Issue 4246: doc: execution model - clear and complete example in documentation (original) (raw)

I'd like to propose adding some complete example regarding scopes and bindings resolving to execution model description. There is no week on pl.comp.lang.python without a question about UnboundLocalError problem. I'm getting tired answering that. ;-) It does not have to look so verbose as my (attached) example, but please add some example, which will clarify this issue.

Your example seem too verbose and diffuse. Perhaps something more focused on what people do wrong would be more helpful. I presume you mean something like this -- with or without x=2 before the def.

def f(): print (x) x = 1

f() Traceback (most recent call last): File "<pyshell#31>", line 1, in f() File "<pyshell#30>", line 2, in f print (x) UnboundLocalError: local variable 'x' referenced before assignment

What are the other ways people get the error.

The FAQ for this was much improved in 2009 (issue 7290): http://docs.python.org/py3k/faq/programming.html#why-am-i-getting-an-unboundlocalerror-when-the-variable-has-a-value

To support the claim that this keeps biting people, at least the following bug reports all came from people misundestanding this: issue 10043 issue 9049 issue 7747 issue 7276 issue 6833 issue 5763 issue 4109 (understood effect of =, surprised by +=) issue 972467 issue 850110 issue 463640 These are just the people who were persistent enough to open a bug (and in most cases managed to produce a minimal example); many more ask on c.l.p, StackOverflow (>50 hits for UnboundLocalError, many of which are this exact issue) etc., or just give up.

[Interesting point: people mostly complain when the unbound reference occurs textually before the assignment (though there's selection bias here), and many of them complain that things happen "out of order". It seems half the misunderstanding is that people expect variables to become localized when first assigned - they don't understand it's a static decision affecting all occurences in a function.]

The central problem I believe is that when people try to modify a non-local var and get UnboundLocalError: local variable foo referenced before assignment their mental model of Python scopes is wrong, so the error message is useless for them (what 'local variable foo' is it talking about?), and have no idea where to look next.

Also, I'm afraid people just won't internalize this issue until it bites them at least once (I've seen a Python course where I had explained this, with a print-before-assignment example, and 2 days later a student was bitten by the exception and was at a loss. Therefore, I think providing a clear learning path from UnboundLocalError is crucial.

==>

I propose (i.e. am working on patch(s)) attacking it at many points:

(1) Expand UnboundLocalError message to point to help('NAMESPACES') [Execution Model → Naming and Binding] and/or the FAQ. (requirement IMHO: use a help() ref that can be followed the from a Python prompt on an offline machine; not sure if FAQ can work this way.)

(1B) Ideally, detect if the var is bound in outer/global scope 
     and suggest help('nonlocal') / help('global') approriately.

(1C) Ideally, improve UnboundLocalError to explain "foo is local 
     throughout the function because it's bound at line 42".

(2) Add an example to Naming and Binding section. Currently, it's a bunch of legalese, defining 7 terms(!) before you get to the meat. Mortal users won't survive that. Being the language spec, the precise legalese must stay there; but it'd be good to prepend it with a human-readable blurb and example regarding this issue.

(3) Improve the tutorial. Sections 4.6 [Defining functions] and 9.2 [Scopes and Namespaces] are relevant. 4.6 mentions the issue of assignment to global but neither section has a clear example. And 9.2 is scary IMHO; I'll see if I can make it any better...

(4) Add examples to documentation of global & nonlocal? Not clear if helpful, probably too much.