You cannot select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
420 lines
16 KiB
Python
420 lines
16 KiB
Python
#! /usr/bin/env python
|
|
|
|
from argparse import ArgumentParser
|
|
from docutils.core import publish_file
|
|
import copy
|
|
import fileinput
|
|
import glob
|
|
import os
|
|
import re
|
|
import shutil
|
|
import subprocess
|
|
import sys
|
|
import yaml
|
|
|
|
os.chdir(os.path.dirname(os.path.abspath(__file__)))
|
|
|
|
stylesheets = {
|
|
"stylesheet_path": ["basic.css", "nature.css", "codehighlight.css"]
|
|
}
|
|
|
|
VERBOSE = False
|
|
|
|
"""
|
|
Read a RST file and replace titles with a different title level if required.
|
|
Args:
|
|
filename: The name of the file being read (for debugging)
|
|
file_stream: The open file stream to read from.
|
|
title_level: The integer which determines the offset to *start* from.
|
|
title_styles: An array of characters detailing the right title styles to use
|
|
e.g. ["=", "-", "~", "+"]
|
|
Returns:
|
|
string: The file contents with titles adjusted.
|
|
Example:
|
|
Assume title_styles = ["=", "-", "~", "+"], title_level = 1, and the file
|
|
when read line-by-line encounters the titles "===", "---", "---", "===", "---".
|
|
This function will bump every title encountered down a sub-heading e.g.
|
|
"=" to "-" and "-" to "~" because title_level = 1, so the output would be
|
|
"---", "~~~", "~~~", "---", "~~~". There is no bumping "up" a title level.
|
|
"""
|
|
def load_with_adjusted_titles(filename, file_stream, title_level, title_styles):
|
|
rst_lines = []
|
|
title_chars = "".join(title_styles)
|
|
title_regex = re.compile("^[" + re.escape(title_chars) + "]{3,}$")
|
|
|
|
prev_line_title_level = 0 # We expect the file to start with '=' titles
|
|
file_offset = None
|
|
prev_non_title_line = None
|
|
for i, line in enumerate(file_stream, 1):
|
|
# ignore anything which isn't a title (e.g. '===============')
|
|
if not title_regex.match(line):
|
|
rst_lines.append(line)
|
|
prev_non_title_line = line
|
|
continue
|
|
# The title underline must match at a minimum the length of the title
|
|
if len(prev_non_title_line) > len(line):
|
|
rst_lines.append(line)
|
|
prev_non_title_line = line
|
|
continue
|
|
|
|
line_title_style = line[0]
|
|
line_title_level = title_styles.index(line_title_style)
|
|
|
|
# Not all files will start with "===" and we should be flexible enough
|
|
# to allow that. The first title we encounter sets the "file offset"
|
|
# which is added to the title_level desired.
|
|
if file_offset is None:
|
|
file_offset = line_title_level
|
|
if file_offset != 0:
|
|
logv((" WARNING: %s starts with a title style of '%s' but '%s' " +
|
|
"is preferable.") % (filename, line_title_style, title_styles[0]))
