sound distribution practices

[Allison Randal]

On 04/08/2012 11:08 AM, Alessandro Ghedini wrote:
> I think I don't understand... NQP and Rakudo *do* ship compiled bytecode, and
> would of course need to be rebuilt when a new Parrot release is uploaded. The
> whole point of parrotapi-* is to make sure that Parrot and its libraries are
> not updated on the users systems (or migrate to testing) until the rebuild of
> the compilers is done. This way we avoid that Parrot-based compilers don't break
> in the period of time between the Parrot upload and the compilers rebuild (which
> may be more or less long).
> 
> But the PBC libraries built from Parrot *do* break the compilers if they are
> updated before the compilers themselves and the other Parrot packages, and that
> is why they need to be "constrained" by the parrotapi-* mechanism too. Making
> them depend on parrotapi-* is just one way, the other would be to make parrot
> striclty depend (as in (= X.Y.Z)) on them.

I'm agreeing with you, +1, we should go ahead. Use parrotapi-* both for
external packages like NQP and Rakudo, and for binary packages built
from the Parrot source package (if they contain compiled bytecode, or
other version-dependent compiled output like dynpmcs, PMC .dump files,
dynops, output of pmc2exe, etc...).

Allison