Content-Type: text/plain; charset="utf-8"
MIME-Version: 1.0
Content-Transfer-Encoding: quoted-printable
Date: Sun, 22 Jan 2012 06:01:41 -0000
To: notmuch mailing list <notmuch@notmuchmail.org>
Message-ID: <20120122060141.16806.41461@thinkbox.jade-hamburg.de>
From: Justus Winter <4winter@informatik.uni-hamburg.de>
Subject: heads up from the python front
Date: Sun, 22 Jan 2012 07:01:41 +0100
Precedence: list

Hey everyone :)

after getting to know nmbug a little better (it's actually very nice
to track bugs and patches this way...) I did some work on the python
bindings. tl;dr version: housekeeping, python3.2 support, fixed nasty
bug.

I familiarized myself with nmbug and went through all the threads
tagged with notmuch::python, added and updated few tags here and there
and started working on the open issues.

Python 3.2 support
------------------

I merged the last patch of a patchset[0] I wrote in december that makes
it possible to use the python bindings with both python2.x and
python3.2.

I do not now how complete the port is, most notably the notmuch.py
script does not work with 3.x. But it is complete enough to make
afew[1] work using python3.2 and the vanilla notmuch bindings.

If you want to help out and have a small script that uses the bindings
I'd like to invite you to try to port your script and report any
issues.

Fix random crashes when using the bindings
------------------------------------------

I found a nasty bug I introduced with a patchset[2] that was supposed
to make the bindings more robust. Annotating pointers returned from
libnotmuch functions called using ctypes allows the ctypes framework
to do typechecking. But I accidentally broke the error handling
code. Citing the commit message:

    Before 3434d1940 the return values of libnotmuch functions were
    declared as c_void_p and the code checking for errors compared the
    returned value to None, which is the ctypes equivalent of a NULL
    pointer.

    But said commit wrapped all the data types in python classes and the
    semantic changed in a subtle way. If a function returns NULL, the
    wrapped python value is falsish, but no longer equal to None.

In fact the minimal test case triggering the bug is:

    import os
    import notmuch

    db_path =3D os.path.expanduser('~/Maildir')

    db_0 =3D notmuch.Database(db_path, mode=3Dnotmuch.Database.MODE.READ_WR=
ITE)
    db_1 =3D notmuch.Database(db_path, mode=3Dnotmuch.Database.MODE.READ_WR=
ITE)

The problem was most apparent when opening the database fails because
it has been locked by someone else. The patch regarding the function
Database.open looks like this:

         res =3D Database._open(_str(path), mode)
 =

-        if res is None:
+        if not res:
             raise NotmuchError(message=3D"Could not open the specified dat=
abase")
         self._db =3D res

The old code fails to notify the callee of the error who causes a
segfault later if he uses that database reference which is in fact
NULL.

I feel kind of bad since I severely broke the error handling for all
pythonic notmuch users out there, is there any chance of a bugfix
release? The patch[3] is really simple and I'd say it's trivial to
backport it to the last release. What do you think?

Justus

0: id:1323860305-15802-1-git-send-email-4winter@informatik.uni-hamburg.de
1: https://github.com/teythoon/afew
2: id:1318198374-926-1-git-send-email-4winter@informatik.uni-hamburg.de
3: 8015cbff263606f009b5750d23b28ee332c25db8