GH-141312: Allow only integers to longrangeiter_setstate state #141317
Conversation
|
@serhiy-storchaka Could you please take a look? |
serhiy-storchaka
left a comment
serhiy-storchaka
left a comment
There was a problem hiding this comment.
Why not simply check state? Normally, it can only be int.
Using PyLong_Check() is not enough -- it would pass for an int subclass with overridden __radd__, __mul__, etc. PyLong_CheckExact() is needed.
|
Should we change |
| if (!PyLong_CheckExact(state)) { | ||
| PyErr_Format(PyExc_TypeError, | ||
| "'%T' object cannot be interpreted as an integer", state); | ||
| return NULL; | ||
| } |
There was a problem hiding this comment.
| if (!PyLong_CheckExact(state)) { | |
| PyErr_Format(PyExc_TypeError, | |
| "'%T' object cannot be interpreted as an integer", state); | |
| return NULL; | |
| } | |
| if (!PyLong_CheckExact(state)) { | |
| PyErr_Format(PyExc_TypeError, "state must be an int, not %T", state); | |
| return NULL; | |
| } |
There was a problem hiding this comment.
This is exactly what I added before checking what rangeiter returns for a wrong state. Is it ok if our messages will be different?
There was a problem hiding this comment.
Oh so that is the message of rangeiter. Ok keep it the same.
There was a problem hiding this comment.
The proposed message is better.
There was a problem hiding this comment.
rangeiter_setstate() supports int subclasses and classes with the __index__ method. If we want to support them here (__index__ is not currently supported), we should use PyNumber_Index().
But there is no good reason to do this. It would only complicate the code without adding any benefit.
Yes we should. It's a bug fix. The rule is that most bugfixes have NEWS. |
Lib/test/test_range.py
Outdated
| with self.assertRaisesRegex(TypeError, msg): | ||
| it = iter(range(10, 100, 2)) | ||
| it.__setstate__(1.0) |
There was a problem hiding this comment.
- Make a separate test method for setstate failures.
- Only assert the exception when calling
setstate. - Check that custom int subclasses are rejected. What is the current behavior with range objects not having exact ints?
There was a problem hiding this comment.
Range iter give following error messages before last changes:
- float:
'float' object cannot be interpreted as an integer - I() (int-like object): no error message
- str:
'str' object cannot be interpreted as an integer
Objects/rangeobject.c
Outdated
| { | ||
| if (!PyLong_CheckExact(state)) { | ||
| PyErr_Format(PyExc_TypeError, | ||
| "'%T' object cannot be interpreted as an integer", state); |
There was a problem hiding this comment.
Such error messages is usually used when the argument can be an int subclass of has the __index__ method. But this is not needed here, only int is supported.
There was a problem hiding this comment.
See my discussion here: #141317 (comment). Should we change rangeiter to match messages?
|
It is ready for review. |
3 participants
Only integers should be allowed for the start value in
longrangeiter.Two extra checks were added:
__setstate__forrangeiterfailslongrangeiter.I made the error message for
longrangeitersimilar to that forrangeiter.compute_range_length: Assertion PyLong_Check(start)' failed#141312