[OE-core] Image history (testlab)

[Koen Kooi]

Op 25 nov. 2011, om 14:52 heeft Richard Purdie het volgende geschreven:

> On Fri, 2011-11-25 at 14:17 +0100, Koen Kooi wrote:
>> Op 25 nov. 2011, om 12:44 heeft Paul Eggleton het volgende geschreven:
>>> Hi Koen / Beth / all,
>>> 
>>> Following on from my work extending packagehistory (which is still under 
>>> review), I'm looking at tracking history of the content of images so we can 
>>> get warnings about regressions. Thanks to Koen we've had this capability via 
>>> testlab.bbclass in OE for some time now, and from there into meta-oe, but what 
>>> I'd like to do now is bring this into OE-core and extend it so that it 
>>> supports rpm (or even better, all of the package backends we support.) This 
>>> could involve abstracting the package backend functionality out into 
>>> rootfs_*.bbclass.
>>> 
>>> It also looks like testlab.bbclass in meta-oe has some license dumping 
>>> functionality. Beth, does that mesh with what you've been working on / 
>>> preparing for Yocto 1.2?
>>> 
>>> Thoughts?
>> 
>> Don't forget its integration into remote git: http://git.angstrom-distribution.org/cgi-bin/cgit.cgi/testlab/?h=yocto 
>> I must admit that it completely sucks when you have >1 builder, but that is fixable. There are a number of other things we need to store somewhere (e.g. package history, PR config, etc), so coming up with a plan to store all that in a single repo would be good.
>> 
>> The are a few issues with its current output:
>> 
>> * License info uses the filename instead of the package name (f'oo_1.2.3_arch.ipk' vs 'foo') which leads to too much changes with small version bumps
>> * Dependencies are not sorted intelligently, so sometimes you get diffs which don't contain any actual change, just ordering junk,
>> * It only reports oe-core rev, not layer revisions, config and shortlog. Ideally it would output something like:
>> 
>> meta: angstrom-staging:b3ffd40a4a1eacab90dac26df8a4a047486f3b96 pulseaudio: update to 1.1, delete 0.9.x
>> meta-oe: master:7019fbb50796fedcedef6ff92981a7f26659a251 mplayer: fix build with -Wl,-as-needed 
>> meta-efl:
>> meta-gnome:
>> meta-xfce:
>> meta-angstrom master:bc72e71ccf7e79cdc8246e0ae0dbe62d11355263 angstrom-2010-preferred-versions.inc : Stop pinning python-cairo
>> 
>> Taking this a step further: it's output should be able to get passed into layer tooling (combo-layer to yocto people, setup-scripts for angstrom people) to reproduce builds.
> 
> So the real question is one of the approach. We have several options and
> its related to how compatible we want/need to be.
> 
> Do we import testlab from meta-oe as is, then add patches (multiple
> package backend support and so on)? How compatible do we need to be? I'm
> conscious its something being actively used.

In this specific case I don't care about backwards compatibility.

> Do we enhance packagehistory reusing code from testlab as appropriate?
> 
> Do we start something new as a hybrid of the two? (shouldn't be needed
> as I'm prepared to say packagehistory can radically change as needed, I
> don't care about compatibility there).
> 
> At the end of this I want to have one great "history" tracking mechanism
> for finding regressions covering testlab's functionality, package
> history and more.
> 
> Just to be clear I don't want to reinvent the wheel :).

For this history tracking both 'testlab' and 'packagehistory' are misnomers, so we can just start a new class that reuses ideas and code if needed.

Hopefully testlab can then fullfill its original goal: fully automated on-target testing. That should be doable with OE-core since that only has qemu machines :)

regards,

Koen
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 163 bytes
Desc: Message signed with OpenPGP using GPGMail
URL: <http://lists.openembedded.org/pipermail/openembedded-core/attachments/20111125/e19707f4/attachment-0002.sig>