diff --git a/Lib/mailbox.py b/Lib/mailbox.py --- a/Lib/mailbox.py +++ b/Lib/mailbox.py @@ -258,19 +258,24 @@ def __init__(self, dirname, factory=None, create=True): """Initialize a Maildir instance.""" Mailbox.__init__(self, dirname, factory, create) + self._paths = { + 'tmp': os.path.join(self._path, 'tmp'), + 'new': os.path.join(self._path, 'new'), + 'cur': os.path.join(self._path, 'cur'), + } if not os.path.exists(self._path): if create: os.mkdir(self._path, 0o700) - os.mkdir(os.path.join(self._path, 'tmp'), 0o700) - os.mkdir(os.path.join(self._path, 'new'), 0o700) - os.mkdir(os.path.join(self._path, 'cur'), 0o700) + for path in self._paths.values(): + os.mkdir(path, 0o700) else: raise NoSuchMailboxError(self._path) self._toc = {} - self._last_read = None # Records last time we read cur/new - # NOTE: we manually invalidate _last_read each time we do any - # modifications ourselves, otherwise we might get tripped up by - # bogus mtime behaviour on some systems (see issue #6896). + self._toc_mtimes = {} + for subdir in ('cur', 'new'): + self._toc_mtimes[subdir] = os.path.getmtime(self._paths[subdir]) + self._last_read = time.time() # Records last time we read cur/new + self._skewfactor = 0.1 # Adjust if os/fs clocks are skewing def add(self, message): """Add message and return assigned key.""" @@ -307,15 +312,11 @@ raise if isinstance(message, MaildirMessage): os.utime(dest, (os.path.getatime(dest), message.get_date())) - # Invalidate cached toc - self._last_read = None return uniq def remove(self, key): """Remove the keyed message; raise KeyError if it doesn't exist.""" os.remove(os.path.join(self._path, self._lookup(key))) - # Invalidate cached toc (only on success) - self._last_read = None def discard(self, key): """If the keyed message exists, remove it.""" @@ -350,8 +351,6 @@ if isinstance(message, MaildirMessage): os.utime(new_path, (os.path.getatime(new_path), message.get_date())) - # Invalidate cached toc - self._last_read = None def get_message(self, key): """Return a Message representation or raise a KeyError.""" @@ -407,8 +406,8 @@ def flush(self): """Write any pending changes to disk.""" # Maildir changes are always written immediately, so there's nothing - # to do except invalidate our cached toc. - self._last_read = None + # to do. + pass def lock(self): """Lock the mailbox.""" @@ -506,36 +505,38 @@ def _refresh(self): """Update table of contents mapping.""" - if self._last_read is not None: - for subdir in ('new', 'cur'): - mtime = os.path.getmtime(os.path.join(self._path, subdir)) - if mtime > self._last_read: - break - else: + # If it has been less than a second since the last _refresh() call, we + # have to unconditionally re-read the mailbox just in case it has been + # modified, because os.path.mtime() only has 1sec resolution. This + # results in a few unnecessary re-reads when _refresh() is called + # multiple times in the same second, but once the clock ticks over, we + # will only re-read as needed. Because the filesystem might be being + # served by an independent system with its own clock, we record and + # compare with the mtimes from the filesystem. Because the other + # system's clock might be skewing relative to our clock, we add an + # extra delta to our wait. The default is one tenth second, but is an + # instance variable and so can be adjusted if dealing with a + # particularly skewed or irregular system. + if time.time() - self._last_read > 1 + self._skewfactor: + refresh = False + for subdir in self._toc_mtimes: + mtime = os.path.getmtime(self._paths[subdir]) + if mtime > self._toc_mtimes[subdir]: + refresh = True + self._toc_mtimes[subdir] = mtime + if not refresh: return - - # We record the current time - 1sec so that, if _refresh() is called - # again in the same second, we will always re-read the mailbox - # just in case it's been modified. (os.path.mtime() only has - # 1sec resolution.) This results in a few unnecessary re-reads - # when _refresh() is called multiple times in the same second, - # but once the clock ticks over, we will only re-read as needed. - now = time.time() - 1 - + # Refresh toc self._toc = {} - def update_dir (subdir): - path = os.path.join(self._path, subdir) + for subdir in self._toc_mtimes: + path = self._paths[subdir] for entry in os.listdir(path): p = os.path.join(path, entry) if os.path.isdir(p): continue uniq = entry.split(self.colon)[0] self._toc[uniq] = os.path.join(subdir, entry) - - update_dir('new') - update_dir('cur') - - self._last_read = now + self._last_read = time.time() def _lookup(self, key): """Use TOC to return subpath for given key, or raise a KeyError.""" diff --git a/Lib/test/test_mailbox.py b/Lib/test/test_mailbox.py --- a/Lib/test/test_mailbox.py +++ b/Lib/test/test_mailbox.py @@ -870,21 +870,18 @@ def test_reread(self): - # Initially, the mailbox has not been read and the time is null. - assert getattr(self._box, '_last_read', None) is None - - # Refresh mailbox; the times should now be set to something. - self._box._refresh() - assert getattr(self._box, '_last_read', None) is not None - # Put the last modified times more than one second into the past - # (because mtime has a one second granularity, a refresh is done - # unconditionally if called for within the same second, just in case - # the mbox has changed). + # (because mtime has a one second granularity) for subdir in ('cur', 'new'): os.utime(os.path.join(self._box._path, subdir), (time.time()-5,)*2) + # Because mtime has a one second granularity, a refresh is done + # unconditionally if called for within one-second-plus-a-bit of the + # last one, just in case the mbox has changed; so now we have to wait + # for that interval to expire. + time.sleep(1.01 + self._box._skewfactor) + # Re-reading causes the ._toc attribute to be assigned a new dictionary # object, so we'll check that the ._toc attribute isn't a different # object. @@ -893,7 +890,7 @@ return self._box._toc is not orig_toc self._box._refresh() - assert not refreshed() + self.assertFalse(refreshed()) # Now, write something into cur and remove it. This changes # the mtime and should cause a re-read. @@ -902,7 +899,7 @@ f.close() os.unlink(filename) self._box._refresh() - assert refreshed() + self.assertTrue(refreshed()) class _TestMboxMMDF(TestMailbox):