dolfin team mailing list archive

["Martin Sandve Alnæs" <martinal@xxxxxxxxx>]

2008/4/3, Anders Logg <logg@xxxxxxxxx>:
> On Thu, Apr 03, 2008 at 07:47:13PM +0200, Martin Sandve Alnæs wrote:
>  > We only talked about create() the other day, copy() can still be
>  > useful! Or does the factory duplicate this functionality?
>
>
> No, but Kent has added an assignment operator:
>
>   GenericVector* y = x.factory().createVector();
>   *y = x;

Ok, but the assignment operator doesn't exist in Python, so some
variant of the copy function may still be needed. Could be just a
wrapper for the two lines above.

But the current default implementation of operator= (just return
*this;) isn't good, it should be abstract or at least raise an error
somehow.

And I the uBlas implementation seems potentially dangerous too,

const uBlasVector& uBlasVector::operator= (const GenericVector& x_)
{
  const uBlasVector* x = dynamic_cast<const uBlasVector*>(&x_);
  if (!x) error("The vector should be of type PETScVector");

  *this = (*x)*1.0;
  return *this;
}

In particular, this line
  *this = (*x)*1.0;
depends on (*x)*1.0 being resolved by the ublas_vector subclass,
triggering operator= in ublas_vector. Implementing operator* in
GenericVector will break this and possibly make an infinite recursion.
Unless I misunderstand something? (I don't know ublas).

--
Martin