|
|
|
|
# Sanity checks: Make sure that this file is obeying the title levels
|
|
# specified and bail if it isn't.
|
|
# The file is allowed to go 1 deeper or any number shallower
|
|
if prev_line_title_level - line_title_level < -1:
|
|
raise Exception(
|
|
("File '%s' line '%s' has a title " +
|
|
"style '%s' which doesn't match one of the " +
|
|
"allowed title styles of %s because the " +
|
|
"title level before this line was '%s'") %
|
|
(filename, (i + 1), line_title_style, title_styles,
|
|
title_styles[prev_line_title_level])
|
|
)
|
|
prev_line_title_level = line_title_level
|
|
|
|
adjusted_level = (
|
|
title_level + line_title_level - file_offset
|
|
)
|
|
|
|
# Sanity check: Make sure we can bump down the title and we aren't at the
|
|
# lowest level already
|
|
if adjusted_level >= len(title_styles):
|
|
raise Exception(
|
|
("Files '%s' line '%s' has a sub-title level too low and it " +
|
|
"cannot be adjusted to fit. You can add another level to the " +
|
|
"'title_styles' key in targets.yaml to fix this.") %
|
|
(filename, (i + 1))
|
|
)
|
|
|
|
if adjusted_level == line_title_level:
|
|
# no changes required
|
|
rst_lines.append(line)
|
|
continue
|
|
|
|
# Adjusting line levels
|
|
logv(
|
|
"File: %s Adjusting %s to %s because file_offset=%s title_offset=%s" %
|
|
(filename, line_title_style, title_styles[adjusted_level],
|
|
file_offset, title_level)
|
|
)
|
|
rst_lines.append(line.replace(
|
|
line_title_style,
|
|
title_styles[adjusted_level]
|
|
))
|
|
|
|
return "".join(rst_lines)
|
|
|
|
|
|
def get_rst(file_info, title_level, title_styles, spec_dir, adjust_titles):
|
|
# string are file paths to RST blobs
|
|
if isinstance(file_info, basestring):
|
|
log("%s %s" % (">" * (1 + title_level), file_info))
|
|
with open(os.path.join(spec_dir, file_info), "r") as f:
|
|
rst = None
|
|
if adjust_titles:
|
|
rst = load_with_adjusted_titles(
|
|
file_info, f, title_level, title_styles
|
|
)
|
|
else:
|
|
rst = f.read()
|
|
if rst[-2:] != "\n\n":
|
|
raise Exception(
|
|
("File %s should end with TWO new-line characters to ensure " +
|
|
"file concatenation works correctly.") % (file_info,)
|
|
)
|
|
return rst
|
|
# dicts look like {0: filepath, 1: filepath} where the key is the title level
|
|
elif isinstance(file_info, dict):
|
|
levels = sorted(file_info.keys())
|
|
rst = []
|
|
for l in levels:
|
|
rst.append(get_rst(file_info[l], l, title_styles, spec_dir, adjust_titles))
|
|
return "".join(rst)
|
|
# lists are multiple file paths e.g. [filepath, filepath]
|
|
elif isinstance(file_info, list):
|
|
rst = []
|
|
for f in file_info:
|
|
rst.append(get_rst(f, title_level, title_styles, spec_dir, adjust_titles))
|
|
return "".join(rst)
|
|
raise Exception(
|
|
"The following 'file' entry in this target isn't a string, list or dict. " +
|
|
"It really really should be. Entry: %s" % (file_info,)
|
|
)
|
|
|
|
|
|
def build_spec(target, out_filename):
|
|
with open(out_filename, "wb") as outfile:
|
|
for file_info in target["files"]:
|
|
section = get_rst(
|
|
file_info=file_info,
|
|
title_level=0,
|
|
title_styles=target["title_styles"],
|
|
spec_dir="../specification/",
|
|
adjust_titles=True
|
|
)
|
|
outfile.write(section)
|
|
|
|
|
|
"""
|
|
Replaces relative title styles with actual title styles.
|
|
|
|
The templating system has no idea what the right title style is when it produces
|
|
RST because it depends on the build target. As a result, it uses relative title
|
|
styles defined in targets.yaml to say "down a level, up a level, same level".
|
|
|
|
This function replaces these relative titles with actual title styles from the
|
|
array in targets.yaml.
|
|
"""
|
|
def fix_relative_titles(target, filename, out_filename):
|
|
title_styles = target["title_styles"]
|
|
relative_title_chars = [
|
|
target["relative_title_styles"]["subtitle"],
|
|
target["relative_title_styles"]["sametitle"],
|
|
target["relative_title_styles"]["supertitle"]
|
|
]
|
|
relative_title_matcher = re.compile(
|
|
"^[" + re.escape("".join(relative_title_chars)) + "]{3,}$"
|
|
)
|
|
title_matcher = re.compile(
|
|
"^[" + re.escape("".join(title_styles)) + "]{3,}$"
|
|
)
|
|
current_title_style = None
|
|
with open(filename, "r") as infile:
|
|
with open(out_filename, "w") as outfile:
|
|
for line in infile.readlines():
|
|
if not relative_title_matcher.match(line):
|
|
if title_matcher.match(line):
|
|
current_title_style = line[0]
|
|
outfile.write(line)
|
|
continue
|
|
line_char = line[0]
|
|
replacement_char = None
|
|
current_title_level = title_styles.index(current_title_style)
|
|
if line_char == target["relative_title_styles"]["subtitle"]:
|
|
if (current_title_level + 1) == len(title_styles):
|
|
raise Exception(
|
|
"Encountered sub-title line style but we can't go " +
|
|
"any lower."
