New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
randrange() mishandles step when stop is None #86938
Comments
When stop is None, the step argument is ignored: >>> randrange(1000, None, 100)
651
>>> randrange(1000, step=100)
673 |
One fix is to move up the code for converting step to istep and then modify the early out test to: if stop is None and istep == 1:
if istart > 0:
return self._randbelow(istart) That would have the downside of slowing down the common case. Another possibility is changing the default step to a sentinel object and just testing for that: if stop is None and step is sentinel:
if istart > 0:
return self._randbelow(istart) That would be user visible in the function signature but it would run fast. We could bite the bullet and fully harmonize the randrange() signature with range(). That would entail switching to *args and no longer accepting keyword arguments: def randrange(self, /, *args):
"Choose random item from range(stop) or range(start, stop[, step])."
return self.choice(range(*args)) This would most closely match user expectations but is potentially a breaking change for existing code that uses keyword arguments. Such code probably exists but is probably not common. For speed, the actual implementation could still have fast paths for common cases. For help() and tooltips, we could add a __text_signature__ to cover-up the *args. However, signature objects currently aren't capable of describing range(). The best we could do is:
That would give help() that looks like this: >>> help(randrange)
Help on method randrange in module Random:
|
Note, we can't actually use "self.choice(range(*args))" because the underlying len() call can Overflow. If it does, the components need to be computed manually. |
Another solution is to use an identity test for the step argument _one = 1
class Random:
def randrange(start, stop=None, step=_one):
...
if stop is None and step is _one:
if istart > 0:
return self._randbelow(istart) This has the advantage of keeping the API unchanged while still keeping the fast path fast. |
Should I restore that optimization in bpo-37319? |
Yes, but to fix the bug it needs to occur earlier in the code (currently line 314): if stop is None and step is _ONE: <-- Line 314
^^^^^^^^^^^^^^^^^ <-- Add this And again later: As a standalone optimization, it wasn't worth it, but as a way to fix the bug without a regression, it is reasonable. |
Upon further reflection, I think these cases should raise a TypeError because the meaning is ambiguous: randrange(1000, step=10) # Could only mean start=0 stop=1000 step=10 but that conflicts with: randrange(1000, None, 100) # Has no reasonable interpretation For comparison, this currently raises a TypeError because None isn't a sensible argument for stop: range(1000, None, 100) |
Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.
Show more details
GitHub fields:
bugs.python.org fields:
The text was updated successfully, but these errors were encountered: