Font-Selection strategies (Re: svn commit: r655281 - in /xmlgraphics/fop/trunk: src/java/org/apache/fop/fo/FOPropertyMapping.java src/java/org/apache/fop/layoutmgr/inline/CharacterLayoutManager.java status.xml)

[Max Berger <ma...@berger.name>]

Andreas,


Am 11.05.2008 um 16:25 schrieb Andreas Delmelle:
>> Support character-by-character font-selection strategy on  
>> fo:character element
>
> Good thinking!
> FWIW, I've also been thinking in a similar direction, but maybe a  
> bit more generic, as utility method in the FontManager maybe (?)

yes, indeed. See the TODO comment in the code. The reason I  
implemented it there were:
- proof-of-concept
- had no idea where it would really belong

> The piece of code that you moved to a separate method here here was  
> indeed a temporary solution of my moving this logic out of the FO  
> tree, since I feel it did not really belong there. I just needed to  
> get the same behavior as before, when it was housed in CommonFont,  
> which always returned the first font-family in the list.

Thanks for that! This really helped getting this started

>> +    private Font selectFontForCharacter(Character fobj) {
>
> Now that we have a FontManager, what if we add something to it like:
>
> public Font selectFontForCharacter(FontTriplet[] triplets, char, int)

+1, what would be the int be for?

> That way the loop in question finally moves to where I think it  
> actually belongs... and a similar approach could then be applied to  
> the TextLayoutManager. As an initial step, we could already make  
> sure that the selected font is always one that can display the first  
> character.
I don't like a "partial" implementation (the first character version)

For the textlayoutmanager: I saw that similar functionality in  there  
multiple times, for paragraphs, page-references, etc.  They all seem  
to create the same result, just based on content (paragraph) or  
context (page-references). So the first step here would be to refactor  
the common functionality out as well.

Now the real question is about the possible font selection stragety.  
We'd have to implement "auto" and "characte-by-character"

they are defined as follow:
auto: is vendor-specific, but should use a "broader context"
char-by-char: select first font-match for evey character.

First the question is about line metrics: In a paragraph, we could use
- the line metrics for the font of the first char for the whole  
paragraph -> easiest implementation, but the result may look bad
- the line metrics of all fonts used in a line (I have no idea how  
that would work) -> best result
- the line metrics maximia for all fonts used in a paragraph (medium  
to implement, medium result)

the next question is: what is auto-selection strategy?

would it be:
- select font based on the first character? now that i think about it,  
I really like it - it would be really fast, and produce a correct  
result IF the symbols are enclosed in their own fo:block
- select a font that can display the whole paragraph (if possible)
- switch for every word? This would probably give good results, but  
may not make so much difference to char-by-char.
- make it the same as character-by-character? This would possibly be  
slower, and produce some overhead, so its a bad idea for "default"  
behavior.

hope this makes sense.


> Andreas

Max

Re: Font-Selection strategies (Re: svn commit: r655281 - in /xmlgraphics/fop/trunk: src/java/org/apache/fop/fo/FOPropertyMapping.java src/java/org/apache/fop/layoutmgr/inline/CharacterLayoutManager.java status.xml)

[Andreas Delmelle <an...@telenet.be>]

On May 13, 2008, at 07:32, Max Berger wrote:

Hi Max

> <snip />
>>> +    private Font selectFontForCharacter(Character fobj) {
>>
>> Now that we have a FontManager, what if we add something to it like:
>>
>> public Font selectFontForCharacter(FontTriplet[] triplets, char, int)
>
> +1, what would be the int be for?

I just noticed that the font-size is being passed somewhere, but  
maybe FontManager does not need it at all. (? to be verified...)

>>
>> That way the loop in question finally moves to where I think it  
>> actually belongs... and a similar approach could then be applied  
>> to the TextLayoutManager. As an initial step, we could already  
>> make sure that the selected font is always one that can display  
>> the first character.
> I don't like a "partial" implementation (the first character version)

All the better ;-)

> For the textlayoutmanager: I saw that similar functionality in   
> there multiple times, for paragraphs, page-references, etc.  They  
> all seem to create the same result, just based on content  
> (paragraph) or context (page-references). So the first step here  
> would be to refactor the common functionality out as well.
>
> Now the real question is about the possible font selection  
> stragety. We'd have to implement "auto" and "characte-by-character"
>
> they are defined as follow:
> auto: is vendor-specific, but should use a "broader context"
> char-by-char: select first font-match for evey character.
>
> First the question is about line metrics: In a paragraph, we could use
> - the line metrics for the font of the first char for the whole  
> paragraph -> easiest implementation, but the result may look bad
> - the line metrics of all fonts used in a line (I have no idea how  
> that would work) -> best result
> - the line metrics maximia for all fonts used in a paragraph  
> (medium to implement, medium result)

The second would, of course, be the better option. I'm hoping Manuel  
chimes in later, as he has already been looking at this, and may have  
some ideas to share.

As far as I remember from earlier discussions, the trickiest part  
will be to keep track of changes in the alignment context as the font  
changes. If we can manage that, the line-layout algorithm should  
automatically take care of the rest.

> the next question is: what is auto-selection strategy?
>
> would it be:
> - select font based on the first character? now that i think about  
> it, I really like it - it would be really fast, and produce a  
> correct result IF the symbols are enclosed in their own fo:block
> - select a font that can display the whole paragraph (if possible)
> - switch for every word? This would probably give good results, but  
> may not make so much difference to char-by-char.
> - make it the same as character-by-character? This would possibly  
> be slower, and produce some overhead, so its a bad idea for  
> "default" behavior.

Auto-selection would, as mentioned, use a 'broader context'. The  
difference is that, with "character-by-character", each character in  
a word could be put in a different font (the first of the specified  
families that has a glyph for it). With "auto" we could do something  
clever like try to determine whether there is a font-family that can  
display whole words (or entire paragraphs), which may lead to much  
nicer-looking results (less font-switching -> better overall  
alignment). If the first font-family only has a glyph for 2 out of 5  
characters in a word, but the second has matching glyphs for all 5,  
with a value of "auto" we could decide to use the second.

That said, getting "character-by-character" to work correctly would  
already be vast improvement over the current situation, where we  
don't look at the characters at all, and just use the first specified  
font.


Cheers

Andreas