Issue #15239: Make mkstringprep.py work again on Python 3.

This commit is contained in:
Serhiy Storchaka 2013-06-09 17:11:48 +03:00
commit 302b8c31ec
2 changed files with 25 additions and 17 deletions

View File

@ -1787,6 +1787,8 @@ Documentation
Tools/Demos
-----------
- Issue #15239: Make mkstringprep.py work again on Python 3.
- Issue #17028: Allowed Python arguments to be supplied to the Windows
launcher.

View File

@ -1,4 +1,5 @@
import re, unicodedata, sys
import re, sys
from unicodedata import ucd_3_2_0 as unicodedata
if sys.maxunicode == 65535:
raise RuntimeError("need UCS-4 Python")
@ -37,16 +38,20 @@ def compact_set(l):
tuple.append((prev,prev+span+1))
else:
single.append(prev)
tuple = " + ".join(["list(range(%d,%d))" % t for t in tuple])
if not single and len(tuple) == 1:
tuple = "range(%d,%d)" % tuple[0]
else:
tuple = " + ".join("list(range(%d,%d))" % t for t in tuple)
if not single:
return "set(%s)" % tuple
if not tuple:
return "set(%s)" % repr(single)
return "set(%s + %s)" % (repr(single),tuple)
return "set(%r)" % (single,)
return "set(%r + %s)" % (single, tuple)
############## Read the tables in the RFC #######################
data = open("rfc3454.txt").readlines()
with open("rfc3454.txt") as f:
data = f.readlines()
tables = []
curname = None
@ -55,8 +60,7 @@ for l in data:
if not l:
continue
# Skip RFC page breaks
if l.startswith("Hoffman & Blanchet") or\
l.startswith("RFC 3454"):
if l.startswith(("Hoffman & Blanchet", "RFC 3454")):
continue
# Find start/end lines
m = re.match("----- (Start|End) Table ([A-Z](.[0-9])+) -----", l)
@ -71,6 +75,8 @@ for l in data:
else:
if not curname:
raise RuntimeError("End without start", l)
if curname != m.group(2):
raise RuntimeError("Unexpected end", l)
curname = None
continue
if not curname:
@ -113,10 +119,10 @@ There are two kinds of tables: sets, for which a member test is provided,
and mappings, for which a mapping function is provided.
\"\"\"
import unicodedata
from unicodedata import ucd_3_2_0 as unicodedata
""")
print("assert unicodedata.unidata_version == %s" % repr(unicodedata.unidata_version))
print("assert unicodedata.unidata_version == %r" % (unicodedata.unidata_version,))
# A.1 is the table of unassigned characters
# XXX Plane 15 PUA is listed as unassigned in Python.
@ -173,15 +179,15 @@ assert name == "B.3"
b3_exceptions = {}
for k,v in table_b2.items():
if map(ord, unichr(k).lower()) != v:
b3_exceptions[k] = u"".join(map(unichr,v))
if list(map(ord, chr(k).lower())) != v:
b3_exceptions[k] = "".join(map(chr,v))
b3 = sorted(b3_exceptions.items())
print("""
b3_exceptions = {""")
for i,(k,v) in enumerate(b3):
print("0x%x:%s," % (k, repr(v)), end=' ')
for i, kv in enumerate(b3):
print("0x%x:%a," % kv, end=' ')
if i % 4 == 3:
print()
print("}")
@ -224,7 +230,7 @@ print("""
def map_table_b2(a):
al = map_table_b3(a)
b = unicodedata.normalize("NFKC", al)
bl = u"".join([map_table_b3(ch) for ch in b])
bl = "".join([map_table_b3(ch) for ch in b])
c = unicodedata.normalize("NFKC", bl)
if b != c:
return c
@ -240,7 +246,7 @@ assert table == {0x20:0x20}
print("""
def in_table_c11(code):
return code == u" "
return code == " "
""")
# C.1.2 is the rest of all space characters
@ -249,12 +255,12 @@ del tables[0]
assert name == "C.1.2"
# table = set(table.keys())
# Zs = set(gen_category(["Zs"])) - set([0x20])
# Zs = set(gen_category(["Zs"])) - {0x20}
# assert Zs == table
print("""
def in_table_c12(code):
return unicodedata.category(code) == "Zs" and code != u" "
return unicodedata.category(code) == "Zs" and code != " "
def in_table_c11_c12(code):
return unicodedata.category(code) == "Zs"