Re: LLVM

[David Chisnall]

On 1 Mar 2008, at 18:07, Nicola Pero wrote:

> They could also fork it, and keep their fork private.  Every 6 months,
> they merge the mainstream changes into their fork, but never open up
> their fork.  So they get your improvements, but you don't get theirs.
> They have been doing exactly this with GCC for years! :-)
>
> The difference is that with GCC they have to release their source code
> when they ship the binaries, so you get a change to see their changes
> and if you want you can merge them back into mainstream yourself.   
> With
> a BSD-like license, they could keep them (or some of them) in their  
> own
> tree forever. :-(

And yet, they are developing LLVM in the main branch of the  
repository, while they were always consigned to a branch (I don't  
intend to go into the politics behind this decision, since it is very  
boring).  Yes, a BSD license does let them be evil.  At some point in  
the future, it might be worth their while to do so.  Considering how  
many other companies are starting to get behind LLVM, however, I  
suspect that it will always be better for them to keep it out in the  
open.

Even without Apple, we still have a lot of people in the commercial  
and academic worlds working on LLVM, and it's very easy for new people  
to get involved.  I had been following the project for a year or so,  
but didn't look at the code until Sunday.  This week, I have had a  
nontrivial patch accepted.  This kind of learning curve simply does  
not exist with GCC.

We could spend all day debating the possible things that could go  
wrong, but it's not productive.  All of the GCC developers could die  
or be hired by Microsoft tomorrow, leaving no one who understands the  
code.  It isn't very likely though.

> Anyway, I understand you see it in a different way.  Good luck with  
> your
> new Objective-C runtime/compiler, and keep us up to date with your  
> progress :-)

The runtime was finished (in as much as any project can ever really be  
declared 'finished') a few months ago, and is being incrementally  
improved.  Quentin has done some work on it, and a couple of other  
people sent me small patches.  I have written a paper about it, which  
has not yet been published (although I can send copies to individuals  
who are interested).

My code for generating Objective-C message sends was merged into clang  
on Friday night, so you can now compile tiny, trivial (not actually  
useful) Objective-C programs with it, as long as your class  
definitions are all from libraries (which are compiled with GCC).  My  
next step will probably be to start working on class definitions.  All  
of this has, so far, been done with the GNU runtime.  I will work on  
code generation for my runtime later, when the GNU one is working.

David