Some questions on latest work

Adam D. Ruppe via Digitalmars-d digitalmars-d at puremagic.com
Tue Apr 26 06:45:04 PDT 2016


On Tuesday, 26 April 2016 at 04:37:03 UTC, deadalnix wrote:
> On 32 bits, nothing fundamental changed, but the compiler
> got a bit smarter about alocating with the no scan flag,
> so that's better.

Indeed, though any large array is liable to be pinned by a false 
pointer on 32 bit, even with no_scan.

I actually recommend malloc/free large arrays with easy lifetimes 
on 32 bit just to avoid those accidental leaks.

64 bit is fine though.

> I don't know for that one, but if that doesn't work with 
> classes, this is indeed a problem worth solving.

I think it kinda works now but not very well because the ref 
countedness gets lost in inheritance. If it implements two 
interfaces for example, it cannot properly express that implicit 
conversion.

Usable though AFAIK, just clunky.


More information about the Digitalmars-d mailing list