lyx_mirror/lib/scripts/layout2layout.py
2013-02-12 10:13:41 -05:00

989 lines
36 KiB
Python

#! /usr/bin/env python
# -*- coding: utf-8 -*-
# file layout2layout.py
# This file is part of LyX, the document processor.
# Licence details can be found in the file COPYING.
# author Georg Baum
# Full author contact details are available in file CREDITS
# This script will update a .layout file to current format
import os, re, string, sys
# Incremented to format 4, 6 April 2007, lasgouttes
# Introduction of generic "Provides" declaration
# Incremented to format 5, 22 August 2007 by vermeer
# InsetLayout material
# Incremented to format 6, 7 January 2008 by spitz
# Requires tag added to layout files
# Incremented to format 7, 24 March 2008 by rgh
# AddToPreamble tag added to layout files
# Incremented to format 8, 25 July 2008 by rgh
# UseModule tag added to layout files
# CopyStyle added to InsetLayout
# Incremented to format 9, 5 October 2008 by rgh
# ForcePlain and CustomPars tags added to InsetLayout
# Incremented to format 10, 6 October 2008 by rgh
# Change format of counters
# Incremented to format 11, 14 October 2008 by rgh
# Add ProvidesModule, ExcludesModule tags
# Incremented to format 12, 10 January 2009 by gb
# Add I18NPreamble tag
# Incremented to format 13, 5 February 2009 by rgh
# Add InToc tag for InsetLayout
# Incremented to format 14, 14 February 2009 by gb
# Rename I18NPreamble to BabelPreamble and add LangPreamble
# Incremented to format 15, 28 May 2009 by lasgouttes
# Add new tag OutputFormat; modules can be conditioned on feature
# "from->to".
# Incremented to format 16, 5 June 2009 by rgh
# Add new tags for Text Class:
# HTMLPreamble, HTMLAddToPreamble
# For Layout:
# HTMLTag, HTMLAttr, HTMLLabel, HTMLLabelAttr, HTMLItem, HTMLItemAttr
# HTMLStyle, and HTMLPreamble
# For InsetLayout:
# HTMLTag, HTMLAttr, HTMLStyle, and HTMLPreamble
# For Floats:
# HTMLType, HTMLClass, HTMLStyle
# Incremented to format 17, 12 August 2009 by rgh
# Add IfStyle and IfCounter tags for layout.
# Incremented to format 18, 27 October 2009 by rgh
# Added some new tags for HTML output.
# Incremented to format 19, 17 November 2009 by rgh
# Added InPreamble tag.
# Incremented to format 20, 17 December 2009 by rgh
# Added ContentAsLabel tag.
# Incremented to format 21, 12 January 2010 by rgh
# Added HTMLTocLayout and HTMLTitle tags.
# Incremented to format 22, 20 January 2010 by rgh
# Added HTMLFormat tag to Counters.
# Incremented to format 23, 13 February 2010 by spitz
# Added Spellcheck tag.
# Incremented to format 24, 5 March 2010 by rgh
# Changed LaTeXBuiltin tag to NeedsFloatPkg and
# added new tag ListCommand.
# Incremented to format 25, 12 March 2010 by rgh
# Added RefPrefix tag for layouts and floats.
# Incremented to format 26, 29 March 2010 by rgh
# Added CiteFormat.
# Incremented to format 27, 4 June 2010 by rgh
# Added RequiredArgs tag.
# Incremented to format 28, 6 August 2010 by lasgouttes
# Added ParbreakIsNewline tag for Layout and InsetLayout.
# Incremented to format 29, 10 August 2010 by rgh
# Changed Custom:Style, CharStyle:Style, and Element:Style
# uniformly to Flex:Style.
# Incremented to format 30, 13 August 2010 by rgh
# Introduced ResetsFont tag for InsetLayout.
# Incremented to format 31, 12 January 2011 by rgh
# Introducted NoCounter tag.
