[ft-devel] Revisiting LSB [2]

[Ulrich]

Sorry for replying that late. I tried to do it in time but obviously my 
message didn't get through. Checking the archive I found this message 
missing which I still had a copy of. I try again. 

I'm no expert in font rasterisation and in FreeType's approach on that 
matter. There are many members of this list who are experts in both 
subjects, first of all, David and Werner, the creators (?) of FreeType. 

George Williams writes: 

> My claim is that there is no single correct answer. There may be many
> good solutions.

This is correct but I still think that this fact doesn't imply that it isn't 
possible to set up a test suite. As far as I understand FreeType will 
generate a certain rasterisation result depending on the parameter settings 
and the environment it detects. There might be and probably will be other 
rasterisation results which are good or even better but they aren't 
generated by FreeType because the algorithms chosen simply don't generate 
them. 

If the cache system were replaced, extended or improved e.g. the 
rasterisation results should still be the same. And exactly this kind of 
fact should be proven by a test suite. There could be other aspects included 
within the test suite like memory used or time consumed. It would be a good 
thing to include those additional parameters to avoid surprises but it is 
not that simple to collect the data and compare them in a meaningful way. 

> Perhaps with truetype, with the bytecode interpreter turned on there is
> a deterministic process which should always produce the same answer. 
>
> But with the autohinter (or postscript hints) I do not believe this is
> the case. Requiring such leaves no room for improvement of hinting.

It seems that we're still talking about slightly different things. So far I 
don't see why some bits of the rasterisation result are randomly set. Which 
part of FreeType does introduce random? 

Of course, after improving the rasterisation algorithm checking the result 
by a human being is required and the data of the test suite which are used 
while comparing the results when the test suite is executed have to be 
replaced by the new samples. But even in that case a test suite might well 
be helpful as not necessarily all generated glyphs get a new shape. Work can 
then be focussed upon the ones that have changed. 

I would appreciate it if one of the experts shed some light on that.