[oe] [RFC] a small bbclass for task recipes

[Richard Purdie]

Hi,

On Sun, 2007-08-12 at 02:48 +0200, pHilipp Zabel wrote:
> Then PACKAGES += "task-b task-c task-d"?
> That will help because it keeps ${PN}-dev and -dbg packages, but still
> doesn't create -dev and -dbg for task-a and task-b ...

Looking at Poky, you do have to add the -dev and -dbg package names to
PACKAGES manually:

http://svn.o-hand.com/view/poky/trunk/meta/packages/tasks/task-poky.bb?rev=2466&view=markup

which is perhaps more ammunition for a task.bbclass. 

> Ah, thanks.
> The reason it didn't work for some packages was that they set RDEPENDS
> instead of RDEPENDS_${PN}.

In the fragment you quoted it does do:

rdepends = explode_deps(bb.data.getVar('RDEPENDS_' + base, d, 1) or
bb.data.getVar('RDEPENDS', d, 1) or "")

which should cover plain RDEPENDS (if RDEPENDS_${PN} isn't set)?

> The reason it didn't work for some other packages is that they use RRECOMMENDS,
> not (only) RDEPENDS. Currently package_depchains()
> only shovels the base packages' RDEPENDS into the flavoured packages'
> RRECOMMENDS, adding the corresponding pre/postfix.
> To make this mechanism useful for task packages, we'd have to scan
> the base packages' RRECOMMENDS, too. Can we do this without negatively
> affecting some existing -dbg or -dev packages?

There were differences in the depchains code in poky which I merged last
night. I think this issue might have been addressed but it needs to be
checked.

> I further noticed that setting ALLOW_EMPTY_${PN}-dbg = "0" like in the
> attached patch has no effect because (not "0") is False in python
> (getVar returns a string).
> Should not bb.data.getVar('ALLOW_EMPTY', localdata) be changed to
> bb.data.getVar('ALLOW_EMPTY', localdata) != "1" in
> package_ipk.bbclass?

Yes, that looks like something it would be good to fix.

Cheers,

Richard