# Incremented to format 32, 30 January 2011 by forenr
# Added Display tag for InsetLayout
# Incremented to format 33, 2 February 2011 by rgh
# Changed NeedsFloatPkg to UsesFloatPkg
# Incremented to format 34, 28 March 2011 by rgh
# Remove obsolete Fill_(Top|Bottom) tags
# Incremented to format 35, 28 March 2011 by rgh
# Try to add "Flex:" to any flex insets that don't have it.
# Incremented to format 36, 7 December 2011, by rgh
# Added HTMLStyles and AddToHTMLStyles tags.
# Incremented to format 37, 29 February 2012 by jrioux
# Implement the citation engine machinery in layouts.
# Change CiteFormat to CiteFormat (default|authoryear|numerical).
# Incremented to format 38, 08 April 2012 by gb
# Introduce LangPreamble and BabelPreamble for InsetLayout.
# Incremented to format 39, 15 April 2012 by sanda
# Introduce styling of branches via "InsetLayout Branch:".
# Incremented to format 40, 10 October 2012 by rgh
# Re-do layout names for layout categories
# Incremented to format 41, 20 November 2012 by spitz
# New Argument syntax
# Incremented to format 42, 22 December 2012 by spitz
# New Style tag "ItemCommand"
# Incremented to format 43, 30 December 2012 by spitz
# Extended InsetCaption format
# Incremented to format 44, 9 February 2013 by rgh
# Remove COUNTER label style; rename as STATIC
# Rename TOP_ENVIRONMENT to ABOVE and CENTERED_TOP_ENVIRONMENT to CENTERED
# Incremented to format 45, 12 February 2013 by rgh
# New Tag "NoInsetLayout"
# Do not forget to document format change in Customization
# Manual (section "Declaring a new text class").
# You might also want to consider running the
# development/tools/updatelayouts.sh script to update all
# layout files to the new format.
currentFormat = 44
def usage(prog_name):
return ("Usage: %s inputfile outputfile\n" % prog_name +
"or %s <inputfile >outputfile" % prog_name)
def error(message):
sys.stderr.write(message + '\n')
sys.exit(1)
def trim_bom(line):
" Remove byte order mark."
if line[0:3] == "\357\273\277":
return line[3:]
else:
return line
def read(source):
" Read input file and strip lineendings."
lines = source.read().splitlines() or ['']
lines[0] = trim_bom(lines[0])
return lines
def write(output, lines):
" Write output file with native lineendings."
output.write(os.linesep.join(lines) + os.linesep)
# Concatenates old and new in an intelligent way:
# If old is wrapped in ", they are stripped. The result is wrapped in ".
def concatenate_label(old, new):
# Don't use strip as long as we support python 1.5.2
if old[0] == '"':
return old[0:-1] + new + '"'
else:
return '"' + old + new + '"'
# appends a string to a list unless it's already there
def addstring(s, l):
if l.count(s) > 0:
return
l.append(s)
def convert(lines):
" Convert to new format."
re_Comment = re.compile(r'^(\s*)#')
re_Counter = re.compile(r'\s*Counter\s*', re.IGNORECASE)
re_Name = re.compile(r'\s*Name\s+(\S+)\s*', re.IGNORECASE)
re_UseMod = re.compile(r'^\s*UseModule\s+(.*)', re.IGNORECASE)
re_Empty = re.compile(r'^(\s*)$')
re_Format = re.compile(r'^(\s*)(Format)(\s+)(\S+)', re.IGNORECASE)
re_Preamble = re.compile(r'^(\s*)Preamble', re.IGNORECASE)
re_EndPreamble = re.compile(r'^(\s*)EndPreamble', re.IGNORECASE)
re_LangPreamble = re.compile(r'^(\s*)LangPreamble', re.IGNORECASE)
