from __future__ import generators import py from py.__.misc import rest from py.__.apigen.linker import relpath import os pypkgdir = py.path.local(py.__file__).dirpath() mypath = py.magic.autopath().dirpath() TIMEOUT_URLOPEN = 5.0 Option = py.test.config.Option option = py.test.config.addoptions("documentation check options", Option('-R', '--checkremote', action="store_true", dest="checkremote", default=False, help="urlopen() remote links found in ReST text files.", ), Option('', '--forcegen', action="store_true", dest="forcegen", default=False, help="force generation of html files even if they appear up-to-date" ), ) def get_apigenpath(): from py.__.conftest import option path = os.environ.get('APIGENPATH') if path is None: path = option.apigenpath return pypkgdir.join(path, abs=True) def get_docpath(): from py.__.conftest import option path = os.environ.get('DOCPATH') if path is None: path = option.docpath return pypkgdir.join(path, abs=True) def get_apigen_relpath(): return relpath(get_docpath().strpath + '/', get_apigenpath().strpath + '/') def deindent(s, sep='\n'): leastspaces = -1 lines = s.split(sep) for line in lines: if not line.strip(): continue spaces = len(line) - len(line.lstrip()) if leastspaces == -1 or spaces < leastspaces: leastspaces = spaces if leastspaces == -1: return s for i, line in py.builtin.enumerate(lines): if not line.strip(): lines[i] = '' else: lines[i] = line[leastspaces:] return sep.join(lines) _initialized = False def checkdocutils(): global _initialized py.test.importorskip("docutils") if not _initialized: from py.__.rest import directive directive.register_linkrole('api', resolve_linkrole) directive.register_linkrole('source', resolve_linkrole) _initialized = True def restcheck(path): localpath = path if hasattr(path, 'localpath'): localpath = path.localpath checkdocutils() import docutils.utils try: cur = localpath for x in cur.parts(reverse=True): confrest = x.dirpath('confrest.py') if confrest.check(file=1): confrest = confrest.pyimport() project = confrest.Project() _checkskip(path, project.get_htmloutputpath(path)) project.process(path) break else: # defer to default processor _checkskip(path) rest.process(path) except KeyboardInterrupt: raise except docutils.utils.SystemMessage: # we assume docutils printed info on stdout py.test.fail("docutils processing failed, see captured stderr") def _checkskip(lpath, htmlpath=None): if not option.forcegen: lpath = py.path.local(lpath) if htmlpath is not None: htmlpath = py.path.local(htmlpath) if lpath.ext == '.txt': htmlpath = htmlpath or lpath.new(ext='.html') if htmlpath.check(file=1) and htmlpath.mtime() >= lpath.mtime(): py.test.skip("html file is up to date, use --forcegen to regenerate") #return [] # no need to rebuild class ReSTSyntaxTest(py.test.collect.Item): def runtest(self): mypath = self.fspath restcheck(py.path.svnwc(mypath)) class DoctestText(py.test.collect.Item): def runtest(self): s = self._normalize_linesep() l = [] prefix = '.. >>> ' mod = py.std.types.ModuleType(self.fspath.purebasename) skipchunk = False for line in deindent(s).split('\n'): stripped = line.strip() if skipchunk and line.startswith(skipchunk): print "skipping", line continue skipchunk = False if stripped.startswith(prefix): try: exec py.code.Source(stripped[len(prefix):]).compile() in \ mod.__dict__ except ValueError, e: if e.args and e.args[0] == "skipchunk": skipchunk = " " * (len(line) - len(line.lstrip())) else: raise else: l.append(line) docstring = "\n".join(l) mod.__doc__ = docstring failed, tot = py.compat.doctest.testmod(mod, verbose=1) if failed: py.test.fail("doctest %s: %s failed out of %s" %( self.fspath, failed, tot)) def _normalize_linesep(self): # XXX quite nasty... but it works (fixes win32 issues) s = self.fspath.read() linesep = '\n' if '\r' in s: if '\n' not in s: linesep = '\r' else: linesep = '\r\n' s = s.replace(linesep, '\n') return s class LinkCheckerMaker(py.test.collect.Collector): def collect(self): l = [] for call, tryfn, path, lineno in genlinkchecks(self.fspath): name = "%s:%d" %(tryfn, lineno) l.append( CheckLink(name, parent=self, args=(tryfn, path, lineno), callobj=call) ) return l class CheckLink(py.test.collect.Function): def repr_metainfo(self): return self.ReprMetaInfo(fspath=self.fspath, lineno=self._args[2], modpath="checklink: %s" % (self._args[0],)) def setup(self): pass def teardown(self): pass class DocfileTests(py.test.collect.File): DoctestText = DoctestText ReSTSyntaxTest = ReSTSyntaxTest LinkCheckerMaker = LinkCheckerMaker def collect(self): return [ self.ReSTSyntaxTest(self.fspath.basename, parent=self), self.LinkCheckerMaker("checklinks", self), self.DoctestText("doctest", self), ] # generating functions + args as single tests def genlinkchecks(path): for lineno, line in py.builtin.enumerate(path.readlines()): line = line.strip() if line.startswith('.. _'): if line.startswith('.. _`'): delim = '`:' else: delim = ':' l = line.split(delim, 1) if len(l) != 2: continue tryfn = l[1].strip() if tryfn.startswith('http:') or tryfn.startswith('https'): if option.checkremote: yield urlcheck, tryfn, path, lineno elif tryfn.startswith('webcal:'): continue else: i = tryfn.find('#') if i != -1: checkfn = tryfn[:i] else: checkfn = tryfn if checkfn.strip() and (1 or checkfn.endswith('.html')): yield localrefcheck, tryfn, path, lineno def urlcheck(tryfn, path, lineno): old = py.std.socket.getdefaulttimeout() py.std.socket.setdefaulttimeout(TIMEOUT_URLOPEN) try: try: print "trying remote", tryfn py.std.urllib2.urlopen(tryfn) finally: py.std.socket.setdefaulttimeout(old) except (py.std.urllib2.URLError, py.std.urllib2.HTTPError), e: if e.code in (401, 403): # authorization required, forbidden py.test.skip("%s: %s" %(tryfn, str(e))) else: py.test.fail("remote reference error %r in %s:%d\n%s" %( tryfn, path.basename, lineno+1, e)) def localrefcheck(tryfn, path, lineno): # assume it should be a file i = tryfn.find('#') if tryfn.startswith('javascript:'): return # don't check JS refs if i != -1: anchor = tryfn[i+1:] tryfn = tryfn[:i] else: anchor = '' fn = path.dirpath(tryfn) ishtml = fn.ext == '.html' fn = ishtml and fn.new(ext='.txt') or fn print "filename is", fn if not fn.check(): # not ishtml or not fn.check(): if not py.path.local(tryfn).check(): # the html could be there py.test.fail("reference error %r in %s:%d" %( tryfn, path.basename, lineno+1)) if anchor: source = unicode(fn.read(), 'latin1') source = source.lower().replace('-', ' ') # aehem anchor = anchor.replace('-', ' ') match2 = ".. _`%s`:" % anchor match3 = ".. _%s:" % anchor candidates = (anchor, match2, match3) print "candidates", repr(candidates) for line in source.split('\n'): line = line.strip() if line in candidates: break else: py.test.fail("anchor reference error %s#%s in %s:%d" %( tryfn, anchor, path.basename, lineno+1)) # ___________________________________________________________ # # hooking into py.test Directory collector's chain ... class DocDirectory(py.test.collect.Directory): DocfileTests = DocfileTests def collect(self): results = super(DocDirectory, self).collect() for x in self.fspath.listdir('*.txt', sort=True): results.append(self.DocfileTests(x, parent=self)) return results Directory = DocDirectory def resolve_linkrole(name, text, check=True): apigen_relpath = get_apigen_relpath() if name == 'api': if text == 'py': return ('py', apigen_relpath + 'api/index.html') else: assert text.startswith('py.'), ( 'api link "%s" does not point to the py package') % (text,) dotted_name = text if dotted_name.find('(') > -1: dotted_name = dotted_name[:text.find('(')] # remove pkg root path = dotted_name.split('.')[1:] dotted_name = '.'.join(path) obj = py if check: for chunk in path: try: obj = getattr(obj, chunk) except AttributeError: raise AssertionError( 'problem with linkrole :api:`%s`: can not resolve ' 'dotted name %s' % (text, dotted_name,)) return (text, apigen_relpath + 'api/%s.html' % (dotted_name,)) elif name == 'source': assert text.startswith('py/'), ('source link "%s" does not point ' 'to the py package') % (text,) relpath = '/'.join(text.split('/')[1:]) if check: pkgroot = py.__pkg__.getpath() abspath = pkgroot.join(relpath) assert pkgroot.join(relpath).check(), ( 'problem with linkrole :source:`%s`: ' 'path %s does not exist' % (text, relpath)) if relpath.endswith('/') or not relpath: relpath += 'index.html' else: relpath += '.html' return (text, apigen_relpath + 'source/%s' % (relpath,)) # legacy ReSTChecker = DocfileTests