The execution process of a Python program includes converting source code into bytecode (i.e. compilation) and executing the bytecode

Question:

We have to write some Python programs every day, either to process some text, or to do some system management work. After the program is written, you only need to type the python command to start the program and start executing it:

$ python some-program.py

So, how is a .py file in text form converted step by step into something that can be executed by the CPU? What about machine instructions? In addition, .pyc files may be generated during program execution. What are the functions of these files?

1. Execution process

Although Python looks more like an interpreted language like Shell script in terms of behavior, in fact, the execution principle of Python program is essentially the same as that of Java or C# and can be summarized For virtual machine and bytecode. Python executes the program in two steps: first compile the program code into bytecode, and then start the virtual machine to execute the bytecode:

Although the Python command is also called the Python interpreter , but it is fundamentally different from other scripting language interpreters. In fact, the Python interpreter consists of compiler and virtual machine. When the Python interpreter is started, it mainly performs the following two steps:

The compiler compiles the Python source code in the .py file into bytecode. The virtual machine executes the bytecode generated by the compiler line by line.

Therefore, the Python statements in the .py file are not directly converted into machine instructions, but into Python bytecode.

2. Bytecode

The compiled result of the Python program is bytecode, which contains a lot of content related to the operation of Python. Therefore, whether it is to have a deeper understanding of the operating mechanism of the Python virtual machine or to optimize the operating efficiency of the Python program, bytecode is the key content. So, what does Python bytecode look like? How can we obtain the bytecode of a Python program? Python provides a built-in function compile for instant compilation of source code. We only need to call the compile function with the source code to be compiled as a parameter to obtain the compilation result of the source code.

3. Source code compilation

Below, we compile a program through the compile function:

The source code is saved in the demo.py file:

PI = 3.14

def circle_area(r):
    return PI * r ** 2

class Person(object):
    def __init__(self, name):
        self.name = name

    def say(self):
        print('i am', self.name)

Compile Previously, the source code needed to be read from the file:

>>> text = open('D:\myspace\code\pythonCode\mix\demo.py').read()
>>> print(text)
PI = 3.14

def circle_area(r):
    return PI * r ** 2

class Person(object):
    def __init__(self, name):
        self.name = name

    def say(self):
        print('i am', self.name)

Then call the compile function to compile the source code:

>>> result = compile(text,'D:\myspace\code\pythonCode\mix\demo.py', 'exec')

There are 3 required parameters for the compile function:

source : Source code to be compiled

filename: file name where the source code is located

mode: compilation mode, exec means compiling the source code as a module

Three compilation modes:

exec: used to compile module source code

single: used to compile a single Python statement (interactively)

eval: used to compile an eval expression

4. PyCodeObject

Through the compile function, we get the final source code compilation result result:

>>> result
<code object <module> at 0x000001DEC2FCF680, file "D:\myspace\code\pythonCode\mix\demo.py", line 1>
>>> result.__class__
<class &#39;code&#39;>

Finally we get a code type object, and its corresponding underlying structure is PyCodeObject

The source code of PyCodeObject is as follows:

/* Bytecode object */
struct PyCodeObject {
    PyObject_HEAD
    int co_argcount;            /* #arguments, except *args */
    int co_posonlyargcount;     /* #positional only arguments */
    int co_kwonlyargcount;      /* #keyword only arguments */
    int co_nlocals;             /* #local variables */
    int co_stacksize;           /* #entries needed for evaluation stack */
    int co_flags;               /* CO_..., see below */
    int co_firstlineno;         /* first source line number */
    PyObject *co_code;          /* instruction opcodes */
    PyObject *co_consts;        /* list (constants used) */
    PyObject *co_names;         /* list of strings (names used) */
    PyObject *co_varnames;      /* tuple of strings (local variable names) */
    PyObject *co_freevars;      /* tuple of strings (free variable names) */
    PyObject *co_cellvars;      /* tuple of strings (cell variable names) */
    /* The rest aren't used in either hash or comparisons, except for co_name,
       used in both. This is done to preserve the name and line number
       for tracebacks and debuggers; otherwise, constant de-duplication
       would collapse identical functions/lambdas defined on different lines.
    */
    Py_ssize_t *co_cell2arg;    /* Maps cell vars which are arguments. */
    PyObject *co_filename;      /* unicode (where it was loaded from) */
    PyObject *co_name;          /* unicode (name, for reference) */
    PyObject *co_linetable;     /* string (encoding addr<->lineno mapping) See
                                   Objects/lnotab_notes.txt for details. */
    void *co_zombieframe;       /* for optimization only (see frameobject.c) */
    PyObject *co_weakreflist;   /* to support weakrefs to code objects */
    /* Scratch space for extra data relating to the code object.
       Type is a void* to keep the format private in codeobject.c to force
       people to go through the proper APIs. */
    void *co_extra;

