if not modules then modules = { } end modules ['font-tfm'] = {
version = 1.001,
comment = "companion to font-ini.mkiv",
author = "Hans Hagen, PRAGMA-ADE, Hasselt NL",
copyright = "PRAGMA ADE / ConTeXt Development Team",
license = "see context related readme files"
}
local utf = unicode.utf8
local next, format, match, lower, gsub = next, string.format, string.match, string.lower, string.gsub
local concat, sortedkeys, utfbyte, serialize = table.concat, table.sortedkeys, utf.byte, table.serialize
local allocate = utilities.storage.allocate
local trace_defining = false trackers.register("fonts.defining", function(v) trace_defining = v end)
local trace_scaling = false trackers.register("fonts.scaling" , function(v) trace_scaling = v end)
local report_define = logs.new("define fonts")
-- tfmdata has also fast access to indices and unicodes
-- to be checked: otf -> tfm -> tfmscaled
--
-- watch out: no negative depths and negative eights permitted in regular fonts
--[[ldx--
Here we only implement a few helper functions.
--ldx]]--
local fonts = fonts
local tfm = fonts.tfm
fonts.loaded = allocate()
fonts.dontembed = allocate()
fonts.triggers = fonts.triggers or { } -- brrr
fonts.initializers = fonts.initializers or { }
fonts.initializers.common = fonts.initializers.common or { }
local set_attribute = node.set_attribute
local fontdata = fonts.ids
local nodecodes = nodes.nodecodes
local disc_code = nodecodes.disc
local glyph_code = nodecodes.glyph
--[[ldx--
The next function encapsulates the standard loader as
supplied by .
--ldx]]--
tfm.resolvevirtualtoo = true -- false
tfm.sharebasekerns = false -- true (.5 sec slower on mk but brings down mem from 410M to 310M, beware: then script/lang share too)
tfm.mathactions = { }
tfm.fontnamemode = "fullpath"
tfm.enhance = tfm.enhance or function() end
fonts.formats.tfm = "type1" -- we need to have at least a value here
function tfm.read_from_tfm(specification)
local fname, tfmdata = specification.filename or "", nil
if fname ~= "" then
if trace_defining then
report_define("loading tfm file %s at size %s",fname,specification.size)
end
tfmdata = font.read_tfm(fname,specification.size) -- not cached, fast enough
if tfmdata then
tfmdata.descriptions = tfmdata.descriptions or { }
if tfm.resolvevirtualtoo then
fonts.logger.save(tfmdata,file.extname(fname),specification) -- strange, why here
fname = resolvers.findbinfile(specification.name, 'ovf')
if fname and fname ~= "" then
local vfdata = font.read_vf(fname,specification.size) -- not cached, fast enough
if vfdata then
local chars = tfmdata.characters
for k,v in next, vfdata.characters do
chars[k].commands = v.commands
end
tfmdata.type = 'virtual'
tfmdata.fonts = vfdata.fonts
end
end
end
tfm.enhance(tfmdata,specification)
end
elseif trace_defining then
report_define("loading tfm with name %s fails",specification.name)
end
return tfmdata
end
--[[ldx--
We need to normalize the scale factor (in scaled points). This has to
do with the fact that uses a negative multiple of 1000 as
a signal for a font scaled based on the design size.
--ldx]]--
local factors = {
pt = 65536.0,
bp = 65781.8,
}
function tfm.setfactor(f)
tfm.factor = factors[f or 'pt'] or factors.pt
end
tfm.setfactor()
function tfm.scaled(scaledpoints, designsize) -- handles designsize in sp as well
if scaledpoints < 0 then
if designsize then
if designsize > tfm.factor then -- or just 1000 / when? mp?
return (- scaledpoints/1000) * designsize -- sp's
else
return (- scaledpoints/1000) * designsize * tfm.factor
end
else
return (- scaledpoints/1000) * 10 * tfm.factor
end
else
return scaledpoints
end
end
--[[ldx--
Before a font is passed to we scale it. Here we also need
to scale virtual characters.