re_EndLangPreamble = re.compile(r'^(\s*)EndLangPreamble', re.IGNORECASE)
re_BabelPreamble = re.compile(r'^(\s*)BabelPreamble', re.IGNORECASE)
re_EndBabelPreamble = re.compile(r'^(\s*)EndBabelPreamble', re.IGNORECASE)
re_MaxCounter = re.compile(r'^(\s*)(MaxCounter)(\s+)(\S+)', re.IGNORECASE)
re_LabelType = re.compile(r'^(\s*)(LabelType)(\s+)(\S+)', re.IGNORECASE)
re_LabelString = re.compile(r'^(\s*)(LabelString)(\s+)(("[^"]+")|(\S+))', re.IGNORECASE)
re_LabelStringAppendix = re.compile(r'^(\s*)(LabelStringAppendix)(\s+)(("[^"]+")|(\S+))', re.IGNORECASE)
re_LatexType = re.compile(r'^(\s*)(LatexType)(\s+)(\S+)', re.IGNORECASE)
re_Style = re.compile(r'^(\s*)(Style)(\s+)(\S+)', re.IGNORECASE)
re_CopyStyle = re.compile(r'^(\s*)(CopyStyle)(\s+)(\S+)', re.IGNORECASE)
re_NoStyle = re.compile(r'^(\s*)(NoStyle)(\s+)(\S+)', re.IGNORECASE)
re_End = re.compile(r'^(\s*)(End)(\s*)$', re.IGNORECASE)
re_Provides = re.compile(r'^(\s*)Provides(\S+)(\s+)(\S+)', re.IGNORECASE)
re_CharStyle = re.compile(r'^(\s*)CharStyle(\s+)(\S+)$', re.IGNORECASE)
re_CiteFormat = re.compile(r'^(\s*)(CiteFormat)(?:(\s*)()|(\s+)(default|authoryear|numerical))', re.IGNORECASE)
re_AMSMaths = re.compile(r'^\s*Input ams(?:math|def)s.inc\s*')
re_AMSMathsPlain = re.compile(r'^\s*Input amsmaths-plain.inc\s*')
re_AMSMathsSeq = re.compile(r'^\s*Input amsmaths-seq.inc\s*')
re_TocLevel = re.compile(r'^(\s*)(TocLevel)(\s+)(\S+)', re.IGNORECASE)
re_I18nPreamble = re.compile(r'^(\s*)I18nPreamble', re.IGNORECASE)
re_EndI18nPreamble = re.compile(r'^(\s*)EndI18nPreamble', re.IGNORECASE)
re_Float = re.compile(r'^\s*Float\s*$', re.IGNORECASE)
re_Type = re.compile(r'\s*Type\s+(\w+)', re.IGNORECASE)
re_Builtin = re.compile(r'^(\s*)LaTeXBuiltin\s+(\w*)', re.IGNORECASE)
re_True = re.compile(r'^\s*(?:true|1)\s*$', re.IGNORECASE)
re_InsetLayout = re.compile(r'^\s*InsetLayout\s+(?:Custom|CharStyle|Element):(\S+)\s*$', re.IGNORECASE)
# with quotes
re_QInsetLayout = re.compile(r'^\s*InsetLayout\s+"(?:Custom|CharStyle|Element):([^"]+)"\s*$', re.IGNORECASE)
re_InsetLayout_CopyStyle = re.compile(r'^\s*CopyStyle\s+(?:Custom|CharStyle|Element):(\S+)\s*$', re.IGNORECASE)
re_QInsetLayout_CopyStyle = re.compile(r'^\s*CopyStyle\s+"(?:Custom|CharStyle|Element):([^"]+)"\s*$', re.IGNORECASE)
re_NeedsFloatPkg = re.compile(r'^(\s*)NeedsFloatPkg\s+(\w+)\s*$', re.IGNORECASE)
re_Fill = re.compile(r'^\s*Fill_(?:Top|Bottom).*$', re.IGNORECASE)
re_InsetLayout2 = re.compile(r'^\s*InsetLayout\s+(\S+)\s*$', re.IGNORECASE)
# with quotes
re_QInsetLayout2 = re.compile(r'^\s*InsetLayout\s+"([^"]+)"\s*$', re.IGNORECASE)
re_IsFlex = re.compile(r'\s*LyXType.*$', re.IGNORECASE)
re_CopyStyle2 = re.compile(r'(\s*CopyStyle\s+)"?([^"]+)"?\s*$')
# for categories
re_Declaration = re.compile(r'^#\s*\\Declare\w+Class.*$')
re_ExtractCategory = re.compile(r'^(#\s*\\Declare\w+Class(?:\[[^]]*?\])?){([^(]+?)\s+\(([^)]+?)\)\s*}\s*$')
ConvDict = {"article": "Articles", "book" : "Books", "letter" : "Letters", "report": "Reports", \
"presentation" : "Presentations", "curriculum vitae" : "Curricula Vitae", "handout" : "Handouts"}
# Arguments
re_OptArgs = re.compile(r'^(\s*)OptionalArgs(\s+)(\d+)\D*$', re.IGNORECASE)
re_ReqArgs = re.compile(r'^(\s*)RequiredArgs(\s+)(\d+)\D*$', re.IGNORECASE)
