273 lines
8.1 KiB
Python
273 lines
8.1 KiB
Python
#!/usr/bin/env python
|
|
# -*- coding: utf-8 -*-
|
|
#
|
|
|
|
import re
|
|
import sys
|
|
import pprint
|
|
import mistune # must be v0.8.4
|
|
|
|
def formspec_escape(text):
|
|
text = text.replace("\\", "")
|
|
text = text.replace("[", "\\\\[")
|
|
text = text.replace("]", "\\\\]")
|
|
text = text.replace(";", "\\\\;")
|
|
text = text.replace(",", "\\\\,")
|
|
text = text.replace('"', '\\"')
|
|
text = text.replace('\n', '\\n')
|
|
#print ">>>>"+text+"<<<<"
|
|
return text
|
|
|
|
def header_escsape(header):
|
|
header = header.lower()
|
|
header = header.replace(" ", "-")
|
|
header = header.replace("/", "")
|
|
return header
|
|
|
|
lTitel = []
|
|
lText = []
|
|
lItemName = []
|
|
lPlanTable = []
|
|
lTocLinks = []
|
|
|
|
def reset():
|
|
global lTitel, lText, lItemName, lPlanTable, lTocLinks
|
|
|
|
lTitel = []
|
|
lText = []
|
|
lItemName = []
|
|
lPlanTable = []
|
|
lTocLinks = []
|
|
|
|
|
|
def lua_table(name, lData):
|
|
lOut = []
|
|
lOut.append("%s = {" % name)
|
|
for line in lData:
|
|
lOut.append(' "%s",' % line)
|
|
lOut.append("}\n\n")
|
|
return "\n".join(lOut)
|
|
|
|
def lua_text_table(name, lData):
|
|
lOut = []
|
|
lOut.append("%s = {" % name)
|
|
for lines in lData:
|
|
for line in lines[:-1]:
|
|
line = line.replace('<br>', '\\n')
|
|
lOut.append(' "%s\\n"..' % line)
|
|
if len(lines) > 0:
|
|
lOut.append(' "%s\\n",' % lines[-1])
|
|
else:
|
|
lOut.append(' "",')
|
|
lOut.append("}\n\n")
|
|
return "\n".join(lOut)
|
|
|
|
class WikiLinkInlineLexer(mistune.InlineLexer):
|
|
def enable_wiki_link(self):
|
|
# add wiki_link rules
|
|
self.rules.wiki_link = re.compile(
|
|
r'\[' # [
|
|
r'([\s\S]+?\|[\s\S]+?)' # name| img-type
|
|
r'\](?!\])' # ]
|
|
)
|
|
|
|
# Add wiki_link parser to default rules
|
|
# you can insert it some place you like
|
|
# but place matters, maybe 3 is not good
|
|
self.default_rules.insert(3, 'wiki_link')
|
|
|
|
def output_wiki_link(self, m):
|
|
text = m.group(1)
|
|
name, itype = text.split('|')
|
|
# you can create an custom render
|
|
# you can also return the html if you like
|
|
return self.renderer.wiki_link(name, itype)
|
|
|
|
class MyRenderer(mistune.Renderer):
|
|
def __init__(self, *args, **kwargs):
|
|
mistune.Renderer.__init__(self, *args, **kwargs)
|
|
self.TextChunck = []
|
|
self.ItemName = ""
|
|
self.PlanTable = ""
|
|
self.is_first_header = True
|
|
|
|
def add_last_paragraph(self):
|
|
"""
|
|
Used to add a text block before the next header or at the end of the document
|
|
"""
|
|
lText.append(self.TextChunck)
|
|
self.TextChunck = []
|
|
lItemName.append(self.ItemName)
|
|
self.ItemName = ""
|
|
lPlanTable.append(self.PlanTable)
|
|
self.PlanTable = ""
|
|
##
|
|
## Block Level
|
|
##
|
|
def block_code(self, code, lang):
|
|
text = formspec_escape(code.strip())
|
|
lines = text.split("\\n")
|
|
lines = [" " + item for item in lines]
|
|
self.TextChunck.extend(lines)
|
|
self.TextChunck.append("")
|
|
return ""
|
|
|
|
# ~ def block_quote(self, text):
|
|
# ~ print "block_quote", text
|
|
# ~ self.TextChunck.append("\n%s\n" % text)
|
|
# ~ return ""
|
|
|
|
def header(self, text, level, raw=None):
|
|
if not self.is_first_header:
|
|
self.add_last_paragraph()
|
|
self.is_first_header = False
|
|
lTitel.append("%u,%s" % (level, formspec_escape(text)))
|
|
lTocLinks.append({"level": level, "header": formspec_escape(text), "link": self.src_name})
|
|
return ""
|
|
|
|
def hrule(self):
|
|
self.TextChunck.append("\n----------------------------------------------------\n")
|
|
return ""
|
|
|
|
def paragraph(self, text):
|
|
lines = text.split("\\n") + [""]
|
|
self.TextChunck.extend(lines)
|
|
return ""
|
|
|
|
def list(self, body, ordered=True):
|
|
lines = body.split("\n")
|
|
self.TextChunck.extend(lines)
|
|
return ""
|
|
|
|
def list_item(self, text):
|
|
return " - %s\n" % text.strip()
|
|
##
|
|
## Span Level
|
|
##
|
|
def emphasis(self, text):
|
|
return "*%s*" % formspec_escape(text)
|
|
|
|
def double_emphasis(self, text):
|
|
