[nil].sum returns 0

[mrkn]

Is this intentional?

Steps to reproduce

$ pwd
/Users/mrkn/work/rails/master/activesupport
$ git rev-parse --short HEAD
151080a
$ ruby -v -I lib -r active_support/core_ext/enumerable -e 'p [nil].sum'
ruby 2.3.1p112 (2016-04-26 revision 54768) [x86_64-darwin15]
0

Expected behavior

I don't have the idea what is expected, ideal behavior.
But [nil].sum raises TypeError on Ruby 2.4.

$ ruby -ve '[nil].sum'
ruby 2.4.0dev (2016-04-28 trunk 54810) [x86_64-darwin15]
-e:1:in `+': nil can't be coerced into Fixnum (TypeError)
    from -e:1:in `sum'
    from -e:1:in `<main>'

Actual behavior

Returns 0.

System configuration

Rails version: 151080a

Ruby version: ruby 2.3.1p112 (2016-04-26 revision 54768) [x86_64-darwin15]

[vipulnsward]

This is what we rely on and what ruby does

ruby -v 
=> 2.3.0
[nil].inject(:+)
=> nil

Since the end result is nil, we fallback to 0- vipulnsward@5241b97

@mrkn I am not sure if that is intentional though.

[vipulnsward]

@mrkn I see we behave differently for inject for [nil] and sum. In inject we just return as is, but with sum try to coerce, is that expected?

➜  ruby git:(trunk) ✗ ./ruby -ve 'puts [nil].inject(:+)'
ruby 2.4.0dev (2016-04-30 trunk 54853) [x86_64-darwin15]
➜  ruby git:(trunk) ✗ ./ruby -ve 'puts [nil, 1].inject(:+)'
ruby 2.4.0dev (2016-04-30 trunk 54853) [x86_64-darwin15]
-e:1:in `inject': undefined method `+' for nil:NilClass (NoMethodError)
    from -e:1:in `<main>'
➜  ruby git:(trunk) ✗ ./ruby -ve '[nil].sum'
ruby 2.4.0dev (2016-04-30 trunk 54853) [x86_64-darwin15]
-e:1:in `+': nil can't be coerced into Fixnum (TypeError)
    from -e:1:in `sum'
    from -e:1:in `<main>'
➜  ruby git:(trunk) ✗

[mrkn]

In Ruby 2.4, sum isn't identical to inject(:+) while AS's sum is identical to it.
So there some differences between them.

[nil].sum is one of them.

The other difference is the initial value effect.
Ruby 2.4's sum basically assumes numeric elements so the default initial value is 0.
In contrast, AS's sum uses the identity parameter only when the enumerable is empty.
I've posted a pull-request to fix this inconsistency in #24795.

I want to resolve the inconsistency of the behavior of [nil].sum, too.

[mrkn]

I want to resolve the inconsistency of the behavior of [nil].sum, too.

In other words, I want to fix this if it isn't the desired behavior.

[vipulnsward]

We don't have a specification for this. cc @jeremy

Ref: https://github.com/vipulnsward/rails/blob/7ad4690b2149fbb23faa179c21698b92ff383c73/activesupport/test/core_ext/enumerable_test.rb#L43-L58

[jeremy]

Great edge case. The identity fallback is meant to catch [].sum. Note that [false].sum is affected in the same way.

[jeremy]

#24795

[jeremy]

Oops! I see #24795 does not address this case.

Agree: behavior here is undefined. Expected behavior is [nil].sum == nil and [false].sum == false, matching inject(:+) results.

I think this is buggy, since [single_element].sum == single_element even when !single_element.respond_to?(:+). If there are multiple elements, it would have raised NoMethodError.

[sgrif]

Is this buggy? I think it's perfectly reasonable for this to match the behavior of .inject(:+). If Ruby 2.4 provides a sum method, we can defer to that in the future.

[vipulnsward]

I think we could handle it. Generally, in our Apps, someone would do [].compact.sum and not care about this nil.
But this condition would be only for a single element. (P.S: we have had this behaviour since 2006.)

[Chiether]

probably related...

before

[].sum #=> 0
[nil].sum #=> 0
[].sum(nil) #=> nil <-- I used this case!
[nil].sum(nil) #=> nil

after:

[].sum #=> 0
[nil].sum #=> 0
[].sum(nil) #=> 0
[nil].sum(nil) #=> 0

That’s life. sigh and replace my code

[junjizhi]

Confirmed that this is still an issue in Rails 5.2.1

Loading development environment (Rails 5.2.1)
irb(main):001:0> [nil].sum
=> 0

But in ruby 2.5:

$ ruby -v
ruby 2.5.0p0 (2017-12-25 revision 61468) [x86_64-darwin17]

irb(main):001:0> [nil].sum
Traceback (most recent call last):
        4: from /usr/local/bin/irb:11:in `<main>'
        3: from (irb):1
        2: from (irb):1:in `sum'
        1: from (irb):1:in `+'
TypeError (nil can't be coerced into Integer)

[loganb]

Just to make sure the variant is covered: [].sum currently throws an exception when it should return 0

[okuramasafumi]

The latest version of Ruby/Rails (5.2.3 and 2.6.3) can both handle [].sum gracefully with returning 0.

[0xYZSR]

Loading development environment (Rails 5.2.6)
irb(main):001:0> [].sum
Traceback (most recent call last):
        1: from (irb):1
NoMethodError (undefined method `each' for nil:NilClass)

[zzak]

@blk0 Can you reproduce this in 6.1? Since 5.2 is no longer maintained for bug fixes it's unlikely this will get fixed.