[OE-core] PV from filename, not reflected in siginfo?

[Richard Purdie]

On Mon, 2016-01-18 at 13:44 +0100, Olof Johansson wrote:
> Hi,
> 
> I'm currently investigating reports of bitbake not correctly
> handling a type of change to a recipe, where the only change is a
> filename rename to update the PV.
> 
> With bitbake-dumpsigs:
> 
>  ...
>  Variable PV value is ${@bb.parse.BBHandler.vars_from_file(d.getVar('
> FILE', False),d)[1] or '1.0'}
>  Variable PN value is ${@bb.parse.BBHandler.vars_from_file(d.getVar('
> FILE', False),d)[0] or 'defaultpkgname'}
>  ...
> 
> ... with nothing containing the resolved value of PV. I would
> expect the computed value of PV to be a part of the siginfo. The
> consquence of this seems to be that bitbake doesn't schedule
> dependents to be rebuilt, i.e. if I rename the recipe
> foo_1.2.3.bb to foo_1.2.4.bb, the recipe foo is rebuilt, but the
> image isn't.
> 
> I tried some changes:
> 
>  - Changing the PV assignment in bitbake.conf to :=. Didn't work,
>    got errors that I suspect are related to not being able to
>    include foo-${PV}.inc files. The value of PV was clobbered.
>    Also played around with using BB_FILENAME instead, but didn't
>    help.
> 
>  - Changes within Bitbake that assigned the version part of the
>    filename to PV when loading new files ending with .bb. This
>    worked but is really ugly.
> 
>  - Moving the PV assignment into the recipe (as an explicit
>    assignment). This works, makes dependents build as expected
>    and bitbake-dumpsigs lists the value, but is also ugly.
> 
> I suspect this isn't usually a problem in OE-Core, since most
> recipe updates requires updates within the recipe as well
> (tarball checksums, for instance), but that's not as true for our
> internal recipes.
> 
> Is this a known issue? Any ideas on how to solve this? Nicer
> workarounds?

It isn't known but I can imagine how this could cause a problem.

I suspect (but am guessing) that:

PV[vardepvalue] = "${PV}"

might happen to fix this...

Cheers,

Richard