codeobject.co_nlocals and codeobject.co_varnames are incorrect for function definition

[Enigma3166]

Bug report

Bug description:

The problem is with non-parameter-local-variable inner-function-references. In the code example, the 'c' in the line '0, a, c' is causing the problem (as described in this issue's title). Replacing '0, a, c' with '0, a' works just fine.

def fn1(a, b):
    c = 0
    d = 0
    def fn2():
        0, a, c
print()
print(fn1.__code__.co_nlocals)                                                                      ### incorrect locals count
print(fn1.__code__.co_varnames)                                                                     ### incorrect locals tuple
print(fn1.__code__.co_cellvars)
print(tuple(sorted(   set(fn1.__code__.co_varnames) | set(fn1.__code__.co_cellvars)   )))           ### correct (sorted) locals tuple
print()

CPython versions tested on:

3.13

Operating systems tested on:

Windows

[Enigma3166]

In the code example:

co_nlocals should be 5 not 4

co_varnames should be ('a', 'b', 'c', 'd', 'fn2') not ('a', 'b', 'd', 'fn2')

[gaogaotiantian]

I believe nlocals includes cell vars.

[Enigma3166]

This is my understanding of what the relationships between nlocals, varnames and cellvars should be:

nlocals == len(varnames)

set(cellvars) <= set(varnames)

In my code example, the equality holds, but the inequality doesn't. Local variable 'c', which is in cellvars, is NOT in varnames and consequently is NOT reflected in the nlocals count.

P.S. How do you shade text in grey?

[gaogaotiantian]

Ah, obviously I was wrong. co_varnames only includes real local variables, not cell vars. So this is an arbitrary definition that has no right or wrong. It's actually lazily loaded which only provided a way to inspect the code object. I don't think it's easily touchable because dis explains bytecodes and it relies on it. I don't believe it's fundemantally incorrect (as long as you can imagine cell vars are special?).

I won't categorize it as a bug, maybe something to discuss and probably there were some justifications in the initial design.

P.S. github uses markdown so you can do `code` to highlight your inline code.

[Enigma3166]

Thanks for the prompt reply.

It's not a major problem for me. Just part of my exploration of Python (early days).

set(co_varnames) | set(co_cellvars) will give all the local variables.

[iritkatriel]

This is working as designed, so I think this issue can be closed.