Right now there's a special case in handling this opcode:
|
/* If value is a unicode object, and there's no fmt_spec, |
|
then we know the result of format(value) is value |
|
itself. In that case, skip calling format(). I plan to |
|
move this optimization in to PyObject_Format() |
|
itself. */ |
|
if (PyUnicode_CheckExact(value) && fmt_spec == NULL) { |
|
/* Do nothing, just transfer ownership to result. */ |
|
result = value; |
|
} else { |
|
/* Actually call format(). */ |
|
result = PyObject_Format(value, fmt_spec); |
|
Py_DECREF(value); |
|
Py_XDECREF(fmt_spec); |
|
ERROR_IF(result == NULL, error); |
It comes from older version of ceval.c:
|
/* If value is a unicode object, and there's no fmt_spec, |
|
then we know the result of format(value) is value |
|
itself. In that case, skip calling format(). I plan to |
|
move this optimization in to PyObject_Format() |
|
itself. */ |
|
if (PyUnicode_CheckExact(value) && fmt_spec == NULL) { |
|
/* Do nothing, just transfer ownership to result. */ |
|
result = value; |
|
} else { |
|
/* Actually call format(). */ |
|
result = PyObject_Format(value, fmt_spec); |
|
Py_DECREF(value); |
|
Py_XDECREF(fmt_spec); |
|
if (result == NULL) { |
|
goto error; |
|
} |
|
} |
But, I think that the needed optimizaion is there:
|
if (format_spec == NULL || PyUnicode_GET_LENGTH(format_spec) == 0) { |
|
if (PyUnicode_CheckExact(obj)) { |
|
return Py_NewRef(obj); |
|
} |
So, I think this can be safely removed.
I will send a PR.
Linked PRs