4. Execution model

4.1. Structure of a program

A Python program is constructed from code blocks. A block is a piece of Python program text that is executed as a unit. The following are blocks: a module, a function body, and a class definition. Each command typed interactively is a block. A script file (a file given as standard input to the interpreter or specified as a command line argument to the interpreter) is a code block. A script command (a command specified on the interpreter command line with the -c option) is a code block. A module run as a top level script (as module __main__) from the command line using a -m argument is also a code block. The string argument passed to the built-in functions eval() and exec() is a code block.

A code block is executed in an execution frame. A frame contains some administrative information (used for debugging) and determines where and how execution continues after the code block’s execution has completed.

4.2. Naming and binding

4.2.1. Binding of names

Names refer to objects. Names are introduced by name binding operations.

The following constructs bind names:

  • formal parameters to functions,

  • class definitions,

  • function definitions,

  • assignment expressions,

  • targets that are identifiers if occurring in an assignment:

    • for loop header,

    • after as in a with statement, except clause, except* clause, or in the as-pattern in structural pattern matching,

    • in a capture pattern in structural pattern matching

  • import statements.

  • type statements.

  • type parameter lists.

The import statement of the form from ... import * binds all names defined in the imported module, except those beginning with an underscore. This form may only be used at the module level.

A target occurring in a del statement is also considered bound for this purpose (though the actual semantics are to unbind the name).

Each assignment or import statement occurs within a block defined by a class or function definition or at the module level (the top-level code block).

If a name is bound in a block, it is a local variable of that block, unless declared as nonlocal or global. If a name is bound at the module level, it is a global variable. (The variables of the module code block are local and global.) If a variable is used in a code block but not defined there, it is a free variable.

Each occurrence of a name in the program text refers to the binding of that name established by the following name resolution rules.

4.2.2. Resolution of names

A scope defines the visibility of a name within a block. If a local variable is defined in a block, its scope includes that block. If the definition occurs in a function block, the scope extends to any blocks contained within the defining one, unless a contained block introduces a different binding for the name.

When a name is used in a code block, it is resolved using the nearest enclosing scope. The set of all such scopes visible to a code block is called the block’s environment.

When a name is not found at all, a NameError exception is raised. If the current scope is a function scope, and the name refers to a local variable that has not yet been bound to a value at the point where the name is used, an UnboundLocalError exception is raised. UnboundLocalError is a subclass of NameError.

If a name binding operation occurs anywhere within a code block, all uses of the name within the block are treated as references to the current block. This can lead to errors when a name is used within a block before it is bound. This rule is subtle. Python lacks declarations and allows name binding operations to occur anywhere within a code block. The local variables of a code block can be determined by scanning the entire text of the block for name binding operations. See the FAQ entry on UnboundLocalError for examples.

If the global statement occurs within a block, all uses of the names specified in the statement refer to the bindings of those names in the top-level namespace. Names are resolved in the top-level namespace by searching the global namespace, i.e. the namespace of the module containing the code block, and the builtins namespace, the namespace of the module builtins. The global namespace is searched first. If the names are not found there, the builtins namespace is searched next. If the names are also not found in the builtins namespace, new variables are created in the global namespace. The global statement must precede all uses of the listed names.

The global statement has the same scope as a name binding operation in the same block. If the nearest enclosing scope for a free variable contains a global statement, the free variable is treated as a global.

The nonlocal statement causes corresponding names to refer to previously bound variables in the nearest enclosing function scope. SyntaxError is raised at compile time if the given name does not exist in any enclosing function scope. Type parameters cannot be rebound with the nonlocal statement.

The namespace for a module is automatically created the first time a module is imported. The main module for a script is always called __main__.

Class definition blocks and arguments to exec() and eval() are special in the context of name resolution. A class definition is an executable statement that may use and define names. These references follow the normal rules for name resolution with an exception that unbound local variables are looked up in the global namespace. The namespace of the class definition becomes the attribute dictionary of the class. The scope of names defined in a class block is limited to the class block; it does not extend to the code blocks of methods. This includes comprehensions and generator expressions, but it does not include annotation scopes, which have access to their enclosing class scopes. This means that the following will fail:

class A:
    a = 42
    b = list(a + i for i in range(10))

However, the following will succeed:

class A:
    type Alias = Nested
    class Nested: pass

print(A.Alias.__value__)  # <type 'A.Nested'>

4.2.3. Annotation scopes

Annotations, type parameter lists and type statements introduce annotation scopes, which behave mostly like function scopes, but with some exceptions discussed below.

Annotation scopes are used in the following contexts:

Annotation scopes differ from function scopes in the following ways:

  • Annotation scopes have access to their enclosing class namespace. If an annotation scope is immediately within a class scope, or within another annotation scope that is immediately within a class scope, the code in the annotation scope can use names defined in the class scope as if it were executed directly within the class body. This contrasts with regular functions defined within classes, which cannot access names defined in the class scope.

  • Expressions in annotation scopes cannot contain yield, yield from, await, or := expressions. (These expressions are allowed in other scopes contained within the annotation scope.)

