Debugging D shared libraries

[ponce via Digitalmars-d-learn]

On Saturday, 19 September 2015 at 10:45:22 UTC, Russel Winder 
wrote:
> Calling D from Python. I have two functions in D, compiled to a 
> shared object on Linux using LDC (but I get same problem using 
> DMD).
>
> The sequential code:
>
>     extern(C)
>     double sequential(const int n, const double delta) {
>       Runtime.initialize();
>       const pi = 4.0 * delta * reduce!(
>             (double t, int i){ immutable x = (i - 0.5) * delta; 
> return t + 1.0 / (1.0 + x * x); })(
>             0.0, iota(1, n + 1));
>       Runtime.terminate();
>       return pi;
>     }
>
> works entirely fine. However the "parallel" code:
>
>     extern(C)
>     double parallel(const int n, const double delta) {
>       Runtime.initialize();
>       const pi = 4.0 * delta * taskPool.reduce!"a + b"(
>           map!((int i){ immutable x = (i - 0.5) * delta; return 
> 1.0 / (1.0 + x * x); })(iota(1, n + 1)));
>       Runtime.terminate();
>       return pi;
>     }
>
> causes an immediate segfault (with LDC and DMD.  I am assuming 
> that the problem is the lack of initialization of the 
> std.parallelism module and hence the use of taskPool is causing 
> a problem. I am betting I am missing something very simple 
> about module initialization, and that this is not actually a 
> bug.
>
> Anyone any proposals?

Try using an explicit TaskPool and destroying it with scope(exit).


Also if using LDC, you can use global ctor/dtor to deal with the 
runtime.


----------------------->8---------------------

        extern (C) {
             pragma(LDC_global_crt_ctor, 0)
             void initRuntime()
             {
                 import core.runtime;
                 Runtime.initialize();
             }
             pragma(LDC_global_crt_dtor, 0)
             void deinitRuntime()
             {
                 import core.runtime;
                 Runtime.terminate();
             }
         }

----------------------->8---------------------