Re: [Qemu-devel] [PATCH] target-mips: fix call to memset in soft reset code

[Christian Borntraeger]

On 05/12/2016 12:41 AM, Peter Maydell wrote:
> On 11 May 2016 at 20:28, Christian Borntraeger <address@hidden> wrote:
>> On 05/09/2016 07:57 PM, Peter Maydell wrote:
>>> On 9 May 2016 at 18:55, Peter Maydell <address@hidden> wrote:
>>>> On 9 May 2016 at 18:53, Stefan Weil <address@hidden> wrote:
>>>>> I suggest to apply this patch to 2.6, if this is still possible
>>>>
>>>> It is not; sorry.
>>
>> I think we have delayed 2.6 already far too long (so please release)
>> but
> 
> Already done :-)
> 
>>> Note that it's only an error if you're building with -Werror, and
>>> releases don't default to -Werror, so users using released QEMU 2.6
>>> shouldn't hit this even with the newer gcc. Developers developing
>>> on trunk shouldn't be unduly inconvenienced if the commit fixing it
>>> is post-2.6-release rather than the actual release.
>>
>> to me it looks like that this is not a compile time error, instead we
>> really do not memset a variable that we are supposed to memset.
>> the only reason to not consider it for 2.6 is that its is really old
>> and it did not seem to cause harm.
> 
> Yes, it is both a code bug and a compile error, but the former
> has been present for many releases so is not a regression, and
> the latter is only an error if you're building with -Werror.
> So in my view it's the kind of bug we'd certainly fix at about
> rc3 or so, but not a bug which is "release critical showstopper".

Maybe a topic for this years QEMU summit could be to talk about
release process and release criterias.

We could 
a: allow more patches , e.g. I thing that this patch would be have 
been taken in the Linux kernel a day before the release, see for 
example what is applied 4 days before a release as network fixes:
https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=4d8bbbff1227bbb27fdb02b6db17f080c06eedad
22 files changed, 258 insertions, 86 deletions

b: as we are strict and only apply hand selected patches, regressions are
very unlikely, so we could release sooner. For example the CVE fixes could 
have just been taken and rc5 being released as final. (which we did anyway
but 3 days later)

c: we consider everything fine and keep the process

d: better ideas