diff options
Diffstat (limited to 'eggs/py-1.4.0-py2.6.egg/py/_path/svnwc.py')
-rw-r--r-- | eggs/py-1.4.0-py2.6.egg/py/_path/svnwc.py | 1231 |
1 files changed, 0 insertions, 1231 deletions
diff --git a/eggs/py-1.4.0-py2.6.egg/py/_path/svnwc.py b/eggs/py-1.4.0-py2.6.egg/py/_path/svnwc.py deleted file mode 100644 index 72769db..0000000 --- a/eggs/py-1.4.0-py2.6.egg/py/_path/svnwc.py +++ /dev/null @@ -1,1231 +0,0 @@ -""" -svn-Command based Implementation of a Subversion WorkingCopy Path. - - SvnWCCommandPath is the main class. - -""" - -import os, sys, time, re, calendar -import py -import subprocess -from py._path import common - -#----------------------------------------------------------- -# Caching latest repository revision and repo-paths -# (getting them is slow with the current implementations) -# -# XXX make mt-safe -#----------------------------------------------------------- - -class cache: - proplist = {} - info = {} - entries = {} - prop = {} - -class RepoEntry: - def __init__(self, url, rev, timestamp): - self.url = url - self.rev = rev - self.timestamp = timestamp - - def __str__(self): - return "repo: %s;%s %s" %(self.url, self.rev, self.timestamp) - -class RepoCache: - """ The Repocache manages discovered repository paths - and their revisions. If inside a timeout the cache - will even return the revision of the root. - """ - timeout = 20 # seconds after which we forget that we know the last revision - - def __init__(self): - self.repos = [] - - def clear(self): - self.repos = [] - - def put(self, url, rev, timestamp=None): - if rev is None: - return - if timestamp is None: - timestamp = time.time() - - for entry in self.repos: - if url == entry.url: - entry.timestamp = timestamp - entry.rev = rev - #print "set repo", entry - break - else: - entry = RepoEntry(url, rev, timestamp) - self.repos.append(entry) - #print "appended repo", entry - - def get(self, url): - now = time.time() - for entry in self.repos: - if url.startswith(entry.url): - if now < entry.timestamp + self.timeout: - #print "returning immediate Etrny", entry - return entry.url, entry.rev - return entry.url, -1 - return url, -1 - -repositories = RepoCache() - - -# svn support code - -ALLOWED_CHARS = "_ -/\\=$.~+%" #add characters as necessary when tested -if sys.platform == "win32": - ALLOWED_CHARS += ":" -ALLOWED_CHARS_HOST = ALLOWED_CHARS + '@:' - -def _getsvnversion(ver=[]): - try: - return ver[0] - except IndexError: - v = py.process.cmdexec("svn -q --version") - v.strip() - v = '.'.join(v.split('.')[:2]) - ver.append(v) - return v - -def _escape_helper(text): - text = str(text) - if py.std.sys.platform != 'win32': - text = str(text).replace('$', '\\$') - return text - -def _check_for_bad_chars(text, allowed_chars=ALLOWED_CHARS): - for c in str(text): - if c.isalnum(): - continue - if c in allowed_chars: - continue - return True - return False - -def checkbadchars(url): - # (hpk) not quite sure about the exact purpose, guido w.? - proto, uri = url.split("://", 1) - if proto != "file": - host, uripath = uri.split('/', 1) - # only check for bad chars in the non-protocol parts - if (_check_for_bad_chars(host, ALLOWED_CHARS_HOST) \ - or _check_for_bad_chars(uripath, ALLOWED_CHARS)): - raise ValueError("bad char in %r" % (url, )) - - -#_______________________________________________________________ - -class SvnPathBase(common.PathBase): - """ Base implementation for SvnPath implementations. """ - sep = '/' - - def _geturl(self): - return self.strpath - url = property(_geturl, None, None, "url of this svn-path.") - - def __str__(self): - """ return a string representation (including rev-number) """ - return self.strpath - - def __hash__(self): - return hash(self.strpath) - - def new(self, **kw): - """ create a modified version of this path. A 'rev' argument - indicates a new revision. - the following keyword arguments modify various path parts:: - - http://host.com/repo/path/file.ext - |-----------------------| dirname - |------| basename - |--| purebasename - |--| ext - """ - obj = object.__new__(self.