return "*%s*" % formspec_escape(text)
|
|
|
|
def codespan(self, text):
|
|
return "'%s'" % formspec_escape(text)
|
|
|
|
def text(self, text):
|
|
return formspec_escape(text)
|
|
|
|
def link(self, link, title, content):
|
|
"""
|
|
Used for plans and images:
|
|
[myimage](/image/)
|
|
[myplan](/plan/)
|
|
"""
|
|
if link == "/image/":
|
|
self.ItemName = content
|
|
elif link == "/plan/":
|
|
self.PlanTable = content
|
|
return ""
|
|
|
|
def wiki_link(self, name, itype):
|
|
"""
|
|
Used for plans and images:
|
|
[myimage|image]
|
|
[myplan|plan]
|
|
"""
|
|
if itype == "image":
|
|
self.ItemName = name
|
|
elif itype == "plan":
|
|
self.PlanTable = name
|
|
return ""
|
|
|
|
def autolink(self, link, is_email=False):
|
|
return link
|
|
|
|
def linebreak(self):
|
|
return "\\n"
|
|
|
|
def newline(self):
|
|
return "\\n"
|
|
|
|
def inline_html(self, text):
|
|
print(text)
|
|
# ~
|
|
# ~ double_emphasis(text)
|
|
# ~ image(src, title, alt_text)
|
|
# ~ link(link, title, content)
|
|
# ~ strikethrough(text)
|
|
# ~ inline_html(text)
|
|
|
|
|
|
def parse_md_file(src_name, mod, manual):
|
|
print("Read Lua file '%s'" % src_name)
|
|
renderer = MyRenderer()
|
|
inline = WikiLinkInlineLexer(renderer)
|
|
# enable the feature
|
|
inline.enable_wiki_link()
|
|
md = mistune.Markdown(renderer=renderer, inline=inline)
|
|
md.renderer.src_name = src_name
|
|
md.render(open(src_name, 'r').read())
|
|
md.renderer.add_last_paragraph()
|
|
|
|
def gen_lua_file(dest_name):
|
|
print("Write Lua file '%s'" % dest_name)
|
|
lOut = ["%s.%s = {}\n\n" % (mod, manual)]
|
|
lOut.append(lua_table("%s.%s.aTitel" % (mod, manual), lTitel))
|
|
lOut.append(lua_text_table("%s.%s.aText" % (mod, manual), lText))
|
|
lOut.append(lua_table("%s.%s.aItemName" % (mod, manual), lItemName))
|
|
lOut.append(lua_table("%s.%s.aPlanTable" % (mod, manual), lPlanTable))
|
|
open(dest_name, "w").write("".join(lOut))
|
|
|
|
def gen_toc_md_file(dest_name, titel, level_range=[1,6]):
|
|
print("Write MD file '%s'" % dest_name)
|
|
lOut = ["# "+ titel]
|
|
lOut.append("")
|
|
for item in lTocLinks:
|
|
if item["level"] in range(*level_range):
|
|
list_item = " " * (item["level"] - level_range[0]) + "-"
|
|
link = "%s#%s" % (item["link"], header_escsape(item["header"]))
|
|
lOut.append("%s [%s](%s)" % (list_item, item["header"], link))
|
|
open(dest_name, "w").write("\n".join(lOut))
|
|
|
|
def gen_file_local_toc(dest_name, level_range=[1,6]):
|
|
lOut = []
|
|
for item in lTocLinks:
|
|
if item["level"] in range(*level_range):
|
|
list_item = " " * (item["level"] - level_range[0]) + "-"
|
|
link = "#%s" % (item["header"].replace(" ", "-").replace("\\", ""))
|
|
lOut.append("%s [%s](%s)" % (list_item, item["header"].replace("\\", ""), link))
|
|
open(dest_name, "w").write("\n".join(lOut))
|
|
|
|
########################### German #########################
|
|
mod = "techage"
|
|
manual = "manual_DE"
|
|
parse_md_file("./manual_DE.md", mod, manual)
|
|
parse_md_file("./manual_ta1_DE.md", mod, manual)
|
|
parse_md_file("./manual_ta2_DE.md", mod, manual)
|
|
parse_md_file("./manual_ta3_DE.md", mod, manual)
|
|
parse_md_file("./manual_ta4_DE.md", mod, manual)
|
|
parse_md_file("./manual_ta5_DE.md", mod, manual)
|
|
gen_lua_file("../doc/manual_DE.lua")
|
|
gen_toc_md_file("./toc_DE.md", "Inhaltsverzeichnis")
|
|
|
|
########################### English #########################
|
|
reset()
|
|
mod = "techage"
|
|
manual = "manual_EN"
|
|
parse_md_file("./manual_EN.md", mod, manual)
|
|
parse_md_file("./manual_ta1_EN.md", mod, manual)
|
|
parse_md_file("./manual_ta2_EN.md", mod, manual)
|
|
parse_md_file("./manual_ta3_EN.md", mod, manual)
|
|
parse_md_file("./manual_ta4_EN.md", mod, manual)
|
|
parse_md_file("./manual_ta5_EN.md", mod, manual)
|
|
gen_lua_file("../doc/manual_EN.lua")
|
|
gen_toc_md_file("./toc_EN.md", "Table of Contents")
|
|
|
|
########################### Lua Manual #########################
|
|
reset()
|
|
parse_md_file("./ta4_lua_controller_EN.md", mod, manual)
|
|
gen_file_local_toc("toc.txt", level_range=[2,4])
|
|
|
|
|