Le 10/09/2012 21:08, Sébastien Brisard a écrit :
> Hi,
Hi Sébastien,
> I thought it was not good practice to rely on exception in
> unexceptional circumstances. In ArrayFieldVector, there are numerous
> occurences of the following pattern
>
> public FieldVector<T> add(FieldVector<T> v)
> throws DimensionMismatchException {
> try {
> return add((ArrayFieldVector<T>) v);
> } catch (ClassCastException cce) {
> checkVectorDimensions(v);
> T[] out = buildArray(data.length);
> for (int i = 0; i < data.length; i++) {
> out[i] = data[i].add(v.getEntry(i));
> }
> return new ArrayFieldVector<T>(field, out, false);
> }
> }
>
> The "catch (ClassCastException cce)" seems uggly to me. Should I file
> a JIRA issue and start replacing with instanceof?
This was done on purpose a long time ago.
At that time, it appeared that it was faster to do it this way, i.e.
attempt by default the fast method which did not rely on getEntry, and
fall back to a loop using getEntry only in the very rare cases the first
fails.
I am not sure this is useful anymore, as JVM optimizers are better and
better (so they may well replace the getEntry on the fly and use direct
array access when they can).
Perhaps you could do some benchmark with a modern JVM and look if this
hack is still useful or not.
best regards,
Luc
> Best regards,
> Sébastien
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
> For additional commands, e-mail: dev-h...@commons.apache.org
>
>
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
For additional commands, e-mail: dev-h...@commons.apache.org
