Logo Questions Linux Laravel Mysql Ubuntu Git Menu
 

Unpacking, extended unpacking and nested extended unpacking

People also ask

What do you mean by packing and unpacking?

Unpacking in Python refers to an operation that consists of assigning an iterable of values to a tuple (or list ) of variables in a single assignment statement. As a complement, the term packing can be used when we collect several values in a single variable using the iterable unpacking operator, * .

What do you mean by packing and unpacking of tuples?

Tuple packing refers to assigning multiple values into a tuple. Tuple unpacking refers to assigning a tuple into multiple variables.

What is tuple expansion?

Multiple assignment (also known as tuple unpacking or iterable unpacking) allows you to assign multiple variables at the same time in one line of code. This feature often seems simple after you've learned about it, but it can be tricky to recall multiple assignment when you need it most.


My apologies for the length of this post, but I decided to opt for completeness.

Once you know a few basic rules, it's not hard to generalize them. I'll do my best to explain with a few examples. Since you're talking about evaluating these "by hand," I'll suggest some simple substitution rules. Basically, you might find it easier to understand an expression if all the iterables are formatted in the same way.

For the purposes of unpacking only, the following substitutions are valid on the right side of the = (i.e. for rvalues):

'XY' -> ('X', 'Y')
['X', 'Y'] -> ('X', 'Y')

If you find that a value doesn't get unpacked, then you'll undo the substitution. (See below for further explanation.)

Also, when you see "naked" commas, pretend there's a top-level tuple. Do this on both the left and the right side (i.e. for lvalues and rvalues):

'X', 'Y' -> ('X', 'Y')
a, b -> (a, b)

With those simple rules in mind, here are some examples:

(a,b), c = "XY", "Z"                 # a = 'X', b = 'Y', c = 'Z'

Applying the above rules, we convert "XY" to ('X', 'Y'), and cover the naked commas in parens:

((a, b), c) = (('X', 'Y'), 'Z')

The visual correspondence here makes it fairly obvious how the assignment works.

Here's an erroneous example:

(a,b), c = "XYZ"

Following the above substitution rules, we get the below:

((a, b), c) = ('X', 'Y', 'Z')

This is clearly erroneous; the nested structures don't match up. Now let's see how it works for a slightly more complex example:

(a,b), c, = [1,2],'this'             # a = '1', b = '2', c = 'this'

Applying the above rules, we get

((a, b), c) = ((1, 2), ('t', 'h', 'i', 's'))

But now it's clear from the structure that 'this' won't be unpacked, but assigned directly to c. So we undo the substitution.

((a, b), c) = ((1, 2), 'this')

Now let's see what happens when we wrap c in a tuple:

(a,b), (c,) = [1,2],'this'           # ERROR -- too many values to unpack

Becomes

((a, b), (c,)) = ((1, 2), ('t', 'h', 'i', 's'))

Again, the error is obvious. c is no longer a naked variable, but a variable inside a sequence, and so the corresponding sequence on the right is unpacked into (c,). But the sequences have a different length, so there's an error.

Now for extended unpacking using the * operator. This is a bit more complex, but it's still fairly straightforward. A variable preceded by * becomes a list, which contains any items from the corresponding sequence that aren't assigned to variable names. Starting with a fairly simple example:

a, *b, c = "X...Y"                   # a = 'X', b = ['.','.','.'], c = 'Y'

This becomes

(a, *b, c) = ('X', '.', '.', '.', 'Y')

The simplest way to analyze this is to work from the ends. 'X' is assigned to a and 'Y' is assigned to c. The remaining values in the sequence are put in a list and assigned to b.

Lvalues like (*a, b) and (a, *b) are just special cases of the above. You can't have two * operators inside one lvalue sequence because it would be ambiguous. Where would the values go in something like this (a, *b, *c, d) -- in b or c? I'll consider the nested case in a moment.

*a = 1                               # ERROR -- target must be in a list or tuple

Here the error is fairly self-explanatory. The target (*a) must be in a tuple.

*a, = (1,2)                          # a = [1,2]

This works because there's a naked comma. Applying the rules...

(*a,) = (1, 2)

Since there are no variables other than *a, *a slurps up all the values in the rvalue sequence. What if you replace the (1, 2) with a single value?

*a, = 1                              # ERROR -- 'int' object is not iterable

becomes

(*a,) = 1

Again, the error here is self-explanatory. You can't unpack something that isn't a sequence, and *a needs something to unpack. So we put it in a sequence

*a, = [1]                            # a = [1]

Which is eqivalent to

(*a,) = (1,)

Finally, this is a common point of confusion: (1) is the same as 1 -- you need a comma to distinguish a tuple from an arithmetic statement.

*a, = (1)                            # ERROR -- 'int' object is not 

Now for nesting. Actually this example wasn't in your "NESTED" section; perhaps you didn't realize it was nested?

(a,b), *c = 'XY', 2, 3               # a = 'X', b = 'Y', c = [2,3]

Becomes

((a, b), *c) = (('X', 'Y'), 2, 3)

The first value in the top-level tuple gets assigned, and the remaining values in the top-level tuple (2 and 3) are assigned to c -- just as we should expect.

(a,b),c = 1,2,3                      # ERROR -- too many values to unpack
*(a,b), c = 1,2,3                    # a = 1, b = 2, c = 3

I've already explained above why the first line throws an error. The second line is silly but here's why it works:

(*(a, b), c) = (1, 2, 3)

As previously explained, we work from the ends. 3 is assigned to c, and then the remaining values are assigned to the variable with the * preceding it, in this case, (a, b). So that's equivalent to (a, b) = (1, 2), which happens to work because there are the right number of elements. I can't think of any reason this would ever appear in working code. Similarly,

*(a, *b), c = 'this'                 # a = 't', b = ['h', 'i'], c = 's'

becomes

(*(a, *b), c) = ('t', 'h', 'i', 's')

Working from the ends, 's' is assigned to c, and ('t', 'h', 'i') is assigned to (a, *b). Working again from the ends, 't' is assigned to a, and ('h', 'i') is assigned to b as a list. This is another silly example that should never appear in working code.


I find the Python 2 tuple unpacking pretty straightforward. Each name on the left corresponds with either an entire sequence or a single item in a sequence on the right. If names correspond to single items of any sequence, then there must be enough names to cover all of the items.

Extended unpacking, however, can certainly be confusing, because it is so powerful. The reality is you should never be doing the last 10 or more valid examples you gave -- if the data is that structured, it should be in a dict or a class instance, not unstructured forms like lists.

Clearly, the new syntax can be abused. The answer to your question is that you shouldn't have to read expressions like that -- they're bad practice and I doubt they'll be used.

Just because you can write arbitrarily complex expressions doesn't mean you should. You could write code like map(map, iterable_of_transformations, map(map, iterable_of_transformations, iterable_of_iterables_of_iterables)) but you don't.


I you think your code may be misleading use other form to express it.

It's like using extra brackets in expressions to avoid questions about operators precedence. I'ts always a good investment to make your code readable.

I prefer to use unpacking only for simple tasks like swap.