__class__) - obj.rev = kw.get('rev', self.rev) - obj.auth = kw.get('auth', self.auth) - dirname, basename, purebasename, ext = self._getbyspec( - "dirname,basename,purebasename,ext") - if 'basename' in kw: - if 'purebasename' in kw or 'ext' in kw: - raise ValueError("invalid specification %r" % kw) - else: - pb = kw.setdefault('purebasename', purebasename) - ext = kw.setdefault('ext', ext) - if ext and not ext.startswith('.'): - ext = '.' + ext - kw['basename'] = pb + ext - - kw.setdefault('dirname', dirname) - kw.setdefault('sep', self.sep) - if kw['basename']: - obj.strpath = "%(dirname)s%(sep)s%(basename)s" % kw - else: - obj.strpath = "%(dirname)s" % kw - return obj - - def _getbyspec(self, spec): - """ get specified parts of the path. 'arg' is a string - with comma separated path parts. The parts are returned - in exactly the order of the specification. - - you may specify the following parts: - - http://host.com/repo/path/file.ext - |-----------------------| dirname - |------| basename - |--| purebasename - |--| ext - """ - res = [] - parts = self.strpath.split(self.sep) - for name in spec.split(','): - name = name.strip() - if name == 'dirname': - res.append(self.sep.join(parts[:-1])) - elif name == 'basename': - res.append(parts[-1]) - else: - basename = parts[-1] - i = basename.rfind('.') - if i == -1: - purebasename, ext = basename, '' - else: - purebasename, ext = basename[:i], basename[i:] - if name == 'purebasename': - res.append(purebasename) - elif name == 'ext': - res.append(ext) - else: - raise NameError("Don't know part %r" % name) - return res - - def __eq__(self, other): - """ return true if path and rev attributes each match """ - return (str(self) == str(other) and - (self.rev == other.rev or self.rev == other.rev)) - - def __ne__(self, other): - return not self == other - - def join(self, *args): - """ return a new Path (with the same revision) which is composed - of the self Path followed by 'args' path components. - """ - if not args: - return self - - args = tuple([arg.strip(self.sep) for arg in args]) - parts = (self.strpath, ) + args - newpath = self.__class__(self.sep.join(parts), self.rev, self.auth) - return newpath - - def propget(self, name): - """ return the content of the given property. """ - value = self._propget(name) - return value - - def proplist(self): - """ list all property names. """ - content = self._proplist() - return content - - def size(self): - """ Return the size of the file content of the Path. """ - return self.info().size - - def mtime(self): - """ Return the last modification time of the file. """ - return self.info().mtime - - # shared help methods - - def _escape(self, cmd): - return _escape_helper(cmd) - - - #def _childmaxrev(self): - # """ return maximum revision number of childs (or self.rev if no childs) """ - # rev = self.rev - # for name, info in self._listdir_nameinfo(): - # rev = max(rev, info.created_rev) - # return rev - - #def _getlatestrevision(self): - # """ return latest repo-revision for this path. """ - # url = self.strpath - # path = self.__class__(url, None) - # - # # we need a long walk to find the root-repo and revision - # while 1: - # try: - # rev = max(rev, path._childmaxrev()) - # previous = path - # path = path.dirpath() - # except (IOError, process.cmdexec.Error): - # break - # if rev is None: - # raise IOError, "could not determine newest repo revision for %s" % self - # return rev - - class Checkers(common.Checkers): - def dir(self): - try: - return self.path.info().kind == 'dir' - except py.error.Error: - return self._listdirworks() - - def _listdirworks(self): - try: - self.path.listdir() - except py.error.ENOENT: - return False - else: - return True - - def file(self): - try: - return self.path.info().kind == 'file' - except py.error.ENOENT: - return False - - def exists(self): - try: - return self.path.info() - except py.error.ENOENT: - return self._listdirworks() - -def parse_apr_time(timestr): - i = timestr.rfind('.') - if i == -1: - raise ValueError("could not parse %s" % timestr) - timestr = timestr[:i] - parsedtime = time.strptime(timestr, "%Y-%m-%dT%H:%M:%S") - return time.mktime(parsedtime) - -class PropListDict(dict): - """ a Dictionary which fetches values (InfoSvnCommand instances) lazily""" - def __init__(self, path, keynames): - dict.__init__(self, [(x, None) for x in keynames]) - self.path = path - - def __getitem__(self, key): - value = dict.__getitem__(self, key) - if value is None: - value = self.path.propget(key) - dict.__setitem__(self, key, value) - return value - -def fixlocale(): - if sys.platform != 'win32': - return 'LC_ALL=C ' - return '' - -# some nasty chunk of code to solve path and url conversion and quoting issues -ILLEGAL_CHARS = '* | \ / : < > ? \t \n \x0b \x0c \r'.split(' ') -if os.sep in ILLEGAL_CHARS: - ILLEGAL_CHARS.remove(os.sep) -ISWINDOWS = sys.platform == 'win32' -_reg_allow_disk = re.compile(r'^([a-z]\:\\)?