rust/util/export.py

127 lines
3.8 KiB
Python
Raw Normal View History

2016-07-12 07:11:18 -05:00
#!/usr/bin/env python
import os
import re
import json
level_re = re.compile(r'''(Forbid|Deny|Warn|Allow)''')
conf_re = re.compile(r'''define_Conf! {\n([^}]*)\n}''', re.MULTILINE)
confvar_re = re.compile(r'''/// Lint: (\w+). (.*).*\n *\("([^"]*)", (?:[^,]*), (.*) => (.*)\),''')
lint_subheadline = re.compile(r'''^\*\*([\w\s]+)[:?.!]\*\*(.*)''')
# TODO: actual logging
def warn(*args): print(args)
def debug(*args): print(args)
def info(*args): print(args)
def parse_path(p="clippy_lints/src"):
d = []
for f in os.listdir(p):
if f.endswith(".rs"):
parse_file(d, os.path.join(p, f))
return (d, parse_conf(p))
def parse_conf(p):
c = {}
with open(p + '/utils/conf.rs') as f:
f = f.read()
m = re.search(conf_re, f)
m = m.groups()[0]
m = re.findall(confvar_re, m)
for (lint, doc, name, default, ty) in m:
c[lint.lower()] = (name, ty, doc, default)
return c
def parseLintDef(level, comment, name):
lint = {}
lint['id'] = name
lint['level'] = level
lint['docs'] = {}
last_section = None
for line in comment:
if len(line.strip()) == 0:
continue
match = re.match(lint_subheadline, line)
if match:
last_section = match.groups()[0]
text = match and match.groups()[1] or line
if not last_section:
warn("Skipping comment line as it was not preceded by a heading")
debug("in lint `%s`, line `%s`" % name, line)
lint['docs'][last_section] = (lint['docs'].get(last_section, "") + "\n" + text).strip()
return lint
def parse_file(d, f):
last_comment = []
comment = True
with open(f) as rs:
for line in rs:
if comment:
if line.startswith("///"):
if line.startswith("/// "):
last_comment.append(line[4:])
else:
last_comment.append(line[3:])
elif line.startswith("declare_lint!"):
comment = False
deprecated = False
restriction = False
elif line.startswith("declare_restriction_lint!"):
comment = False
deprecated = False
restriction = True
elif line.startswith("declare_deprecated_lint!"):
comment = False
deprecated = True
else:
last_comment = []
if not comment:
l = line.strip()
m = re.search(r"pub\s+([A-Z_][A-Z_0-9]*)", l)
if m:
name = m.group(1).lower()
# Intentionally either a never looping or infinite loop
while not deprecated and not restriction:
m = re.search(level_re, line)
if m:
level = m.group(0)
break
line = next(rs)
if deprecated:
level = "Deprecated"
elif restriction:
level = "Allow"
info("found %s with level %s in %s" % (name, level, f))
d.append(parseLintDef(level, last_comment, name=name))
last_comment = []
comment = True
if "}" in l:
warn("Warning: Missing Lint-Name in", f)
comment = True
def main():
(lints, config) = parse_path()
info("got %s lints" % len(lints))
with open("util/gh-pages/lints.json", "w") as file:
json.dump(lints, file, indent=2)
info("wrote JSON for great justice")
if __name__ == "__main__":
main()