Invisible function attributes

[Olivier Lefevre]

Python 2.3.... foo.a = 1
....
So it would appear that function attributes are not really
there until the first call to the function. If that is the
intended behaviour, it is really weird. I couldn't find any
explicit discussion of this topic in the LRM.

Thanks if anyone can shed some light on this,

-- O.L.

 

[Tom Lee]

Python 2.3


... foo.a = 1
...


{'a': 1}


So it would appear that function attributes are not really
there until the first call to the function. If that is the
intended behaviour, it is really weird. I couldn't find any
explicit discussion of this topic in the LRM.

Thanks if anyone can shed some light on this,

-- O.L.

Makes sense to me.

The foo.a = 1 line should never be executed until foo() is executed.
Thus, foo.a is never set before the call to foo.

- TL

 

[Peter Otten]

def foo():
try:
foo.a += 1 # executed every time you call the function
except AttributeError:
foo.a = 1 # set to one if it's not already there

foo.b = 1 # executed once

print vars(foo) # function body not yet called {'b': 1}

for i in range(3):
foo()
print vars(foo)

Loop output is

{'a': 1, 'b': 1}
{'a': 2, 'b': 1}
{'a': 3, 'b': 1}

That is all perfectly sane as code in the function body is never executed
unless you call the function, whereas code on the module level is executed
immediately as the module is imported. So put foo.attr = ... into the
function body iff you want it to execute every time the function is
invoked; otherwise put it into the module startup code, i. e. do not indent
it.

Peter

 

[Terry Reedy]

Python 2.3
... foo.a = 1

If you want foo to be attributed before it is called, move the setter
outside the function.
{'a': 1}

Terry J Reedy

 

[Bob Gailer]

Of course the inevitable question is why do you want to do this? Consider
creating a class instead (with a __call__ method if you want to call the
instance as a function).

class Foo:
a = 1 # pre-initialized property
def __call__(self):
print self.a

foo = Foo()
foo()
# prints 1

Bob Gailer
(e-mail address removed)
303 442 2625

 

[John J. Lee]

So it would appear that function attributes are not really
there until the first call to the function. If that is the

Not function attributes in general, just those that are first assigned
to in the function body. No special rule here, though, because...

intended behaviour, it is really weird. I couldn't find any
explicit discussion of this topic in the LRM.

Thanks if anyone can shed some light on this,

....what I'm guessing you haven't figured out yet is that everything
works like this in Python. For example, what might be called a 'class
declaration' in other languages isn't really a declaration in Python,
it's code that gets executed at runtime. Same is true of functions:

if WANT_SPAM:
def sayhello(): print "spam"
else:
def sayhello(): print "eggs"

sayhello()


And your foo.a = 1 isn't a declaration (Python doesn't have them,
really), it's just an attribute assignment.


John

 

[Olivier Lefevre]

Thanks to all those who replied.

...what I'm guessing you haven't figured out yet is that everything
works like this in Python.

Very possibly. I am coming to python from Java and I want to investigate
the weird stuff precisely because either it's a one-off (in which case
I'll make a note to myself to ignore it and not use it) or it holds the
key to what is specific about the language. I seem to have hit pay dirt
with this one ;-)

Nudged by the dot syntax, I was thinking of this function attribute as
if it were a sort of class member (i.e., pretending for a while this
function is a class) and, since functions can't have instances, treating
it as a sort of static member of the function, which should be available
as soon as declared. Obviously I got it all wrong. Instead, they work
like local variables except that they "persist" after the function has
exited. That still feels weird to me. What are they used for? Give me
a compelling reason to have such a beast in the language.

OTOH, does this behaviour have anything to do with so-called "futures"?

-- O.L.

 

[John Roth]

Thanks to all those who replied.


