mirror of
https://github.com/esphome/esphome-docs.git
synced 2024-12-27 17:37:45 +01:00
275 lines
8.7 KiB
Python
275 lines
8.7 KiB
Python
import csv
|
|
from itertools import zip_longest
|
|
import os
|
|
import re
|
|
import string
|
|
|
|
from docutils import nodes, utils
|
|
from docutils.parsers.rst import directives
|
|
from docutils.parsers.rst.directives.tables import Table
|
|
|
|
|
|
def libpr_role(name, rawtext, text, lineno, inliner, options=None, content=None):
|
|
ref = "https://github.com/esphome/esphome-core/pull/{}".format(text)
|
|
return [make_link_node(rawtext, "core#{}".format(text), ref, options)], []
|
|
|
|
|
|
def yamlpr_role(name, rawtext, text, lineno, inliner, options=None, content=None):
|
|
ref = "https://github.com/esphome/esphome/pull/{}".format(text)
|
|
return [make_link_node(rawtext, "esphome#{}".format(text), ref, options)], []
|
|
|
|
|
|
def docspr_role(name, rawtext, text, lineno, inliner, options=None, content=None):
|
|
ref = "https://github.com/esphome/esphome-docs/pull/{}".format(text)
|
|
return [make_link_node(rawtext, "docs#{}".format(text), ref, options)], []
|
|
|
|
|
|
def ghuser_role(name, rawtext, text, lineno, inliner, options=None, content=None):
|
|
ref = "https://github.com/{}".format(text)
|
|
return [make_link_node(rawtext, "@{}".format(text), ref, options)], []
|
|
|
|
|
|
value_re = re.compile(r"^(.*)\s*<(.*)>$")
|
|
DOXYGEN_LOOKUP = {}
|
|
for s in string.ascii_lowercase + string.digits:
|
|
DOXYGEN_LOOKUP[s] = s
|
|
for s in string.ascii_uppercase:
|
|
DOXYGEN_LOOKUP[s] = "_{}".format(s.lower())
|
|
DOXYGEN_LOOKUP[":"] = "_1"
|
|
DOXYGEN_LOOKUP["_"] = "__"
|
|
DOXYGEN_LOOKUP["."] = "_8"
|
|
|
|
|
|
def split_text_value(value):
|
|
match = value_re.match(value)
|
|
if match is None:
|
|
return None, value
|
|
return match.group(1), match.group(2)
|
|
|
|
|
|
def encode_doxygen(value):
|
|
value = value.split("/")[-1]
|
|
try:
|
|
return "".join(DOXYGEN_LOOKUP[s] for s in value)
|
|
except KeyError:
|
|
raise ValueError("Unknown character in doxygen string! '{}'".format(value))
|
|
|
|
|
|
def apiref_role(name, rawtext, text, lineno, inliner, options=None, content=None):
|
|
text, value = split_text_value(text)
|
|
if text is None:
|
|
text = "API Reference"
|
|
ref = "/api/{}.html".format(encode_doxygen(value))
|
|
return [make_link_node(rawtext, text, ref, options)], []
|
|
|
|
|
|
def apiclass_role(name, rawtext, text, lineno, inliner, options=None, content=None):
|
|
text, value = split_text_value(text)
|
|
if text is None:
|
|
text = value
|
|
ref = "/api/classesphome_1_1{}.html".format(encode_doxygen(value))
|
|
return [make_link_node(rawtext, text, ref, options)], []
|
|
|
|
|
|
def apistruct_role(name, rawtext, text, lineno, inliner, options=None, content=None):
|
|
text, value = split_text_value(text)
|
|
if text is None:
|
|
text = value
|
|
ref = "/api/structesphome_1_1{}.html".format(encode_doxygen(value))
|
|
return [make_link_node(rawtext, text, ref, options)], []
|
|
|
|
|
|
def ghedit_role(name, rawtext, text, lineno, inliner, options=None, content=None):
|
|
path = os.path.relpath(
|
|
inliner.document.current_source, inliner.document.settings.env.app.srcdir
|
|
)
|
|
ref = "https://github.com/esphome/esphome-docs/blob/current/{}".format(path)
|
|
return [make_link_node(rawtext, "Edit this page on GitHub", ref, options)], []
|
|
|
|
|
|
def make_link_node(rawtext, text, ref, options=None):
|
|
options = options or {}
|
|
node = nodes.reference(rawtext, utils.unescape(text), refuri=ref, **options)
|
|
return node
|
|
|
|
|
|
# https://stackoverflow.com/a/3415150/8924614
|
|
def grouper(n, iterable, fillvalue=None):
|
|
"""Pythonic way to iterate over sequence, 4 items at a time.
