tor/scripts/maint/redox.py
Nick Mathewson 703ad69587 Deal with the aftermath of sorting contrib
This basically amounts to grepping for every file that mentioned
contrib and adjusting its references to refer to the right place.
2014-04-28 11:59:55 -04:00

229 lines
7.2 KiB
Python
Executable File

#!/usr/bin/python
#
# Copyright (c) 2008-2013, The Tor Project, Inc.
# See LICENSE for licensing information.
#
# Hi!
# I'm redox.py, the Tor redocumentation tool!
# I am a horrible hack!
# I read the output of doxygen from stderr, and add missing DOCDOC comments
# to tell you where documentation should go!
# To use me, edit the stuff below...
# ...and run 'make doxygen 2>doxygen.stderr' ...
# ...and run ./scripts/maint/redox.py < doxygen.stderr !
# I'll make a bunch of new files by adding missing DOCDOC comments to your
# source. Those files will have names like ./src/common/util.c.newdoc.
# You will want to look over the changes by hand before checking them in.
#
# So, here's your workflow:
#
# 0. Make sure you're running a bourne shell for the redirects below.
# 1. make doxygen 1>doxygen.stdout 2>doxygen.stderr.
# 2. grep Warning doxygen.stderr | grep -v 'is not documented' | less
# [This will tell you about all the bogus doxygen output you have]
# 3. python ./scripts/maint/redox.py <doxygen.stderr
# [This will make lots of .newdoc files with DOCDOC comments for
# whatever was missing documentation.]
# 4. Look over those .newdoc files, and see which docdoc comments you
# want to merge into the main file. If it's all good, just run
# "mv fname.c.newdoc fname.c". Otherwise, you'll need to merge
# the parts you like by hand.
# Which files should we ignore warning from? Mostly, these are external
# files that we've snarfed in from somebody else, whose C we do no intend
# to document for them.
SKIP_FILES = [ "OpenBSD_malloc_Linux.c",
"eventdns.c",
"eventdns.h",
"strlcat.c",
"strlcpy.c",
"sha256.c",
"sha256.h",
"aes.c",
"aes.h" ]
# What names of things never need javadoc
SKIP_NAME_PATTERNS = [ r'^.*_c_id$',
r'^.*_H_ID$' ]
# Which types of things should get DOCDOC comments added if they are
# missing documentation? Recognized types are in KINDS below.
ADD_DOCDOCS_TO_TYPES = [ 'function', 'type', 'typedef' ]
ADD_DOCDOCS_TO_TYPES += [ 'variable', ]
# ====================
# The rest of this should not need hacking.
import re
import sys
KINDS = [ "type", "field", "typedef", "define", "function", "variable",
"enumeration" ]
NODOC_LINE_RE = re.compile(r'^([^:]+):(\d+): (\w+): (.*) is not documented\.$')
THING_RE = re.compile(r'^Member ([a-zA-Z0-9_]+).*\((typedef|define|function|variable|enumeration|macro definition)\) of (file|class) ')
SKIP_NAMES = [re.compile(s) for s in SKIP_NAME_PATTERNS]
def parsething(thing):
"""I figure out what 'foobar baz in quux quum is not documented' means,
and return: the name of the foobar, and the kind of the foobar.
"""
if thing.startswith("Compound "):
tp, name = "type", thing.split()[1]
else:
m = THING_RE.match(thing)
if not m:
print thing, "???? Format didn't match."
return None, None
else:
name, tp, parent = m.groups()
if parent == 'class':
if tp == 'variable' or tp == 'function':
tp = 'field'
return name, tp
def read():
"""I snarf doxygen stderr from stdin, and parse all the "foo has no
documentation messages. I return a map from filename to lists
of tuples of (alleged line number, name of thing, kind of thing)
"""
errs = {}
for line in sys.stdin:
m = NODOC_LINE_RE.match(line)
if m:
file, line, tp, thing = m.groups()
assert tp.lower() == 'warning'
name, kind = parsething(thing)
errs.setdefault(file, []).append((int(line), name, kind))
return errs
def findline(lines, lineno, ident):
"""Given a list of all the lines in the file (adjusted so 1-indexing works),
a line number that ident is alledgedly on, and ident, I figure out
the line where ident was really declared."""
lno = lineno
for lineno in xrange(lineno, 0, -1):
try:
if ident in lines[lineno]:
return lineno
except IndexError:
continue
return None
FUNC_PAT = re.compile(r"^[A-Za-z0-9_]+\(")
def hascomment(lines, lineno, kind):
"""I return true if it looks like there's already a good comment about
the thing on lineno of lines of type kind. """
if "*/" in lines[lineno-1]:
return True
if kind == 'function' and FUNC_PAT.match(lines[lineno]):
if "*/" in lines[lineno-2]:
return True
return False
def hasdocdoc(lines, lineno, kind):
"""I return true if it looks like there's already a docdoc comment about
the thing on lineno of lines of type kind."""
try:
if "DOCDOC" in lines[lineno]:
return True
except IndexError:
pass
try:
if "DOCDOC" in lines[lineno-1]:
return True
except IndexError:
pass
if kind == 'function' and FUNC_PAT.match(lines[lineno]):
if "DOCDOC" in lines[lineno-2]:
return True
return False
def checkf(fn, errs):
"""I go through the output of read() for a single file, and build a list
of tuples of things that want DOCDOC comments. Each tuple has:
the line number where the comment goes; the kind of thing; its name.
"""
for skip in SKIP_FILES:
if fn.endswith(skip):
print "Skipping",fn
return
comments = []
lines = [ None ]
try:
lines.extend( open(fn, 'r').readlines() )
except IOError:
return
for line, name, kind in errs:
if any(pat.match(name) for pat in SKIP_NAMES):
continue
if kind not in ADD_DOCDOCS_TO_TYPES:
continue
ln = findline(lines, line, name)
if ln == None:
print "Couldn't find the definition of %s allegedly on %s of %s"%(
name, line, fn)
else:
if hasdocdoc(lines, line, kind):
# print "Has a DOCDOC"
# print fn, line, name, kind
# print "\t",lines[line-2],
# print "\t",lines[line-1],
# print "\t",lines[line],
# print "-------"
pass
else:
if kind == 'function' and FUNC_PAT.match(lines[ln]):
ln = ln - 1
comments.append((ln, kind, name))
return comments
def applyComments(fn, entries):
"""I apply lots of comments to the file in fn, making a new .newdoc file.
"""
N = 0
lines = [ None ]
try:
lines.extend( open(fn, 'r').readlines() )
except IOError:
return
# Process the comments in reverse order by line number, so that
# the line numbers for the ones we haven't added yet remain valid
# until we add them. Standard trick.
entries.sort()
entries.reverse()
for ln, kind, name in entries:
lines.insert(ln, "/* DOCDOC %s */\n"%name)
N += 1
outf = open(fn+".newdoc", 'w')
for line in lines[1:]:
outf.write(line)
outf.close()
print "Added %s DOCDOCs to %s" %(N, fn)
e = read()
for fn, errs in e.iteritems():
print `(fn, errs)`
comments = checkf(fn, errs)
if comments:
applyComments(fn, comments)