[Openembedded-architecture] Recipe Specific Sysroots

[Richard Purdie]

On Mon, 2016-11-28 at 23:04 -0500, Trevor Woerner wrote:
> In theory, if more resources were directed towards fixing the issues
> found in
> Martin's "Failed dependencies..." logs (e.g.
> http://lists.openembedded.org/pipermail/openembedded-core/2016-Novemb
> er/128521.html)
> would your proposal be necessary?

This assumes Martin's approach catches all problems and that we'd
continue running such tests forever on all combinations. There is a
subset of our userbase who want/need truly reproducible builds so I
believe that some portion of our users will want this. OE prides itself
on being one of the most functionally complete systems out there and
this is a bit of a drawback.

So yes and no.

> On Mon 2016-11-28 @ 06:11:39 PM, Richard Purdie wrote:
> > 
> > Known issues/Todo:
> > * Once a dependency is installed into a recipe specific sysroot,
> > it 
> >   can't be removed. In particular, if you rebuild that dependency
> > (and 
> >   its sstate checksum changes), it isn't reinstalled.
> If I'm building and image that includes recipeA, and recipeA DEPENDS
> on recipeB, then in recipeA's sysroot I will find artifacts from both
> recipeA and recipeB?

You'd find just artefacts from recipeB since you're building recipeA.

> But if I then bump the version of, or tweak the recipe for, recipeB,
> will recipeA's sysroot be updated to include the newer version of
> recipeB before recipeA gets rebuilt?

With the branch I shared, no. I realise this is a problem and its one
of my next things to fix. It can't merge until this gets addressed.


> Will my final image include recipeA artifacts that have been rebuilt
> against the new recipeB? I assume my final image will include the
> updated recipeB's artifacts?

That is certainly the plan. The deploy artefacts (packages) are
unchanged so this piece works as before.

Cheers,

Richard