texlive-base/texlive-20200327-dviasm-py3.patch
Tom spot Callaway 876c81a4c5 missing patch
2020-12-30 13:56:45 -05:00

319 lines
13 KiB
Diff

diff -up texlive-base-20200327/source/texk/texlive/linked_scripts/dviasm/dviasm.py.py3fix texlive-base-20200327/source/texk/texlive/linked_scripts/dviasm/dviasm.py
--- texlive-base-20200327/source/texk/texlive/linked_scripts/dviasm/dviasm.py.py3fix 2019-11-27 17:03:14.000000000 -0500
+++ texlive-base-20200327/source/texk/texlive/linked_scripts/dviasm/dviasm.py 2020-09-18 17:30:44.000000000 -0400
@@ -5,8 +5,8 @@
#
# Copyright (C) 2007-2008 by Jin-Hwan Cho <chofchof@ktug.or.kr>
# Copyright (C) 2011-2017 by Khaled Hosny <khaledhosny@eglug.org>
-# Copyright (C) 2019 by Arthur Reutenauer <arthur@reutenauer.eu>
-# Copyright (C) 2019 by Hironobu Yamashita <h.y.acetaminophen@gmail.com>
+# Copyright (C) 2019 by Arthur Reutenauer <arthur@reutenauer.eu>
+# Copyright (C) 2019-2020 by Hironobu Yamashita <h.y.acetaminophen@gmail.com>
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
@@ -180,22 +180,26 @@ def GetInt(s):
except: return -1
def GetStrASCII(s): # used in Parse()
- if len(s) > 1 and ((s[0] == "'" and s[-1] == "'") or (s[0] == '"' and s[-1] == '"')): return [ord(c) for c in s[1:-1].decode('unicode_escape')]
+ if len(s) > 1 and ((s[0] == "'" and s[-1] == "'") or (s[0] == '"' and s[-1] == '"')):
+ return [ord(c) for c in s[1:-1].decode('unicode_escape')]
else: return ''
def UCS2toJIS(c):
- s = c.encode('iso2022-jp')
+ try:
+ s = c.encode('iso2022-jp')
+ except UnicodeEncodeError:
+ s = c.encode('raw_unicode_escape')
if len(s) == 1: return ord(s)
else: return (s[3] << 8) + s[4]
def GetStrUTF8(s): # used in Parse()
if len(s) > 1 and ((s[0] == "'" and s[-1] == "'") or (s[0] == '"' and s[-1] == '"')):
- t = s[1:-1]
+ t = s[1:-1].encode('raw_unicode_escape').decode('unicode_escape')
if is_ptex: return [UCS2toJIS(c) for c in t]
else: return [ord(c) for c in t]
else: return ''
-def PutStrASCII(t): # unsed in Dump()
+def PutStrASCII(t): # used in Dump()
s = ''
for o in t:
if o == 92: s += '\\\\'
@@ -206,7 +210,7 @@ def PutStrASCII(t): # unsed in Dump()
warning('Not support characters > 65535; may skip %d.\n' % o)
return "'%s'" % s
-def PutStrLatin1(t): # unsed in Dump()
+def PutStrLatin1(t): # used in Dump()
s = ''
for o in t:
if o == 92: s += '\\\\'
@@ -221,14 +225,14 @@ def DecodeISO2022JP(c):
try:
s = bytes.fromhex("1b 24 42 %02x %02x" % (c//256, c%256)).decode('iso2022-jp')
except UnicodeDecodeError:
- s = ''
+ s = chr(c)
return s
-def PutStrUTF8(t): # unsed in Dump()
+def PutStrUTF8(t): # used in Dump()
s = ''
if is_subfont:
for o in t:
- s += chr((subfont_idx << 8) + o).encode('utf8')
+ s += chr((subfont_idx << 8) + o)
else: # not the case of subfont
for o in t:
if o == 92: s += '\\\\'
@@ -239,16 +243,6 @@ def PutStrUTF8(t): # unsed in Dump()
else: s += chr(o)
return "'%s'" % s
-def PutStrSJIS(t): # unsed in Dump()
- s = ''
- for o in t:
- if o == 92: s += '\\\\'
- elif 32 <= o < 127: s += chr(o)
- elif o < 128: s += ('\\x%02x' % o)
- else:
- s += DecodeISO2022JP(o).encode('sjis')
- return "'%s'" % s
-
def IsFontChanged(f, z):
global cur_font, cur_ssize, subfont_idx, is_subfont
for n in subfont_list:
@@ -461,7 +455,7 @@ class DVI(object):
if o == SET_RULE:
s.append([SET_RULE, [p, SignedQuad(fp)]])
elif o in (PUT1, PUT2, PUT3, PUT4):
- s.append([PUT1, p])
+ s.append([PUT1, [p]])
elif o == PUT_RULE:
s.append([PUT_RULE, [p, SignedQuad(fp)]])
elif o == NOP:
@@ -498,7 +492,7 @@ class DVI(object):
elif o < FNT_NUM_0 + 64 or o in (FNT1, FNT2, FNT3, FNT4):
s.append([FNT1, p])
elif o in (XXX1, XXX2, XXX3, XXX4):
- q = fp.read(p).decode('utf8')
+ q = fp.read(p)
s.append([XXX1, q])
elif o in (FNT_DEF1, FNT_DEF2, FNT_DEF3, FNT_DEF4):
self.DefineFont(p, fp)
@@ -622,11 +616,11 @@ class DVI(object):
if cmd[0] == SET1:
for o in cmd[1]:
if o < 128: s.append(bytes.fromhex('%02x' % (SET_CHAR_0 + o)))
- else: s.append(self.CmdPair([SET1, o]))
+ else: s.append(self.CmdPairU([SET1, o]))
elif cmd[0] in (SET_RULE, PUT_RULE):
s.append(bytes.fromhex('%02x' % cmd[0]) + PutSignedQuad(cmd[1][0]) + PutSignedQuad(cmd[1][1]))
elif cmd[0] == PUT1:
- s.append(self.CmdPair([PUT1, cmd[1][0]]))
+ s.append(self.CmdPairU([PUT1, cmd[1][0]]))
elif cmd[0] in (RIGHT1, DOWN1):
s.append(self.CmdPair(cmd))
elif cmd[0] in (W0, X0, Y0, Z0):
@@ -648,12 +642,22 @@ class DVI(object):
z = cmd[1]; s.append(self.CmdPair(cmd))
elif cmd[0] == FNT1:
if cmd[1] < 64: s.append(bytes.fromhex('%02x' % (FNT_NUM_0 + cmd[1])))
- else: s.append(self.CmdPair(cmd))
+ else: s.append(self.CmdPairU(cmd))
elif cmd[0] == XXX1:
- cmd1 = cmd[1].encode('utf8')
- l = len(cmd[1])
- if l < 256: s.append(bytes.fromhex('%02x' % XXX1) + bytes.fromhex('%02x' % l) + cmd1)
- else: s.append(bytes.fromhex('%02x' % XXX4) + PutSignedQuad(l) + cmd1)
+ if options.xxx_encoding == "none":
+ l = len(cmd[1]) # leave encoding untouched
+ else:
+ cmd1 = cmd[1].encode(options.xxx_encoding)
+ l = len(cmd1)
+ if l < 256:
+ s.append(bytes.fromhex('%02x' % XXX1) + bytes.fromhex('%02x' % l))
+ else:
+ s.append(bytes.fromhex('%02x' % XXX4) + PutSignedQuad(l))
+ if options.xxx_encoding == "none":
+ for o in cmd[1]:
+ s.append(bytes.fromhex('%02x' % ord(o)))
+ else:
+ s.append(cmd1)
elif cmd[0] == DIR:
s.append(bytes.fromhex('%02x' % DIR) + bytes.fromhex('%02x' % cmd[1]))
elif cmd[0] == BEGIN_REFLECT:
@@ -685,7 +689,8 @@ class DVI(object):
def WriteFontDefinitions(self, fp):
s = []
for e in sorted(self.font_def.keys()):
- if self.font_def[e]['native']:
+ try:
+ self.font_def[e]['native']
flags = self.font_def[e]['flags']
s.append(PutByte(NATIVE_FONT_DEF))
s.append(PutSignedQuad(e))
@@ -698,7 +703,7 @@ class DVI(object):
if flags & XDV_FLAG_EXTEND: s.append(PutSignedQuad(self.font_def[e]['extend']))
if flags & XDV_FLAG_SLANT: s.append(PutSignedQuad(self.font_def[e]['slant']))