--ldx]]--
function tfm.getvirtualid(tfmdata)
-- since we don't know the id yet, we use 0 as signal
if not tfmdata.fonts then
tfmdata.type = "virtual"
tfmdata.fonts = { { id = 0 } }
return 1
else
tfmdata.fonts[#tfmdata.fonts+1] = { id = 0 }
return #tfmdata.fonts
end
end
function tfm.checkvirtualid(tfmdata, id)
if tfmdata and tfmdata.type == "virtual" then
if not tfmdata.fonts or #tfmdata.fonts == 0 then
tfmdata.type, tfmdata.fonts = "real", nil
else
local vfonts = tfmdata.fonts
for f=1,#vfonts do
local fnt = vfonts[f]
if fnt.id and fnt.id == 0 then
fnt.id = id
end
end
end
end
end
--[[ldx--
Beware, the boundingbox is passed as reference so we may not overwrite it
in the process; numbers are of course copies. Here 65536 equals 1pt. (Due to
excessive memory usage in CJK fonts, we no longer pass the boundingbox.)
--ldx]]--
fonts.trace_scaling = false
-- the following hack costs a bit of runtime but safes memory
--
-- basekerns are scaled and will be hashed by table id
-- sharedkerns are unscaled and are be hashed by concatenated indexes
--~ function tfm.check_base_kerns(tfmdata)
--~ if tfm.sharebasekerns then
--~ local sharedkerns = tfmdata.sharedkerns
--~ if sharedkerns then
--~ local basekerns = { }
--~ tfmdata.basekerns = basekerns
--~ return sharedkerns, basekerns
--~ end
--~ end
--~ return nil, nil
--~ end
--~ function tfm.prepare_base_kerns(tfmdata)
--~ if tfm.sharebasekerns and not tfmdata.sharedkerns then
--~ local sharedkerns = { }
--~ tfmdata.sharedkerns = sharedkerns
--~ for u, chr in next, tfmdata.characters do
--~ local kerns = chr.kerns
--~ if kerns then
--~ local hash = concat(sortedkeys(kerns), " ")
--~ local base = sharedkerns[hash]
--~ if not base then
--~ sharedkerns[hash] = kerns
--~ else
--~ chr.kerns = base
--~ end
--~ end
--~ end
--~ end
--~ end
-- we can have cache scaled characters when we are in node mode and don't have
-- protruding and expansion: hash == fullname @ size @ protruding @ expansion
-- but in practice (except from mk) the otf hash will be enough already so it
-- makes no sense to mess up the code now
local charactercache = { }
-- The scaler is only used for otf and afm and virtual fonts. If
-- a virtual font has italic correction make sure to set the
-- has_italic flag. Some more flags will be added in the future.
--[[ldx--
The reason why the scaler was originally split, is that for a while we experimented
with a helper function. However, in practice the calls are too slow to
make this profitable and the based variant was just faster. A days
wasted day but an experience richer.
--ldx]]--
tfm.autocleanup = true
local lastfont = nil
-- we can get rid of the tfm instance when we have fast access to the
-- scaled character dimensions at the tex end, e.g. a fontobject.width
--
-- flushing the kern and ligature tables from memory saves a lot (only
-- base mode) but it complicates vf building where the new characters
-- demand this data .. solution: functions that access them
-- we don't need the glyph data as we can use the description .. but we will
-- have to wait till we can access the internal tfm table efficiently in which
-- case characters will become a metatable afterwards
function tfm.cleanuptable(tfmdata) -- we need a cleanup callback, now we miss the last one
if tfm.autocleanup then -- ok, we can hook this into everyshipout or so ... todo
if tfmdata.type == 'virtual' or tfmdata.virtualized then
for k, v in next, tfmdata.characters do
if v.commands then v.commands = nil end
-- if v.kerns then v.kerns = nil end
end
else
-- for k, v in next, tfmdata.characters do
-- if v.kerns then v.kerns = nil end
-- end
end
end
end
function tfm.cleanup(tfmdata) -- we need a cleanup callback, now we miss the last one
end
