namespaces

[Paolino]

Why descriptor mechanism doesn't apply to modules?

I suspect another anomaly in the namespaces implementations.
class Mosse(object):
def __get__(self,*_):
print 'HERE'

m=Mosse()
m # doesn't work

import new
mod=new.module('hopeless')
mod.m=Mosse()
assert 'm' in mod.__dict__ # this is said to be the key for descriptors
# to be called
mod.m # doesn't work

Thanks Paolino
___________________________________
Yahoo! Messenger: chiamate gratuite in tutto il mondo
http://it.beta.messenger.yahoo.com

[Peter Otten]

Paolino wrote:

Why descriptor mechanism doesn't apply to modules?

Because modules are instances of the module class and the descriptor has to
be defined in the class in order to work with the instance. E. g.:

import sys
class Descr(object): .... def __get__(*args): print "__get__%r" % (args,)
.... module = type(sys)
class Module(module): .... descr = Descr()
.... m = Module("noname")
m.descr __get__(<__main__.Descr object at 0x4029372c>, <module 'noname' (built-in)>,
<class '__main__.Module'>)

Peter

[Paolino]

Peter Otten wrote:

Paolino wrote:

Why descriptor mechanism doesn't apply to modules?

Because modules are instances of the module class and the descriptor has to
be defined in the class in order to work with the instance. E. g.:

Got it,thanks.
Then there is no way of having descriptors at module level,as 'module'
class is not updatable.

___________________________________
Yahoo! Mail: gratis 1GB per i messaggi e allegati da 10MB
http://mail.yahoo.it

[Scott David Daniels]

Paolino wrote:

Peter Otten wrote:

Paolino wrote:

Why descriptor mechanism doesn't apply to modules?

Because modules are instances of the module class and the descriptor
has to be defined in the class in order to work with the instance....

Then there is no way of having descriptors at module level,as 'module'
class is not updatable.

Well, an entry in the dictionary "sys.modules" is what is returned by
imports, so you could either pre-replace or post-replace a module by
an object which uses the descriptor technique to get to some (but not
necessarily all) of the attributes. Think of this technique as a
hack to get to a goal, rather than a good technique to use; good for
debugging, not so nice for production work. If you still don't know
how to do this from this admittedly sketchy description, I'd suggest
you avoid trying it altogether.

--Scott David Daniels

[Bengt Richter]

On Tue, 09 Aug 2005 08:35:38 -0700, Scott David Daniels <Sc***********@Acm.Org> wrote:

Paolino wrote:

Peter Otten wrote:

Paolino wrote:
Why descriptor mechanism doesn't apply to modules?
Because modules are instances of the module class and the descriptor
has to be defined in the class in order to work with the instance....

Then there is no way of having descriptors at module level,as 'module'
class is not updatable.

Well, an entry in the dictionary "sys.modules" is what is returned by
imports, so you could either pre-replace or post-replace a module by
an object which uses the descriptor technique to get to some (but not
necessarily all) of the attributes. Think of this technique as a
hack to get to a goal, rather than a good technique to use; good for
debugging, not so nice for production work. If you still don't know
how to do this from this admittedly sketchy description, I'd suggest
you avoid trying it altogether.

I had the thought a while ago that it might be interesting to have
an import variant that could produce a subclassed module, where you
indicate the import name as usual (for __import__) and also supply
the subclass source text as an argument (string or file).
E.g., something like

subclass_source = """\
class MyModule(math):
twopi = property(lambda: math.pi*2.0')
"""

module_with_property = subclassing_import('math', subclass=subclass_source)

Another thought/bf would be a way to extend the attribute namespace of an arbitrary object
by chaining the attribute name spaces of a sequence of objects (this would be a language mod)
e.g., (sort of a dynamic instance attribute mixin)

obj ..= a, b, c # a sequence of objects

then

obj.x # looks for obj.x, then a.x then b.x then c.x before giving up with attribute error
obj ..=() # clears chained attribute name space ?

Then you could add a property to the namespace of a module by adding an object whose class defines
the property, like

mod ..= objhavingproperty # same tuple ambiguity as with 'somestr' % x

something analogous to += on immutables would have to be done for builtin objects I suppose.

Just adding more mulch/worms to the idea garden ;-)

Regards,
Bengt Richter

[Aahz]

In article <ma***************************************@python. org>,
Paolino <pa*************@tiscali.it> wrote:

Why descriptor mechanism doesn't apply to modules?

The real answer: Because Guido doesn't want it to
--
Aahz (aa**@pythoncraft.com) <*> http://www.pythoncraft.com/

The way to build large Python applications is to componentize and
loosely-couple the hell out of everything.

[Simon Burton]

I've been needing a module level __getattr__ for some c library
wrapping. This has solved the problem:

# mymod.py:
import sys
from new import module
class ModuleProxy(module):
def __init__( self, name, master ):
module.__init__( self, name )
self._master = master
self.__dict__["__all__"] = dir(master)
def __getattr__(self, name):
attr = getattr( self._master, name )
return attr

# ... end of file:
sys.modules["mymod"] = ModuleProxy("mymod",sys.modules["mymod"])
--Simon Burton

[Paolino]

Scott David Daniels wrote:

Well, an entry in the dictionary "sys.modules" is what is returned by
imports, so you could either pre-replace or post-replace a module by
an object which uses the descriptor technique to get to some (but not
necessarily all) of the attributes. Think of this technique as a
hack to get to a goal, rather than a good technique to use; good for
debugging, not so nice for production work. If you still don't know
how to do this from this admittedly sketchy description, I'd suggest
you avoid trying it altogether.

Thanks.Is don't know if this is *the* way to wrap the module ?

import new,sys
class Free(new.module):
def __init__(self,moduleName):
new.module.__init__(self,moduleName)
assert moduleName in sys.modules
self.__dict__.update(sys.modules[moduleName].__dict__)
sys.modules[moduleName]=self
def desc(self):
pass

Free(__name__) #'no errors'
but 'desc' is not defined in this namespace.

Paolino

___________________________________
Yahoo! Mail: gratis 1GB per i messaggi e allegati da 10MB
http://mail.yahoo.it

[Paolino]

Bengt Richter wrote:

Another thought/bf would be a way to extend the attribute namespace of an arbitrary object
by chaining the attribute name spaces of a sequence of objects (this would be a language mod)
e.g., (sort of a dynamic instance attribute mixin)

obj ..= a, b, c # a sequence of objects

then

obj.x # looks for obj.x, then a.x then b.x then c.x before giving up with attribute error
obj ..=() # clears chained attribute name space ?

Something like a __lookup__ attribute for all instances which are
namespaces.
Are all objects namespaces or have all objects a namespace?

Also syntax:
lookup(obj).append(x)

I suppose bound methods will not be found,but in that case some tolerant
unbound methods could do,naturally if the PEP on eliminating them will
be accepted.That check they do on the first parameter they push in the
function (aka 'self') is really contrasting dinamycal Python IMO.
Then you could add a property to the namespace of a module by adding an object whose class defines
the property, like

mod ..= objhavingproperty # same tuple ambiguity as with 'somestr' % x

something analogous to += on immutables would have to be done for builtin objects I suppose.

This is a little hard for me.Has it something to do with extensions also?

Regards Paolino

___________________________________
Yahoo! Mail: gratis 1GB per i messaggi e allegati da 10MB
http://mail.yahoo.it