Re: [Vala] Plugable Code Generators

[Andrea Del Signore <sejerpz gmail com>]

On Wed, Aug 8, 2012 at 1:58 PM, Luca Bruno <lethalman88 gmail com> wrote:
> On Wed, Aug 8, 2012 at 12:57 PM, Andrea Del Signore <sejerpz gmail com>
> wrote:
>>
>> can you elaborate a little why the transform branch should be merged
>> before?
>> What's its purpose?
>>
>> Anyway we are talking about experimental stuff and we shouldn't be
>> worried about compatibility and such, so why not include a simple
>> patch like mine and after go for the full solution?
>
>
> I don't think you're able to resurrect posix and dova with only that patch
> plus a plugin, am I wrong?

Tecnically yes, we can do just that, but this isn't the real point:
with that simple patch we can experiment we specialized backends of
which Dova and Posix are just 2 examples.

For "specialized backends" I mean a codegen suited for micros with
just a few Kb of memory, where memory and processing power is the
first constraint.

As an example we can have a type system that isn't thread safe just
because we will not ever have threads on the target cpu or because we
don't have atomics operations available, nor an OS available etc...
To be clear I'm not stating that vala should support this model, but
that it can be interesting to develop something highly specialized and
see how it come. I'm not even sure that vala is suited for this kind
of development :)

> The point behind the transform branch is to transform the code tree and not
> write codegen backends. Thus plugins will be visitors that transform the
> code tree, not that add new inconsistent codegen backends.

This will fix the "airy" code that I see in the codegen for sure...
meanwhile, since we are not committed to any api stability, we can
just experiment with yet another inconsistent codegen backend don't
you think?

Ciao,
Andrea