function tfm.calculatescale(tfmtable, scaledpoints)
if scaledpoints < 0 then
scaledpoints = (- scaledpoints/1000) * tfmtable.designsize -- already in sp
end
local units = tfmtable.units or 1000
local delta = scaledpoints/units -- brr, some open type fonts have 2048
return scaledpoints, delta, units
end
function tfm.scale(tfmtable, scaledpoints, relativeid)
-- tfm.prepare_base_kerns(tfmtable) -- optimalization
local t = { } -- the new table
local scaledpoints, delta, units = tfm.calculatescale(tfmtable, scaledpoints, relativeid)
t.units_per_em = units or 1000
local hdelta, vdelta = delta, delta
-- unicoded unique descriptions shared cidinfo characters changed parameters indices
for k,v in next, tfmtable do
if type(v) == "table" then
-- print(k)
else
t[k] = v
end
end
local extend_factor = tfmtable.extend_factor or 0
if extend_factor ~= 0 and extend_factor ~= 1 then
hdelta = hdelta * extend_factor
t.extend = extend_factor * 1000
else
t.extend = 1000
end
local slant_factor = tfmtable.slant_factor or 0
if slant_factor ~= 0 then
t.slant = slant_factor * 1000
else
t.slant = 0
end
-- status
local isvirtual = tfmtable.type == "virtual" or tfmtable.virtualized
local hasmath = (tfmtable.math_parameters ~= nil and next(tfmtable.math_parameters) ~= nil) or (tfmtable.MathConstants ~= nil and next(tfmtable.MathConstants) ~= nil)
local nodemode = tfmtable.mode == "node"
local hasquality = tfmtable.auto_expand or tfmtable.auto_protrude
local hasitalic = tfmtable.has_italic
local descriptions = tfmtable.descriptions or { }
--
if hasmath then
t.has_math = true -- this will move to elsewhere
end
--
t.parameters = { }
t.characters = { }
t.MathConstants = { }
-- fast access
t.unscaled = tfmtable -- the original unscaled one (temp)
t.unicodes = tfmtable.unicodes
t.indices = tfmtable.indices
t.marks = tfmtable.marks
t.goodies = tfmtable.goodies
t.colorscheme = tfmtable.colorscheme
-- t.embedding = tfmtable.embedding
t.descriptions = descriptions
if tfmtable.fonts then
t.fonts = table.fastcopy(tfmtable.fonts) -- hm also at the end
end
local tp = t.parameters
local mp = t.math_parameters
local tfmp = tfmtable.parameters -- let's check for indexes
--
tp.slant = (tfmp.slant or tfmp[1] or 0)
tp.space = (tfmp.space or tfmp[2] or 0)*hdelta
tp.space_stretch = (tfmp.space_stretch or tfmp[3] or 0)*hdelta
tp.space_shrink = (tfmp.space_shrink or tfmp[4] or 0)*hdelta
tp.x_height = (tfmp.x_height or tfmp[5] or 0)*vdelta
tp.quad = (tfmp.quad or tfmp[6] or 0)*hdelta
tp.extra_space = (tfmp.extra_space or tfmp[7] or 0)*hdelta
local protrusionfactor = (tp.quad ~= 0 and 1000/tp.quad) or 0
local tc = t.characters
local characters = tfmtable.characters
local nameneeded = not tfmtable.shared.otfdata --hack
local changed = tfmtable.changed or { } -- for base mode
local ischanged = changed and next(changed)
local indices = tfmtable.indices
local luatex = tfmtable.luatex
local tounicode = luatex and luatex.tounicode
local defaultwidth = luatex and luatex.defaultwidth or 0
local defaultheight = luatex and luatex.defaultheight or 0
local defaultdepth = luatex and luatex.defaultdepth or 0
-- experimental, sharing kerns (unscaled and scaled) saves memory
-- local sharedkerns, basekerns = tfm.check_base_kerns(tfmtable)
-- loop over descriptions (afm and otf have descriptions, tfm not)
-- there is no need (yet) to assign a value to chr.tonunicode
local scaledwidth = defaultwidth * hdelta
local scaledheight = defaultheight * vdelta
local scaleddepth = defaultdepth * vdelta
local stackmath = tfmtable.ignore_stack_math ~= true
local private = fonts.privateoffset
local sharedkerns = { }
for k,v in next, characters do
local chr, description, index
if ischanged then
-- basemode hack
local c = changed[k]
if c then
description = descriptions[c] or v
