Re: [ft] Off by one error in freetype (or in programmer?)

[Dave Arnold]

> If I look at the glyph /H in Source Sans Pro in a font editor, the top right 
> bound is a point at (562,656) and the bottom left point is at (90,0). If I were 
> to think about this as a bitmap, I can address row 0 and row 656, so there are 
> 657 addressable rows.
>
> But when I render it to a bitmap in freetype, I get bitmap.rows == 656, 
> bitmap.width == 472 (562 minus the LSB of 90 is 472).

I would add a caution: be careful in assuming a relationship between outline 
coordinates and bitmaps coordinates. The grid-fitting and scan conversion 
processes are not simple, especially with hinting (you don't say if you are 
rendering SourceSansPro hinted). For example, don't rely on an outline bounding 
box to protect access to a bitmap.

-Dave


On 4/5/2018 11:55 PM, Lawrence D'Oliveiro wrote:
> On Fri, 06 Apr 2018 08:40:03 +0200 (CEST), Werner LEMBERG wrote:
>
> > Of course there will be stair-stepping!  Just have a look into the
> > additional documentation files that accompany the old TrueType
> > specification (both Apple and MS versions).  You'll see rectangles
> > everywhere talking about B/W rasterization.
> I know, I used Macs in those days. Most of them still had only
> one-bit-per-pixel screens, and people didn’t really know any better.
>
> > A graphics system might apply filters to display such bitmaps on
> > screens, but this is not covered by the TT specification.
> If you are going to be filtering out frequencies above the Nyquist
> limit (antialiasing), then a rectangular shape is the wrong thing to
> start from, since you have already thrown away even harmonics below the
> limit.
>
> You start from a dimensionless point, i.e. an “ideally-sampled” pixel.
>
> _______________________________________________
> Freetype mailing list
> address@hidden
> https://lists.nongnu.org/mailman/listinfo/freetype