Re: [Vala] GSOC Idea: LLVM Backend?

[Luca Bruno <lethalman88 gmail com>]

Il 29/04/2013 06:13, Kenny Micklas ha scritto:
> Hi all,
>
> I am a student at Brown University interested in participating in Google
> Summer of Code. I am a Vala enthusiast and I have always been interested in
> the idea of writing an LLVM code generator for valac. The principle
> benefits of this would be decoupling Vala from C, increased flexibility in
> the generated code, vastly better compile times, and potentially better
> runtime performance too.
>
> I know this is not one of the suggested GSOC ideas for GNOME, but I was
> wondering if the Vala team would think this is a good project, and whether
> or not there would be an appropriate mentor. If so, I will write up a full
> proposal for GSOC.
Hi,
thanks for your interest in vala.
I'm not a possible mentor, but I have to say that an llvm code generator 
might not be of high value.
The points you rised are rather marketing ideas rather than real useful 
gains.
First of all, why flexibility in generated code?
Then, about compile times, you can just compile the generated C code 
with clang... same goes for runtime performance.

On the other hand, there's a wip/transform branch in vala. The purpose 
of the branch is to introduce an intermediate step for transforming the 
ast. For example, the current gdbus module works more on the C level, 
while with this branch you rewrite the vala ast into another vala ast 
calling dbus methods.
That is you write vala inside the vala compiler itself.
Potential: allow other transformers similar to [DBus], for example 
[RPC], or [Protobuf], or [DBTable], or whatever. That opens to a 
possible plugin system where each plugin can change the behavior of the 
vala code.
The branch itself works quite well, it needs to be tested on existing 
vala projects to ensure there are no regressions, as well as adding new 
features and a plugin system.
I don't know if this could be a gsoc project or not.

Best regards,