On Friday, September 09, 2011 23:58:56 Timon Gehr wrote:
> On 09/09/2011 11:44 PM, Dmitry Olshansky wrote:
> > On 10.09.2011 0:55, Timon Gehr wrote:
> >> On 09/09/2011 10:30 PM, Timon Gehr wrote:
> >>> void main(){
> >>> mixin({
> >>> string r;
> >>> foreach(i;0..12000) r~=q{mixin(q{{enum x="";}});};
> >>> return r;
> >>> }());
> >>> }
> >>>
> >>> When I compile this program, DMD sometimes uses about 1GB of memory,
> >>> sometimes about 2GB of memory and sometimes it makes my OS crash.
> >>>
> >>> Can anyone reproduce this?
> >>
> >> Similarly this:
> >>
> >> template TT(string s){enum TT=s;}
> >>
> >> void main(){
> >> mixin({
> >> string r;
> >> foreach(i;0..12000) r~=q{{enum x=TT!"";};};
> >> return r;
> >> }());
> >> }
> >>
> >> That does use a somewhat smaller amount of memory, but it is still
> >> huge
> >> and non-deterministic in size.
> >
> > I guess issue 6498, i.e. it's CTFE doing memory allocation on each
> > append in foreach loop.
> > http://d.puremagic.com/issues/show_bug.cgi?id=6498
>
> Well, I don't get the issue if I don't mixin mixins or template
> instantiations.
>
> void main(){
> mixin({
> string r;
> foreach(i;0..12000) r~="{}{}{}{}{}{}{}{}";
>
> return r;
> }());
> }
>
> > As for as non-determinism, maybe heap fragmentation?
>
> That might well be it. Thanks!
Excessive memory consumption during compilation does not surprise me. While
some bugs in that area have been fixed, there are still several issues with dmd
using way too much memory (one of the main issues being that it never actually
releases any of it). The non-determinism, however, is a bit odd. Heap
fragmentation may be a good explanation though. I don't know.
- Jonathan M Davis
