Re: Does anybody try to compile the 2.3 kernels at all ? :((

[Khimenko Victor]

In <[e-mail]> Arjan van de Ven ([e-mail]) wrote:
AV> Hi,

AV> On Sat, 12 Feb 2000, Khimenko Victor wrote:
AV> [snip]
>> > Can you please send me your .config then (the one that doesn't work)?
>> > 2.2.14 should compile for every[1] .config.

>> Not for every but for most natural .config's :-)

AV> The footnote named 3 known problems. If you have any other problem, please
AV> let me know! (Preferably by sending me the .config)

I've not seen any other problem yet, thnx.

>> Yes, 2.2.14 is MUCH better here. Since it's basically Cox's kernel :-)

AV> No it is not that way.

Ok. I know for sure that 2.2.5ac5 was MUCH MORE compileable then vanilla 2.2.5
(I was unable to compile vanilla 2.2.5 for almost every  configuration I tried;
2.2.5ac5 needed only few small fixes). The same is true for 2.2.10ac10.
And so on.

>> Usually situation is the following:
>> Linus's kernels can not be compiled in WAST range of configurations (like
>> "everything in modules, everything enabled") while Cox's kernels can do this
>> (ok, sometimes there are glitches to but usually compileability of Cox's kernels
>> are order of magnitude better then compileability of Linus's kernels).

AV> It's not a Linus versus Alan thing.

Perhaps.

AV> During the late 2.1.x series (x>125), I and other
AV> people fixed a lot of those bugs. Not all fixes were incorporated then,
AV> simply because they were not of a good enough quality for the kernel.

AV> Linus managed the first 2.2.x releases, but when 2.3.x activity became
AV> more hectic, Alan took over (as he did for the 2.0 series). The stable
AV> kernels usually only change because of bugfixes or new drivers, and
AV> several of the known "2.1 leftovers" were fixed. But new drivers contain
AV> also new bugs. For 2.2.14, a serious effort was made to make sure that it
AV> compiles (or gives a descriptive #error message) for every .config.

AFAIK Cox always tries to fix such "minor" issues if he can. Linus did not.
It's he's right: he has more serious thing to do. But when supposedly stable
kernel can not be even compiled it looks... hmm... strange at least.

AV> At the time of the 2.2.14 release, only three .config combinations were
AV> known to be problematic. The hfmodem-case is obvious and will not be fixed
AV> due to lack of a maintainer, the IrLPT one was not fixed because the
AV> correct fix came to late (and would not be useful for a long time as IrLPT
AV> was about to be changed in 2.2.15pre1) and finally the IPv6/network bug is
AV> a VERY rare (but still serious) bug in the network-Config.in scripts.
AV> I monitor the 2.2.15pre releases for new bugs on a regular basis, and I
AV> want to try (in cooperation with Alan, Linus and others) to prevent new
AV> problems in the official 2.2.x releases, and in the 2.4.x releases.

>> It's more or less normal for development kernels but it's NOT Ok for stable
>> ones.

AV> Agreed. And sometimes, compilation is broken on purpose to find "bad"
AV> situations in the development series. Once the major changes settle down,
AV> I (and others of course) will start fixing the "leftover cases".

I'm not talking about development kernel. There such thing are unavoidable :-/
Sad but true.

>> I understood that Linus has LOTS OF things to worry about apart of
>> compileability of kernels with weird .config but then perhaps Cox should
>> release stable kernels from the very beginning ?

AV> Again, it is not Linus versus Alan. It's a maturity thing. Fixing takes
AV> time, and starting too soon would hinder progress in the development
AV> series.

I'm NOT talking about development series. Even for development series it looks
strange and stupid when something can not be compiled just since there are
extra "}" somewhere. But it's development after all. I'm talking about stable
and -pre series.