`datetime.timedelta` returns unexpected value when passed a bool `True` for days arg

[danish-wani]

Bug report

Bug description:

Today I stumbled upon a strange issue where I was supposed to pass an integer value to datetime.timedelta for days arg but I was passing a boolean True value to this function as a positional argument. Below is what I was doing;

import datetime

datetime.datetime.now() -  datetime.timedelta(True)

# instead of

start_datetimestamp = datetime.datetime.now() -  datetime.timedelta(10)

to my surprise, when I checked the value start_datetimestamp it was not what I expected. Upon debugging it further I realized I was passing the boolean True instead of integer 10 as a positional argument to this function.

Since in my case, the days arg was initialized as True the below statement evaluated to True + 0 = 1, hence assigning 1 to days instead of raising a ValueError.

days += weeks*7 => True += 0 *7 => 1

Proposed better handling of this behaviour:

Adding a type check on all the arguments should possibly avoid this unexpected behaviour;

        d = s = us = 0
        assert isinstance(days, (int, float))
        assert isinstance(seconds, (int, float))
        assert isinstance(microseconds, (int, float))
        assert isinstance(milliseconds, (int, float))
        assert isinstance(minutes, (int, float))
        assert isinstance(hours, (int, float))
        assert isinstance(weeks, (int, float))

When a bool True is passed to datetime.timedelta

import datetime

datetime.timedelta(True)

# output
# datetime.timedelta(days=1)

CPython versions tested on:

3.10

Operating systems tested on:

macOS

[ericvsmith]

As a rule we don't add type checks like this. And in any event, as you discovered True is an instance of int, so the above test wouldn't work, anyway. You'd need to explicitly exclude bool as a valid type.

[danish-wani]

Fair enough, are you suggesting adding this check before calling timedelta in the caller function? Shouldn't this explicitly exclude bool as a valid type. be added inside the source code? @ericvsmith

[bswck]

I wouldn't consider this a bug, but an example use case of the Liskov substitution principle.

[bswck]

Fair enough, are you suggesting adding this check before calling timedelta in the caller function? Shouldn't this explicitly exclude bool as a valid type. be added inside the source code? @ericvsmith

I'm afraid such a check is shortsighted (despite being breaking).
What about enum int flags? What about Windows-only subprocess.Handle? They all also don't fit conceptually.
You can't create a complete blacklist.
And someone might actually want to use bool there, for instance

# is_one_day_event=True -> timedelta(days=1), is_one_day_event=False -> timedelta(0)
event_duration = timedelta(days=is_one_day_event)

I've seen a common practice of using booleans as integers for string multiplication (a string becomes empty when multiplied by False). I guess that you found a feature, not a bug.
I'd consider closing this.

[JelleZijlstra]

The fact that bools are a subclass of int is arguably a design bug in Python but not one we can fix now.

[danish-wani]

Appreciate your response and clarifications. Thank you @bswck @JelleZijlstra