[^:]+$', re.I) -def _check_path(path): - illegal = ILLEGAL_CHARS[:] - sp = path.strpath - if ISWINDOWS: - illegal.remove(':') - if not _reg_allow_disk.match(sp): - raise ValueError('path may not contain a colon (:)') - for char in sp: - if char not in string.printable or char in illegal: - raise ValueError('illegal character %r in path' % (char,)) - -def path_to_fspath(path, addat=True): - _check_path(path) - sp = path.strpath - if addat and path.rev != -1: - sp = '%s@%s' % (sp, path.rev) - elif addat: - sp = '%s@HEAD' % (sp,) - return sp - -def url_from_path(path): - fspath = path_to_fspath(path, False) - quote = py.std.urllib.quote - if ISWINDOWS: - match = _reg_allow_disk.match(fspath) - fspath = fspath.replace('\\', '/') - if match.group(1): - fspath = '/%s%s' % (match.group(1).replace('\\', '/'), - quote(fspath[len(match.group(1)):])) - else: - fspath = quote(fspath) - else: - fspath = quote(fspath) - if path.rev != -1: - fspath = '%s@%s' % (fspath, path.rev) - else: - fspath = '%s@HEAD' % (fspath,) - return 'file://%s' % (fspath,) - -class SvnAuth(object): - """ container for auth information for Subversion """ - def __init__(self, username, password, cache_auth=True, interactive=True): - self.username = username - self.password = password - self.cache_auth = cache_auth - self.interactive = interactive - - def makecmdoptions(self): - uname = self.username.replace('"', '\\"') - passwd = self.password.replace('"', '\\"') - ret = [] - if uname: - ret.append('--username="%s"' % (uname,)) - if passwd: - ret.append('--password="%s"' % (passwd,)) - if not self.cache_auth: - ret.append('--no-auth-cache') - if not self.interactive: - ret.append('--non-interactive') - return ' '.join(ret) - - def __str__(self): - return "<SvnAuth username=%s ...>" %(self.username,) - -rex_blame = re.compile(r'\s*(\d+)\s*(\S+) (.*)') - -class SvnWCCommandPath(common.PathBase): - """ path implementation offering access/modification to svn working copies. - It has methods similar to the functions in os.path and similar to the - commands of the svn client. - """ - sep = os.sep - - def __new__(cls, wcpath=None, auth=None): - self = object.__new__(cls) - if isinstance(wcpath, cls): - if wcpath.__class__ == cls: - return wcpath - wcpath = wcpath.localpath - if _check_for_bad_chars(str(wcpath), - ALLOWED_CHARS): - raise ValueError("bad char in wcpath %s" % (wcpath, )) - self.localpath = py.path.local(wcpath) - self.auth = auth - return self - - strpath = property(lambda x: str(x.localpath), None, None, "string path") - rev = property(lambda x: x.info(usecache=0).rev, None, None, "revision") - - def __eq__(self, other): - return self.localpath == getattr(other, 'localpath', None) - - def _geturl(self): - if getattr(self, '_url', None) is None: - info = self.info() - self._url = info.url #SvnPath(info.url, info.rev) - assert isinstance(self._url, py.builtin._basestring) - return self._url - - url = property(_geturl, None, None, "url of this WC item") - - def _escape(self, cmd): - return _escape_helper(cmd) - - def dump(self, obj): - """ pickle object into path location""" - return self.localpath.dump(obj) - - def svnurl(self): - """ return current SvnPath for this WC-item. """ - info = self.info() - return py.path.svnurl(info.url) - - def __repr__(self): - return "svnwc(%r)" % (self.strpath) # , self._url) - - def __str__(self): - return str(self.localpath) - - def _makeauthoptions(self): - if self.auth is None: - return '' - return self.auth.makecmdoptions() - - def _authsvn(self, cmd, args=None): - args = args and list(args) or [] - args.append(self._makeauthoptions()) - return self._svn(cmd, *args) - - def _svn(self, cmd, *args): - l = ['svn %s' % cmd] - args = [self._escape(item) for item in args] - l.extend(args) - l.append('"%s"' % self._escape(self.strpath)) - # try fixing the locale because we can't otherwise parse - string = fixlocale() + " ".join(l) - try: - try: - key = 'LC_MESSAGES' - hold = os.environ.get(key) - os.environ[key] = 'C' - out = py.process.cmdexec(string) - finally: - if hold: - os.environ[key] = hold - else: - del os.environ[key] - except py.process.cmdexec.Error: - e = sys.exc_info()[1] - strerr = e.err.lower() - if