|
|
)
|
|
replacement_char = title_styles[current_title_level + 1]
|
|
elif line_char == target["relative_title_styles"]["sametitle"]:
|
|
replacement_char = title_styles[current_title_level]
|
|
elif line_char == target["relative_title_styles"]["supertitle"]:
|
|
if (current_title_level - 1) < 0:
|
|
raise Exception(
|
|
"Encountered super-title line style but we can't go " +
|
|
"any higher."
|
|
)
|
|
replacement_char = title_styles[current_title_level - 1]
|
|
else:
|
|
raise Exception(
|
|
"Unknown relative line char %s" % (line_char,)
|
|
)
|
|
|
|
outfile.write(
|
|
line.replace(line_char, replacement_char)
|
|
)
|
|
|
|
|
|
|
|
def rst2html(i, o):
|
|
with open(i, "r") as in_file:
|
|
with open(o, "w") as out_file:
|
|
publish_file(
|
|
source=in_file,
|
|
destination=out_file,
|
|
reader_name="standalone",
|
|
parser_name="restructuredtext",
|
|
writer_name="html",
|
|
settings_overrides=stylesheets
|
|
)
|
|
|
|
|
|
def run_through_template(input, set_verbose):
|
|
tmpfile = './tmp/output'
|
|
try:
|
|
with open(tmpfile, 'w') as out:
|
|
args = [
|
|
'python', 'build.py',
|
|
"-i", "matrix_templates",
|
|
"-o", "../scripts/tmp",
|
|
"../scripts/"+input
|
|
]
|
|
if set_verbose:
|
|
args.insert(2, "-v")
|
|
log("EXEC: %s" % " ".join(args))
|
|
log(" ==== build.py output ==== ")
|
|
print subprocess.check_output(
|
|
args,
|
|
stderr=out,
|
|
cwd="../templating"
|
|
)
|
|
except subprocess.CalledProcessError as e:
|
|
print e.output
|
|
with open(tmpfile, 'r') as f:
|
|
sys.stderr.write(f.read() + "\n")
|
|
raise
|
|
|
|
|
|
"""
|
|
Extract and resolve groups for the given target in the given targets listing.
|
|
Args:
|
|
targets_listing (str): The path to a YAML file containing a list of targets
|
|
target_name (str): The name of the target to extract from the listings.
|
|
Returns:
|
|
dict: Containing "filees" (a list of file paths), "relative_title_styles"
|
|
(a dict of relative style keyword to title character) and "title_styles"
|
|
(a list of characters which represent the global title style to follow,
|
|
with the top section title first, the second section second, and so on.)
|
|
"""
|
|
def get_build_target(targets_listing, target_name):
|
|
build_target = {
|
|
"title_styles": [],
|
|
"relative_title_styles": {},
|
|
"files": []
|
|
}
|
|
with open(targets_listing, "r") as targ_file:
|
|
all_targets = yaml.load(targ_file.read())
|
|
|
|
build_target["title_styles"] = all_targets["title_styles"]
|
|
build_target["relative_title_styles"] = all_targets["relative_title_styles"]
|
|
target = all_targets["targets"].get(target_name)
|
|
if not target:
|
|
raise Exception(
|
|
"No target by the name '" + target_name + "' exists in '" +
|
|
targets_listing + "'."
|
|
)
|
|
if not isinstance(target.get("files"), list):
|
|
raise Exception(
|
|
"Found target but 'files' key is not a list."
