken because I have compiled
> and am using egcs/pgcc 2.2.3 kernels on several machines.
linux' use of extern inline is broken. asm usage is broken in some drivers
(this is being fixed, though).
The fact that gcc-2.7.2.3 and pgcc-1.1.1 and egcs and any snapshots
compile ok in most configurations does not prove anything. In fact, the
kernel became much safer with respect to gcc/egcs/pgcc (no, not pgcc got
safer), but this is not due to linus short sightedness but due to the many
people (david miller, bernd schmidt and many others) that were strong
enough to get though his insults.
> Your statement that Linus supports only 2.7.2 is slightly aside.
He claims this, and he did this repeatedly. I can only believe in what he
says, not what you think he might have meant, but not said.
> the egcs/pgcc family. 2.7.2.3 is the lowest cc capable of decently
> compiling the kernel. 2.7.2.3 is "supported" by Linus because it is the
> lowest workable compiler.
And Linus has repeatedly said the he does not support other versions. Thats a
fact, and I adhere.
The root of the problem is that, from time to time, bugs in egcs and
linux come up. If they are in linux, the first thing that linus does is
accuse the egcs developers of a broken compiler. This is easy for him, but
extremely unfriendly.
> And yes, using higher optimisations for the kernel does certainly improve
> it's performance.
Why?
> A short bit later in a discussion regarding c++ a gentleman said things are
> broken and you said "AFAIK there are problems with both kde and
> linux-kernel, NOT with egcs or pgcc" and I disagree. I picked up a binary
> tgz of pgcc a bit back and after a few weeks of frustration found that it
> [c++ includes] was compiled to look for everything under /usr/local instead
> of /usr. Once I symlinked everything all was well.
This might be a bug in the pgcc binary distribution indeed. This has
nothing to do with pgcc itself. Its also known that in most cases this
error is a local one.
When in doubt, take pgcc and compile it yourself.
> I don't claim egcs is full of bugs, I expect mistakes, mistakes happen.
> Saying "It's not my fault -- bug Linus!" doesn't get us very far.
Its better than what linus does. Its true in most cases, btw, and since
Linus publicly said he won't support it I do the same.
> that filename.c will cause a spilled register, try patching like so... will
> get us somewhere.
Linus knows about all problems, there is no need to reiterate that all the
time. Either he listens or he doesn't.
Please understand that it is NOT my responsibility to fix bugs in linux.
> "The consensus is that Linus tries to read the documentation before
> flaming and the egcs developers try to help the kernel by supporting more
> interfaces in the future.
>
> You make a very positive statement that many should follow. In general,
> egcs/pgcc bring light to grey areas in the kernel that need touching and the
> kernel excercises the bounds of the compiler. It is somwhat symbiotic in
> nature.
It certainly is. The key point, however, is to understand that linus
is only one user (an important one, but still only one) of gcc. There
are many others. In addition, most egcs developers have a far greater
experience with software development than linus has. Even worse, gcc is a
much more complicated product than linux.
> Linus certainly isn't perfect and neither are the rest of us. Spreading
> discordanant vibes in email just doesn't help really.
Who does that?
> As to Linus' flaming publicly/privately, you're doing just that. It's not
> getting us anywhere but riling our emotional side.
I haven't started this in the first place. All I originally did was to
explain that the linux kernel does not work reliably and I don't want to
have bug-reports or bad publicity because of unfounded claims.
> I griped for a good time about the messiness of the 2.1 kernel when I
> switched to pgcc. People can get pretty busy and stress goes up.
> Eventually many of the signed v.s. unsigned and etc warnings have been
> corrected.
>
> "the point is that I don't want to hear bug reports regarding the kernel,
> as my (official) position is that the kernel is buggy. I tried to be
> cooperative with Linus (and his problems), but thats over once and for
> all.
>
> Here's a bugger. Communications are stopping. This is a Bad Thing (tm).
> Perhaps my opinion wasn't quite so wrong :P
It is a bad thing. But this is not my problem at all. Linus stopped
communications by his claims.
He publicly and repeatedly said that he does not support non-gcc-2.7.2.3
compilers. Please go and take this issue up with him.
Yes, my statement was over-reacting and misleading. I actually will listen to
Linus and linux problems. But I'm also extremely annoyed at the bad publicity
linus makes about non-gcc compilers.
BTW, what I say here is in my position as pgcc maintainer, which allows me
more freedom than my position as egcs committee member. As egcs SC member
I will listen to the linux problems in the same way as i/we/they ever did.
> "Inofficially, I want the kernel to work with pgcc, of course, but until the
> current faq maintainer does hard work and sorts out all these issues in the
> faq I'll tell all people "don't do that!".
>
> hmm?
>
> I asked you specifically what is broken in the kernel. Everyone is
> unofficially the FAQ maintainer by providing correct input. Tell me what
> you think is so terribly broken and let's get things fixed.
Nothing is "terribly" broken. What is broken is the attitude when new bugs
are found. the current bugs are pretty minor (after all, it works in many
cases) and might only show up with specific snapshots. This does not
change the fact that these are bugs in the kernel.
I guess there are also many bugs in pgcc, but I do claim "its a bug in
linux" in cases where I am sure they are.
> Quite possible that I'm wrong on this but regardless, I really love
> these guys. They cost significantly less than a pentium and perform
this is very true. intel demands huge amounts of money for only a little
improvement. If you "only" want a fats machine then AMD, at least at the
moment (and probably in the future) is the way to go.
I'm very fond of the K7 architecture, and I think it has great potential
of overtaking intel.
> gain a different view on how broken the kernel is. I have use egcs/pgcc
> built kernels on industrial sized production machines for two years now.
this is impressive.
> I don't have problems. I have uptime and I have performance.
I do have problems, but each revision gets indeed better. Networking works
again in 2.2.3 (imagine! ;), even when using pgcc.
PS: this whole thread is getting too long, we should cut it down to
reasonable sizes (but I'm sure you think the same).
--
-----==- |
----==-- _ |
---==---(_)__ __ ____ __ Marc Lehmann +--
--==---/ / _ \/ // /\ \/ / pcg AT goof DOT com |e|
-=====/_/_//_/\_,_/ /_/\_\ XX11-RIPE --+
The choice of a GNU generation |
|
- Raw text -