if flags & XDV_FLAG_EMBOLDEN: s.append(PutSignedQuad(self.font_def[e]['embolden']))
- else:
+ except KeyError:
l, q = PutUnsigned(e)
s.append(PutByte(FNT_DEF1 + l))
s.append(q)
@@ -710,6 +715,10 @@ class DVI(object):
s.append(self.font_def[e]['name'].encode('utf8'))
fp.write(b''.join(s))
+ def CmdPairU(self, cmd):
+ l, q = PutUnsigned(cmd[1])
+ return bytes.fromhex('%02x' % (cmd[0] + l)) + q
+
def CmdPair(self, cmd):
l, q = PutSigned(cmd[1])
return bytes.fromhex('%02x' % (cmd[0] + l)) + q
@@ -718,7 +727,7 @@ class DVI(object):
# Parse: Text -> Internal Format
##########################################################
def Parse(self, fn, encoding=''):
- fp = open(fn, 'r')
+ fp = open(fn, 'r', encoding=encoding)
s = fp.read()
fp.close()
self.ParseFromString(s, encoding=encoding)
@@ -807,7 +816,10 @@ class DVI(object):
else:
self.cur_page.append([SET1, ol])
elif key == 'put':
- self.cur_page.append([PUT1, GetStr(val)])
+ ol = GetStr(val)
+ if len(ol) != 1:
+ warning('only one character is allowed for put!')
+ self.cur_page.append([PUT1, ol])
elif key == 'setrule':
v = val.split(' ')
if len(v) != 2:
@@ -895,7 +907,7 @@ class DVI(object):
# Dump: Internal Format -> Text
##########################################################
def Dump(self, fn, tabsize=2, encoding=''):
- fp = open(fn, 'w')
+ fp = open(fn, 'w', encoding=encoding)
self.DumpToFile(fp, tabsize=tabsize, encoding=encoding)
fp.close()
@@ -903,7 +915,6 @@ class DVI(object):
global PutStr
if encoding == 'ascii': PutStr = PutStrASCII
elif encoding == 'latin1': PutStr = PutStrLatin1
- elif encoding == 'sjis': PutStr = PutStrSJIS
else: PutStr = PutStrUTF8
# DumpPreamble
fp.write("[preamble]\n")
@@ -921,10 +932,10 @@ class DVI(object):
# DumpFontDefinitions
fp.write("\n[font definitions]\n")
for e in sorted(self.font_def.keys()):
- fp.write("fntdef: %s" % self.font_def[e]['name'])
+ fp.write("fntdef: %s " % self.font_def[e]['name'])
if self.font_def[e]['design_size'] != self.font_def[e]['scaled_size']:
- fp.write(" (%s) " % self.byconv(self.font_def[e]['design_size']))
- fp.write(" at %s\n" % self.byconv(self.font_def[e]['scaled_size']))
+ fp.write("(%s) " % self.byconv(self.font_def[e]['design_size']))
+ fp.write("at %s\n" % self.byconv(self.font_def[e]['scaled_size']))
# DumpPages
for page in self.pages:
fp.write("\n[page" + (" %d"*10 % tuple(page['count'])) + "]\n")
@@ -939,7 +950,10 @@ class DVI(object):
fp.write("push:\n")
indent += tabsize
elif cmd[0] == XXX1:
- fp.write("xxx: %s\n" % repr(cmd[1]))
+ if options.xxx_encoding == "none":
+ fp.write("xxx: %s\n" % PutStrASCII(cmd[1])) # leave encoding untouched
+ else:
+ fp.write("xxx: '%s'\n" % cmd[1].decode(options.xxx_encoding))
elif cmd[0] == DIR:
fp.write("dir: %d\n" % cmd[1])
elif cmd[0] == BEGIN_REFLECT:
@@ -1092,7 +1106,6 @@ class DVI(object):
f['slant'] = slant
f['embolden'] = embolden
else:
- f['native'] = False
f['name'] = n
if q[:2] == "at": q = q[2:]
@@ -1144,16 +1157,18 @@ def ProcessOptions():