|
|
|
|
grouper(3, 'ABCDEFG', 'x') --> ABC DEF Gxx
|
|
"""
|
|
args = [iter(iterable)] * n
|
|
return zip_longest(fillvalue=fillvalue, *args)
|
|
|
|
|
|
# Based on https://www.slideshare.net/doughellmann/better-documentation-through-automation-creating-docutils-sphinx-extensions
|
|
class ImageTableDirective(Table):
|
|
|
|
option_spec = {
|
|
"columns": directives.positive_int,
|
|
}
|
|
|
|
def run(self):
|
|
cols = self.options.get("columns", 3)
|
|
|
|
items = []
|
|
|
|
data = list(csv.reader(self.content))
|
|
for row in data:
|
|
if not row:
|
|
continue
|
|
name, page, image = row[0:3]
|
|
link = page.strip()
|
|
if link.startswith("http"):
|
|
pass
|
|
else:
|
|
if not link.startswith("/"):
|
|
link = "/{}".format(link)
|
|
if ".html" not in link:
|
|
link += ".html"
|
|
items.append(
|
|
{
|
|
"name": name.strip(),
|
|
"link": link,
|
|
"image": "/images/{}".format(image.strip()),
|
|
"category": row[3] if len(row) >= 4 else None
|
|
}
|
|
)
|
|
|
|
title, messages = self.make_title()
|
|
table = nodes.table()
|
|
table["classes"].append("table-center")
|
|
table["classes"].append("colwidths-given")
|
|
|
|
# Set up column specifications based on widths
|
|
tgroup = nodes.tgroup(cols=cols)
|
|
table += tgroup
|
|
tgroup.extend(nodes.colspec(colwidth=1) for _ in range(cols))
|
|
|
|
tbody = nodes.tbody()
|
|
tgroup += tbody
|
|
rows = []
|
|
for value in grouper(cols, items):
|
|
trow = nodes.row()
|
|
for cell in value:
|
|
entry = nodes.entry()
|
|
if cell is None:
|
|
entry += nodes.paragraph()
|
|
trow += entry
|
|
continue
|
|
name = cell["name"]
|
|
link = cell["link"]
|
|
image = cell["image"]
|
|
reference_node = nodes.reference(refuri=link)
|
|
img = nodes.image(uri=directives.uri(image), alt=name)
|
|
img["classes"].append("component-image")
|
|
reference_node += img
|
|
para = nodes.paragraph()
|
|
para += reference_node
|
|
entry += para
|
|
trow += entry
|
|
rows.append(trow)
|
|
|
|
trow = nodes.row()
|
|
for cell in value:
|
|
entry = nodes.entry()
|
|
if cell is None:
|
|
entry += nodes.paragraph()
|
|
trow += entry
|
|
continue
|
|
name = cell["name"]
|
|
link = cell["link"]
|
|
ref = nodes.reference(name, name, refuri=link)
|
|
para = nodes.paragraph()
|
|
para += ref
|
|
entry += para
|
|
cat_text = cell["category"]
|
|
if cat_text:
|
|
cat = nodes.paragraph(text=cat_text)
|
|
entry += cat
|
|
trow += entry
|
|
rows.append(trow)
|
|
tbody.extend(rows)
|
|
|
|
self.add_name(table)
|
|
if title:
|
|
table.insert(0, title)
|
|
|
|
return [table] + messages
|
|
|
|
|
|
class PinTableDirective(Table):
|
|
option_spec = {}
|
|
|
|
def run(self):
|
|
items = []
|
|
|
|
data = list(csv.reader(self.content))
|
|
for row in data:
|
|
if not row:
|
|
continue
|
|
if len(row) == 3:
|
|
items.append((row[0], row[1], True))
|
|
else:
|
|
items.append((row[0], row[1], False))
|
|
|
|
col_widths = self.get_column_widths(2)
|
|
title, messages = self.make_title()
|
|
table = nodes.table()
|
|
|
|
# Set up column specifications based on widths
|
|
tgroup = nodes.tgroup(cols=2)
|
|
table += tgroup
|
|
tgroup.extend(nodes.colspec(colwidth=col_width) for col_width in col_widths)
|
|
|
|
thead = nodes.thead()
|
|
tgroup += thead
|
|
trow = nodes.row()
|
|
thead += trow
|
|
trow.extend(
|
|
nodes.entry(h, nodes.paragraph(text=h)) for h in ("Pin", "Function")
|
|
)
|
|
|
|
tbody = nodes.tbody()
|
|
tgroup += tbody
|
|
for name, func, important in items:
|
|
trow = nodes.row()
|
|
entry = nodes.entry()
|
|
para = nodes.paragraph()
|
|
para += nodes.literal(text=name)
|
|
entry += para
|
|
trow += entry
|
|
|
|
entry = nodes.entry()
|
|
if important:
|
|
para = nodes.paragraph()
|
|
para += nodes.strong(text=func)
|
|
else:
|
|
para = nodes.paragraph(text=func)
|
|
entry += para
|
|
trow += entry
|
|
tbody += trow
|
|
|
|
self.add_name(table)
|
|
if title:
|
|
table.insert(0, title)
|
|
|
|
return [table] + messages
|
|
|
|
|
|
def setup(app):
|
|
app.add_role("libpr", libpr_role)
|
|
app.add_role("corepr", libpr_role)
|
|
app.add_role("yamlpr", yamlpr_role)
|
|
app.add_role("esphomepr", yamlpr_role)
|
|
app.add_role("docspr", docspr_role)
|
|
app.add_role("ghuser", ghuser_role)
|
|
app.add_role("apiref", apiref_role)
|
|
app.add_role("apiclass", apiclass_role)
|
|
app.add_role("apistruct", apistruct_role)
|
|
app.add_role("ghedit", ghedit_role)
|
|
app.add_directive("imgtable", ImageTableDirective)
|
|
app.add_directive("pintable", PinTableDirective)
|
|
return {"version": "1.0.0", "parallel_read_safe": True, "parallel_write_safe": True}
|