|
|
)
|
|
|
|
def get_group(group_id, depth):
|
|
group_name = group_id[len("group:"):]
|
|
group = all_targets.get("groups", {}).get(group_name)
|
|
if not group:
|
|
raise Exception(
|
|
"Tried to find group '%s' but it doesn't exist." % group_name
|
|
)
|
|
if not isinstance(group, list):
|
|
raise Exception(
|
|
"Expected group '%s' to be a list but it isn't." % group_name
|
|
)
|
|
# deep copy so changes to depths don't contaminate multiple uses of this group
|
|
group = copy.deepcopy(group)
|
|
# swap relative depths for absolute ones
|
|
for i, entry in enumerate(group):
|
|
if isinstance(entry, dict):
|
|
group[i] = {
|
|
(rel_depth + depth): v for (rel_depth, v) in entry.items()
|
|
}
|
|
return group
|
|
|
|
resolved_files = []
|
|
for file_entry in target["files"]:
|
|
# file_entry is a group id
|
|
if isinstance(file_entry, basestring) and file_entry.startswith("group:"):
|
|
group = get_group(file_entry, 0)
|
|
# The group may be resolved to a list of file entries, in which case
|
|
# we want to extend the array to insert each of them rather than
|
|
# insert the entire list as a single element (which is what append does)
|
|
if isinstance(group, list):
|
|
resolved_files.extend(group)
|
|
else:
|
|
resolved_files.append(group)
|
|
# file_entry is a dict which has more file entries as values
|
|
elif isinstance(file_entry, dict):
|
|
resolved_entry = {}
|
|
for (depth, entry) in file_entry.iteritems():
|
|
if not isinstance(entry, basestring):
|
|
raise Exception(
|
|
"Double-nested depths are not supported. Entry: %s" % (file_entry,)
|
|
)
|
|
if entry.startswith("group:"):
|
|
resolved_entry[depth] = get_group(entry, depth)
|
|
else:
|
|
# map across without editing (e.g. normal file path)
|
|
resolved_entry[depth] = entry
|
|
resolved_files.append(resolved_entry)
|
|
continue
|
|
# file_entry is just a plain ol' file path
|
|
else:
|
|
resolved_files.append(file_entry)
|
|
build_target["files"] = resolved_files
|
|
return build_target
|
|
|
|
def log(line):
|
|
print "gendoc: %s" % line
|
|
|
|
def logv(line):
|
|
if VERBOSE:
|
|
print "gendoc:V: %s" % line
|
|
|
|
|
|
def prepare_env():
|
|
try:
|
|
os.makedirs("./gen")
|
|
except OSError:
|
|
pass
|
|
try:
|
|
os.makedirs("./tmp")
|
|
except OSError:
|
|
pass
|
|
|
|
|
|
def cleanup_env():
|
|
shutil.rmtree("./tmp")
|
|
|
|
|
|
def main(target_name, keep_intermediates):
|
|
prepare_env()
|
|
log("Building spec [target=%s]" % target_name)
|
|
target = get_build_target("../specification/targets.yaml", target_name)
|
|
build_spec(target=target, out_filename="tmp/templated_spec.rst")
|
|
run_through_template("tmp/templated_spec.rst", VERBOSE)
|
|
fix_relative_titles(
|
|
target=target, filename="tmp/templated_spec.rst",
|
|
out_filename="tmp/full_spec.rst"
|
|
)
|
|
shutil.copy("../supporting-docs/howtos/client-server.rst", "tmp/howto.rst")
|
|
run_through_template("tmp/howto.rst", False) # too spammy to mark -v on this
|
|
rst2html("tmp/full_spec.rst", "gen/specification.html")
|
|
rst2html("tmp/howto.rst", "gen/howtos.html")
|
|
if not keep_intermediates:
|
|
cleanup_env()
|
|
|
|
|
|
if __name__ == '__main__':
|
|
parser = ArgumentParser(
|
|
"gendoc.py - Generate the Matrix specification as HTML to the gen/ folder."
|
|
)
|
|
parser.add_argument(
|
|
"--nodelete", "-n", action="store_true",
|
|
help="Do not delete intermediate files. They will be found in tmp/"
|
|
)
|
|
parser.add_argument(
|
|
"--target", "-t", default="main",
|
|
help="Specify the build target to build from specification/targets.yaml"
|
|
)
|
|
parser.add_argument(
|
|
"--verbose", "-v", action="store_true",
|
|
help="Turn on verbose mode."
|
|
)
|
|
args = parser.parse_args()
|
|
if not args.target:
|
|
parser.print_help()
|
|
sys.exit(1)
|
|
VERBOSE = args.verbose
|
|
main(args.target, args.nodelete)
|