Image of closure

Introduction

This question (not a question on Qiita)

I understand that closures can refer to outer variables, but I'm not convinced. Doesn't the variable [^ 1](including arguments) in the function disappear when the function ends? The variable may be held somewhere, but where is it specifically?

[^ 1]: To be exact, the object referenced by the variable

On the contrary

"First you have to think that's what it is. If you want to know more, you have to read the Python source. By the way, it's tied to a function (closure)."

I answered, but when I thought about it a little more,

"Where did you learn from the concept of variables disappearing when the function ends (although they disappear)? It is assumed that you have no knowledge of language implementation (that is, knowledge that variables are secured on the stack [^ 2])."

[^ 2]: For more information about this area [here](https://qiita.com/junjis0203/items/7d3e63253a3d291a04c6#%E9%96%A2%E6%95%B0%E3%81%AE%E5% AE% 9F% E8% A1% 8C% E6% 96% B9% E6% B3% 95% E3% 83% AD% E3% 83% BC% E3% 82% AB% E3% 83% AB% E5% A4% See also 89% E6% 95% B0)

I was worried about that. That's why this time I will explain the concept around here with pictures.

Functions and variables and the end of functions

Before the closure is a normal function first.

Ordinary function


def add(a, b):
    return a + b

add(2, 3)

The last line calls ʻadd with ʻa as 2 and b as 3. If you draw this in the figure, it will be as follows.

func_exec_state1.png

The "execution state of add" is because of the "behavior when the function ends" which will be explained later (the story becomes inaccurate if "add" is used), but anyway, for the object "2". From the execution state, the arrow of ʻa and the arrow of b` appear for the object "3". Don't worry too much because a and b are upside down to match the figure you will draw later.

ʻWhen add ends, the "add execution state" disappears. Then, of course, the ʻa and b arrows that are out of the execution state also disappear. This is the situation where the variable disappears when the function ends.

func_exec_state2.png

Closures and variables

So far, we have explained "the image of variables when the function is executed" and "the image of variables when the function ends". With this in mind, let's move on to closures. There is a useless b variable, but it's convenient.

closure


def make_adder(a, b):
    def adder(x):
        return x + a
    return adder

adder3 = make_adder(2, 3)
adder3(4)

When the make_adder (2, 3) call is made and the time before the return of the make_adder (that is, after the ʻadder is defined) is drawn in the figure as follows. In other words, the "2" pointed to by ʻa is referenced by both make_adder (execution state) and ʻadder`. [^ 3]

func_exec_state3.png

[^ 3]: The arrow from "make_adder execution state" and the one pointed to are ʻadder`, but this is not an error because "the function itself has the name adder".

When make_adder finishes, the execution state disappears like a normal variable. However, the ʻadder body written in blue has been returned, so it will not disappear. And ʻa (pointed to" 2 ") referenced by ʻadder` does not disappear either.

func_exec_state4.png

In this way, ʻa (the one referenced by) remains, so when ʻadder3 (4) is called after that, the saved "2" can be retrieved and used.

closure_exec_state.png

That's it. Oh, I think nonlocal is a little more annoying, but I haven't checked it, so I'll omit it.

reference

[Python] What is closure (function closure) It describes in detail how variables are stored.

[Read Python / Closure](http://dzeta.jp/~junjis/code_reading/index.php?Python%2F%E3%82%AF%E3%83%AD%E3%83%BC%E3%82 % B8% E3% 83% A3) This is a memo when I read "Details of Closure Implementation" about two and a half years ago.

Recommended Posts

Image of closure
[PyTorch] Image classification of CIFAR-10
Application of CNN2 image recognition
closure
Image capture of firefox using python
Extract feature points of an image
Image recognition of fruits using VGG16
A memorandum of closure survey contents
Basics of binarized image processing with Python
Closure implementation
Python: Application of image recognition using CNN
[Python] Calculation of image similarity (Dice coefficient)
Grayscale by matrix-Reinventor of Python image processing-
Image denoising
Drawing with Matrix-Reinventor of Python Image Processing-
Analysis of X-ray microtomography image by Python
Image recognition
Image crawler
[Image classification] Facial expression analysis of dogs
Matrix Convolution Filtering-Reinventor of Python Image Processing-
I tried image recognition of CIFAR-10 with Keras-Learning-
Extract dominant color of image by k-means clustering
Color page judgment of scanned image with python
I tried to correct the keystone of the image
Image processing? The story of starting Python for
Judge Yosakoi Naruko by image classification of Tensorflow.
Super (concise) summary of image classification by ArcFace
Basic principles of image recognition technology (for beginners)
Implementation of Deep Learning model for image recognition
ROS Lecture 119 Correct the color of the camera image
I tried using the image filter of OpenCV
[Small story] Download the image of Ghibli immediately
[Python + OpenCV] Whiten the transparent part of the image