Re: [Caml-list] whither portability?

[whitequark <whitequark@whitequark.org>]

On 2015-09-27 20:19, Raoul Duke wrote:
>> Right now, LLVM IR (of which the bitcode is a serialization) is not
>> flexible enough to express the invariants and metadata required
>> for the OCaml GC to function[1]. Even if someone released the 
>> necessary,
>> highly nontrivial changes to both LLVM and OCaml today, it would take
>> many months for them to be reviewed, merged & propagated into Apple's
>> LLVM fork--all for a small improvement in a select few numeric 
>> workloads
>> that can benefit from using the instruction selector tailored for
>> the particular device, and a substantial reduction in being able to
>> debug your code[2].
> 
> Thanks for the details. I am (only) a little bit aware of the history
> of LLVM vs. GCs, so I can believe what you report. I would much prefer
> to be able to debug, as well.
> 
> (Given Apple, I wouldn't be surprised if an a year they required
> bitcode, however.)

It is already required for Apple Watch. Given that bitcode allows to
include inline assembly, I wonder if one could just take OCaml's output,
put it in one giant module-level inline assembly block, and ship it like
that.

-- 
whitequark