Bug #2756: Issues with Math and Complex behavior on 1.9

• Status: Rejected
• Assignee: mrkn (Kenta Murata)
• Target version: -
• ruby -v: ruby 1.9.2dev (2010-02-18 trunk 26704) [i386-darwin9.8.0]
• Backport: 2.3: UNKNOWN, 2.4: UNKNOWN

[brixen (Brian Shirai)]

This ticket aggregates several issues with Math methods on 1.9. There are related tickets that either have not yet or do not, in my opinion, resolve these issues in a satisfactory manner. (see http://redmine.ruby-lang.org/issues/show/1708, and related to the behavior of Math http://redmine.ruby-lang.org/issues/show/2189 and to 1.8 behavior http://redmine.ruby-lang.org/issues/show/2754)

1. There are behaviors that are inconsistent with 1.8

# On 1.8, the argument is coerced
$ ruby1.8.7 -v -e 'o = Object.new; def o.to_f; 0.5; end; p Math.atanh(o)'ruby 1.8.7 (2009-12-24 patchlevel 248) [i686-darwin9.8.0]
0.549306144334055
 
# On 1.9, the argement is not coerced
$ ruby1.9 -v -e 'o = Object.new; def o.to_f; 0.5; end; p Math.atanh(o)'
ruby 1.9.2dev (2010-02-18 trunk 26704) [i386-darwin9.8.0]
-e:1:in `atanh': can't convert Object into Float (TypeError)
        from -e:1:in `<main>'

Q. Should 1.9 coerce arguments to Math methods?

# On 1.8, an ArgmentError is raised
$ ruby1.8.7 -v -e 'p Math.atanh("str")'
ruby 1.8.7 (2009-12-24 patchlevel 248) [i686-darwin9.8.0]
-e:1:in `atanh': invalid value for Float(): "str" (ArgumentError)
        from -e:1

# On 1.9, a TypeError is raised
$ ruby1.9 -v -e 'p Math.atanh("str")'
ruby 1.9.2dev (2010-02-18 trunk 26704) [i386-darwin9.8.0]
-e:1:in `atanh': can't convert String into Float (TypeError)
        from -e:1:in `<main>'

Q. In this case, TypeError would appear more correct, so can the 1.8.7 behavior be changed? Also note that changing the 1.8.7 behavior would make it consistent with the behavior of atanh when requiring Complex (see http://redmine.ruby-lang.org/issues/show/2754)

2. There are behaviors that are inconsistent when requiring lib/complex.rb

# The original method raise a TypeError
$ ruby1.9 -v -e 'p Math.atanh(nil)'
ruby 1.9.2dev (2010-02-18 trunk 26704) [i386-darwin9.8.0]
-e:1:in `atanh': can't convert nil into Float (TypeError)
        from -e:1:in `<main>'

# The new method attempts an undefined operation and consequently raises a NoMethodError
$ ruby1.9 -v -rcomplex -e 'p Math.atanh(nil)'
ruby 1.9.2dev (2010-02-18 trunk 26704) [i386-darwin9.8.0]
lib/complex.rb is deprecated
/Users/brian/devel/ruby19/install/lib/ruby/1.9.1/cmath.rb:196:in `atanh': undefined method `real?' for nil:NilClass (NoMethodError)
        from -e:1:in `<main>'

The same behavior is observed when passing a String.

Q. Should the behavior of atanh after requiring lib/complex.rb be the same for non-Complex inputs as before?

Also, requiring lib/complex.rb on 1.9 causes a warning: "lib/complex.rb is deprecated". But this is not entirely true. As best as I can understand from http://redmine.ruby-lang.org/issues/show/1708, it was never decided whether complex.rb should require cmath.rb. It appears that there are some behaviors acquired via lib/complex.rb that are not deprecated. In that case, this warning is confusing and misleading.

Q. Is lib/complex.rb deprecated or not? If it is, why is it deprecated and not removed? 1.9 already removes many libraries. Why is this one special and allowed to cause such confusion?

Q. Is there a definitive document that explains the policy and behavior of Math and Complex in 1.9?

---

To summarize the questions in this ticket?

• Q. Should 1.9 coerce arguments to Math methods?
• Q. Can we change the 1.8.7 behavior when raising exceptions to be both internally consistent and consistent with the behavior of 1.9 (Note that numerous changes to the exception raised have already been made in 1.8.5 -> 1.8.6 -> 1.8.7, so this request is not without precedent.) (see http://redmine.ruby-lang.org/issues/show/2754)
• Q. Should the behavior of atanh after requiring lib/complex.rb be the same for non-Complex inputs as before?
• Q. Is lib/complex.rb deprecated or not? If it is, why is it deprecated and not removed?
• Q. Is there a definitive document that explains the policy and behavior of Math and Complex in 1.9?

Thanks,
Brian

[mame (Yusuke Endoh)]

=begin
Hi!

