Re: librump hits again...

[Michal Suchanek <hramrach%centrum.cz@localhost>]

On 14 May 2010 15:25, Antti Kantee <pooka%netbsd.org@localhost> wrote:
> On Thu, May 13, 2010 at 10:06:30AM +0300, Andreas Gustafsson wrote:
>> Kurt Schreiner wrote:
>> > with -current source updated some minutes ago build.sh. ... distribution
>> > stops with:
>> >
>> > dependall ===> usr.sbin/puffs/rump_cd9660
>> > Â Â Â Âlink Ârump_cd9660/rump_cd9660
>> > /u/NetBSD/arch/alpha/dest/usr/lib/librump.so: undefined reference to 
>> > `rumpns_uvm_km_va_drain'
>> > collect2: ld returned 1 exit status
>>
>> FWIW, the failure appeared with the following commit:
>>
>> Â Â2010.05.12.03.43.46 rmind src/sys/kern/subr_pool.c,v 1.184
>>
>> I mailed rmind about it yesterday but haven't seen a response.
>
> Fixed. Â*sigh*, I find it a little disturbing that ./build.sh rumptest
> is too difficult to type.
>
> Well, actually, I find it more disturbing that people still commit
> changes without running "anita test", but considering it takes 3h
> on my machine as opposed to the 2min buildtest, I can see the merit
> if you want to rush in untested changes.
>
> For the n'th time, my usual M.O. per day / set of related changes is:
> 1) make changes in working tree
> 2) test
> 3) commit
> 4) checkout to clean tree
> 5) build.sh release
> 6) anita test
> 7) have beer or, if necessary, fix any unforeseen regressions
>
> I really wish that would be made required policy for all MI /
> non-hardware changes. ÂThe only thing it really requires is that
> a developer finishes committing changes a few hours before going
> to bed or afk or whatever.
>
> (yes, anita test and/or atf could use some further improvements
> for even better reporting of regressions, but it's god damn impressive
> already as-is. Âkudos to gson and jmmv for making it happen).
>

Or better yet, have the commits go to a queue and only accept ones
that don't break builds ;-)

Thanks

Michal