    /* Per opcodes just-in-time cache
     *
     * To reduce cache size, we use indirect mapping from opcode index to
     * cache object:
     *   cache = co_opcache[co_opcache_map[next_instr - first_instr] - 1]
     */

    // co_opcache_map is indexed by (next_instr - first_instr).
    //  * 0 means there is no cache for this opcode.
    //  * n > 0 means there is cache in co_opcache[n-1].
    unsigned char *co_opcache_map;
    _PyOpcache *co_opcache;
    int co_opcache_flag;  // used to determine when create a cache.
    unsigned char co_opcache_size;  // length of co_opcache.
};

The code object PyCodeObject is used to store the compilation results, including bytecodes and constants, names, etc. involved in the code. Key fields include:

##co_argcountNumber of parametersco_kwonlyargcountNumber of keyword parametersco_nlocalsPartial Number of variablesco_stacksizeStack space required to execute the codeco_flagsIdentificationco_firstlinenoThe first line number of the code blockco_codeInstruction operation code, that is, bytecode co_constsConstant listco_namesName listco_varnamesLocal variable name list

下面打印看一下这些字段对应的数据：

通过co_code字段获得字节码：

>>> result.co_code
b'd\x00Z\x00d\x01d\x02\x84\x00Z\x01G\x00d\x03d\x04\x84\x00d\x04e\x02\x83\x03Z\x03d\x05S\x00'

通过co_names字段获得代码对象涉及的所有名字：

>>> result.co_names
('PI', 'circle_area', 'object', 'Person')

通过co_consts字段获得代码对象涉及的所有常量：

>>> result.co_consts
(3.14, <code object circle_area at 0x0000023D04D3F310, file "D:\myspace\code\pythonCode\mix\demo.py", line 3>, &#39;circle_area&#39;, <code object Person at 0x0000023D04D3F5D0, file "D:\myspace\code\pythonCode\mix\demo.py", line 6>, &#39;Person&#39;, None)

可以看到，常量列表中还有两个代码对象，其中一个是circle_area函数体，另一个是Person类定义体。对应Python中作用域的划分方式，可以自然联想到：每个作用域对应一个代码对象。如果这个假设成立，那么Person代码对象的常量列表中应该还包括两个代码对象：init函数体和say函数体。下面取出Person类代码对象来看一下：

>>> person_code = result.co_consts[3]
>>> person_code
<code object Person at 0x0000023D04D3F5D0, file "D:\myspace\code\pythonCode\mix\demo.py", line 6>
>>> person_code.co_consts
(&#39;Person&#39;, <code object __init__ at 0x0000023D04D3F470, file "D:\myspace\code\pythonCode\mix\demo.py", line 7>, &#39;Person.__init__&#39;, <code object say at 0x0000023D04D3F520, file "D:\myspace\code\pythonCode\mix\demo.py", line 10>, &#39;Person.say&#39;, None)

因此，我们得出结论：Python源码编译后，每个作用域都对应着一个代码对象，子作用域代码对象位于父作用域代码对象的常量列表里，层级一一对应。

至此，我们对Python源码的编译结果——代码对象PyCodeObject有了最基本的认识，后续会在虚拟机、函数机制、类机制中进一步学习。

5. 反编译

字节码是一串不可读的字节序列，跟二进制机器码一样。如果想读懂机器码，可以将其反汇编，那么字节码可以反编译吗？

通过dis模块可以将字节码反编译：

>>> import dis
>>> dis.dis(result.co_code)
 0 LOAD_CONST               0 (0)
 2 STORE_NAME               0 (0)
 4 LOAD_CONST               1 (1)
 6 LOAD_CONST               2 (2)
 8 MAKE_FUNCTION            0
10 STORE_NAME               1 (1)
12 LOAD_BUILD_CLASS
14 LOAD_CONST               3 (3)
16 LOAD_CONST               4 (4)
18 MAKE_FUNCTION            0
20 LOAD_CONST               4 (4)
22 LOAD_NAME                2 (2)
24 CALL_FUNCTION            3
26 STORE_NAME               3 (3)
28 LOAD_CONST               5 (5)
30 RETURN_VALUE

字节码反编译后的结果和汇编语言很类似。其中，第一列是字节码的偏移量，第二列是指令，第三列是操作数。以第一条字节码为例，LOAD_CONST指令将常量加载进栈，常量下标由操作数给出，而下标为0的常量是：

>>> result.co_consts[0]3.14

这样，第一条字节码的意义就明确了：将常量3.14加载到栈。

由于代码对象保存了字节码、常量、名字等上下文信息，因此直接对代码对象进行反编译可以得到更清晰的结果：

>>>dis.dis(result)
  1           0 LOAD_CONST               0 (3.14)
              2 STORE_NAME               0 (PI)

