dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Fri, Apr 11, 2008 at 02:24:13PM +0200, Dag Lindbo wrote:
> Anders Logg wrote:
> > On Fri, Apr 11, 2008 at 01:34:24PM +0200, Martin Sandve Alnæs wrote:
> >> 2008/4/11, Anders Logg <logg@xxxxxxxxx>:
> >>> On Fri, Apr 11, 2008 at 12:23:15PM +0100, Garth N. Wells wrote:
> >>>  >
> >>>  >
> >>>  > Anders Logg wrote:
> >>>  > > Which one do we want to use for element access in vectors and
> >>>  > > matrices. It looks like operator() is now implemented for uBlasVector,
> >>>  > > but isn't operator[] more natural to use?
> >>>  > >
> >>>  >
> >>>  > Much the same to me. We've used (.,.) for matrices, so it seems natural
> >>>  > to use (.) for vectors.
> >>>  >
> >>>  > Garth
> >>>
> >>>
> >>> We could use [] for matrices also I guess.
> >>>
> >>>  Then it will look the same as numpy.
> >> It's useful to index vectors with [], since the code will be similar
> >> for regular arrays.
> > 
> > I forgot something important, namely that operator[] expects exactly
> > one argument, so we can't do A[i, j] for matrices.
> > 
> > Maybe then it's better to have () both for vectors and matrices?
> > 
> 
> Whatever you do with matrices, I don't have an opinion.
> 
> However, I think operator[] should be present for vectors. One can argue 
> that it does not _need_ to be there, but the fact is that it _is_ in the 
> 0.7.2 interface (it worked until yesterday for uBlasVector, but it was 
> maybe handled by uBlas itself?).
> 
> /Dag

Other opinions? Should we have both?

Since we have A(i, j), it will be expected that x(i) works.

But it will also be expected that x[i] works (since it works in numpy
and it has worked before for uBlasVector).

-- 
Anders