strerr.find('file not found') != -1: - raise py.error.ENOENT(self) - if (strerr.find('file exists') != -1 or - strerr.find('file already exists') != -1 or - strerr.find("can't create directory") != -1): - raise py.error.EEXIST(self) - raise - return out - - def switch(self, url): - """ switch to given URL. """ - self._authsvn('switch', [url]) - - def checkout(self, url=None, rev=None): - """ checkout from url to local wcpath. """ - args = [] - if url is None: - url = self.url - if rev is None or rev == -1: - if (py.std.sys.platform != 'win32' and - _getsvnversion() == '1.3'): - url += "@HEAD" - else: - if _getsvnversion() == '1.3': - url += "@%d" % rev - else: - args.append('-r' + str(rev)) - args.append(url) - self._authsvn('co', args) - - def update(self, rev='HEAD', interactive=True): - """ update working copy item to given revision. (None -> HEAD). """ - opts = ['-r', rev] - if not interactive: - opts.append("--non-interactive") - self._authsvn('up', opts) - - def write(self, content, mode='w'): - """ write content into local filesystem wc. """ - self.localpath.write(content, mode) - - def dirpath(self, *args): - """ return the directory Path of the current Path. """ - return self.__class__(self.localpath.dirpath(*args), auth=self.auth) - - def _ensuredirs(self): - parent = self.dirpath() - if parent.check(dir=0): - parent._ensuredirs() - if self.check(dir=0): - self.mkdir() - return self - - def ensure(self, *args, **kwargs): - """ ensure that an args-joined path exists (by default as - a file). if you specify a keyword argument 'directory=True' - then the path is forced to be a directory path. - """ - p = self.join(*args) - if p.check(): - if p.check(versioned=False): - p.add() - return p - if kwargs.get('dir', 0): - return p._ensuredirs() - parent = p.dirpath() - parent._ensuredirs() - p.write("") - p.add() - return p - - def mkdir(self, *args): - """ create & return the directory joined with args. """ - if args: - return self.join(*args).mkdir() - else: - self._svn('mkdir') - return self - - def add(self): - """ add ourself to svn """ - self._svn('add') - - def remove(self, rec=1, force=1): - """ remove a file or a directory tree. 'rec'ursive is - ignored and considered always true (because of - underlying svn semantics. - """ - assert rec, "svn cannot remove non-recursively" - if not self.check(versioned=True): - # not added to svn (anymore?), just remove - py.path.local(self).remove() - return - flags = [] - if force: - flags.append('--force') - self._svn('remove', *flags) - - def copy(self, target): - """ copy path to target.""" - py.process.cmdexec("svn copy %s %s" %(str(self), str(target))) - - def rename(self, target): - """ rename this path to target. """ - py.process.cmdexec("svn move --force %s %s" %(str(self), str(target))) - - def lock(self): - """ set a lock (exclusive) on the resource """ - out = self._authsvn('lock').strip() - if not out: - # warning or error, raise exception - raise Exception(out[4:]) - - def unlock(self): - """ unset a previously set lock """ - out = self._authsvn('unlock').strip() - if out.startswith('svn:'): - # warning or error, raise exception - raise Exception(out[4:]) - - def cleanup(self): - """ remove any locks from the resource """ - # XXX should be fixed properly!!! - try: - self.unlock() - except: - pass - - def status(self, updates=0, rec=0, externals=0): - """ return (collective) Status object for this file. """ - # http://svnbook.red-bean.com/book.html#svn-ch-3-sect-4.3.1 - # 2201 2192 jum test - # XXX - if externals: - raise ValueError("XXX cannot perform status() " - "on external items yet") - else: - #1.2 supports: externals = '--ignore-externals' - externals = '' - if rec: - rec= '' - else: - rec = '--non-recursive' - - # XXX does not work on all subversion versions - #if not externals: - # externals = '--ignore-externals' - - if updates: - updates = '-u' - else: - updates = '' - - try: - cmd = 'status -v --xml --no-ignore %s %s %s' % ( - updates, rec, externals) - out = self._authsvn(cmd) - except py.process.cmdexec.Error: - cmd = 'status -v --no-ignore %s %s %s' % ( - updates, rec, externals) - out = self._authsvn(cmd) - rootstatus = WCStatus(self).fromstring(out, self) - else: - rootstatus = XMLWCStatus(self).fromstring(out, self) - return rootstatus - - def diff(self, rev=None): - """ return a diff of the current path against revision rev (defaulting - to the last one). - """ - args = [] - if rev is not None: - args.append("-r %d" % rev) - out = self._authsvn('diff', args) - return out - - def blame(self): - """ return a list of tuples of three elements: - (revision, commiter, line) - """ - out = self._svn('blame') - result = [] - blamelines = out.splitlines() - reallines = py.path.svnurl(self.url).readlines() - for i, (blameline, line) in enumerate( - zip(blamelines, reallines)): - m = rex_blame.match(blameline) - if not m: - raise ValueError("output line %r of svn blame does not match " - "expected format" % (line, )) - rev, name, _ = m.groups() - result.append((int(rev), name, line)) - return result - - _rex_commit = re.compile(r'.*Committed revision (\d+)\.$', re.DOTALL) - def commit(self, msg='', rec=1): - """ commit with support for non-recursive commits """ - # XXX i guess escaping should be done better here?!? - cmd = 'commit -m "%s" --force-log' % (msg.replace('"', '\\"'),) - if not rec: - cmd += ' -N' - out = self._authsvn(cmd) - try: - del cache.info[self] - except KeyError: - pass - if out: - m = self._rex_commit.match(out) - return int(m.group(1)) - - def propset(self, name, value, *args): - """ set property name to value on this path. """ - d = py.path.local.mkdtemp() - try: - p = d.join('value') - p.write(value) - self._svn('propset', name, '--file', str(p), *args) - finally: - d.remove() - - def propget(self, name): - """ get property name on this path. """ - res = self._svn('propget', name) - return res[:-1] # strip trailing newline - - def propdel(self, name): - """ delete property name on this path. """ - res = self._svn('propdel', name) - return res[:-1] # strip trailing newline - - def proplist(self, rec=0): - """ return a mapping of property names to property values. -If rec is True, then return a dictionary mapping sub-paths to such mappings. -""" - if rec: - res = self._svn('proplist -R') - return make_recursive_propdict(self, res) - else: - res = self._svn('proplist') - lines = res.split('\n') - lines = [x.strip() for x in lines[1:]] - return PropListDict(self, lines) - - def revert(self, rec=0): - """ revert the local changes of this path. if rec is True, do so -recursively. """ - if rec: - result = self._svn('revert -R') - else: - result = self._svn('revert') - return result - - def new(self, **kw): - """ create a modified version of this path. A 'rev' argument - indicates a new revision. - the following keyword arguments modify various path parts: - - http://host.com/repo/path/file.ext - |-----------------------| dirname - |------| basename - |--| purebasename - |--| ext - """ - if kw: - localpath = self.localpath.new(**kw) - else: - localpath = self.localpath - return self.__class__(localpath, auth=self.auth) - - def join(self, *args, **kwargs): - """ return a new Path (with the same revision) which is composed - of the self Path followed by 'args' path components. - """ - if not args: - return self - localpath = self.localpath.join(*args, **kwargs) - return self.__class__(localpath, auth=self.auth) - - def info(self, usecache=1): - """ return an Info structure with svn-provided information. """ - info = usecache and cache.info.get(self) - if not info: - try: - output = self._svn('info') - except py.process.cmdexec.Error: - e = sys.exc_info()[1] - if e.err.find('Path is not a working copy directory') != -1: - raise py.error.ENOENT(self, e.err) - elif e.err.find("is not under version control") != -1: - raise py.error.ENOENT(self, e.err) - raise - # XXX SVN 1.3 has output on stderr instead of stdout (while it does - # return 0!), so a bit nasty, but we assume no output is output - # to stderr... - if (output.strip() == '' or - output.lower().find('not a versioned resource') != -1): - raise py.error.ENOENT(self, output) - info = InfoSvnWCCommand(output) - - # Can't reliably compare on Windows without access to win32api - if py.std.sys.platform != 'win32': - if info.path != self.localpath: - raise py.error.ENOENT(self, "not a versioned resource:" + - " %s != %s" % (info.path, self.localpath)) - cache.info[self] = info - return info - - def listdir(self, fil=None, sort=None): - """ return a sequence of Paths. - - listdir will return either a tuple or a list of paths - depending on implementation choices. - """ - if isinstance(fil, str): - fil = common.FNMatcher(fil) - # XXX unify argument naming with LocalPath.listdir - def notsvn(path): - return path.basename != '.svn' - - paths = [] - for localpath in self.localpath.listdir(notsvn): - p = self.