Very possibly. I am coming to python from Java and I want to investigate
the weird stuff precisely because either it's a one-off (in which case
I'll make a note to myself to ignore it and not use it) or it holds the
key to what is specific about the language. I seem to have hit pay dirt
with this one ;-)

Nudged by the dot syntax, I was thinking of this function attribute as
if it were a sort of class member (i.e., pretending for a while this
function is a class) and, since functions can't have instances, treating
it as a sort of static member of the function, which should be available
as soon as declared. Obviously I got it all wrong. Instead, they work
like local variables except that they "persist" after the function has
exited. That still feels weird to me. What are they used for? Give me
a compelling reason to have such a beast in the language.

I don't know of a really compelling reason, other than it simply
works that way. Like everything else in the language, functions
are objects, which means that they have a dictionary at their
core. Therefore, functions can have attributes.

The only use I can think of would be definitely advanced
programming. Functions are first class objects, which means
they can be rebound anywhere you want them. If you find
a good reason to do that, then as an extension you might find
a reason to add attributes to classify what you've got so you
can manage the process.

As I said, I'm reaching with this one...

OTOH, does this behaviour have anything to do with so-called "futures"?

No. Future is a feature so that experimental features can be added to the
language in one release, and then made standard in a future release.

John Roth

 

[Jacek Generowicz]

Nudged by the dot syntax, I was thinking of this function attribute as
if it were a sort of class member (i.e., pretending for a while this
function is a class) and, since functions can't have instances, treating
it as a sort of static member of the function, which should be available
as soon as declared.

There is no declaration in Python; you just create attributes by
binding objects to names.

Obviously I got it all wrong. Instead, they work like local
variables except that they "persist" after the function has exited.

Hmmm. No. Functions are first class objects, and you can dynamically
add attributes to them as you go along (as is the case for many, but
not all, other types of objects).

Do you realize that the foo identifier you used in your example is not
inextricably linked to any function whose name is foo? The name of a
function, and the variables to which it is bound are two different
concepts.

Consequently, the foo.a in the function body only _coincidentally_
refers to an attribute of the function in which it appears.

Consider:
Traceback (most recent call last):
File "<stdin>", line 1, in ?

The error at "bar(4)" in the above should hint at the fact that the
"foo" in "foo.a" in the definition of the function foo, does not refer
to the function itself but to whatever object which happens to be
bound to the name "foo" at the time the function is being exectuted.

What are they used for? Give me a compelling reason to have such a
beast in the language.

Now there's a good question I haven't found a use for these
myself ... but then I haven't looked very hard for one.

Given that function attributes were added to the language (in version
2.1 or 2.2 ?), I guess someone felt a need for them, and Guido agreed.

 

[Olivier Lefevre]

Python has no declarations, only executable statements.

I think this was the key to my confusion in this case.

Traceback (most recent call last):
TypeError: 'int' object has only read-only attributes (assign to .a)

The way I read this, it says that a was bound to the name foo,
not to the function foo stood for at the time of that function's
definition; you are saying as much. Thus "foo.a" has to be looked
up and resolved anew for each call. This must be costly. Why was
it done this way?

OTOH I read the func attr PEP and it says that they are
implemented via a dict inside the function object. If so,
shouldn't they be bound to the function object rather than
to its name?? Or is func_dict itself an attribute?

-- O.L.

 

[John J. Lee]

Crumbs, is this thread still running?

I think this was the key to my confusion in this case.

Certainly part of it.

The way I read this, it says that a was bound to the name foo,

I'm not sure what's in your mind, and I'm only looking at this one
message (the rest aren't in my newsreader) but that traceback says you
tried to assign to an attribute on an integer, like so:

3.a = "bananas"


Doesn't make a lot of sense!

(Of course, you probably were assigning to a named integer, not a
literal

foo = 3
foo.a = "bananas"
)

[...]

OTOH I read the func attr PEP and it says that they are
implemented via a dict inside the function object. If so,
shouldn't they be bound to the function object rather than
to its name??

They are.

Or is func_dict itself an attribute?

It is.


John