Re: [ft] Creating an [OT]TF font from BDF font

On Sat, 2005-12-10 at 00:42, Werner LEMBERG wrote:
> All of this looks very promising.  I think the final decision on the
> table format can only be done after converting a bunch of BDFs forth
> and back.
Ok. I have a version of fontforge which puts bdf properties into a 
'BDF ' table in an sfnt, and then reads them back and writes them out
into bdf. The data seem to survive the round trip.

I've posted a source tarball
	http://fontforge.sf.net/fontforge_full-20051213.tar.bz2

I've also posted an otb file containing 3 bitmap strikes and a BDF table
	http://fontforge.sf.net/hidden/FixedMedium.otb

On Thu, 2005-12-08 at 16:46, Keith Packard wrote: 
> I need this + a utility to regenerate BDF files from the TTF so I can
> validate a lossless round-trip for the existing BDF files. 
A couple of caveats about "lossless" round-trips:
* BDF format supports an (x,y) advance for both horizontal & vertical
metrics
	(ie. it could support Urdu where there is a vertical advance
	 as well as horizontal)
	EBLC metrics only support one advance for each writing
	 direction.
	hmtx/vmtx also only support one advance
 In other words for those rare fonts with an advance vector rather than 
  a simple advance an otb font can't retain that info.(I don't think
  that's an issue for X)

* BDF format supports advances, pixelsizes > 255 pixels
	EBLC metrics don't

* FontForge doesn't really support VVector
   So this data will be lost (I don't think that's an issue for X)

* Each otb file should be single resolution:
   X/BDF can distinguish between
	-gww-caslon-medium-r-normal--12-120-75-75-p-150-iso8859-1
	-gww-caslon-medium-r-normal--12-100-100-100-p-150-iso8859-1
   (same pixel size, but potentially different bitmap patterns because
    designed for different point-sizes at different resolutions)
   But EBLC can only have one strike per pixelsize.

* If FontForge can't recognize the encoding of a font it doesn't know
  how to map glyphs into the cmap table.
=====================================================================
I changed my BDF table proposal, taking out the idea of an array type
and reverting to David's suggestion of treating them as atoms. So here
is my current spec:
/* Format:
	USHORT	version	    : 'BDF' table version number, must be 0x0001
	USHORT	strikeCount : number of strikes in table
	ULONG	stringTable : offset (from start of BDF table) to string
                                 table

followed by an array of 'strikeCount' descriptors that look like: 
	USHORT	ppem	    : vertical pixels-per-EM for this strike
	USHORT	num_items   : number of items (properties and
                                   atoms), max is 255

this array is followed by 'strikeCount' value sets. Each "value set" is 
an array of (num_items) items that look like:
	ULONG	item_name	: offset in string table to item name
	USHORT	item_type	: item type: 0 => non-property string
					 (e.g. COMMENT)
 					     1 => non-property atom
					 (e.g. FONT)
					 (also SIZE even though not
                                           really an atom)
					     2 => non-property int32
					     3 => non-property uint32
					  0x10 => flag for a property, 
						  ored with above value
						  types)
	ULONG	item_value	: item value. 
				strings	 => an offset into the string
					  table of the corresponding
					  string, without the 					  surrounding double-quotes

				atoms	 => an offset into the string
					  table

				integers => the corresponding 32-bit
					  value
Then the string table of null terminated strings. These strings should
be in ASCII.
==================================================================
Comments?