AIM-PIbd-32-Kurbanova-A-A/aimenv/Lib/site-packages/fontTools/afmLib.py

440 lines
13 KiB
Python
Raw Permalink Normal View History

2024-10-02 22:15:59 +04:00
"""Module for reading and writing AFM (Adobe Font Metrics) files.
Note that this has been designed to read in AFM files generated by Fontographer
and has not been tested on many other files. In particular, it does not
implement the whole Adobe AFM specification [#f1]_ but, it should read most
"common" AFM files.
Here is an example of using `afmLib` to read, modify and write an AFM file:
>>> from fontTools.afmLib import AFM
>>> f = AFM("Tests/afmLib/data/TestAFM.afm")
>>>
>>> # Accessing a pair gets you the kern value
>>> f[("V","A")]
-60
>>>
>>> # Accessing a glyph name gets you metrics
>>> f["A"]
(65, 668, (8, -25, 660, 666))
>>> # (charnum, width, bounding box)
>>>
>>> # Accessing an attribute gets you metadata
>>> f.FontName
'TestFont-Regular'
>>> f.FamilyName
'TestFont'
>>> f.Weight
'Regular'
>>> f.XHeight
500
>>> f.Ascender
750
>>>
>>> # Attributes and items can also be set
>>> f[("A","V")] = -150 # Tighten kerning
>>> f.FontName = "TestFont Squished"
>>>
>>> # And the font written out again (remove the # in front)
>>> #f.write("testfont-squished.afm")
.. rubric:: Footnotes
.. [#f1] `Adobe Technote 5004 <https://www.adobe.com/content/dam/acom/en/devnet/font/pdfs/5004.AFM_Spec.pdf>`_,
Adobe Font Metrics File Format Specification.
"""
import re
# every single line starts with a "word"
identifierRE = re.compile(r"^([A-Za-z]+).*")
# regular expression to parse char lines
charRE = re.compile(
r"(-?\d+)" # charnum
r"\s*;\s*WX\s+" # ; WX
r"(-?\d+)" # width
r"\s*;\s*N\s+" # ; N
r"([.A-Za-z0-9_]+)" # charname
r"\s*;\s*B\s+" # ; B
r"(-?\d+)" # left
r"\s+"
r"(-?\d+)" # bottom
r"\s+"
r"(-?\d+)" # right
r"\s+"
r"(-?\d+)" # top
r"\s*;\s*" # ;
)
# regular expression to parse kerning lines
kernRE = re.compile(
r"([.A-Za-z0-9_]+)" # leftchar
r"\s+"
r"([.A-Za-z0-9_]+)" # rightchar
r"\s+"
r"(-?\d+)" # value
r"\s*"
)
# regular expressions to parse composite info lines of the form:
# Aacute 2 ; PCC A 0 0 ; PCC acute 182 211 ;
compositeRE = re.compile(
r"([.A-Za-z0-9_]+)" # char name
r"\s+"
r"(\d+)" # number of parts
r"\s*;\s*"
)
componentRE = re.compile(
r"PCC\s+" # PPC
r"([.A-Za-z0-9_]+)" # base char name
r"\s+"
r"(-?\d+)" # x offset
r"\s+"
r"(-?\d+)" # y offset
r"\s*;\s*"
)
preferredAttributeOrder = [
"FontName",
"FullName",
"FamilyName",
"Weight",
"ItalicAngle",
"IsFixedPitch",
"FontBBox",
"UnderlinePosition",
"UnderlineThickness",
"Version",
"Notice",
"EncodingScheme",
"CapHeight",
"XHeight",
"Ascender",
"Descender",
]
class error(Exception):
pass
class AFM(object):
_attrs = None
_keywords = [
"StartFontMetrics",
"EndFontMetrics",
"StartCharMetrics",
"EndCharMetrics",
"StartKernData",
"StartKernPairs",
"EndKernPairs",
"EndKernData",
"StartComposites",
"EndComposites",
]
def __init__(self, path=None):
"""AFM file reader.
