Re: pango question

[Sayamindu Dasgupta <sayamindu clai net>]

On Tue, 2004-04-06 at 10:21 -0400, Owen Taylor wrote:
> On Mon, 2004-04-05 at 14:26, Eric Mader wrote:
> > I'm the original author of the current Pango Indic code. I made a single 
> > module because much of the functionality for each script was the same, 
> > and I wanted to avoid all of the problems associated with duplicate code.
> > 
> > Having worked with this code now for about five years, I've found that 
> > the approach I took makes it too fragile - it's too easy for the fix to 
> > one script to break other scripts. I now believe that the functionality 
> > should be split apart to avoid this problem. I think there may still be 
> > a way to do this that minimizes code duplication, but I haven't yet 
> > taken the time to look at this in detail.
> > 
> > The code I contributed to Pango is based on the ICU LayoutEngine code. I 
> >   am currently scheduled to make this change to the ICU code for release 
> > 3.2, which will be finished at the end of this year. My 3.2 schedule 
> > won't be firm until after ICU 3.0 is released, which is currently 
> > shceduled for mid-June of this year.
> 
> I'd certainly support this change; it's been quite hard for me to make
> bug fixes to the Indic code in the current form, because while I can
> trust that a bug reporter will be proposing a change that makes his
> script render better, it's quite likely that it will cause a regression
> elsewhere.
> 
> Other than splitting things up, the other thing that could really help
> here would be a set of automated or even semi-automated test cases.
> (By semi-automated, I mean someone would have to look at the result
> and compare to stored bitmap images.)

IMHO, it would be nice if each and every active (Indic) language group
got together and created a reference text - along with a bitmapped
representation of what the ideal rendering may look like. Something in
the lines of 

char1 + char2 + char3 -> conjunct

I am ready to submit such a fileset on behalf of the Bangla team, and I
think Mahesh here will be able to submit something for the Malayalam.

Usually, there are only a *few* special cases for each script that
causes problems (khanda-ta, ryaa in Bangla for instance) - and I think
if those special cases are rendered correctly, the developers should be
able to feel pretty confident :-). This would also mean that you won't
have to crosscheck a huge text file with a huge image each time you
change something in the code.

-thanks-
sdg