DVIasm is a Python script to support changing or creating DVI files
via disassembling into text, editing, and then reassembling into
binary format. It is fully documented at
+ http://tug.org/TUGboat/Articles/tb28-2/tb89cho.pdf
+ http://ajt.ktug.kr/assets/2008/5/1/0201cho.pdf
-http://tug.org/TUGboat/Articles/tb28-2/tb89cho.pdf
-http://ajt.ktug.kr/assets/2008/5/1/0201cho.pdf"""
+Please report bugs to
+ https://github.com/aminophen/dviasm"""
+
+ version = """This is %prog-20200918
- version = """This is %prog-20191126
-
Copyright (C) 2007-2008 by Jin-Hwan Cho <chofchof@ktug.or.kr>
Copyright (C) 2011-2017 by Khaled Hosny <khaledhosny@eglug.org>
-Copyright (C) 2019 by Arthur Reutenauer <arthur@reutenauer.eu>
-Copyright (C) 2019 by Hironobu Yamashita <h.y.acetaminophen@gmail.com>
+Copyright (C) 2019 by Arthur Reutenauer <arthur@reutenauer.eu>
+Copyright (C) 2019-2020 by Hironobu Yamashita <h.y.acetaminophen@gmail.com>
This is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
@@ -1173,6 +1188,10 @@ the Free Software Foundation, either ver
action="store", type="string", dest="encoding",
metavar="STR",
help="encoding for input/output [default=%default]")
+ parser.add_option("-x", "--xxx-encoding",
+ action="store", type="string", dest="xxx_encoding",
+ metavar="STR",
+ help="encoding for interpreting xxx strings [default=%default]")
parser.add_option("-t", "--tabsize",
action="store", type="int", dest="tabsize",
metavar="INT",
@@ -1184,18 +1203,20 @@ the Free Software Foundation, either ver
action="append", type="string", dest="subfont",
metavar="STR",
help="the list of fonts with UCS2 subfont scheme (comma separated); disable internal subfont list if STR is empty")
- parser.set_defaults(unit='pt', encoding='utf8', tabsize=2)
+ parser.set_defaults(unit='pt', encoding='utf8', xxx_encoding='none', tabsize=2)
(options, args) = parser.parse_args()
if not options.unit in ['sp', 'pt', 'bp', 'mm', 'cm', 'in']:
parser.error("invalid unit name '%s'!" % options.unit)
- if options.tabsize < 0:
+ if options.tabsize < 0:
parser.error("negative tabsize!")
- if not options.encoding in ['ascii', 'latin1', 'utf8', 'sjis']:
+ if not options.xxx_encoding in ['none', 'utf8', 'sjis', 'eucjp']:
+ parser.error("invalid xxx-encoding '%s'!" % options.xxx_encoding)
+ if not options.encoding in ['ascii', 'latin1', 'utf8', 'sjis', 'eucjp']:
parser.error("invalid encoding '%s'!" % options.encoding)
if options.ptex:
global is_ptex
is_ptex = True
- if not options.encoding in ['utf8', 'sjis']:
+ if not options.encoding in ['utf8', 'sjis', 'eucjp']:
parser.error("invalid encoding '%s' for Japanese pTeX!" % options.encoding)
if options.subfont:
global subfont_list
@@ -1235,4 +1256,4 @@ if __name__ == '__main__':
else: # dump -> dvi
aDVI.Parse(args[0], encoding=options.encoding)
if options.output: aDVI.Save(options.output)
- else: aDVI.SaveToFile(sys.stdout)
+ else: aDVI.SaveToFile(sys.stdout.buffer)