Paul Sandoz wrote:
> 'Add to table' is currently only supported for strings and not for
> encoding algorithm data.
>
> When adding to the table it is necessary to determine whether a String
> has already been added. The most efficient general way is to use the
> String in the native form. Encoding to UTF-8 and then checking the
> bytes would be slower as this would have to be performed for each
> string to ascertain whether it is already indexed.
>
Let me make sure I understand. FI allows items added to a table to be
sent encoded instead of as text, but the current implementation is only
sending it as text? If so, that's good.
Perhaps maintain a quick lookup which string, but send the encoded pattern.
> Mixing generic approaches and specific approaches for adding stuff to
> the table is tricky. Maybe the application needs more control over this?
>
>
> For encoding algorithm data this is more tricky, it depends on whether
> one is converting from an object or already has the sequence of
> encoded octets. Seems like for efficiency the 'add to table' checking
> for encoding algorithm data would need to be used. Then there is the
> issue of managing more than one type of data per table.
>
true... I'm still working through a few bugs. Let's table this for a
moment and I'll come back to it when I get around to parsing speed time
trials.