  • Names defined in annotation scopes cannot be rebound with nonlocal statements in inner scopes. This includes only type parameters, as no other syntactic elements that can appear within annotation scopes can introduce new names.

  • While annotation scopes have an internal name, that name is not reflected in the qualified name of objects defined within the scope. Instead, the __qualname__ of such objects is as if the object were defined in the enclosing scope.

Added in version 3.12: Annotation scopes were introduced in Python 3.12 as part of PEP 695.

Changed in version 3.13: Annotation scopes are also used for type parameter defaults, as introduced by PEP 696.

Changed in version 3.14: Annotation scopes are now also used for annotations, as specified in PEP 649 and PEP 749.

4.2.4. Lazy evaluation

Most annotation scopes are lazily evaluated. This includes annotations, the values of type aliases created through the type statement, and the bounds, constraints, and default values of type variables created through the type parameter syntax. This means that they are not evaluated when the type alias or type variable is created, or when the object carrying annotations is created. Instead, they are only evaluated when necessary, for example when the __value__ attribute on a type alias is accessed.

Example:

>>> type Alias = 1/0
>>> Alias.__value__
Traceback (most recent call last):
  ...
ZeroDivisionError: division by zero
>>> def func[T: 1/0](): pass
>>> T = func.__type_params__[0]
>>> T.__bound__
Traceback (most recent call last):
  ...
ZeroDivisionError: division by zero

Here the exception is raised only when the __value__ attribute of the type alias or the __bound__ attribute of the type variable is accessed.

This behavior is primarily useful for references to types that have not yet been defined when the type alias or type variable is created. For example, lazy evaluation enables creation of mutually recursive type aliases:

from typing import Literal

type SimpleExpr = int | Parenthesized
type Parenthesized = tuple[Literal["("], Expr, Literal[")"]]
type Expr = SimpleExpr | tuple[SimpleExpr, Literal["+", "-"], Expr]

Lazily evaluated values are evaluated in annotation scope, which means that names that appear inside the lazily evaluated value are looked up as if they were used in the immediately enclosing scope.

Added in version 3.12.

4.2.5. Builtins and restricted execution

CPython implementation detail: Users should not touch __builtins__; it is strictly an implementation detail. Users wanting to override values in the builtins namespace should import the builtins module and modify its attributes appropriately.

The builtins namespace associated with the execution of a code block is actually found by looking up the name __builtins__ in its global namespace; this should be a dictionary or a module (in the latter case the module’s dictionary is used). By default, when in the __main__ module, __builtins__ is the built-in module builtins; when in any other module, __builtins__ is an alias for the dictionary of the builtins module itself.

4.2.6. Interaction with dynamic features

Name resolution of free variables occurs at runtime, not at compile time. This means that the following code will print 42:

i = 10
def f():
    print(i)
i = 42
f()

The eval() and exec() functions do not have access to the full environment for resolving names. Names may be resolved in the local and global namespaces of the caller. Free variables are not resolved in the nearest enclosing namespace, but in the global namespace. [1] The exec() and eval() functions have optional arguments to override the global and local namespace. If only one namespace is specified, it is used for both.

4.3. Exceptions

Exceptions are a means of breaking out of the normal flow of control of a code block in order to handle errors or other exceptional conditions. An exception is raised at the point where the error is detected; it may be handled by the surrounding code block or by any code block that directly or indirectly invoked the code block where the error occurred.

The Python interpreter raises an exception when it detects a run-time error (such as division by zero). A Python program can also explicitly raise an exception with the raise statement. Exception handlers are specified with the try … except statement. The finally clause of such a statement can be used to specify cleanup code which does not handle the exception, but is executed whether an exception occurred or not in the preceding code.

Python uses the “termination” model of error handling: an exception handler can find out what happened and continue execution at an outer level, but it cannot repair the cause of the error and retry the failing operation (except by re-entering the offending piece of code from the top).

When an exception is not handled at all, the interpreter terminates execution of the program, or returns to its interactive main loop. In either case, it prints a stack traceback, except when the exception is SystemExit.

Exceptions are identified by class instances. The except clause is selected depending on the class of the instance: it must reference the class of the instance or a non-virtual base class thereof. The instance can be received by the handler and can carry additional information about the exceptional condition.

Note

Exception messages are not part of the Python API. Their contents may change from one version of Python to the next without warning and should not be relied on by code which will run under multiple versions of the interpreter.

See also the description of the try statement in section The try statement and raise statement in section The raise statement.

4.4. Runtime Components

4.4.1. General Computing Model

Python’s execution model does not operate in a vacuum. It runs on a host machine and through that host’s runtime environment, including its operating system (OS), if there is one. When a program runs, the conceptual layers of how it runs on the host look something like this:

host machine
process (global resources)
thread (runs machine code)

Each process represents a program running on the host. Think of each process itself as the data part of its program. Think of the process’ threads as the execution part of the program. This distinction will be important to understand the conceptual Python runtime.

