Re: [Caml-list] Why List.map does not be implemented tail-recursively?

[Goswin von Brederlow <goswin-v-b@web.de>]

On Tue, Jun 02, 2015 at 01:04:46PM +0100, Stephen Dolan wrote:
> On Mon, Jun 1, 2015 at 1:02 PM, Jon Harrop <jon@ffconsultancy.com> wrote:
> > What happens when the first allocation in the creation of a cyclic immutable structure incurs a minor collection?
> >
> > let rec xs = 0::ys and ys = 1::xs
> 
> Such allocations are handled specially. Essentially, it boils down to
> having a write barrier on the final mutation that ties the knot on the
> cyclic data structure. This isn't multicore-specific: see
> caml_alloc_dummy and caml_update_dummy in byterun/alloc.c for the gory
> details.
> 
> In multicore, this mutation needs to promote (copy to the shared heap)
> the cyclic data structure if the head node has already been collected.
> 
> Stephen

I believe ocaml already optimizes this into a single allocation for
both objects. So there can't be a GC run between them. And even if
there is, both xs and ys are alive in the local stackframe and can not
be collected just yet.

MfG
	Goswin