Hi,

Due to Timer's object creation behavior, it is impossible to subclass it. This kind of declaration will result to an exception raising:

class A(Timer): pass

Works for me:

>>> from timeit import Timer
>>> class A(Timer): pass
... 

(Tested with Python 3.1 and 3.2 on OSX.)

I've tested it with Python 3.1.2 under Windows 7 32 bits. It raises the following TypeError exception "function() argument 1 must be code, not str"

I can't reproduce the error. Do you have a script that shows the issue?
What is the complete traceback?

Ah, got it. It's about threading.Timer, which looks like a class, but is actually a function.

It seems to be by design: from the documentation:
http://docs.python.org/py3k/library/threading.html
"Timer is a subclass of Thread", and a Thread subclass should "only override the __init__() and run() methods".

What are you trying to do here? overriding run() is probably wrong; and overriding __init__ is better done by passing the correct parameters to threading.Timer().

Yes that's it. I Should precise it. I've taken a screenshot from my python's interpreter to spot it.

AFAIK this is by design. Not that I agree with this decision anyway.

Adding Guido, who checked the module in, to nosy.

Guido: Could you tell us whether the fake classes in threading.py should stay as is or can be fixed?  The whole _Verbose business and Thing/_Thing indirection seem a bit outdated and unneeded.

IIRC:

The design started out this way because it predates new-style classes. When this was put in one couldn't subclass extension types, and there were plans/hopes to replace some of the lock types with platform-specific built-in versions on some platforms.

Since it is now possible to write subclassable extension types the Thing/_Thing design is no longer needed.

I'm not sure about the _Verbose hacks, if it is deemed not useful I'm fine with letting it go.

See also issue 5831.  That should probably be closed as a dup of this since this has an explanation.

One concern expressed on a duplicate report by Martijn van Oosterhout:

> Note this is a behaviour change. Under the old scheme (Foo is a class)
>
> Foo.timerclass = Timer
>
> created a method, whereas now it will just assign the class as an
> attribute. To work around this you had to use _Timer. Will that dummy
> class remain as an alias to avoid breaking code (in 2.7 at least)?

Stable versions (2.7, 3.1, soon 3.2) won’t get this change.  They may get a doc patch to warn people about the fake class/factory thing.


In the py3k documentation for threading, some of the fake classes are documented as factories (Event) and other ones as classes (Timer); do you people think there would be harm in cleaning up all those outdated shenanigans for 3.3, with due versionchanged notes in the doc?

Attached patch removes the indirection functions; the _Verbose shenanigans are left alone.  The test suite passes; I haven’t edited the doc yet.

I’ve committed the cleanup to my 3.3 clone and will push this week.

Here’s a doc patch.  Before my patch, the various classes were documented in two parts: one entry with the factory function (e.g. Thread), without index reference, and one section (e.g. “Thread Objects”), which used a class directive (and thus index target) for most but not all classes.

After my patch, all classes are documented with a class directive, in their section (i.e. “Thread Objects”), with a versionchanged note informing that the name used to be that of a factory function.  The only remaining glitch is that the “X Objects” sections start with a description of the class’ use, which references methods with constructs like :meth:`run`, which cannot be turned into links as Sphinx lacks context: the class directive only happens after.  I could move the class directives right after the heading (“X Objects”), so that the meth roles get turned into links.

You ought to be able to use either a context directive (I forget its name and syntax), or the full reference syntax (:meth:`~threading.Thread.run`) to make those links work without moving things around.

> You ought to be able to use either a context directive (I forget its
> name and syntax),

Hm, do you mean something similar to currentmodule?

> or the full reference syntax (:meth:`~threading.Thread.run`) to make
> those links work without moving things around.

Yes, Thread.run would work, but I find that inelegant.

New changeset 7f606223578a by Éric Araujo in branch 'default':
Remove indirection in threading (issue #10968).
http://hg.python.org/cpython/rev/7f606223578a

@eric: The doc has to be updated:

http://docs.python.org/dev/library/threading.html#threading.activeCount

"threading.Condition()
A factory function that returns a new condition variable object. A condition variable allows one or more threads to wait until they are notified by another thread.

See Condition Objects."

--

See also (maybe) issue #12960.

Victor: Yes, I know the doc needs an update, that’s why I posted a patch six weeks ago and asked for a review.

> that’s why I posted a patch six weeks ago and asked for a review

Oh ok, cool, I missed the patches.

Ping.

New changeset 98499371ca53 by R David Murray in branch '3.3':
#10968: commit threading doc changes and corresponding whatsnew entry.
http://hg.python.org/cpython/rev/98499371ca53

New changeset ad435964fc9c by R David Murray in branch 'default':
merge #10968: commit threading doc changes and corresponding whatsnew entry.
http://hg.python.org/cpython/rev/ad435964fc9c

I have now committed (a revised version of) the doc changes.

Like I said in the commit message, it is unfortunate that the underscore names were not kept as aliases and that RLock wasn't also converted to a class, but it is too late to fix that in 3.3.  If someone wants to do RLock in 3.4 they should open a new issue.