Q. Should 1.9 coerce arguments to Math methods?

I really want to agree, but it seems intended change for
prohibiting string:

$ ruby18 -ve 'p Math.atanh("0.5")'
ruby 1.8.8dev (2010-02-01 revision 26536) [i686-linux]
0.549306144334055

$ ./ruby -ve 'p Math.atanh("0.5")'
ruby 1.9.2dev (2010-02-18 trunk 26703) [i686-linux]
-e:1:in atanh': can't convert String into Float (TypeError) from -e:1:in '

The root cause is that explicit float conversion and implicit
one are not distinguished. There are to_i, to_int and to_f,
but no to_flo.
I think this can be fixed by adding to_flo in theory, but it
may change behavior of many methods that take Float as arg.
Needs much time to discuss, and the change is too big for 1.9
series.

Q. Can we change the 1.8.7 behavior when raising exceptions to be both internally consistent and consistent with the behavior of 1.9 (Note that numerous changes to the exception raised have already been made in 1.8.5 -> 1.8.6 -> 1.8.7, so this request is not without precedent.)

Shyouhei, please decide it.
IMO, TEENY version may accept such a small spec change, but
patch-level might not.

Q. Should the behavior of atanh after requiring lib/complex.rb be the same for non-Complex inputs as before?

I agree in this case.
Ruby library generally tends to ignore such a type difference.
(see http://redmine.ruby-lang.org/issues/show/2495)
But cmath.rb attempts to replace the core Math module, so I
think it has a responsibility to behave as similarly as
possible. I'll work later.

Q. Is lib/complex.rb deprecated or not? If it is, why is it deprecated and not removed?

The main features of lib/complex.rb are embedded to core in 1.9,
but some features (e.g., Numeric#im and complex-sensitive Math)
are not. Complex-sensitive Math (just CMath, not replacing core
Math) are moved to cmath.rb. Numeric#im and replacing core Math
seemed to be deprecated. So lib/complex.rb is now just for
compatibility.

This is just the fact. I don't know its reason.

Q. Is there a definitive document that explains the policy and behavior of Math and Complex in 1.9?

Of course, no it isn't, I guess ;-(

--
Yusuke ENDOH [e-mail]

=end

[shyouhei (Shyouhei Urabe)]

=begin
Hi,

Yusuke ENDOH wrote:

Q. Can we change the 1.8.7 behavior when raising exceptions to be both internally consistent and consistent with the behavior of 1.9 (Note that numerous changes to the exception raised have already been made in 1.8.5 -> 1.8.6 -> 1.8.7, so this request is not without precedent.)

Shyouhei, please decide it.
IMO, TEENY version may accept such a small spec change, but
patch-level might not.

Agreed. I'd want to see this a spec change between versions.

Attachment: signature.asc
=end

[brixen (Brian Shirai)]

=begin
Hi,

On Wed, Feb 17, 2010 at 10:23 PM, Urabe Shyouhei [e-mail] wrote:

Hi,

Yusuke ENDOH wrote:

Q. Can we change the 1.8.7 behavior when raising exceptions to be both internally consistent and consistent with the behavior of 1.9 (Note that numerous changes to the exception raised have already been made in 1.8.5 -> 1.8.6 -> 1.8.7, so this request is not without precedent.)

Shyouhei, please decide it.
IMO, TEENY version may accept such a small spec change, but
patch-level might not.

Agreed. I'd want to see this a spec change between versions.

I would argue that the behavior of lib/complex.rb be considered a bug
because the existing method is overridden but the existing method's
behavior is changed incompatibly. Bugs can be fixed in patchlevels.

OTOH, while 1.8.8 is not well defined (at least I don't know where to
look for a definition), it could be the target for this change. Of the
two options, I'd really like to see it changed in 1.8.7.

Cheers,
Brian

=end

[mame (Yusuke Endoh)]

• Target version changed from 1.9.2 to 2.0.0

=begin

=end

[naruse (Yui NARUSE)]

• Status changed from Open to Assigned
• Assignee set to mame (Yusuke Endoh)

[mame (Yusuke Endoh)]

• Assignee changed from mame (Yusuke Endoh) to mrkn (Kenta Murata)

Hello,

Q. Should 1.9 coerce arguments to Math methods?
Q. Can we change the 1.8.7 behavior when raising exceptions to be both internally consistent and consistent with the behavior of 1.9 (Note that numerous changes to the exception raised have already been made in 1.8.5 -> 1.8.6 -> 1.8.7, so this request is not without precedent.) (see http://redmine.ruby-lang.org/issues/show/2754)
Q. Should the behavior of atanh after requiring lib/complex.rb be the same for non-Complex inputs as before?
Q. Is lib/complex.rb deprecated or not? If it is, why is it deprecated and not removed?
Q. Is there a definitive document that explains the policy and behavior of Math and Complex in 1.9?

Though there is no official maintainer for both lib/complex.rb and
lib/cmath.rb, I guess mrkn, keiju and tadf have a thorough knowledge
of them. So pass this ticket to mrkn.

--
Yusuke Endoh [e-mail]

[mrkn (Kenta Murata)]

• Target version changed from 2.0.0 to 2.6

[mrkn (Kenta Murata)]

• Description updated (diff)

[naruse (Yui NARUSE)]

• Target version deleted (2.6)

[mrkn (Kenta Murata)]

• Status changed from Assigned to Rejected

I'm rejecting this issue because this is meaningless now.