[oe] Package Maintenance
Mike (mwester)
mwester at dls.net
Wed Mar 25 18:13:30 UTC 2009
Koen Kooi wrote:
> On 25-03-09 17:05, Mike (mwester) wrote:
>> Koen Kooi wrote:
>>> On 25-03-09 09:51, Frans Meulenbroeks wrote:
>>>
>>>> and a last last thing: about the vetting: not sure whether I fully
>>>> understand the idea, but is a lot of this info on what builds on what
>>>> not already present in tinderbox?
>>>
>>> The tinderbox data is purged periodically, so don't depend on it.
>>
>> In addition, you cannot determine which builds in tinderbox were done
>> with an empty TMPDIR -- and unless you have an empty TMPDIR, there's
>> really no complete test.
>
> Empty TMPDIR is only once case we should test for, I've seen many times
> that a 'populated' TMPDIR creates bugs, e.g. midori picking up libhildon
> to mention a recent one. So 'always build from empty TMPDIR' will leave
> serious and harder to trackdown bugs unnoticed.
> Another example: building qt/e for ppc405 will OOM[1] if you have a
> populated TMPDIR, it will work fine from scratch.
> We should certainly do builds from scratch to track down dependency and
> process bugs, but let's not forget about non-empty TMPDIR bugs.
>
> regards,
>
> Koen
>
> [1] It took me a while to track that one down, since the OOM would
> either take sshd down as well, or just lock up the box.
Fair point -- which brings up a related area, which is parallel-enabled
builds. Recipes and dependencies may behave differently due to Makefile
bugs, or inadequately-stated dependencies.
I guess we can't test everything, but we should agree about some minimum
standards to determine the "goodness" of a recipe. I just don't know if
we can all agree on that minimum! :-)
Mike (mwester)
More information about the Openembedded-devel
mailing list