[Mono-list] GCC .NET

Zaphod j0k3rin@yahoo.co.in
Tue, 3 Sep 2002 00:16:54 +0530


On Fri, Aug 30, 2002 at 05:14:08PM +0000, Cyber Wiz wrote:

> that we should modify the front ends c++ etc. to create gcj compatible tree 
> output etc.

That would be ripping open the GCC's standard bridge of RTL between the
frontend and backend ... and converting that to an AST bridge ...
(I'm not a pessimist, but IMHO it needs an intimate knowledge of the
totally undocumented gcc internals...)

> won't be unstandard in il bytecode. (Gaurav could i get your opinion 
> on this).

Afterall trying to emulate a register machine on a stack machine is not
the 'standard' code .... Well we could do it, the triples of each operation
could be operated on the stack , but the operations would increase 4 fold.
This is the so called RTL mode of codegen...

But grabbing the AST and processing from it is possible and might be
the only real way to do .. 

> And mono could get jilc fully operational, after all mono has a highly 
> skilled team of hackers.

Hmm.... that might be possible ... But the JILC model has nothing to
do with gcj model ... I know one of the guys who started it, I asked
him and he told me it depended on a very simple tree data model which 
was not much good for anything else than linear loops through it. 
(In his opinion this why he chose it for disassembling).

The real idea should be to rip out RTL and design a new transfer bridge
which can convey stack based semantics as well ... the rest of the ideas
are just hacks ..