Instantiating an object with a path name will cause the file to be opened,
read, and parsed. Alternatively the path can be left unspecified, and a
file can be parsed later with the :meth:`read` method."""
self._attrs = {}
self._chars = {}
self._kerning = {}
self._index = {}
self._comments = []
self._composites = {}
if path is not None:
self.read(path)
def read(self, path):
"""Opens, reads and parses a file."""
lines = readlines(path)
for line in lines:
if not line.strip():
continue
m = identifierRE.match(line)
if m is None:
raise error("syntax error in AFM file: " + repr(line))
pos = m.regs[1][1]
word = line[:pos]
rest = line[pos:].strip()
if word in self._keywords:
continue
if word == "C":
self.parsechar(rest)
elif word == "KPX":
self.parsekernpair(rest)
elif word == "CC":
self.parsecomposite(rest)
else:
self.parseattr(word, rest)
def parsechar(self, rest):
m = charRE.match(rest)
if m is None:
raise error("syntax error in AFM file: " + repr(rest))
things = []
for fr, to in m.regs[1:]:
things.append(rest[fr:to])
charname = things[2]
del things[2]
charnum, width, l, b, r, t = (int(thing) for thing in things)
self._chars[charname] = charnum, width, (l, b, r, t)
def parsekernpair(self, rest):
m = kernRE.match(rest)
if m is None:
raise error("syntax error in AFM file: " + repr(rest))
things = []
for fr, to in m.regs[1:]:
things.append(rest[fr:to])
leftchar, rightchar, value = things
value = int(value)
self._kerning[(leftchar, rightchar)] = value
def parseattr(self, word, rest):
if word == "FontBBox":
l, b, r, t = [int(thing) for thing in rest.split()]
self._attrs[word] = l, b, r, t
elif word == "Comment":
self._comments.append(rest)
else:
try:
value = int(rest)
except (ValueError, OverflowError):
self._attrs[word] = rest
else:
self._attrs[word] = value
def parsecomposite(self, rest):
m = compositeRE.match(rest)
if m is None:
raise error("syntax error in AFM file: " + repr(rest))
charname = m.group(1)
ncomponents = int(m.group(2))
rest = rest[m.regs[0][1] :]
components = []
while True:
m = componentRE.match(rest)
if m is None:
raise error("syntax error in AFM file: " + repr(rest))
basechar = m.group(1)
xoffset = int(m.group(2))
yoffset = int(m.group(3))
components.append((basechar, xoffset, yoffset))
rest = rest[m.regs[0][1] :]
if not rest:
break
assert len(components) == ncomponents
self._composites[charname] = components
def write(self, path, sep="\r"):
"""Writes out an AFM font to the given path."""
import time
lines = [
"StartFontMetrics 2.0",
"Comment Generated by afmLib; at %s"
% (time.strftime("%m/%d/%Y %H:%M:%S", time.localtime(time.time()))),
]
# write comments, assuming (possibly wrongly!) they should
# all appear at the top
for comment in self._comments:
lines.append("Comment " + comment)
# write attributes, first the ones we know about, in
# a preferred order
attrs = self._attrs
for attr in preferredAttributeOrder:
if attr in attrs:
value = attrs[attr]
if attr == "FontBBox":
value = "%s %s %s %s" % value
lines.append(attr + " " + str(value))
# then write the attributes we don't know about,
# in alphabetical order
items = sorted(attrs.items())
for attr, value in items:
if attr in preferredAttributeOrder:
continue
lines.append(attr + " " + str(value))
# write char metrics
lines.append("StartCharMetrics " + repr(len(self._chars)))
items = [
(charnum, (charname, width, box))
for charname, (charnum, width, box) in self._chars.items()
]
def myKey(a):
"""Custom key function to make sure unencoded chars (-1)
end up at the end of the list after sorting."""
