block scope?

[Neal Becker]

One thing I sometimes miss, which is common in some other languages (c++),
is idea of block scope. It would be useful to have variables that did not
outlive their block, primarily to avoid name clashes. This also leads to
more readable code. I wonder if this has been discussed?

 

[James Stroud]

One thing I sometimes miss, which is common in some other languages (c++),
is idea of block scope. It would be useful to have variables that did not
outlive their block, primarily to avoid name clashes. This also leads to
more readable code. I wonder if this has been discussed?

Probably, with good code, block scope would be overkill, except that I
would welcome list comprehensions to have a new scope:


py> i
------------------------------------------------------------
Traceback (most recent call last):
File "<ipython console>", line 1, in <module>
<type 'exceptions.NameError'>: name 'i' is not defined

py> [i for i in xrange(4)]
[0, 1, 2, 3]
py> i # hoping for NameError
3

 

[Paul Rubin]

Probably, with good code, block scope would be overkill, except that I
would welcome list comprehensions to have a new scope:

Block scope is a win because it gets rid of the uncertainty of whether
the variable is used outside the block or not. The "good code" theory
(just manually don't use the variable elsewhere) doesn't always hold
up under release deadline pressure and so on and doesn't make sense
anyway. What's the point of NOT having block scope if you don't want
to allow for creating variables in inner blocks and using them in
other blocks? I think it's best to require creating the variable
in a mutually enclosing scope if you want to use it that way.

 

[John Nagle]

Block scope is a win because it gets rid of the uncertainty of whether
the variable is used outside the block or not.

In a language with few declarations, it's probably best not to
have too many different nested scopes. Python has a reasonable
compromise in this area. Functions and classes have a scope, but
"if" and "for" do not. That works adequately.

Javascript got it wrong. They have declarations, but the default,
in the absence of a declaration, is global, not local or an error.
Bad design. It's a result of retrofitting declarations to a language,
which usually has painful aftereffects.

John Nagle

 

[Paul Rubin]

In a language with few declarations, it's probably best not to
have too many different nested scopes. Python has a reasonable
compromise in this area. Functions and classes have a scope, but
"if" and "for" do not. That works adequately.

I think Perl did this pretty good. If you say "my $i" that declares
$i to have block scope, and it's considered good practice to do this,
but it's not required. You can say "for (my $i=0; $i < 5; $i++) { ... }"
and that gives $i the same scope as the for loop. Come to think of it
you can do something similar in C++.

 

[James Stroud]

I think Perl did this pretty good. If you say "my $i" that declares
$i to have block scope, and it's considered good practice to do this,
but it's not required. You can say "for (my $i=0; $i < 5; $i++) { ... }"
and that gives $i the same scope as the for loop. Come to think of it
you can do something similar in C++.

How then might one define a block? All lines at the same indent level
and the lines nested within those lines?

i = 5
for my i in xrange(4):
if i: # skips first when i is 0
my i = 100
if i:
print i # of course 100
break
print i # i is between 0 & 3 here
print i # i is 5 here


Doesn't leave a particularly bad taste in one's mouth, I guess (except
for the intended abuse).

James

 

[Neal Becker]

How then might one define a block? All lines at the same indent level
and the lines nested within those lines?

i = 5
for my i in xrange(4):
if i: # skips first when i is 0
my i = 100
if i:
print i # of course 100
break
print i # i is between 0 & 3 here
print i # i is 5 here


Doesn't leave a particularly bad taste in one's mouth, I guess (except
for the intended abuse).

James

Yes, the above is pretty much what I had in mind. +1.

 

[irstas]

Probably, with good code, block scope would be overkill, except that I
would welcome list comprehensions to have a new scope:

Generator expressions have a new scope, and in Python 3.0 list
comprehensions will have one as well (according to http://www.python.org/dev/peps/pep-0289/
). It's a fix that might break existing code so it couldn't be
introduced in "minor" versions like 2.4 and 2.5.

 

[Alex Martelli]

Yes, the above is pretty much what I had in mind. +1.

I prefer Java's approach (14.4.2 in the JLS 2nd edition): forbid "inner"
blocks from shadowing variables in "outer" ones. I quote:
"""
If a declaration of an identifier as a local variable of the same
method, constructor, or initializer block appears within the scope of a
parameter or local variable of the same name, a compile-time error
occurs.
Thus the following example does not compile:

class Test {
public static void main(String[] args) {
int i;
for (int i = 0; i < 10; i++)
System.out.println(i);
}
}
This restriction helps to detect some otherwise very obscure bugs.
"""
I entirely agree with the JLS here, having fought just such bugs in C++
and other languages that lack the restriction in question. I just wish
Python had adopted the same restriction regarding nested functions, when
proper lexical scoping was introduced -- I argued for it at the time,
but backwards compatibility blocked its introduction. There are
definitely NOT many Java-specific language characteristics that I like,
but this one is a winner!-) [[but, I disagree with the lack in Java of
a similar restriction against shadowing between instance variables and
local variables, and the weak rationale for that in the JLS]].


Alex

 

[Steve Holden]

Yes, the above is pretty much what I had in mind. +1.