# various changes associated with changing how chapters are handled
re_LabelTypeIsCounter = re.compile(r'^(\s*)LabelType(\s*)Counter\s*$', re.IGNORECASE)
re_TopEnvironment = re.compile(r'^(\s*)LabelType(\s+)Top_Environment\s*$', re.IGNORECASE)
re_CenteredEnvironment = re.compile(r'^(\s*)LabelType(\s+)Centered_Top_Environment\s*$', re.IGNORECASE)
re_ChapterStyle = re.compile(r'^\s*Style\s+Chapter\s*$', re.IGNORECASE)
# counters for sectioning styles (hardcoded in 1.3)
counters = {"part" : "\\Roman{part}",
"chapter" : "\\arabic{chapter}",
"section" : "\\arabic{section}",
"subsection" : "\\arabic{section}.\\arabic{subsection}",
"subsubsection" : "\\arabic{section}.\\arabic{subsection}.\\arabic{subsubsection}",
"paragraph" : "\\arabic{section}.\\arabic{subsection}.\\arabic{subsubsection}.\\arabic{paragraph}",
"subparagraph" : "\\arabic{section}.\\arabic{subsection}.\\arabic{subsubsection}.\\arabic{paragraph}.\\arabic{subparagraph}"}
# counters for sectioning styles in appendix (hardcoded in 1.3)
appendixcounters = {"chapter" : "\\Alph{chapter}",
"section" : "\\Alph{section}",
"subsection" : "\\arabic{section}.\\arabic{subsection}",
"subsubsection" : "\\arabic{section}.\\arabic{subsection}.\\arabic{subsubsection}",
"paragraph" : "\\arabic{section}.\\arabic{subsection}.\\arabic{subsubsection}.\\arabic{paragraph}",
"subparagraph" : "\\arabic{section}.\\arabic{subsection}.\\arabic{subsubsection}.\\arabic{paragraph}.\\arabic{subparagraph}"}
# Value of TocLevel for sectioning styles
toclevels = {"part" : 0,
"chapter" : 0,
"section" : 1,
"subsection" : 2,
"subsubsection" : 3,
"paragraph" : 4,
"subparagraph" : 5}
i = 0
only_comment = 1
counter = ""
toclevel = ""
label = ""
labelstring = ""
labelstringappendix = ""
space1 = ""
labelstring_line = -1
labelstringappendix_line = -1
labeltype_line = -1
latextype = ""
latextype_line = -1
style = ""
maxcounter = 0
format = 1
formatline = 0
usemodules = []
flexstyles = []
opts = 0
reqs = 0
inchapter = False
while i < len(lines):
# Skip comments and empty lines
if (re_Comment.match(lines[i]) or re_Empty.match(lines[i])):
# We need to deal with this conversion here, because it happens
# inside the initial comment block.
if only_comment and format == 39:
match = re_ExtractCategory.match(lines[i])
if match:
lpre = match.group(1)
lcat = match.group(2)
lnam = match.group(3)
if lcat in ConvDict:
lcat = ConvDict[lcat]
lines[i] = lpre + "{" + lnam + "}"
lines.insert(i+1, "# \\DeclareCategory{" + lcat + "}")
i += 1
i += 1
continue
# insert file format if not already there
if (only_comment):
match = re_Format.match(lines[i])
if match:
formatline = i
format = int(match.group(4))
if format > 1 and format < currentFormat:
lines[i] = "Format %d" % (format + 1)
only_comment = 0
elif format == currentFormat:
# nothing to do
return format
else:
error('Cannot convert file format %s' % format)
else:
lines.insert(i, "Format 2")
only_comment = 0
continue
# Don't get confused by LaTeX code
if re_Preamble.match(lines[i]):
i += 1
while i < len(lines) and not re_EndPreamble.match(lines[i]):
i += 1
continue
if re_LangPreamble.match(lines[i]):
i += 1
while i < len(lines) and not re_EndLangPreamble.match(lines[i]):
i += 1
continue
if re_BabelPreamble.match(lines[i]):
i += 1
while i < len(lines) and not re_EndBabelPreamble.match(lines[i]):
i += 1
continue
if format == 44:
# nothing to do.
i += 1
continue
if format == 43:
match = re_LabelTypeIsCounter.match(lines[i])
if match:
if inchapter:
lines[i] = match.group(1) + "LabelType" + match.group(2) + "Above"
else:
lines[i] = match.group(1) + "LabelType" + match.group(2) + "Static"
match = re_TopEnvironment.match(lines[i])
if match:
lines[i] = match.group(1) + "LabelType" + match.group(2) + "Above"
match = re_CenteredEnvironment.match(lines[i])
if match:
lines[i] = match.group(1) + "LabelType" + match.group(2) + "Centered"
if inchapter:
match = re_Style.match(lines[i])
if match:
inchapter = False
else:
match = re_ChapterStyle.match(lines[i])
if match:
inchapter = True
i += 1
continue
if format == 42:
if lines[i] == "InsetLayout Caption":
lines[i] = "InsetLayout Caption:Standard"
i += 1
continue
if format == 41:
# nothing to do.
i += 1
continue
if format == 40:
# reset counters on Style beginning
match = re_Style.match(lines[i])
if match:
opts = 0
reqs = 0
i += 1
continue
match = re_OptArgs.match(lines[i])
if match:
# Save number of optional arguments
space1 = match.group(1)
opts = int(match.group(3))
# OptionalArgs 0 > ResetArgs 1
if opts == 0:
lines[i] = space1 + "ResetArgs\t1"
i += 1
else:
del lines[i]
continue
match = re_ReqArgs.match(lines[i])
if match:
# Save number of required arguments
space1 = match.group(1)
reqs = int(match.group(3))
del lines[i]
continue
# Insert the required number of arguments at the end of the style definition
match = re_End.match(lines[i])
if match:
newarg = ['']
# First the optionals (this is the required order pre 2.1)
if opts > 0:
if opts == 1:
newarg = [ '%sArgument 1' % (space1),
'%s\tLabelString\t\"Optional Layout Argument\"' % (space1),
'%sEndArgument' % (space1)]
elif opts > 1:
actopt = 1
while actopt < (opts + 1):
newarg += [ '%sArgument %d' % (space1, actopt),
'%s\tLabelString\t\"Optional Layout Argument %d\"' % (space1, actopt),
'%sEndArgument' % (space1)]
actopt += 1
# Now the mandatories
if reqs > 0:
actopt = opts + 1
while actopt < (opts + reqs + 1):
newarg += [ '%sArgument %d' % (space1, actopt),
'%s\tLabelString\t"Required Layout Argument %d"' % (space1, actopt - opts),
'%s\tMandatory\t1' % (space1),
'%sEndArgument' % (space1)]
actopt += 1
# Since we replace the "End" line, re-add this line
if len(newarg) > 1:
newarg += ['End']
lines[i:i+1] = newarg
i += len(newarg)
# Reset the counters
opts = 0
reqs = 0
i += 1
continue
if format == 39:
# There is a conversion with format 40, but it is done within the
# initial comment block and so is above.
i += 1
continue
if format == 37 or format == 38:
i += 1
continue
if format == 36:
match = re_CiteFormat.match(lines[i]);
if match and match.group(4) == "":
lines[i] = match.group(0) + " default"
i += 1
continue
if format == 35:
i += 1
continue
if format == 34:
match = re_QInsetLayout2.match(lines[i])
if not match:
match = re_InsetLayout2.match(lines[i])
if not match:
match = re_CopyStyle2.match(lines[i])
if not match:
i += 1
continue
style = match.group(2)
if flexstyles.count(style):
lines[i] = match.group(1) + "\"Flex:" + style + "\""
i += 1
continue
name = match.group(1)
names = name.split(":", 1)
if len(names) > 1 and names[0] == "Flex":
i += 1
continue
isflex = False
for j in range(i + 1, len(lines)):
if re_IsFlex.match(lines[j]):
isflex = True
break
if re_End.match(lines[j]):
break
if isflex:
flexstyles.append(name)
lines[i] = "InsetLayout \"Flex:" + name + "\""
i += 1
continue
if format == 33:
m = re_Fill.match(lines[i])
if m:
lines[i] = ""
i += 1
continue
if format == 32:
match = re_NeedsFloatPkg.match(lines[i])
if match:
space = match.group(1)
val = match.group(2)
lines[i] = space + "UsesFloatPkg " + val
newval = 'true'
if val == '1' or val.lower() == 'true':
newval = 'false'
lines.insert(i, space + "IsPredefined " + newval)
i += 1
i += 1
continue
# Only new features
if format >= 29 and format <= 31:
i += 1
continue
if format == 28:
match = re_InsetLayout.match(lines[i])
if match:
lines[i] = "InsetLayout Flex:" + match.group(1)
else:
match = re_QInsetLayout.match(lines[i])
if match:
lines[i] = "InsetLayout \"Flex:" + match.group(1) + "\""
else:
match = re_InsetLayout_CopyStyle.match(lines[i])
if match:
lines[i] = "\tCopyStyle Flex:" + match.group(1)
else:
match = re_QInsetLayout_CopyStyle.match(lines[i])
if match:
lines[i] = "\tCopyStyle \"Flex:" + match.group(1) + "\""
i += 1
continue
# Only new features
if format >= 24 and format <= 27:
i += 1
continue
if format == 23:
match = re_Float.match(lines[i])
i += 1
if not match:
continue
# we need to do two things:
# (i) Convert Builtin to NeedsFloatPkg
# (ii) Write ListCommand lines for the builtin floats table and figure
builtin = False
cmd = ""
while True and i < len(lines):
m1 = re_End.match(lines[i])
if m1:
if builtin and cmd:
line = " ListCommand " + cmd
lines.insert(i, line)
i += 1
break
m2 = re_Builtin.match(lines[i])
if m2:
builtin = True
ws1 = m2.group(1)
arg = m2.group(2)
newarg = ""
if re_True.match(arg):
newarg = "false"
else:
newarg = "true"
lines[i] = ws1 + "NeedsFloatPkg " + newarg
m3 = re_Type.match(lines[i])
if m3:
fltype = m3.group(1)
fltype = fltype.lower()
if fltype == "table":
cmd = "listoftables"
elif fltype == "figure":
cmd = "listoffigures"
# else unknown, which is why we're doing this
i += 1
continue
# This just involved new features, not any changes to old ones
if format >= 14 and format <= 22:
i += 1
continue
# Rename I18NPreamble to BabelPreamble
if format == 13:
match = re_I18nPreamble.match(lines[i])
if match:
lines[i] = match.group(1) + "BabelPreamble"
i += 1
match = re_EndI18nPreamble.match(lines[i])
while i < len(lines) and not match:
i += 1
match = re_EndI18nPreamble.match(lines[i])
lines[i] = match.group(1) + "EndBabelPreamble"
i += 1
continue
# These just involved new features, not any changes to old ones
if format == 11 or format == 12:
i += 1
continue
if format == 10:
match = re_UseMod.match(lines[i])
if match:
module = match.group(1)
lines[i] = "DefaultModule " + module
i += 1
continue
if format == 9:
match = re_Counter.match(lines[i])
if match:
counterline = i
i += 1
while i < len(lines):
namem = re_Name.match(lines[i])
if namem:
name = namem.group(1)
lines.pop(i)
lines[counterline] = "Counter %s" % name
# we don't need to increment i
continue
endem = re_End.match(lines[i])
if endem:
i += 1
break
i += 1
i += 1
continue
if format == 8:
