[oe] bitbake vs incremental builds

[Richard Purdie]

On Wed, 2008-06-18 at 15:40 -0700, Rich Pixley wrote:
> If C depends on B depends on A...
> 
> A first build will build A, B, C.
> A second build will build nothing since nothing has changed.  (this is 
> correct behavior).
> However, if I change A and rebuild, only A will be rebuilt, not B or C.
> 
> This is a nuisance for developers who are using bitbake as a build 
> system.  It's a problem for continuous build systems since they must 
> necessarily build everything over from source in order to test whether 
> builds work.
> 
> I'm thinking that it should be fairly easy to adjust which recipe+task 
> pairs are scheduled for execution by looking not only at stamp file 
> existence, but also at the comparative difference in stamp file times.  
> If the stamp file for A is more recent than the stamp file for B, then B 
> needs to be scheduled for execution as does C.
> 
> Does anyone know of any reason why this cannot or should not be done?

It wasn't done for a long time simply because nobody had taken the time
to implement it. It has however recently been implemented, simply set:

BB_STAMP_POLICY = "full" 
or
BB_STAMP_POLICY = "whitelist" 

Whitelist allows you to exempt certain packages from stamp checking, see
packaged-staging.bbclass for an example.

Cheers,

Richard