mirror of
https://github.com/esphome/esphome-docs.git
synced 2024-11-04 09:10:02 +01:00
234810e6f5
* updates * updates * add verify missing descriptions * updates * schema handling updates * add action and condition * add more checks for missing props * updates * support typed schema * better inner props handling and mqtt * updates * clean up * add links to titles on markdown * remove spelling not used * format black * format black * Better titles
195 lines
4.8 KiB
Python
195 lines
4.8 KiB
Python
#!/usr/bin/env python
|
|
# reStructuredText (RST) to GitHub-flavored Markdown converter
|
|
|
|
import re
|
|
import sys
|
|
from docutils import core, nodes, writers
|
|
from urllib import parse
|
|
|
|
|
|
class Translator(nodes.NodeVisitor):
|
|
def __init__(self, base_url, document):
|
|
nodes.NodeVisitor.__init__(self, document)
|
|
self.output = ""
|
|
self.indent = 0
|
|
self.preserve_newlines = False
|
|
self.base_url = base_url
|
|
|
|
def write(self, text):
|
|
self.output += text.replace("\n", "\n" + " " * self.indent)
|
|
|
|
def visit_document(self, node):
|
|
pass
|
|
|
|
def depart_document(self, node):
|
|
pass
|
|
|
|
def visit_section(self, node):
|
|
pass
|
|
|
|
def depart_section(self, node):
|
|
# Skip all sections except the first one.
|
|
raise nodes.StopTraversal
|
|
|
|
def visit_title(self, node):
|
|
self.version = re.match(r"(\d+\.\d+\.\d+).*", node.children[0]).group(1)
|
|
raise nodes.SkipChildren
|
|
|
|
def visit_title_reference(self, node):
|
|
raise Exception(
|
|
f'Markdown conversion: Improper format or not supported feature in "{node.parent.rawsource}". Check "`"'
|
|
)
|
|
|
|
def depart_title(self, node):
|
|
pass
|
|
|
|
def visit_Text(self, node):
|
|
if not self.preserve_newlines:
|
|
node = node.replace("\n", " ")
|
|
self.write(node)
|
|
|
|
def depart_Text(self, node):
|
|
pass
|
|
|
|
def visit_bullet_list(self, node):
|
|
self.write("\n")
|
|
pass
|
|
|
|
def depart_bullet_list(self, node):
|
|
pass
|
|
|
|
def visit_list_item(self, node):
|
|
self.write("* ")
|
|
self.indent += 2
|
|
|
|
def depart_list_item(self, node):
|
|
self.indent -= 2
|
|
self.write("\n\n")
|
|
|
|
def visit_paragraph(self, node):
|
|
pass
|
|
|
|
def depart_paragraph(self, node):
|
|
pass
|
|
|
|
def visit_reference(self, node):
|
|
# if not is_github_ref(node):
|
|
self.write("[")
|
|
|
|
def depart_reference(self, node):
|
|
# if not is_github_ref(node):
|
|
# self.write('](' + node['refuri'] + ')')
|
|
refid = node.get("refid")
|
|
if refid:
|
|
self.write("](" + parse.urljoin(self.base_url, "#" + refid) + ")")
|
|
else:
|
|
refuri = node.get("refuri")
|
|
if refuri:
|
|
# if refuri.startswith('../'):
|
|
# refuri = parse.urljoin(self.base_url, refuri)
|
|
# self.write('](' + refuri + ')')
|
|
self.write("](" + parse.urljoin(self.base_url, refuri) + ")")
|
|
|
|
def visit_emphasis(self, node):
|
|
self.write("*")
|
|
pass
|
|
|
|
def depart_emphasis(self, node):
|
|
self.write("*")
|
|
pass
|
|
|
|
def visit_target(self, node):
|
|
pass
|
|
|
|
def depart_target(self, node):
|
|
pass
|
|
|
|
def visit_literal(self, node):
|
|
self.write("`")
|
|
|
|
def depart_literal(self, node):
|
|
self.write("`")
|
|
|
|
def visit_literal_block(self, node):
|
|
self.write("\n\n```")
|
|
if "c++" in node["classes"]:
|
|
self.write("c++")
|
|
self.write("\n")
|
|
self.preserve_newlines = True
|
|
|
|
def depart_literal_block(self, node):
|
|
self.write("\n```\n")
|
|
self.preserve_newlines = False
|
|
|
|
def visit_inline(self, node):
|
|
pass
|
|
|
|
def depart_inline(self, node):
|
|
pass
|
|
|
|
def visit_image(self, node):
|
|
self.write("![](" + node["uri"] + ")")
|
|
|
|
def depart_image(self, node):
|
|
pass
|
|
|
|
def write_row(self, row, widths):
|
|
for i, entry in enumerate(row):
|
|
text = entry[0][0] if len(entry) > 0 else ""
|
|
if i != 0:
|
|
self.write("|")
|
|
self.write("{:{}}".format(text, widths[i]))
|
|
self.write("\n")
|
|
|
|
def visit_table(self, node):
|
|
table = node.children[0]
|
|
colspecs = table[:-2]
|
|
thead = table[-2]
|
|
tbody = table[-1]
|
|
widths = [int(cs["colwidth"]) for cs in colspecs]
|
|
sep = "|".join(["-" * w for w in widths]) + "\n"
|
|
self.write("\n\n")
|
|
self.write_row(thead[0], widths)
|
|
self.write(sep)
|
|
for row in tbody:
|
|
self.write_row(row, widths)
|
|
raise nodes.SkipChildren
|
|
|
|
def depart_table(self, node):
|
|
pass
|
|
|
|
def visit_strong(self, node):
|
|
self.write("**")
|
|
pass
|
|
|
|
def depart_strong(self, node):
|
|
self.write("**")
|
|
pass
|
|
|
|
def visit_block_quote(self, node):
|
|
pass
|
|
|
|
def depart_block_quote(self, node):
|
|
pass
|
|
|
|
|
|
class MDWriter(writers.Writer):
|
|
"""GitHub-flavored markdown writer"""
|
|
|
|
supported = ("md",)
|
|
"""Formats this writer supports."""
|
|
|
|
def translate(self):
|
|
translator = Translator(self.document)
|
|
self.document.walkabout(translator)
|
|
self.output = (translator.output, translator.version)
|
|
|
|
|
|
def convert(rst_path):
|
|
"""Converts RST file to Markdown."""
|
|
return core.publish_file(source_path=rst_path, writer=MDWriter())
|
|
|
|
|
|
if __name__ == "__main__":
|
|
convert(sys.argv[1])
|