# We want to scan for ams-type includes and, if we find them,
# add corresponding UseModule tags to the layout.
match = re_AMSMaths.match(lines[i])
if match:
addstring("theorems-ams", usemodules)
addstring("theorems-ams-extended", usemodules)
addstring("theorems-sec", usemodules)
lines.pop(i)
continue
match = re_AMSMathsPlain.match(lines[i])
if match:
addstring("theorems-starred", usemodules)
lines.pop(i)
continue
match = re_AMSMathsSeq.match(lines[i])
if match:
addstring("theorems-ams", usemodules)
addstring("theorems-ams-extended", usemodules)
lines.pop(i)
continue
i += 1
continue
# These just involved new features, not any changes to old ones
if format >= 5 and format <= 7:
i += 1
continue
if format == 4:
# Handle conversion to long CharStyle names
match = re_CharStyle.match(lines[i])
if match:
lines[i] = "InsetLayout CharStyle:%s" % (match.group(3))
i += 1
lines.insert(i, "\tLyXType charstyle")
i += 1
lines.insert(i, "")
lines[i] = "\tLabelString %s" % (match.group(3))
i += 1
continue
if format == 3:
# convert 'providesamsmath x', 'providesmakeidx x', 'providesnatbib x', 'providesurl x' to
# 'provides amsmath x', 'provides makeidx x', 'provides natbib x', 'provides url x'
# x is either 0 or 1
match = re_Provides.match(lines[i])
if match:
lines[i] = "%sProvides %s%s%s" % (match.group(1), match.group(2).lower(),
match.group(3), match.group(4))
i += 1
continue
if format == 2:
caption = []
# delete caption styles
match = re_Style.match(lines[i])
if match:
style = string.lower(match.group(4))
if style == "caption":
del lines[i]
while i < len(lines) and not re_End.match(lines[i]):
caption.append(lines[i])
del lines[i]
if i == len(lines):
error('Incomplete caption style.')
else:
del lines[i]
continue
# delete undefinition of caption styles
match = re_NoStyle.match(lines[i])
if match:
style = string.lower(match.group(4))
if style == "caption":
del lines[i]
continue
# replace the CopyStyle statement with the definition of the real
# style. This may result in duplicate statements, but that is OK
# since the second one will overwrite the first one.
match = re_CopyStyle.match(lines[i])
if match:
style = string.lower(match.group(4))
if style == "caption":
if len(caption) > 0:
lines[i:i+1] = caption
else:
# FIXME: This style comes from an include file, we
# should replace the real style and not this default.
lines[i:i+1] = [' Margin First_Dynamic',
' LatexType Command',
' LatexName caption',
' NeedProtect 1',
' LabelSep xx',
' ParSkip 0.4',
' TopSep 0.5',
' Align Center',
' AlignPossible Center',
' LabelType Sensitive',
' LabelString "Senseless!"',
' OptionalArgs 1',
' LabelFont',
' Series Bold',
' EndFont']
i += 1
continue
# Delete MaxCounter and remember the value of it
match = re_MaxCounter.match(lines[i])
if match:
level = match.group(4)
if string.lower(level) == "counter_chapter":
maxcounter = 0
elif string.lower(level) == "counter_section":
maxcounter = 1
elif string.lower(level) == "counter_subsection":
maxcounter = 2
elif string.lower(level) == "counter_subsubsection":
maxcounter = 3
elif string.lower(level) == "counter_paragraph":
maxcounter = 4
elif string.lower(level) == "counter_subparagraph":
maxcounter = 5
elif string.lower(level) == "counter_enumi":
maxcounter = 6
elif string.lower(level) == "counter_enumii":
maxcounter = 7
elif string.lower(level) == "counter_enumiii":
maxcounter = 8
del lines[i]
continue
# Replace line
#
# LabelType Counter_EnumI
#
# with two lines
#
# LabelType Counter
# LabelCounter EnumI
#
match = re_LabelType.match(lines[i])
if match:
label = match.group(4)
# Remember indenting space for later reuse in added lines
space1 = match.group(1)
