Re: [cedet-semantic] predictive-mode

[Toby C. <ts...@ca...>]

As the author of predictive mode, I can make a few comments about this.
Firstly, as you'll see from the "Future Improvements" section of the
predictive user guide, integration with semantic is on the to-do list. I
even discussed it a while ago on this list with Eric. But it's a fairly
major undertaking, and I don't have time at the moment to do it.

In fact, properly integrating predictive-mode with semantic wouldn't
only give UI improvements, but would also enhance the completion process
itself, in two ways:

1) Predictive mode stores "dictionaries" of completion candidates in a
data structure that is optimised for finding completions of a prefix
string, so finding completions is very fast, even in huge dictionaries.
(The data structure is basically built on a ternary search tree, but
with caching of time-consuming completions in a hash table.) In fact,
the data structures are implemented in a separate library, called
dict-tree.el, so with a bit of coding it would be possible to use them
as a semantic database. This is probably the correct way to start
integrating predictive mode with semantic.

2) Completion with predictive mode is...predictive. That is, it learns
how often you use each completion, and prioritises them accordingly.
This is obviously of most benefit when the number of possible
completions is very large (e.g. when completing English words). But this
will often be the case in large programming projects too, or languages
with huge collections of libraries (Java, PHP...).


The predictive user interface is separated off into a completely
independent library, called completion-UI. Completion-ui can be used
without predictive mode, so the UI benefits can be had without any need
for predictive mode itself (though you will of course lose the benefits
described above). Using completion-UI with semantic is fairly easy: it
just requires a writing a wrapper function around semantic's completion
function, in order to pass the list of completion candidates to
completion-UI in the required form. (To get it working well, there are a
few other things to do...) I believe there's even some code floating
around, either the Emacs wiki or the emacs-sources mailing list, that
makes a stab at this.

If anyone wants to take one of these projects on, I'm happy to answer
any questions about the guts of completion-UI, dict-tree.el and
predictive mode. The completion-UI integration should be the easiest,
followed by using the dict-tree.el data structures as semantic database,
leaving full integration of predictive mode as the most difficult.

Toby


Arne Schmitz wrote:
> I just read about predictive-mode [1], and thought it might give some ideas, 
> too, how to improve on semantic's code completion. Although more from a UI 
> point of view.
> 
> Arne
> 
> [1] http://www.emacswiki.org/cgi-bin/emacs?PredictiveMode