Some python cleanup from Thibault.

This commit is contained in:
Richard Kimberly Heck 2020-06-21 11:37:23 -04:00
parent 95eccf0c53
commit 6f084c4c30
2 changed files with 14 additions and 11 deletions

View File

@ -1747,11 +1747,13 @@ ruby_inset_def = [
r'End',
]
def convert_ruby_module(document):
"""Use ruby module instead of local module definition"""
if document.del_local_layout(ruby_inset_def):
document.add_module("ruby")
def revert_ruby_module(document):
"""Replace ruby module with local module definition"""
if document.del_module("ruby"):
@ -1768,6 +1770,7 @@ def convert_utf8_japanese(document):
or (lang == "japanese-cjk" and inputenc == "utf8-cjk")):
document.set_parameter("inputencoding", "utf8")
def revert_utf8_japanese(document):
"""Use Japanese utf8 variants with Japanese documents."""
inputenc = get_value(document.header, "\\inputencoding")

View File

@ -392,7 +392,7 @@ def convert(lines, end_format):
# for categories
re_Declaration = re.compile(b'^#\\s*\\Declare\\w+Class.*$')
re_ExtractCategory = re.compile(b'^(#\\s*\\Declare\\w+Class(?:\\[[^]]*?\\])?){([^(]+?)\\s+\\(([^)]+?)\\)\\s*}\\s*$')
ConvDict = {"article": "Articles", "book" : "Books", "letter" : "Letters", "report": "Reports", \
ConvDict = {"article": "Articles", "book" : "Books", "letter" : "Letters", "report": "Reports",
"presentation" : "Presentations", "curriculum vitae" : "Curricula Vitae", "handout" : "Handouts"}
# Arguments
re_OptArgs = re.compile(b'^(\\s*)OptionalArgs(\\s+)(\\d+)\\D*$', re.IGNORECASE)
@ -485,12 +485,12 @@ def convert(lines, end_format):
continue
# insert file format if not already there
if (only_comment):
if only_comment:
match = re_Format.match(lines[i])
if match:
formatline = i
format = int(match.group(4))
if format > 1 and format < end_format:
if 1 < format < end_format:
lines[i] = b"Format %d" % (format + 1)
only_comment = 0
elif format == end_format:
@ -545,7 +545,7 @@ def convert(lines, end_format):
i += 1
continue
if format >= 60 and format <= 62:
if 60 <= format <= 62:
# nothing to do.
i += 1
continue
@ -578,7 +578,7 @@ def convert(lines, end_format):
i += 1
continue
if format >= 50 and format <= 56:
if 50 <= format <= 56:
# nothing to do.
i += 1
continue
@ -676,7 +676,7 @@ def convert(lines, end_format):
continue
if format >= 44 and format <= 47:
if 44 <= format <= 47:
# nothing to do.
i += 1
continue
@ -795,7 +795,7 @@ def convert(lines, end_format):
continue
if format == 36:
match = re_CiteFormat.match(lines[i]);
match = re_CiteFormat.match(lines[i])
if match and match.group(4) == b"":
lines[i] = match.group(0) + b" default"
i += 1
@ -864,7 +864,7 @@ def convert(lines, end_format):
continue
# Only new features
if format >= 29 and format <= 31:
if 29 <= format <= 31:
i += 1
continue
@ -888,7 +888,7 @@ def convert(lines, end_format):
continue
# Only new features
if format >= 24 and format <= 27:
if 24 <= format <= 27:
i += 1
continue
@ -934,7 +934,7 @@ def convert(lines, end_format):
continue
# This just involved new features, not any changes to old ones
if format >= 14 and format <= 22:
if 14 <= format <= 22:
i += 1
continue
@ -1011,7 +1011,7 @@ def convert(lines, end_format):
continue
# These just involved new features, not any changes to old ones
if format >= 5 and format <= 7:
if 5 <= format <= 7:
i += 1
continue