# Remember the line for adding the LabelCounter later.
# We can't do it here because it could shift latextype_line etc.
labeltype_line = i
if string.lower(label[:8]) == "counter_":
counter = string.lower(label[8:])
lines[i] = re_LabelType.sub(r'\1\2\3Counter', lines[i])
# Remember the LabelString line
match = re_LabelString.match(lines[i])
if match:
labelstring = match.group(4)
labelstring_line = i
# Remember the LabelStringAppendix line
match = re_LabelStringAppendix.match(lines[i])
if match:
labelstringappendix = match.group(4)
labelstringappendix_line = i
# Remember the LatexType line
match = re_LatexType.match(lines[i])
if match:
latextype = string.lower(match.group(4))
latextype_line = i
# Remember the TocLevel line
match = re_TocLevel.match(lines[i])
if match:
toclevel = string.lower(match.group(4))
# Reset variables at the beginning of a style definition
match = re_Style.match(lines[i])
if match:
style = string.lower(match.group(4))
counter = ""
toclevel = ""
label = ""
space1 = ""
labelstring = ""
labelstringappendix = ""
labelstring_line = -1
labelstringappendix_line = -1
labeltype_line = -1
latextype = ""
latextype_line = -1
if re_End.match(lines[i]):
# Add a line "LatexType Bib_Environment" if LabelType is Bibliography
# (or change the existing LatexType)
if string.lower(label) == "bibliography":
if (latextype_line < 0):
lines.insert(i, "%sLatexType Bib_Environment" % space1)
i += 1
else:
lines[latextype_line] = re_LatexType.sub(r'\1\2\3Bib_Environment', lines[latextype_line])
# Change "LabelType Static" to "LabelType Itemize" for itemize environments
if latextype == "item_environment" and string.lower(label) == "static":
lines[labeltype_line] = re_LabelType.sub(r'\1\2\3Itemize', lines[labeltype_line])
# Change "LabelType Counter_EnumI" to "LabelType Enumerate" for enumerate environments
if latextype == "item_environment" and string.lower(label) == "counter_enumi":
lines[labeltype_line] = re_LabelType.sub(r'\1\2\3Enumerate', lines[labeltype_line])
# Don't add the LabelCounter line later
counter = ""
# Replace
#
# LabelString "Chapter"
#
# with
#
# LabelString "Chapter \arabic{chapter}"
#
# if this style has a counter. Ditto for LabelStringAppendix.
# This emulates the hardcoded article style numbering of 1.3
#
if counter != "":
if style in counters:
if labelstring_line < 0:
lines.insert(i, '%sLabelString "%s"' % (space1, counters[style]))
i += 1
else:
new_labelstring = concatenate_label(labelstring, counters[style])
lines[labelstring_line] = re_LabelString.sub(
r'\1\2\3%s' % new_labelstring.replace("\\", "\\\\"),
lines[labelstring_line])
if style in appendixcounters:
if labelstringappendix_line < 0:
lines.insert(i, '%sLabelStringAppendix "%s"' % (space1, appendixcounters[style]))
i += 1
else:
new_labelstring = concatenate_label(labelstring, appendixcounters[style])
lines[labelstringappendix_line] = re_LabelStringAppendix.sub(
r'\1\2\3%s' % new_labelstring.replace("\\", "\\\\"),
lines[labelstringappendix_line])
# Now we can safely add the LabelCounter line
lines.insert(labeltype_line + 1, "%sLabelCounter %s" % (space1, counter))
i += 1
# Add the TocLevel setting for sectioning styles
if toclevel == "" and style in toclevels and maxcounter <= toclevels[style]:
lines.insert(i, '%s\tTocLevel %d' % (space1, toclevels[style]))
i += 1
i += 1
if only_comment:
lines.insert(i, "Format 2")
if usemodules:
i = formatline + 1
for mod in usemodules:
lines.insert(i, "UseModule " + mod)
i += 1
return format + 1
def main(argv):
# Open files
if len(argv) == 1:
source = sys.stdin
output = sys.stdout
elif len(argv) == 3:
source = open(argv[1], 'rb')
output = open(argv[2], 'wb')
else:
error(usage(argv[0]))
# Do the real work
lines = read(source)
format = 1
while (format < currentFormat):
format = convert(lines)
write(output, lines)
# Close files
if len(argv) == 3:
source.close()
output.close()
return 0
if __name__ == "__main__":
main(sys.argv)