ARM rebuild

[Matthias Klose]

On 18.11.2009 08:58, Matt Zimmerman wrote:
> On Mon, Nov 16, 2009 at 10:42:34AM -0600, Steve Langasek wrote:
>> On Sat, Nov 14, 2009 at 12:52:58PM -0600, Matt Zimmerman wrote:
>>>> We should be a little careful about how we phrase our commitment here.
>>>> As yet, we don't have the capability to do binary-only rebuilds on a
>>>> single architecture, so the only way to rebuild all armel binaries would
>>>> be to reupload every source package in the archive. This is a pretty
>>>> good way to lose Ubuntu mirrors, and in the past we've decided that we
>>>> didn't want to do that after all.
>>
>>>> I think it should be adequate to identify a core set of packages and
>>>> ensure that all of those get rebuilt, either during the initial merge
>>>> from Debian or separately.
>>
>>> I realize we've been conservative about this in the past, and for good
>>> reason, but I think there are risks to this approach as well.  Unless we
>>> rebuild everything, we don't know if it builds and works with the new
>>> compilation defaults.
>>
>> If this is the main concern, then I think we're much better served by an
>> archive test rebuild that uses its own output, since this lets us rebuild
>> everything without a need for per-package human intervention.  To do a full
>> archive rebuild, someone will have to do a sourceful upload of each package,
>> which I don't think makes sense if the goal is only to test the toolchain
>> changes.
>
> A test rebuild, as I understand it has been done in the past, will not tell
> us whether the software works, only whether it builds.  Testing the
> toolchain requires functionally testing its output as well, not just the
> toolchain itself, no?

except for packages where a testsuite is run during the package build. but yes, 
it would be good to keep the packages built and  build a CD from these packages 
and test this one as well.

   Matthias