Date: Sat, 23 Jul 2011 14:36:02 +0100
From: Patrick Totzke <patricktotzke@googlemail.com>
To: David Bremner <david@tethera.net>
Subject: Re: xapian exceptions not caught in python bindings?
Message-ID: <20110723133602.GA30241@brick.lan>
References: <20110626202733.GA26837@brick> <20110717193538.GA16615@brick.lan>
	<87r55o65yq.fsf@zancas.localnet>
MIME-Version: 1.0
Content-Type: multipart/signed; micalg=pgp-sha1;
	protocol="application/pgp-signature"; boundary="fdj2RfSjLxBAspz7"
Content-Disposition: inline
In-Reply-To: <87r55o65yq.fsf@zancas.localnet>
User-Agent: Mutt/1.5.21 (2010-09-15)
Cc: Patrick Totzke <patricktotzke@googlemail.com>, notmuch@notmuchmail.org
Precedence: list


--fdj2RfSjLxBAspz7
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

Hi all,

I hope the patch I send is correctly formated, I'm still fumbling with
git send-email and the --in-reply-to option.
Anyhow, forgive my language, of course I didn't mean to be condescending in=
 any
way by calling these prints garbage! It's just that it's highly unusual and=
 very
'non-pythonic' that a module directly prints to stderr instead of raising e=
xceptions
and if you work directly with a curseslike interface on a terminal these
errormessages litter my screen.

The patch I send is a suggestion how to fix the behaviour described in my l=
ast post.
It introduces a ContextManager class that can be used to raise messages fro=
m stderr
as Xapian exceptions like this:

> from notmuch.globals import RaiseStderrErrors
> with RaiseStderrErrors():
>  do_stuff()

Now, if one executes:=20
------------
> from notmuch import Database
> bad_querystring =3D "test AND"
> query =3D Database().create_query(bad_querystring
---------------
one gets a nice Xapian exception
-----------------------
  File "syntax.py", line 4, in <module>
    query =3D Database().create_query(bad_querystring)
  File "/usr/local/lib/python2.7/dist-packages/notmuch/database.py", line 4=
32, in create_query
    return Query(self, querystring)
  File "/usr/local/lib/python2.7/dist-packages/notmuch/database.py", line 5=
14, in __init__
    self.create(db, querystr)
  File "/usr/local/lib/python2.7/dist-packages/notmuch/database.py", line 5=
47, in create
    Query._count_messages(self._query)
  File "/usr/local/lib/python2.7/dist-packages/notmuch/globals.py", line 12=
3, in __exit__
    raise NotmuchError(STATUS.XAPIAN_EXCEPTION, message=3Derr)
notmuch.globals.NotmuchError: A Xapian exception occurred: Syntax: <express=
ion> AND <expression>
Query string was: test AND
----------------------------

There are two problems with this suggestion however.
First, one needs to redirect sdterr to a tempfile: Using StringIO doesn't w=
ork since
just replacing sys.stderr with something else is not "low level" enough, th=
e messages will
still get printed. An alternative is using os.dup2, which replaces the file=
 descriptor
of stderr directly, but cStringIO objects dont have .fileno()..
see [0,1].
The second problem is, that creating a query object with a malformed querys=
tring doesn't
print anything to stderr, only if you use that object errors get printed (n=
otmuch/lib/query.cc).
Thus, I call count_messages() once after initialising a new notmuch.databas=
=2EQuery object (at the end of  Query.create). This ensures that Query.__in=
it__() raises an exception when given bad=20
querystrings, but at the cost of triggering an unnecessary count_messages().

best,
/p



----
[0]: http://stackoverflow.com/questions/5903501/attributeerror-stringio-ins=
tance-has-no-attribute-fileno
[1]: http://code.activestate.com/recipes/577564-context-manager-for-low-lev=
el-redirection-of-stdou/


On Sun, Jul 17, 2011 at 04:51:41PM -0300, David Bremner wrote:
> On Sun, 17 Jul 2011 20:35:38 +0100, Patrick Totzke <patricktotzke@googlem=
ail.com> wrote:
> > If you run the following snippet, you notice that not only do we get
> > xapian-garbage on stderr but we don't really get any exceptions at the
> > position where it would make sense:
>=20
> I wouldn't call that "xapian-garbage" since it is output from
> libnotmuch.
>=20
> d

--fdj2RfSjLxBAspz7
Content-Type: application/pgp-signature; name="signature.asc"
Content-Description: Digital signature

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)

iEYEARECAAYFAk4qzkIACgkQlDQDZ9fWxar8EACgrSVyAbS67aXjWLzeBpoJuzBb
rwYAn0EJ3Sfqsw6zTlew+oGuubWgRuXz
=tQEF
-----END PGP SIGNATURE-----

--fdj2RfSjLxBAspz7--