v = characters[c] or v
index = (indices and indices[c]) or c
else
description = descriptions[k] or v
index = (indices and indices[k]) or k
end
else
description = descriptions[k] or v
index = (indices and indices[k]) or k
end
local width = description.width
local height = description.height
local depth = description.depth
if width then width = hdelta*width else width = scaledwidth end
if height then height = vdelta*height else height = scaledheight end
-- if depth then depth = vdelta*depth else depth = scaleddepth end
if depth and depth ~= 0 then
depth = delta*depth
if nameneeded then
chr = {
name = description.name,
index = index,
height = height,
depth = depth,
width = width,
}
else
chr = {
index = index,
height = height,
depth = depth,
width = width,
}
end
else
-- this saves a little bit of memory time and memory, esp for big cjk fonts
if nameneeded then
chr = {
name = description.name,
index = index,
height = height,
width = width,
}
else
chr = {
index = index,
height = height,
width = width,
}
end
end
-- if trace_scaling then
-- report_define("t=%s, u=%s, i=%s, n=%s c=%s",k,chr.tounicode or k,description.index,description.name or '-',description.class or '-')
-- end
if tounicode then
local tu = tounicode[index] -- nb: index!
if tu then
chr.tounicode = tu
end
end
if hasquality then
-- we could move these calculations elsewhere (saves calculations)
local ve = v.expansion_factor
if ve then
chr.expansion_factor = ve*1000 -- expansionfactor, hm, can happen elsewhere
end
local vl = v.left_protruding
if vl then
chr.left_protruding = protrusionfactor*width*vl
end
local vr = v.right_protruding
if vr then
chr.right_protruding = protrusionfactor*width*vr
end
end
-- todo: hasitalic
if hasitalic then
local vi = description.italic or v.italic
if vi and vi ~= 0 then
chr.italic = vi*hdelta
end
end
-- to be tested
if hasmath then
-- todo, just operate on descriptions.math
local vn = v.next
if vn then
chr.next = vn
--~ if v.vert_variants or v.horiz_variants then
--~ report_define("glyph 0x%05X has combination of next, vert_variants and horiz_variants",index)
--~ end
else
local vv = v.vert_variants
if vv then
local t = { }
for i=1,#vv do
local vvi = vv[i]
t[i] = {
["start"] = (vvi["start"] or 0)*vdelta,
["end"] = (vvi["end"] or 0)*vdelta,
["advance"] = (vvi["advance"] or 0)*vdelta,
["extender"] = vvi["extender"],
["glyph"] = vvi["glyph"],
}
end
chr.vert_variants = t
--~ local ic = v.vert_italic_correction
--~ if ic then
--~ chr.italic = ic * hdelta
--~ print(format("0x%05X -> %s",k,chr.italic))
--~ end
else
local hv = v.horiz_variants
if hv then
local t = { }
for i=1,#hv do
local hvi = hv[i]
t[i] = {
["start"] = (hvi["start"] or 0)*hdelta,
["end"] = (hvi["end"] or 0)*hdelta,
["advance"] = (hvi["advance"] or 0)*hdelta,
["extender"] = hvi["extender"],
["glyph"] = hvi["glyph"],
}
end
chr.horiz_variants = t
end
end
end
local vt = description.top_accent
if vt then
chr.top_accent = vdelta*vt
end
if stackmath then
local mk = v.mathkerns
if mk then
local kerns = { }
local v = mk.top_right if v then local k = { } for i=1,#v do local vi = v[i]
k[i] = { height = vdelta*vi.height, kern = vdelta*vi.kern }
end kerns.top_right = k end
local v = mk.top_left if v then local k = { } for i=1,#v do local vi = v[i]
k[i] = { height = vdelta*vi.height, kern = vdelta*vi.kern }
end kerns.top_left = k end
local v = mk.bottom_left if v then local k = { } for i=1,#v do local vi = v[i]
k[i] = { height = vdelta*vi.height, kern = vdelta*vi.kern }
end kerns.bottom_left = k end
local v = mk.bottom_right if v then local k = { } for i=1,#v do local vi = v[i]
k[i] = { height = vdelta*vi.height, kern = vdelta*vi.kern }
end kerns.bottom_right = k end
chr.mathkern = kerns -- singular
end
end
end