I prefer Java's approach (14.4.2 in the JLS 2nd edition): forbid "inner"
blocks from shadowing variables in "outer" ones. I quote:
"""
If a declaration of an identifier as a local variable of the same
method, constructor, or initializer block appears within the scope of a
parameter or local variable of the same name, a compile-time error
occurs.
Thus the following example does not compile:

class Test {
public static void main(String[] args) {
int i;
for (int i = 0; i < 10; i++)
System.out.println(i);
}
}
This restriction helps to detect some otherwise very obscure bugs.
"""
I entirely agree with the JLS here, having fought just such bugs in C++
and other languages that lack the restriction in question. I just wish
Python had adopted the same restriction regarding nested functions, when
proper lexical scoping was introduced -- I argued for it at the time,
but backwards compatibility blocked its introduction. There are
definitely NOT many Java-specific language characteristics that I like,
but this one is a winner!-) [[but, I disagree with the lack in Java of
a similar restriction against shadowing between instance variables and
local variables, and the weak rationale for that in the JLS]].

What do you think the chances are of this being accepted for Python 3.0?
It is indeed about the most rational approach, though of course it does
cause problems with dynamic namespaces.

regards
Steve

 

[Paul Rubin]

Thus the following example does not compile:
class Test {
public static void main(String[] args) {
int i;
for (int i = 0; i < 10; i++)

I'm ok with this; at the minimum, I think such nesting should produce
a warning message.

 

[Alex Martelli]

What do you think the chances are of this being accepted for Python 3.0?
It is indeed about the most rational approach, though of course it does
cause problems with dynamic namespaces.

What problems do you have in mind? The compiler already determines the
set of names that are local variables for a function; all it needs to do
is diagnose an error or warning if the set of names for a nested
function overlaps with that of an outer one.

I shamefully admit that I haven't followed Python 3.0 discussions much
lately, so I don't really know what's planned on this issue.


Alex

 

[Alex Martelli]

Thus the following example does not compile:
class Test {
public static void main(String[] args) {
int i;
for (int i = 0; i < 10; i++)

I'm ok with this; at the minimum, I think such nesting should produce
a warning message.

Yes, a warning could surely be a reasonable compromise.


Alex

 

[Aahz]

What problems do you have in mind? The compiler already determines the
set of names that are local variables for a function; all it needs to do
is diagnose an error or warning if the set of names for a nested
function overlaps with that of an outer one.

exec?

 

[John Nagle]

I think Perl did this pretty good. If you say "my $i" that declares
$i to have block scope, and it's considered good practice to do this,
but it's not required. You can say "for (my $i=0; $i < 5; $i++) { ... }"
and that gives $i the same scope as the for loop. Come to think of it
you can do something similar in C++.

Those languages have local declarations. "my" is a local
declaration. If you have explicit declarations, explict block
scope is no problem. Without that, there are problems. Consider

def foo(s, sname) :
if s is None :
result = ""
else :
result = s
msg = "Value of %s is %s" % (sname, result)
return(msg)

It's not that unusual in Python to initialize a variable on
two converging paths. With block scope, you'd break
code that did that.


John Nagle

 

[Alex Martelli]

exec?

option 1: that just runs the compiler a bit later -- thus transforming
ClashingVariableError into a runtime issue, exactly like it already does
for SyntaxError.

option 2: since a function containing any exec statement does not
benefit from the normal optimization of local variables, let it also
forgo the normal diagnosis of shadowed/clashing names.

option 3: extend the already-existing prohibition of mixing exec with
nested functions:
.... def inner(): return x
.... exec('x=23')
.... return inner()
....
File "<stdin>", line 3
SyntaxError: unqualified exec is not allowed in function 'outer' it
contains a nested function with free variables

to prohibit any mixing of exec and nested functions (not just those
cases where the nested function has free variables).


My personal favorite is option 3.


Alex

 

[Paul Rubin]

option 1: that just runs the compiler a bit later ...

Besides exec, there's also locals(), i.e.
locals['x'] = 5
can shadow a variable. Any bad results are probably deserved

 

[MRAB]

How then might one define a block? All lines at the same indent level
and the lines nested within those lines?

i = 5
for my i in xrange(4):
if i: # skips first when i is 0
my i = 100
if i:
print i # of course 100
break
print i # i is between 0 & 3 here
print i # i is 5 here

Doesn't leave a particularly bad taste in one's mouth, I guess (except
for the intended abuse).

How about something like this instead:

i = 5
block:
for i in xrange(4):
if i: # skips first when i is 0
block:
i = 100
if i:
print i # of course 100
break
print i # i is between 0 & 3 here
print i # i is 5 here

Any variable that's assigned to within a block would be local to that
block, as it is in functions.

 

[Alex Martelli]

option 1: that just runs the compiler a bit later ...

Besides exec, there's also locals(), i.e.
locals['x'] = 5
can shadow a variable. Any bad results are probably deserved

locals['x']=5

Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TypeError: 'builtin_function_or_method' object does not support item
assignment

I suspect you want to index the results of calling locals(), rather than
the builtin function itself. However:
.... locals()['x'] = 5
.... return x
.... Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "<stdin>", line 3, in f
NameError: global name 'x' is not defined

No "shadowing", as you see: the compiler knows that x is NOT local,
because it's not assigned to (the indexing of locals() does not count:
the compiler's not expected to detect that), so it's going to look it up
as a global variable (and not find it in this case).

I think that ideally there should be a runtime error when assigning an
item of locals() with a key that's not a local variable name (possibly
excepting functions containing exec, which are kind of screwy anyway).


Alex

 

[Paul Rubin]

locals['x']=5

Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TypeError: 'builtin_function_or_method' object does not support item
assignment

Whoops, yeah, meant "locals()['x'] = 5".

I think that ideally there should be a runtime error when assigning an
item of locals() with a key that's not a local variable name (possibly
excepting functions containing exec, which are kind of screwy anyway).

I have no opinion of this, locals() has always seemed like a crazy
part of the language to me and I never use it. I'd be happy to see it
gone since it makes compiling a lot easier.