Issue #17017 has been updated by Eregon (Benoit Daloze).


marcandre (Marc-Andre Lafortune) wrote in #note-15:
> Unless I'm mistaken, this behavior change was not approved by Matz (or anybody else), changes a behavior that dates back to Ruby 1.8, breaks (that we know of) `activemodel` and `rubocop`, doesn't even raise the right error and isn't tested, ...

Which change is that?
The first PR above, https://github.com/ruby/ruby/pull/3306 ?
That seems to have some tests.

> Yet when you propose that that commit remains as is and that someone else has to bring up the situation with Matz, no other committer seems confused by this.

I think the first PR should already have matz's or at least someone's else review.

Range is full of edge and special cases though, and there seems to be no global vision for it.
Maybe a good step here would be try to define the semantics we'd like for it. Obviously it's quite some work.

If reverting the first PR helps, maybe we should do that first?

----------------------------------------
Bug #17017: Range#max & Range#minmax incorrectly use Float end as max
https://bugs.ruby-lang.org/issues/17017#change-86604

* Author: sambostock (Sam Bostock)
* Status: Open
* Priority: Normal
* ruby -v: ruby 2.8.0dev (2020-07-14T04:19:55Z master e60cd14d85) [x86_64-darwin17]
* Backport: 2.5: UNKNOWN, 2.6: UNKNOWN, 2.7: UNKNOWN
----------------------------------------
While continuing to add edge cases to [`Range#minmax` specs](https://github.com/ruby/spec/pull/777), I discovered the following bug:

```ruby
(1..3.1).to_a        == [1, 2, 3] # As expected

(1..3.1).to_a.max    == 3         # As expected
(1..3.1).to_a.minmax == [1, 3]    # As expected

(1..3.1).max    == 3.1            # Should be 3, as above
(1..3.1).minmax == [1, 3.1]       # Should be [1, 3], as above
```

One way to detect this scenario might be to do (whatever the C equivalent is of)

```ruby
range_end.is_a?(Numeric)                      // Is this a numeric range?
  && (range_end - range_begin).modulo(1) == 0 // Can we reach the range_end using the standard step size (1)
```

As for how to handle it, a couple options come to mind:

- We could error out and do something similar to what we do for exclusive ranges

```ruby
raise TypeError, 'cannot exclude non Integer end value'
```

- We might be able to calculate the range end by doing something like

```ruby
num_steps = (range_end / range_beg).to_i - 1 # one fewer steps than would exceed the range_end
max = range_beg + num_steps                  # take that many steps all at once
```

- We could delegate to `super` and enumerate the range to find the max

```ruby
super
```

- We could update the documentation to define the max for this case as the `range_end`, similarly to how the documentation for `include?` says it behaves like `cover?` for numeric ranges.



-- 
https://bugs.ruby-lang.org/

Unsubscribe: <mailto:ruby-core-request / ruby-lang.org?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>