if not nodemode then
local vk = v.kerns
if vk then
--~ if sharedkerns then
--~ local base = basekerns[vk] -- hashed by table id, not content
--~ if not base then
--~ base = {}
--~ for k,v in next, vk do base[k] = v*hdelta end
--~ basekerns[vk] = base
--~ end
--~ chr.kerns = base
--~ else
--~ local tt = {}
--~ for k,v in next, vk do tt[k] = v*hdelta end
--~ chr.kerns = tt
--~ end
local s = sharedkerns[vk]
if not s then
s = { }
for k,v in next, vk do s[k] = v*hdelta end
sharedkerns[vk] = s
end
chr.kerns = s
end
local vl = v.ligatures
if vl then
if true then
chr.ligatures = vl -- shared
else
local tt = { }
for i,l in next, vl do
tt[i] = l
end
chr.ligatures = tt
end
end
end
if isvirtual then
local vc = v.commands
if vc then
-- we assume non scaled commands here
-- tricky .. we need to scale pseudo math glyphs too
-- which is why we deal with rules too
local ok = false
for i=1,#vc do
local key = vc[i][1]
if key == "right" or key == "down" then
ok = true
break
end
end
if ok then
local tt = { }
for i=1,#vc do
local ivc = vc[i]
local key = ivc[1]
if key == "right" then
tt[i] = { key, ivc[2]*hdelta }
elseif key == "down" then
tt[i] = { key, ivc[2]*vdelta }
elseif key == "rule" then
tt[i] = { key, ivc[2]*vdelta, ivc[3]*hdelta }
else -- not comment
tt[i] = ivc -- shared since in cache and untouched
end
end
chr.commands = tt
else
chr.commands = vc
end
end
end
tc[k] = chr
end
-- t.encodingbytes, t.filename, t.fullname, t.name: elsewhere
t.size = scaledpoints
t.factor = delta
t.hfactor = hdelta
t.vfactor = vdelta
if t.fonts then
t.fonts = table.fastcopy(t.fonts) -- maybe we virtualize more afterwards
end
if hasmath then
-- mathematics.extras.copy(t) -- can be done elsewhere if needed
local ma = tfm.mathactions
for i=1,#ma do
ma[i](t,tfmtable,delta,hdelta,vdelta) -- what delta?
end
end
-- needed for \high cum suis
local tpx = tp.x_height
if hasmath then
if not tp[13] then tp[13] = .86*tpx end -- mathsupdisplay
if not tp[14] then tp[14] = .86*tpx end -- mathsupnormal
if not tp[15] then tp[15] = .86*tpx end -- mathsupcramped
if not tp[16] then tp[16] = .48*tpx end -- mathsubnormal
if not tp[17] then tp[17] = .48*tpx end -- mathsubcombined
if not tp[22] then tp[22] = 0 end -- mathaxisheight
if t.MathConstants then t.MathConstants.AccentBaseHeight = nil end -- safeguard
end
t.tounicode = 1
t.cidinfo = tfmtable.cidinfo
-- we have t.name=metricfile and t.fullname=RealName and t.filename=diskfilename
-- when collapsing fonts, luatex looks as both t.name and t.fullname as ttc files
-- can have multiple subfonts
if hasmath then
if trace_defining then
report_define("math enabled for: name '%s', fullname: '%s', filename: '%s'",t.name or "noname",t.fullname or "nofullname",t.filename or "nofilename")
end
else
if trace_defining then
report_define("math disabled for: name '%s', fullname: '%s', filename: '%s'",t.name or "noname",t.fullname or "nofullname",t.filename or "nofilename")
end
t.nomath, t.MathConstants = true, nil
end
if not t.psname then
-- name used in pdf file as well as for selecting subfont in ttc/dfont
t.psname = t.fontname or (t.fullname and fonts.names.cleanname(t.fullname))
end
if trace_defining then
report_define("used for accessing (sub)font: '%s'",t.psname or "nopsname")
report_define("used for subsetting: '%s'",t.fontname or "nofontname")
end
-- this will move up (side effect of merging split call)
t.factor = delta
t.ascender = delta*(tfmtable.ascender or 0)
t.descender = delta*(tfmtable.descender or 0)
t.shared = tfmtable.shared or { }
t.unique = table.fastcopy(tfmtable.unique or {})
tfm.cleanup(t)
-- print(t.fontname,table.serialize(t.MathConstants))
return t
end
--[[ldx--
Analyzers run per script and/or language and are needed in order to
process features right.
