On 17/08/2017 18:37, Peter Maydell wrote: > On 17 August 2017 at 17:31, Paolo Bonzini <[email protected]> wrote: >> On 17/08/2017 18:17, Peter Maydell wrote: >>> On 17 August 2017 at 17:16, Paolo Bonzini <[email protected]> wrote: >>>> On 17/08/2017 18:14, Igor Mammedov wrote: >>>>> travis builds fail at HEAD at rc3 master with >>>>> >>>>> block/nbd-client.c: In function ‘nbd_read_reply_entry’: >>>>> block/nbd-client.c:110:8: error: ‘ret’ may be used uninitialized in >>>>> this function [-Werror=uninitialized] >>>>> >>>>> fix it by initializing 'ret' to 0 >>>> >>>> This is a false positive, but it's understandably impossible for the >>>> compiler to figure it out. >>>> >>>> Even though we disable -Werror on release builds, it may be worth fixing >>>> this in 2.10 if it doesn't delay the release. Peter, what do you think >>>> about applying this on top of -rc3 without doing a fourth candidate? >>> >>> I don't like doing releases which haven't had an rc, >>> but we've had abbreviated "just a couple of days" rc-to-final >>> cycles before. >> >> It's just a matter of "looking unpolished". It doesn't deserve -rc4, >> not even for just a couple of days. > > The purpose of having an rc4 is to avoid the chance of > messing up the change (which is possible, even if it's a pretty > remote chance). Having an rc4 gives us a window to catch and > fix that kind of error -- once we've tagged something as the > final release we don't get to do it over.
Understood, I meant we can release with the issue unfixed. Paolo