The process, as the data part, is the execution context in which the program runs. It mostly consists of the set of resources assigned to the program by the host, including memory, signals, file handles, sockets, and environment variables.

Processes are isolated and independent from one another. (The same is true for hosts.) The host manages the process’ access to its assigned resources, in addition to coordinating between processes.

Each thread represents the actual execution of the program’s machine code, running relative to the resources assigned to the program’s process. It’s strictly up to the host how and when that execution takes place.

From the point of view of Python, a program always starts with exactly one thread. However, the program may grow to run in multiple simultaneous threads. Not all hosts support multiple threads per process, but most do. Unlike processes, threads in a process are not isolated and independent from one another. Specifically, all threads in a process share all of the process’ resources.

The fundamental point of threads is that each one does run independently, at the same time as the others. That may be only conceptually at the same time (“concurrently”) or physically (“in parallel”). Either way, the threads effectively run at a non-synchronized rate.

Note

That non-synchronized rate means none of the process’ memory is guaranteed to stay consistent for the code running in any given thread. Thus multi-threaded programs must take care to coordinate access to intentionally shared resources. Likewise, they must take care to be absolutely diligent about not accessing any other resources in multiple threads; otherwise two threads running at the same time might accidentally interfere with each other’s use of some shared data. All this is true for both Python programs and the Python runtime.

The cost of this broad, unstructured requirement is the tradeoff for the kind of raw concurrency that threads provide. The alternative to the required discipline generally means dealing with non-deterministic bugs and data corruption.

4.4.2. Python Runtime Model

The same conceptual layers apply to each Python program, with some extra data layers specific to Python:

host machine
process (global resources)
Python global runtime (state)
Python interpreter (state)
thread (runs Python bytecode and “C-API”)
Python thread state

At the conceptual level: when a Python program starts, it looks exactly like that diagram, with one of each. The runtime may grow to include multiple interpreters, and each interpreter may grow to include multiple thread states.

Note

A Python implementation won’t necessarily implement the runtime layers distinctly or even concretely. The only exception is places where distinct layers are directly specified or exposed to users, like through the threading module.

Note

The initial interpreter is typically called the “main” interpreter. Some Python implementations, like CPython, assign special roles to the main interpreter.

Likewise, the host thread where the runtime was initialized is known as the “main” thread. It may be different from the process’ initial thread, though they are often the same. In some cases “main thread” may be even more specific and refer to the initial thread state. A Python runtime might assign specific responsibilities to the main thread, such as handling signals.

As a whole, the Python runtime consists of the global runtime state, interpreters, and thread states. The runtime ensures all that state stays consistent over its lifetime, particularly when used with multiple host threads.

The global runtime, at the conceptual level, is just a set of interpreters. While those interpreters are otherwise isolated and independent from one another, they may share some data or other resources. The runtime is responsible for managing these global resources safely. The actual nature and management of these resources is implementation-specific. Ultimately, the external utility of the global runtime is limited to managing interpreters.

In contrast, an “interpreter” is conceptually what we would normally think of as the (full-featured) “Python runtime”. When machine code executing in a host thread interacts with the Python runtime, it calls into Python in the context of a specific interpreter.

Note

The term “interpreter” here is not the same as the “bytecode interpreter”, which is what regularly runs in threads, executing compiled Python code.

In an ideal world, “Python runtime” would refer to what we currently call “interpreter”. However, it’s been called “interpreter” at least since introduced in 1997 (CPython:a027efa5b).

Each interpreter completely encapsulates all of the non-process-global, non-thread-specific state needed for the Python runtime to work. Notably, the interpreter’s state persists between uses. It includes fundamental data like sys.modules. The runtime ensures multiple threads using the same interpreter will safely share it between them.

A Python implementation may support using multiple interpreters at the same time in the same process. They are independent and isolated from one another. For example, each interpreter has its own sys.modules.

For thread-specific runtime state, each interpreter has a set of thread states, which it manages, in the same way the global runtime contains a set of interpreters. It can have thread states for as many host threads as it needs. It may even have multiple thread states for the same host thread, though that isn’t as common.

Each thread state, conceptually, has all the thread-specific runtime data an interpreter needs to operate in one host thread. The thread state includes the current raised exception and the thread’s Python call stack. It may include other thread-specific resources.

Note

The term “Python thread” can sometimes refer to a thread state, but normally it means a thread created using the threading module.

Each thread state, over its lifetime, is always tied to exactly one interpreter and exactly one host thread. It will only ever be used in that thread and with that interpreter.

Multiple thread states may be tied to the same host thread, whether for different interpreters or even the same interpreter. However, for any given host thread, only one of the thread states tied to it can be used by the thread at a time.

Thread states are isolated and independent from one another and don’t share any data, except for possibly sharing an interpreter and objects or other resources belonging to that interpreter.

Once a program is running, new Python threads can be created using the threading module (on platforms and Python implementations that support threads). Additional processes can be created using the os, subprocess, and multiprocessing modules. Interpreters can be created and used with the interpreters module. Coroutines (async) can be run using asyncio in each interpreter, typically only in a single thread (often the main thread).

Footnotes