__class__(localpath, auth=self.auth) - if notsvn(p) and (not fil or fil(p)): - paths.append(p) - self._sortlist(paths, sort) - return paths - - def open(self, mode='r'): - """ return an opened file with the given mode. """ - return open(self.strpath, mode) - - def _getbyspec(self, spec): - return self.localpath._getbyspec(spec) - - class Checkers(py.path.local.Checkers): - def __init__(self, path): - self.svnwcpath = path - self.path = path.localpath - def versioned(self): - try: - s = self.svnwcpath.info() - except (py.error.ENOENT, py.error.EEXIST): - return False - except py.process.cmdexec.Error: - e = sys.exc_info()[1] - if e.err.find('is not a working copy')!=-1: - return False - if e.err.lower().find('not a versioned resource') != -1: - return False - raise - else: - return True - - def log(self, rev_start=None, rev_end=1, verbose=False): - """ return a list of LogEntry instances for this path. -rev_start is the starting revision (defaulting to the first one). -rev_end is the last revision (defaulting to HEAD). -if verbose is True, then the LogEntry instances also know which files changed. -""" - assert self.check() # make it simpler for the pipe - rev_start = rev_start is None and "HEAD" or rev_start - rev_end = rev_end is None and "HEAD" or rev_end - if rev_start == "HEAD" and rev_end == 1: - rev_opt = "" - else: - rev_opt = "-r %s:%s" % (rev_start, rev_end) - verbose_opt = verbose and "-v" or "" - locale_env = fixlocale() - # some blather on stderr - auth_opt = self._makeauthoptions() - #stdin, stdout, stderr = os.popen3(locale_env + - # 'svn log --xml %s %s %s "%s"' % ( - # rev_opt, verbose_opt, auth_opt, - # self.strpath)) - cmd = locale_env + 'svn log --xml %s %s %s "%s"' % ( - rev_opt, verbose_opt, auth_opt, self.strpath) - - popen = subprocess.Popen(cmd, - stdout=subprocess.PIPE, - stderr=subprocess.PIPE, - shell=True, - ) - stdout, stderr = popen.communicate() - stdout = py.builtin._totext(stdout, sys.getdefaultencoding()) - minidom,ExpatError = importxml() - try: - tree = minidom.parseString(stdout) - except ExpatError: - raise ValueError('no such revision') - result = [] - for logentry in filter(None, tree.firstChild.childNodes): - if logentry.nodeType == logentry.ELEMENT_NODE: - result.append(LogEntry(logentry)) - return result - - def size(self): - """ Return the size of the file content of the Path. """ - return self.info().size - - def mtime(self): - """ Return the last modification time of the file. """ - return self.info().mtime - - def __hash__(self): - return hash((self.strpath, self.__class__, self.auth)) - - -class WCStatus: - attrnames = ('modified','added', 'conflict', 'unchanged', 'external', - 'deleted', 'prop_modified', 'unknown', 'update_available', - 'incomplete', 'kindmismatch', 'ignored', 'locked', 'replaced' - ) - - def __init__(self, wcpath, rev=None, modrev=None, author=None): - self.wcpath = wcpath - self.rev = rev - self.modrev = modrev - self.author = author - - for name in self.attrnames: - setattr(self, name, []) - - def allpath(self, sort=True, **kw): - d = {} - for name in self.attrnames: - if name not in kw or kw[name]: - for path in getattr(self, name): - d[path] = 1 - l = d.keys() - if sort: - l.sort() - return l - - # XXX a bit scary to assume there's always 2 spaces between username and - # path, however with win32 allowing spaces in user names there doesn't - # seem to be a more solid approach :( - _rex_status = re.compile(r'\s+(\d+|-)\s+(\S+)\s+(.+?)\s{2,}(.*)') - - def fromstring(data, rootwcpath, rev=None, modrev=None, author=None): - """ return a new WCStatus object from data 's' - """ - rootstatus = WCStatus(rootwcpath, rev, modrev, author) - update_rev = None - for line in data.split('\n'): - if not line.strip(): - continue - #print "processing %r" % line - flags, rest = line[:8], line[8:] - # first column - c0,c1,c2,c3,c4,c5,x6,c7 = flags - #if '*' in line: - # print "flags", repr(flags), "rest", repr(rest) - - if c0 in '?XI': - fn = line.split(None, 1)[1] - if c0 == '?': - wcpath = rootwcpath.join(fn, abs=1) - rootstatus.unknown.append(wcpath) - elif c0 == 'X': - wcpath = rootwcpath.