[M3commit] CVS Update: cm3

[Jay K]

 > LLVM is well specified IR
Someone still has to read and understand the specification. :)


 > GCC IR being unspecified and prone to major changes at every minor release


I'm not sure of this.
Tony will know better, having done more upgrades of the backend.
I don't think our gcc backend has been much affected by it.
What we are affected by:
   - we add our own IR node for nested functions, and that was affected 
    - our IR has never been well formed (even to this day!) 
      The main problem is our use of overly low level and lacking types and typeful opertions (member references).
    - seemingly nobody has tried configure -enable-checking, made it work, kept it working
       This is still needed.


Granted, tree.h and tree.def might not be as good a specification as LLVM has.


> licensing

Understood.
Yeah, it bugs me we can't statically link.
Though the separate executable provides nice testability/debugability -- a separate entry point. :)


 > intent is to control through GPL


Agreed. :( 


 - Jay

----------------------------------------
> From: dragisha at m3w.org
> Date: Fri, 5 Nov 2010 16:40:56 +0100
> To: jay.krell at cornell.edu
> CC: m3commit at elegosoft.com
> Subject: Re: [M3commit] CVS Update: cm3
>
> FreePascal people are doing their own backend... I don't think it's big luck, but it is free and they are Wirthian (meaning at least nested scopes and strong typing, and IIRC they are some kind of strange body in GCC).
>
> LLVM is well specified IR. Their intent is to enable people to use it, and GCC's intent is to control through GPL. I am big fan of whole FSF idea, projects, everything... But sometimes they are just over the edge. They killed GNU Pascal with GCC IR being unspecified and prone to major changes at every minor release.
>
> I am not telling you to stop your C effort. All your points are valid. Except some assumptions about LLVM. LLVM is not GCC, and not GPL, and so on...
>
>
> On Nov 5, 2010, at 4:32 PM, Jay K wrote:
>
> > I shouldn't bite the hand that feeds, but I presume LLVM is a moving target roughly as much as gcc is.
> > gcc's moving target has indeed been slightly painful. But it is not the only complaint I have.
> > Plus, you don't see Tony complaining about whatever he had to do to move from 3.x to 4.x.
> > I'm a whiner.
> > LLVM is probably large and slow to build, similar to gcc.
> > But we don't get too many complaints about that.
> >
> >
> > gcc's reach is impressive.
> > C/C++'s reach is even better.
> > I remain keen on VMS/Alpha, VMS/IA64, Linux/ia64.
> > I know they are viable with gcc/c/C++. Unknown about LLVM.
> >
> >
> > I'm also still interested in our own backend maybe.
>