[devel] Problems with kerning support for PFR fonts in FreeType2 v.2.1.4.

[Adam Piotrowski]

Dear All,

This is my "first time" on any Open Source developer's list,
so do not hesitate to point out if I did something wrong.

Company I work for uses FreeType2 in SetTopBox software.
I have had 3 problems with retrieving kerning data out of PFR font file,
after applying solutions described below I have got sensible kerning pairs
from the PFR file.

1. function FT_Get_PFR_Kerning() always returns error code = 132.
    I think the reason of the problem is in the file src/base/ftpfr.c, line
38 -
    This is the place were the function ft_pfr_check() looks for the string
"pfr",
    but the index were the terminating null is looked for is 4.
    After changing index to 3 the string "pfr" is matched correctly.

2. Kerning pair data are processed incorrectly, due to the type definition
in:
    src\pfr\pfrtypes.h, line 282.
    The type defines flags in the extraItem carrying kerning pair data;
these flags
    indicates if charcter codes or kerning adjustment, respectively, are
encoded on 1 or 2 bytes.
    Values of this enum type are applied as bit masks to the byte with flags
in order
    to extract each flag. In my opinion enum values are swapped which
resulted in
    2-byte codes being interpreted as single-byte, and single-byte
adjustment as 2-byte long,
    this lead to a total mess in retrieved kerning data - I could retrieve
only
    2 corrupted pairs instead of 950 present in the font file.
    (see the description of PFR format:
    http://www.davic.org/Download/Spec1_4_1/141_part09.pdf - Information
Representation.
    Annex A - Coding of the Outline Fonts, par. A.10 Kerning data)

    The type is defined as follows:

  typedef enum PFR_KernFlags_
  {
    PFR_KERN_2BYTE_ADJ      = 0x01,
    PFR_KERN_2BYTE_CHAR  = 0x02
  } PFR_KernFlags;

  I have changed it to the following:

  typedef enum PFR_KernFlags_
  {
    PFR_KERN_2BYTE_CHAR  = 0x01,
    PFR_KERN_2BYTE_ADJ      = 0x02
  } PFR_KernFlags;

3. In the PFR font kerning pairs consist of unicode codes instead of glyph
indices.
    Right now I supply FT_Get_PFR_Kerning() with unicode codes of the
characters,
    (this is contrary to the FreeType Reference Manual, which says that the
function should accept glyph indices)
    but for maybe consistency it would be better to supply
FT_Get_PFR_Kerning()
    with glyph indices and do some mapping of indices to unicode somewhere
in the font driver.
    I know too little about internals of FreeType to propose some sensible
solution,
    probably someone more experienced could comment on this?

Best Regards.

Adam