__class__( - rootwcpath.localpath.join(fn, abs=1), - auth=rootwcpath.auth) - rootstatus.external.append(wcpath) - elif c0 == 'I': - wcpath = rootwcpath.join(fn, abs=1) - rootstatus.ignored.append(wcpath) - - continue - - #elif c0 in '~!' or c4 == 'S': - # raise NotImplementedError("received flag %r" % c0) - - m = WCStatus._rex_status.match(rest) - if not m: - if c7 == '*': - fn = rest.strip() - wcpath = rootwcpath.join(fn, abs=1) - rootstatus.update_available.append(wcpath) - continue - if line.lower().find('against revision:')!=-1: - update_rev = int(rest.split(':')[1].strip()) - continue - if line.lower().find('status on external') > -1: - # XXX not sure what to do here... perhaps we want to - # store some state instead of just continuing, as right - # now it makes the top-level external get added twice - # (once as external, once as 'normal' unchanged item) - # because of the way SVN presents external items - continue - # keep trying - raise ValueError("could not parse line %r" % line) - else: - rev, modrev, author, fn = m.groups() - wcpath = rootwcpath.join(fn, abs=1) - #assert wcpath.check() - if c0 == 'M': - assert wcpath.check(file=1), "didn't expect a directory with changed content here" - rootstatus.modified.append(wcpath) - elif c0 == 'A' or c3 == '+' : - rootstatus.added.append(wcpath) - elif c0 == 'D': - rootstatus.deleted.append(wcpath) - elif c0 == 'C': - rootstatus.conflict.append(wcpath) - elif c0 == '~': - rootstatus.kindmismatch.append(wcpath) - elif c0 == '!': - rootstatus.incomplete.append(wcpath) - elif c0 == 'R': - rootstatus.replaced.append(wcpath) - elif not c0.strip(): - rootstatus.unchanged.append(wcpath) - else: - raise NotImplementedError("received flag %r" % c0) - - if c1 == 'M': - rootstatus.prop_modified.append(wcpath) - # XXX do we cover all client versions here? - if c2 == 'L' or c5 == 'K': - rootstatus.locked.append(wcpath) - if c7 == '*': - rootstatus.update_available.append(wcpath) - - if wcpath == rootwcpath: - rootstatus.rev = rev - rootstatus.modrev = modrev - rootstatus.author = author - if update_rev: - rootstatus.update_rev = update_rev - continue - return rootstatus - fromstring = staticmethod(fromstring) - -class XMLWCStatus(WCStatus): - def fromstring(data, rootwcpath, rev=None, modrev=None, author=None): - """ parse 'data' (XML string as outputted by svn st) into a status obj - """ - # XXX for externals, the path is shown twice: once - # with external information, and once with full info as if - # the item was a normal non-external... the current way of - # dealing with this issue is by ignoring it - this does make - # externals appear as external items as well as 'normal', - # unchanged ones in the status object so this is far from ideal - rootstatus = WCStatus(rootwcpath, rev, modrev, author) - update_rev = None - minidom, ExpatError = importxml() - try: - doc = minidom.parseString(data) - except ExpatError: - e = sys.exc_info()[1] - raise ValueError(str(e)) - urevels = doc.getElementsByTagName('against') - if urevels: - rootstatus.update_rev = urevels[-1].getAttribute('revision') - for entryel in doc.getElementsByTagName('entry'): - path = entryel.getAttribute('path') - statusel = entryel.getElementsByTagName('wc-status')[0] - itemstatus = statusel.getAttribute('item') - - if itemstatus == 'unversioned': - wcpath = rootwcpath.join(path, abs=1) - rootstatus.unknown.append(wcpath) - continue - elif itemstatus == 'external': - wcpath = rootwcpath.__class__( - rootwcpath.localpath.join(path, abs=1), - auth=rootwcpath.auth) - rootstatus.external.append(wcpath) - continue - elif itemstatus == 'ignored': - wcpath = rootwcpath.join(path, abs=1) - rootstatus.ignored.append(wcpath) - continue - elif itemstatus == 'incomplete': - wcpath = rootwcpath.join(path, abs=1) - rootstatus.incomplete.append(wcpath) - continue - - rev = statusel.getAttribute('revision') - if itemstatus == 'added' or itemstatus == 'none': - rev = '0' - modrev = '?' - author = '?' - date = '' - else: - #print entryel.toxml() - commitel = entryel.getElementsByTagName('commit')[0] - if commitel: - modrev = commitel.getAttribute('revision') - author = '' - author_els = commitel.getElementsByTagName('author') - if author_els: - for c in author_els[0].childNodes: - author += c.nodeValue - date = '' - for c in commitel.getElementsByTagName('date')[0]\ - .childNodes: - date += c.nodeValue - - wcpath = rootwcpath.join(path, abs=1) - - assert itemstatus != 'modified' or wcpath.check(file=1), ( - 'did\'t expect a directory with changed content