--ldx]]--
fonts.analyzers = fonts.analyzers or { }
local analyzers = fonts.analyzers
analyzers.aux = analyzers.aux or { }
analyzers.methods = analyzers.methods or { }
analyzers.initializers = analyzers.initializers or { }
-- todo: analyzers per script/lang, cross font, so we need an font id hash -> script
-- e.g. latin -> hyphenate, arab -> 1/2/3 analyze
-- an example analyzer (should move to font-ota.lua)
local state = attributes.private('state')
function analyzers.aux.setstate(head,font)
local useunicodemarks = analyzers.useunicodemarks
local tfmdata = fontdata[font]
local characters = tfmdata.characters
local descriptions = tfmdata.descriptions
local first, last, current, n, done = nil, nil, head, 0, false -- maybe make n boolean
while current do
local id = current.id
if id == glyph_code and current.font == font then
local char = current.char
local d = descriptions[char]
if d then
if d.class == "mark" or (useunicodemarks and categories[char] == "mn") then
done = true
set_attribute(current,state,5) -- mark
elseif n == 0 then
first, last, n = current, current, 1
set_attribute(current,state,1) -- init
else
last, n = current, n+1
set_attribute(current,state,2) -- medi
end
else -- finish
if first and first == last then
set_attribute(last,state,4) -- isol
elseif last then
set_attribute(last,state,3) -- fina
end
first, last, n = nil, nil, 0
end
elseif id == disc_code then
-- always in the middle
set_attribute(current,state,2) -- midi
last = current
else -- finish
if first and first == last then
set_attribute(last,state,4) -- isol
elseif last then
set_attribute(last,state,3) -- fina
end
first, last, n = nil, nil, 0
end
current = current.next
end
if first and first == last then
set_attribute(last,state,4) -- isol
elseif last then
set_attribute(last,state,3) -- fina
end
return head, done
end
function tfm.replacements(tfm,value)
-- tfm.characters[0x0022] = table.fastcopy(tfm.characters[0x201D])
-- tfm.characters[0x0027] = table.fastcopy(tfm.characters[0x2019])
-- tfm.characters[0x0060] = table.fastcopy(tfm.characters[0x2018])
-- tfm.characters[0x0022] = tfm.characters[0x201D]
tfm.characters[0x0027] = tfm.characters[0x2019]
-- tfm.characters[0x0060] = tfm.characters[0x2018]
end
-- checking
function tfm.checkedfilename(metadata,whatever)
local foundfilename = metadata.foundfilename
if not foundfilename then
local askedfilename = metadata.filename or ""
if askedfilename ~= "" then
foundfilename = resolvers.findbinfile(askedfilename,"") or ""
if foundfilename == "" then
report_define("source file '%s' is not found",askedfilename)
foundfilename = resolvers.findbinfile(file.basename(askedfilename),"") or ""
if foundfilename ~= "" then
report_define("using source file '%s' (cache mismatch)",foundfilename)
end
end
elseif whatever then
report_define("no source file for '%s'",whatever)
foundfilename = ""
end
metadata.foundfilename = foundfilename
-- report_define("using source file '%s'",foundfilename)
end
return foundfilename
end
-- status info
statistics.register("fonts load time", function()
return statistics.elapsedseconds(fonts)
end)