del in Python

Posted on by Chris Warburton

In a language like Scheme we can write an expression like this:

(+ 5 (let ((x 7)) (+ x x)))

The let expression defines a variable x which is only in scope for the (+ x x) sub-expression; that binding of x can’t be used outside that sub-expression. In this case, the expression will evaluate like this:

(+ 5 (let ((x 7)) (+ x x)))

;; Substitute in the value of `x`
(+ 5 (let () (+ 7 7)))

;; Unwrap the redundant `let`
(+ 5 (+ 7 7))

;; Perform the inner addition
(+ 5 14)

;; Perform the outer addition
19

By using a let expression we can be sure that (a) our expression is using the correct value of x that we intended and (b) we’re not breaking the value of x in any other expressions.

Consider the following:

(+ x (let ((x 7)) (+ x x)))

Now we have two different variables, which are both called x. Again, evaluation still ensures (a) and (b):

(+ x (let ((x 7)) (+ x x)))

;; Substitute in the value of the "inner" `x`
(+ x (let () (+ 7 7)))

;; Unwrap the redundant `let`
(+ x (+ 7 7))

;; Perform the inner addition
(+ x 14)

By using let, we’ve avoided breaking the other usage of x; our “inner” x has disappeared, since its job is done.

Python doesn’t have let to delimit variable scopes like this (at least, in early versions; it may get added!). Instead, Python’s scopes are tied to other language constructs, like function bodies, loop bodies, etc.

There are ways to fake let using with, although that seems to use dynamic scope rather than “proper” lexical scope, and keep in mind that Python was around for a while before with was added.

We can actually emulate a proper lexically-scoped let in Python by using “one-shot” functions (this is actually quite common in Javascript, at least before let got added to that language):

x + (lambda x=7: x + x)()

Here the x argument of lambda is delimited to the body of that function, which we call immediately.

This is pretty horrible to write because Python has strict rules regarding function definitions, e.g. lambda can only contain one expression; whilst def must name the function, and is a statement rather than an expression.

To alleviate this, we can usually avoid opening a new scope by using an assignment statement like x = 7. This isn’t a drop-in replacement though, since Python allows statements to contain expressions, but doesn’t allow expressions to contain statements. Hence we can’t say:

x + (x = 7; x + x)

Instead, we have to pull the statement out of the expression:

x = 7
x + (x + x)

This rearrangement actually changes the semantics of the code though, since it’s now equivalent to (lambda x=7: x + (x + x))() rather than x + (lambda x=7: x + x)()

To recover the original meaning, we have to calculate each part of the expression separately (basically re-inventing Python’s order of execution), introduce fresh names to keep the intermediate results separate (re-inventing Python’s scoping), and so on. There’s no way to write this in general, since it depends on what the existing variable bindings are.

In any case, the result will still not be equivalent to a “real” let, or a “one-shot” lambda, since all of our intermediate variables will still be in-scope.

To avoid that, we can use the del keyword to get rid of them.

Hence if we’re writing some Python code which we’d like to simplify by splitting up the scope, but we don’t want to introduce new functions, etc. (since that would complicate the code), then it can often be enough to introduce fresh variables with assignments, use them for whatever would have gone into a scope’s body, then use del to “clean them up”, so their presence doesn’t complicate our understanding of the ubsequent code.