[reportlab-users] Chasing pyHnj error

[Robin Becker]

Dinu Gherman wrote:

> Robin Becker:

> 

>> As for new paragraph classes I have been messing with Kuchling's 

>> tex_wrap and have managed to get the run time down to something 

>> reasonable. In python I have the time down from over 3.06" to 0.5". 

>> This just corresponds to the simplest of line breakings. His version 

>> doesn't do hyphenation or even follow Knuth's algorithm all that 

>> carefully.

>>

>> I believe it should be possible to improve much further. I translated 

>> simply to pyrex and have the time down again to 0.23". With more 

>> agressive C style coding that should improve further (perhaps another 

>> factor of 10 since my pyrex version is still using lots of python 

>> variables in the main loop).

> 

> Well, I first strive for having a paragraph class which:

> 

> - is human-readable

> - has a pylint coefficient of => 7.5/10 (instead of the current

>   value of 0.75/10 for RL's paragraph.py ;-)

> - has hyphenation

> - has inline images, and

> - has an extension mechanism for custom tags

> 

> Once I'll have that I'll come back to you for ideas on performance.

> ;-)


well the current paragraph is slowly getting inline images and has had a 
mechanism for non-character things for a long time; that's used for 
anchor targets etc etc. Custom tags presumably means an adjustable 
parser ie all non-standard tags get mapped to a user function; that 
shouldn't be too hard if there's some agreement about what a tag is 
allowed to produce. Hyphenation is just difficult; we need to define 
words before we can find hyphenation points. The current paragraph 
doesn't actually see words very well. Presumably

     "hyp<font color=red>hena</font>te"

is a word, but for efficiency we don't actually see the "hyphenate" all 
in one place so extracting the word and looking for hyphenation points 
is left as an exercise.
-- 
Robin Becker