dmd 1.045 / 2.030 release

[Robert Jacques]

On Tue, 12 May 2009 12:41:50 -0400, dsimcha <dsimcha at yahoo.com> wrote:

> == Quote from Tomas Lindquist Olsen (tomas.l.olsen at gmail.com)'s article
>> Is there a reason for the missing announcement ?
>> http://digitalmars.com/d/1.0/changelog.html#new1_045
>> http://www.digitalmars.com/d/2.0/changelog.html#new2_030
>> and what happened to 1.044 ?
>> -Tomas
>
> Probably because it doesn't quite work yet.  I'm waiting for some crufty  
> old 2.029
> code to run, so I thought I'd try out shared a little.  Here are the  
> results:
>
> import std.stdio, std.perf, core.thread;
>
> shared uint foo;
>
> void main() {
>     auto t = new Thread({stuff();});
>     t.start;
>     scope pc = new PerformanceCounter;
>     pc.start;
>     foreach(i; 0..10_000_000) {
>         foo++;
>     }
>     t.join;
>     pc.stop;
>     writeln(pc.milliseconds);
>     uint bar = foo;
>     writeln(bar);  // Prints some pseudorandom, wrong number.
> }
>
> void stuff() {
>     foreach(i; 0..10_000_000) {
>         foo++;
>     }
> }
>
> Or is the automatic synchronization of shared variables part not  
> supposed to be
> implemented yet?

Bartosz hasn't talked about D's shared system yet. Anyways, what you are  
seeing is a classic read/write race, and is expected even if foo is  
sequential consistent or locked. What you'd need to test is the Peterson  
lock  
(http://bartoszmilewski.wordpress.com/2008/11/11/who-ordered-sequential-consistency/)

Remember foo++ ->
r1 = foo;
r1++;
foo = r1;

So
lock(foo);
r1 = foo;
unlock(foo);
r1++;
lock(foo);
foo = r1;
unlock(foo);