We don't want random require calls in specs. This should either be moved to the top of the file, or the whole test should be moved into its own file.

@brixen Any thoughts on this, is this something we perhaps want in rubysl-bigdecimal instead?

Unfortunately, Ruby does not have a built-in float-like object. Moving this test into BigDecimal would no longer be testing the actual implementation.

The rule to not require standard lib in core lib specs is broken in a few places but those need to be fixed.

The code does not depend on BigDecimal and the specs should not either. The specs should describe the API and protocol (ie which methods are expected and which methods will be used to coerce values).

@brixen What would be the best way of testing float-like numbers in core? These bugs are happening because the code eagerly checks for integer-ness instead of float-ness, but I can't reproduce them without a separate Float implementation.

Also in general, is there a good way of detecting integer-ness vs float-ness? Seems like MRI doesn't do a good job (rational often behaves like integers).

@fengb your proposed patch adds conditionals for is_a?(Integer) and respond_to?(:to_f). Those cases should be possible to cover with specs that neither need BigDecimal, nor an alternate Float implementation.

Am I missing something else?