here') - - itemattrname = { - 'normal': 'unchanged', - 'unversioned': 'unknown', - 'conflicted': 'conflict', - 'none': 'added', - }.get(itemstatus, itemstatus) - - attr = getattr(rootstatus, itemattrname) - attr.append(wcpath) - - propsstatus = statusel.getAttribute('props') - if propsstatus not in ('none', 'normal'): - rootstatus.prop_modified.append(wcpath) - - if wcpath == rootwcpath: - rootstatus.rev = rev - rootstatus.modrev = modrev - rootstatus.author = author - rootstatus.date = date - - # handle repos-status element (remote info) - rstatusels = entryel.getElementsByTagName('repos-status') - if rstatusels: - rstatusel = rstatusels[0] - ritemstatus = rstatusel.getAttribute('item') - if ritemstatus in ('added', 'modified'): - rootstatus.update_available.append(wcpath) - - lockels = entryel.getElementsByTagName('lock') - if len(lockels): - rootstatus.locked.append(wcpath) - - return rootstatus - fromstring = staticmethod(fromstring) - -class InfoSvnWCCommand: - def __init__(self, output): - # Path: test - # URL: http://codespeak.net/svn/std.path/trunk/dist/std.path/test - # Repository UUID: fd0d7bf2-dfb6-0310-8d31-b7ecfe96aada - # Revision: 2151 - # Node Kind: directory - # Schedule: normal - # Last Changed Author: hpk - # Last Changed Rev: 2100 - # Last Changed Date: 2003-10-27 20:43:14 +0100 (Mon, 27 Oct 2003) - # Properties Last Updated: 2003-11-03 14:47:48 +0100 (Mon, 03 Nov 2003) - - d = {} - for line in output.split('\n'): - if not line.strip(): - continue - key, value = line.split(':', 1) - key = key.lower().replace(' ', '') - value = value.strip() - d[key] = value - try: - self.url = d['url'] - except KeyError: - raise ValueError("Not a versioned resource") - #raise ValueError, "Not a versioned resource %r" % path - self.kind = d['nodekind'] == 'directory' and 'dir' or d['nodekind'] - self.rev = int(d['revision']) - self.path = py.path.local(d['path']) - self.size = self.path.size() - if 'lastchangedrev' in d: - self.created_rev = int(d['lastchangedrev']) - if 'lastchangedauthor' in d: - self.last_author = d['lastchangedauthor'] - if 'lastchangeddate' in d: - self.mtime = parse_wcinfotime(d['lastchangeddate']) - self.time = self.mtime * 1000000 - - def __eq__(self, other): - return self.__dict__ == other.__dict__ - -def parse_wcinfotime(timestr): - """ Returns seconds since epoch, UTC. """ - # example: 2003-10-27 20:43:14 +0100 (Mon, 27 Oct 2003) - m = re.match(r'(\d+-\d+-\d+ \d+:\d+:\d+) ([+-]\d+) .*', timestr) - if not m: - raise ValueError("timestring %r does not match" % timestr) - timestr, timezone = m.groups() - # do not handle timezone specially, return value should be UTC - parsedtime = time.strptime(timestr, "%Y-%m-%d %H:%M:%S") - return calendar.timegm(parsedtime) - -def make_recursive_propdict(wcroot, - output, - rex = re.compile("Properties on '(.*)':")): - """ Return a dictionary of path->PropListDict mappings. """ - lines = [x for x in output.split('\n') if x] - pdict = {} - while lines: - line = lines.pop(0) - m = rex.match(line) - if not m: - raise ValueError("could not parse propget-line: %r" % line) - path = m.groups()[0] - wcpath = wcroot.join(path, abs=1) - propnames = [] - while lines and lines[0].startswith(' '): - propname = lines.pop(0).strip() - propnames.append(propname) - assert propnames, "must have found properties!" - pdict[wcpath] = PropListDict(wcpath, propnames) - return pdict - - -def importxml(cache=[]): - if cache: - return cache - from xml.dom import minidom - from xml.parsers.expat import ExpatError - cache.extend([minidom, ExpatError]) - return cache - -class LogEntry: - def __init__(self, logentry): - self.rev = int(logentry.getAttribute('revision')) - for lpart in filter(None, logentry.childNodes): - if lpart.nodeType == lpart.ELEMENT_NODE: - if lpart.nodeName == 'author': - self.author = lpart.firstChild.nodeValue - elif lpart.nodeName == 'msg': - if lpart.firstChild: - self.msg = lpart.firstChild.nodeValue - else: - self.msg = '' - elif lpart.nodeName == 'date': - #2003-07-29T20:05:11.598637Z - timestr = lpart.firstChild.nodeValue - self.date = parse_apr_time(timestr) - elif lpart.nodeName == 'paths': - self.strpaths = [] - for ppart in filter(None, lpart.childNodes): - if ppart.nodeType == ppart.ELEMENT_NODE: - self.strpaths.append(PathEntry(ppart)) - def __repr__(self): - return '<Logentry rev=%d author=%s date=%s>' % ( - self.rev, self.author, self.date) - - |