Am 25.02.2008 um 15:10 schrieb Abdelrazak Younes:
Stefan Schimanski wrote:
Am 25.02.2008 um 14:45 schrieb Abdelrazak Younes:
The math completion works but not the text, is that normal?
Well, depends on what you call working. First I disabled the
automatic completion for text because it might be annoying for many
people.\
What is the purpose of the "Automatically show in text mode" then?
If you select this, it will be shown. But by default, the checkbox is
not selected.
So you have to press tab after entering the first character of a
word.
It doesn't seem to work: I hit tab and I get nothing.
So it does not work. I guess it's connected to the bug you had before.
You are using Linux? Wondering to setup a VmWare Ubuntu...
Second the data collection is very poor because I wanted to get the
backend, WordList, working first in a proper way. So it only adds
words when closing them with a space. I.e. type lalalala<space> and
the word is added to the WordList.
The next step is to fill this word list in a better way. One idea
is to take the longer words from the document.
Yes, at initial file parsing would be good.
Adding words to the word list as it is not, in addition to initial
file parsing, is also a good idea IMO. But I am not sure how one could
handle deletion of words. I mean imagine you make a spelling error in
a new word. It is added to the list. Then you realise the mistake and
you correct it. Should the wrong word stay in the completion? That
would be confusing IMO. One could add some kind of reference counting
to words and erase them as soon as the counter goes to zero. Not sure
though how complicated this is.
Another would be to use a dictionary word list.
Would be nice too.
Indeed. And it is quite easy to implement.
Btw, one thing I wanted to mention to you: using the weighted_btree
one could get rid of the loop in the scrolling code which adds the
heights of paragraphs. You have to use the height (or some guessed
value of those without known metrics) as the weight in the
weighted_btree. Then the summing up is logarithmic in the paragraph
number, not linear as it is now. You can even update the weight (i.e.
height) of a paragraph in logarithmic time (I have to add an
update_weight method to the weighted_btree. But it's easy). There is
never the need to go through all paragraphs then.
No idea if this gives a noticeable speedup for the scrolling. For very
long documents I am quite sure it would. Do you have any numbers how
long this takes e.g. for the user manual?
Stefan
