Re: [ft-devel] [RFC] FT_LCD_FILTER_HEAVY

[Infinality]

One thing to keep in mind is that behind the scences, the Freetype FIR 
filter is implemented as an array of 5 FT_Bytes, which means the values 
are only 0 through 255 for each FIR component.  So, the floats could 
ultimately be getting rounded somewhat, belying their precision.  But, I 
think we'd also want to avoid requiring people to use hex notation or 
values 0 through 255 ;)  So, float might be best.

Freetype actually uses a 5 byte array even for the FIR filter (light 
LCD) which really only has 3 meaningful values.  So, as far as Freetype 
goes, I think 5 makes perfect sense.  I can't speak for anything else 
though.

I have heard of people using an asymmetrical FIR filter, but it's 
probably so rarely used that it would justify not implementing if it 
were considerably more difficult to do.  I think that just specifying 5 
values is the easiest and least-confusing thing to do.  If someone is in 
there tinkering with these settings, they are probably smart enough to 
understand decimal notation. ;)

My thoughts,
Erik


On 11/20/2011 04:50 PM, Eric Rannaud wrote:
> Sorry, Gmail mangled the message. Repost, hopefully readable this time.
>
> So I wrote the attached patch for fontconfig, that I will submit to
> the fontconfig ML.
>
> With the patch and freetype2, lcdfilter=lcddefault would correspond to:
>
>        <match target="font">
>          <edit mode="assign" name="lcdfilter">
>            <const>lcdcustom</const>
>          </edit>
>        </match>
>        <match target="font">
>          <edit mode="assign" name="lcdfiltercustom">
>            <string>.0625 .25 .4375 .25 .0625</string>
>          </edit>
>        </match>
>
> But it's not perfect, as users of the fontconfig API would have to
> parse the lcdfiltercustom string themselves. I will ask if the
> fontconfig developers see a better way of specifying a list of double.
>
> Questions:
>
> - Would there ever be a good reason to not have a symmetric FIR?
> Should be the option be specified as ".4375 .25 .0625" (first half,
> reverse order) and assume symmetry for the other weights?
>
> - Could the size of the filter be different? Or is 5 likely to be
> fairly universal, even for other renderers?
>
> - When specified from the center value ".4375 .25 .0625", missing
> values can be assumed to be 0, and extra values can be ignored. I'm
> tempted to specify the option this way instead, unless asymmetric
> filters are useful.
>
> Thanks.