Questions on writing an input method

Derek Pressnall dpressnall at gmail.com
Mon Oct 15 22:16:30 CEST 2007


Since there has been a lot of discussion on input methods, I thought
I'd take a crack at writing my own.  Actually I was thinking of
putting together more of a framework that multiple input method styles
can be plugged into.  But I have a couple questions to get started:
1) How does the input method register itself to the openmoko framework
so that it shows up as a valid user selectable method?  For example,
on Qtopia, I believe it has to be compiled as a dynamic shared object,
with specific functions / symbols defined, and then plopped in a
particular directory so that it gets picked by the framework.  Where
can I find documentation on how this is to be done in openmoko?
I've looked at the included keyboard app, but it appears to be written
as a regular X11 app and not using any gtk or libmokoui calls, and I
can't find where it registers itself.

2) Does the current UI pop up the default input method automatically,
or is it up to the user
to bring up the soft keyboard manually? I can't tell  by looking at
the code (last time I brought the code up in the emulator was previous
to the 2007.2 changes, and haven't had much luck getting the current
code to build -- problems on my end).

3) If the answer to 2) is yes, then can an app specify any preferences
to the input method that is popped up?  For example if an app field
only accepts numeric input, then it would make more sense to pop up a
10-key pad vs. a full keyboard.

4) How would the input method communicate the inputted data to the app
-- does it have to use X11 specific calls to inject characters into
the keyboard buffer, or is there (currently or plans for) a libmokoui
call to deliver a character as input to the current app?

5) Finally, what which gtk widgets would be best used for building a
screen keyboard?  One method would be to pack shaped buttons (each
representing a key) into a box using specific coordinates.  But this
would require that a key be pressed & released before the action is
delivered -- the button widgets don't support clicking on one key and
then dragging to other keys (or dragging to specific areas of a key).
However it would be good enough for a simple pickboard input method.
An alternative would be generating two pixmaps, one to represent the
virtual keyboard / input area, and a second one to be used as a mask
that identifies the various hot spots.  Then load the main pixmap into
a gtk_drawing_area widget and grab the coordinates that are
clicked/dragged on from there, and compare with the value at those
coordinates in the second pixmap to determine which virtual keys were
hit.  This would work perfectly for some of the more advanced input
methods that were discussed (i.e. QwikScript).

Note that I'm a fairly good C programmer but haven't done a lot with
GTK yet.  However once I get the GTK basics down, I'd like to put
together something that can be utilized as a clear reference input
method app, so I'd probably lean towards code readability vs. making
it the best possible.
Also, I'd like to stick to GTK and libmoko* calls if possible, since
it is possible that I might code up a version using a small
interpreted language which I'm writing up GTK bindings for.



More information about the openmoko-devel mailing list