if a[0] == -1:
a = (0xFFFF,) + a[1:] # 0xffff is an arbitrary large number
return a
items.sort(key=myKey)
for charnum, (charname, width, (l, b, r, t)) in items:
lines.append(
"C %d ; WX %d ; N %s ; B %d %d %d %d ;"
% (charnum, width, charname, l, b, r, t)
)
lines.append("EndCharMetrics")
# write kerning info
lines.append("StartKernData")
lines.append("StartKernPairs " + repr(len(self._kerning)))
items = sorted(self._kerning.items())
for (leftchar, rightchar), value in items:
lines.append("KPX %s %s %d" % (leftchar, rightchar, value))
lines.append("EndKernPairs")
lines.append("EndKernData")
if self._composites:
composites = sorted(self._composites.items())
lines.append("StartComposites %s" % len(self._composites))
for charname, components in composites:
line = "CC %s %s ;" % (charname, len(components))
for basechar, xoffset, yoffset in components:
line = line + " PCC %s %s %s ;" % (basechar, xoffset, yoffset)
lines.append(line)
lines.append("EndComposites")
lines.append("EndFontMetrics")
writelines(path, lines, sep)
def has_kernpair(self, pair):
"""Returns `True` if the given glyph pair (specified as a tuple) exists
in the kerning dictionary."""
return pair in self._kerning
def kernpairs(self):
"""Returns a list of all kern pairs in the kerning dictionary."""
return list(self._kerning.keys())
def has_char(self, char):
"""Returns `True` if the given glyph exists in the font."""
return char in self._chars
def chars(self):
"""Returns a list of all glyph names in the font."""
return list(self._chars.keys())
def comments(self):
"""Returns all comments from the file."""
return self._comments
def addComment(self, comment):
"""Adds a new comment to the file."""
self._comments.append(comment)
def addComposite(self, glyphName, components):
"""Specifies that the glyph `glyphName` is made up of the given components.
The components list should be of the following form::
[
(glyphname, xOffset, yOffset),
...
]
"""
self._composites[glyphName] = components
def __getattr__(self, attr):
if attr in self._attrs:
return self._attrs[attr]
else:
raise AttributeError(attr)
def __setattr__(self, attr, value):
# all attrs *not* starting with "_" are consider to be AFM keywords
if attr[:1] == "_":
self.__dict__[attr] = value
else:
self._attrs[attr] = value
def __delattr__(self, attr):
# all attrs *not* starting with "_" are consider to be AFM keywords
if attr[:1] == "_":
try:
del self.__dict__[attr]
except KeyError:
raise AttributeError(attr)
else:
try:
del self._attrs[attr]
except KeyError:
raise AttributeError(attr)
def __getitem__(self, key):
if isinstance(key, tuple):
# key is a tuple, return the kernpair
return self._kerning[key]
else:
# return the metrics instead
return self._chars[key]
def __setitem__(self, key, value):
if isinstance(key, tuple):
# key is a tuple, set kernpair
self._kerning[key] = value
else:
# set char metrics
self._chars[key] = value
def __delitem__(self, key):
if isinstance(key, tuple):
# key is a tuple, del kernpair
del self._kerning[key]
else:
# del char metrics
del self._chars[key]
def __repr__(self):
if hasattr(self, "FullName"):
return "<AFM object for %s>" % self.FullName
else:
return "<AFM object at %x>" % id(self)
def readlines(path):
with open(path, "r", encoding="ascii") as f:
data = f.read()
return data.splitlines()
def writelines(path, lines, sep="\r"):
with open(path, "w", encoding="ascii", newline=sep) as f:
f.write("\n".join(lines) + "\n")
if __name__ == "__main__":
import EasyDialogs
path = EasyDialogs.AskFileForOpen()
if path:
afm = AFM(path)
char = "A"
if afm.has_char(char):
print(afm[char]) # print charnum, width and boundingbox
pair = ("A", "V")
if afm.has_kernpair(pair):
print(afm[pair]) # print kerning value for pair
print(afm.Version) # various other afm entries have become attributes
print(afm.Weight)
# afm.comments() returns a list of all Comment lines found in the AFM
print(afm.comments())
# print afm.chars()
# print afm.kernpairs()
print(afm)
afm.write(path + ".muck")