Re: [DotGNU]Reminder - Weekly DotGNU Meeting

[James Michael DuPont]

--- Rhys Weatherley <address@hidden> wrote:
> James Michael DuPont wrote:
> 
> > Ahww Go on!
> > I aggree with you, we should use the tree in the gcc.
> > generate native treecc structures that are 100% the same as the gcc
> > trees, and then add in abilities into treecc to do tree
> > transformations.
> > So you go :
> > 
> > GCC ---> GCC Style TreeCC --> CSCC Style TreeCC --> IL generate.
> 
> The idea of using treecc with gcc came from Bradley Kuhn.  He was
> more thinking along the lines of:
> 
> New Language --> Treecc Definition --> GCC trees --> RTL --> native
> code

OK, that would also be fine. Just the First two steps.

> GCC ---> GCC Style TreeCC
I mean to produce a treecc input that is simlar to the gcc trees.

> New Language --> Treecc Definition --> GCC trees
Do you mean to generate an tree.def and the relavant tree information
out of the treecc? Or to you mean to use the treecc output instead of
the old style tree.

> i.e. no cscc code involved.  The treecc definition would contain a
> new list of nodes, designed around the requirements of gcc itself.

That of course would benefit the gcc directly, yes. 

> A "GenTree" operation on the nodes would produce a gcc tree, ready
> to be fed into the rest of the compiler.  
The GenTree on the GCC style treecc generated node would produce an old
school tree node? Ok.

>All of this would happen
> in memory (i.e. no external files or scripting engines that might
> compromise gcc).
Yes of course, all those things are just scaffolding, ie : the
ast-optimizers-branch has a graphviz dumper, that helps debug the asts,
but is not really needed in production.
> 
> One of the hardest problems facing the gcc crowd right now is that
> writing new front-end languages is hard.  

It is hard also because of the lack of explicit declarations of ht e

> Treecc may be the answer.
> The goal is to produce a clean and simple system to help people
> develop gcc.  

>Bolting cscc onto gcc isn't really the best approach
> to acheive this.
Yes, you are right there, when we where discussing this, I was thinking
on how to boost the cscc output or give pnet more opportunities. 
Using the treecc for gcc is something that we also dicussed, I think it
is a good idea. 
I have mapped out all the core tree nodes, both in a meta-description
and refactored that to be OO. that is used to create the SQL, perl,
java classes that accept and store the gcc tree nodes.

I would  next transform this into a treecc definition and generate via
tree the c, c++ and c# bindings.
This is something where I can directly help and it makes sense, what do
you think rhys?

mike


=====
James Michael DuPont
http://introspector.sourceforge.net/

__________________________________________________
Do you Yahoo!?
Y! Web Hosting - Let the expert host your web site
http://webhosting.yahoo.com/