[LLVMdev] ocaml+llvm

[Alain Frisch]

Chris Lattner wrote:
> On Aug 14, 2007, at 3:24 AM, Gordon Henriksen wrote:

>> I'm not sure you want to incorporate ocaml's runtime per se; its  
>> object model is very unique. :) How do you envision LLVM GC support  
>> going?
> 
> No idea - I know little about the ocaml GC.  I assumed it was generic  
> enough to be reused in other contexts.  If not, then disregard my  
> comments.

The OCaml GC is quite generic. Basically it exposes a model where values 
  all uniformly represented by a N-bit machine word (N=32 or 64) which 
can denote either a (N-1)-bit integer (with the lsb set to 1), a pointer 
to a block controlled by the GC, or a pointer outside the GC area 
(though the last possibility is considered bad style).

A GC block comes with a 1-word header that describes its length and also 
contain a small 8-bit tag (plus a few bits for the GC). In OCaml, the 
tag is used in particular to store the constructor for union types 
(discriminated sums), but the runtime system is really agnostic about 
the meaning of the tag except for a few special values which denote 
special kinds of blocks such as:
- byte buffers (not traced by the GC, used to represent OCaml strings), 
- custom blocks (native data not controlled by the GC, plus a pointer to 
a table of custom functions to implement finalization and generic 
operations like equality/hashing/serialization),
- lazy pointers (the GC is able to shortcut chains of forwarding 
pointers to forced values),
- arrays of weak pointers,...

Of course, this model makes some assumptions, but it could certainly be 
useful to support languages other than OCaml!


-- Alain