Re: Linux 4.0-rc1 out..

[Christian Borntraeger <borntraeger@de.ibm.com>]

Am 24.02.2015 um 03:34 schrieb Mike Galbraith:
> On Mon, 2015-02-23 at 16:43 +0100, Christian Borntraeger wrote:
>> Am 23.02.2015 um 04:06 schrieb Linus Torvalds:
>>> .. let's see how much, if anything, breaks due to the version number.
>>> Probably less than during the 3.0 timeframe, but I can just imagine
>>> somebody checking for meaningful versions.
>>>
>>> Because the people have spoken, and while most of it was complete
>>> gibberish, numbers don't lie. People preferred 4.0, and 4.0 it shall
>>> be. Unless somebody can come up with a good argument against it.
>>
>> The only argument that I can come up with is "we do not break userspace".
>> For example there is this "gem" in configure.ac of valgrind:
>>
>>
>>         case "${kernel}" in
>>              2.6.*|3.*)
>>                     AC_MSG_RESULT([2.6.x/3.x family (${kernel})])
>>                     AC_DEFINE([KERNEL_2_6], 1, [Define to 1 if you're using Linux 2.6.x or Linux 3.x])
>>                     ;;
>>
>>              2.4.*)
>>                     AC_MSG_RESULT([2.4 family (${kernel})])
>>                     AC_DEFINE([KERNEL_2_4], 1, [Define to 1 if you're using Linux 2.4.x])
>>                     ;;
>>
>>              *)
>>                     AC_MSG_RESULT([unsupported (${kernel})])
>>                     AC_MSG_ERROR([Valgrind works on kernels 2.4, 2.6])
>>                     ;;
> 
> 
> Heh, if this is an argument, we have one hell of a lot of reverting to
> do :)   Crash for example breaks at much higher resolution, and indeed
> just broke yet again.  Tough titty for userspace methinks.

Well crash is not a good example as it by design goes beyond the user ABI
and directly touches the kernel data structures ;-)

I am not requesting to go back to 3.*, I was just pointing out that if we apply
strict rules on "we dont break userspace", the move to 3.* and 4.* was a mistake.
We do provide uname26 as a workaround, so this is ok and the switch to 4 should
be a lot smoother. 

But better end the discussion here :-)

Christian

FWIW, valgrind svn is fixed as of yesterday (for good, so Linux 5.* 6.*.. should
also work)