  3           4 LOAD_CONST               1 (<code object circle_area at 0x0000023D04D3F310, file "D:\myspace\code\pythonCode\mix\demo.py", line 3>)
              6 LOAD_CONST               2 (&#39;circle_area&#39;)
              8 MAKE_FUNCTION            0
             10 STORE_NAME               1 (circle_area)

  6          12 LOAD_BUILD_CLASS
             14 LOAD_CONST               3 (<code object Person at 0x0000023D04D3F5D0, file "D:\myspace\code\pythonCode\mix\demo.py", line 6>)
             16 LOAD_CONST               4 (&#39;Person&#39;)
             18 MAKE_FUNCTION            0
             20 LOAD_CONST               4 (&#39;Person&#39;)
             22 LOAD_NAME                2 (object)
             24 CALL_FUNCTION            3
             26 STORE_NAME               3 (Person)
             28 LOAD_CONST               5 (None)
             30 RETURN_VALUE

Disassembly of <code object circle_area at 0x0000023D04D3F310, file "D:\myspace\code\pythonCode\mix\demo.py", line 3>:
  4           0 LOAD_GLOBAL              0 (PI)
              2 LOAD_FAST                0 (r)
              4 LOAD_CONST               1 (2)
              6 BINARY_POWER
              8 BINARY_MULTIPLY
             10 RETURN_VALUE

Disassembly of <code object Person at 0x0000023D04D3F5D0, file "D:\myspace\code\pythonCode\mix\demo.py", line 6>:
  6           0 LOAD_NAME                0 (__name__)
              2 STORE_NAME               1 (__module__)
              4 LOAD_CONST               0 (&#39;Person&#39;)
              6 STORE_NAME               2 (__qualname__)

  7           8 LOAD_CONST               1 (<code object __init__ at 0x0000023D04D3F470, file "D:\myspace\code\pythonCode\mix\demo.py", line 7>)
             10 LOAD_CONST               2 (&#39;Person.__init__&#39;)
             12 MAKE_FUNCTION            0
             14 STORE_NAME               3 (__init__)

 10          16 LOAD_CONST               3 (<code object say at 0x0000023D04D3F520, file "D:\myspace\code\pythonCode\mix\demo.py", line 10>)
             18 LOAD_CONST               4 (&#39;Person.say&#39;)
             20 MAKE_FUNCTION            0
             22 STORE_NAME               4 (say)
             24 LOAD_CONST               5 (None)
             26 RETURN_VALUE

Disassembly of <code object __init__ at 0x0000023D04D3F470, file "D:\myspace\code\pythonCode\mix\demo.py", line 7>:
  8           0 LOAD_FAST                1 (name)
              2 LOAD_FAST                0 (self)
              4 STORE_ATTR               0 (name)
              6 LOAD_CONST               0 (None)
              8 RETURN_VALUE

Disassembly of <code object say at 0x0000023D04D3F520, file "D:\myspace\code\pythonCode\mix\demo.py", line 10>:
 11           0 LOAD_GLOBAL              0 (print)
              2 LOAD_CONST               1 (&#39;i am&#39;)
              4 LOAD_FAST                0 (self)
              6 LOAD_ATTR                1 (name)
              8 CALL_FUNCTION            2
             10 POP_TOP
             12 LOAD_CONST               0 (None)
             14 RETURN_VALUE

操作数指定的常量或名字的实际值在旁边的括号内列出，此外，字节码以语句为单位进行了分组，中间以空行隔开，语句的行号在字节码前面给出。例如PI = 3.14这个语句就被会变成了两条字节码：

  1           0 LOAD_CONST               0 (3.14)
              2 STORE_NAME               0 (PI)

6. pyc

如果将demo作为模块导入，Python将在demo.py文件所在目录下生成.pyc文件：

>>> import demo

pyc文件会保存经过序列化处理的代码对象PyCodeObject。这样一来，Python后续导入demo模块时，直接读取pyc文件并反序列化即可得到代码对象，避免了重复编译导致的开销。只有demo.py有新修改（时间戳比.pyc文件新），Python才会重新编译。

因此，对比Java而言：Python中的.py文件可以类比Java中的.java文件，都是源码文件；而.pyc文件可以类比.class文件，都是编译结果。只不过Java程序需要先用编译器javac命令来编译，再用虚拟机java命令来执行；而Python解释器把这两个过程都完成了。

Field	Purpose

The above is the detailed content of The execution process of a Python program includes converting source code into bytecode (i.e. compilation) and executing the bytecode. For more information, please follow other related articles on the PHP Chinese website!