[oe] strange error when coding a bbclass

[Richard Purdie]

On Tue, 2007-06-12 at 16:50 +0200, Koen Kooi wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Richard Purdie schreef:
> > On Tue, 2007-06-12 at 11:18 -0300, Rodrigo Vivi wrote:
> >> On 6/12/07, Richard Purdie <rpurdie at rpsys.net> wrote:
> >>> On Mon, 2007-06-11 at 16:49 -0300, Rodrigo Vivi wrote:
> >>>
> >>> addtask inserts a dependency so if you add
> >>>
> >>> addtask blah before do_configure
> >>>
> >>> You are saying that say "patch" must run, followed by "blah" followed by
> >>> "configure". Bitbake will check the task dependencies and build them if
> >>> they're missing. If a subsequent task has a timestamp older than its
> >>> dependencies it will rerun them too (just on a per recipe basis at
> >>> present). There have been requests to extend that so if you rebuild gtk
> >>> +, all the gtk apps will rebuild and we will probably optionally allow
> >>> that in due course (it should be easy with the new bitbake core).
> >> I've noticed that and it makes sense. What I really cannot understand
> >> is how a very simple function that does anything but a print can break
> >> that packages?
> > 
> > Try "bitbake busybox", then "bitbake busybox -c package -f". You will
> > find that either it will fail outright or the busybox packages will
> > become corrupt (empty). This is because install must always be (re)run
> > before package. I suspect you're seeing side-affects from that area of
> > breakage (and you're not the only one)...
> 
> Didn't we have a proposed fix for that?

My original suggestion broke do_deploy tasks if I recall correctly. I
therefore proposed copying files rather than moving them but I've not
written that patch yet :/.

Cheers,

Richard