Bytecode and frames

Bytecode: what the interpreter actually executes

CPython compiles your source to bytecode — a stack-machine instruction list — then executes it in a loop. Disassemble with the dis module:

import dis

def add_squares(xs):
    return sum(x * x for x in xs)

dis.dis(add_squares)

Reading bytecode answers real questions:

• Why is x += 1 on an instance attribute three operations? LOAD the attribute, add, STORE the attribute — interleaving lives between them.
• Why is a loop body that rebuilds an expression slow? Every iteration re-executes the same LOAD_GLOBAL/LOAD_ATTR instructions; hoisting a lookup out of the loop removes bytecode from the hot path.
• LOAD_CONST / LOAD_FAST / LOAD_GLOBAL — const pool, local frame slots, module dict lookups, in increasing cost order.
• Small-int caching and string interning are implementation details, not language semantics — is on integers is undefined behavior territory; == is the comparison you mean.

A frame is the runtime state of one function call: its locals, its value stack, the bytecode pointer, and the reference to the calling frame. Tracebacks are just the chain of frames — that is what inspect.currentframe() walks.