Bug #18671: ruby2_keywords_hash not #equal? to itself with __send__

• Status: Rejected
• Assignee: -
• Target version: -
• ruby -v: ruby 3.0.3p157 (2021-11-24 revision 3fb7d2cadc) [x86_64-linux]
• Backport: 2.6: UNKNOWN, 2.7: UNKNOWN, 3.0: UNKNOWN, 3.1: UNKNOWN

[Eregon (Benoit Daloze)]

h = {a: 1}
h = Hash.ruby2_keywords_hash(h)
args = [h]

p h.equal?(h) # => true
p h.__send__(:equal?, *args) # => false

[jeremyevans0 (Jeremy Evans)]

• Status changed from Open to Rejected

This is expected. When you use *args, it passes the flagged hash as keywords (new hash), it doesn't pass it through directly. This has nothing to do with __send__, you get the same false result using p h.equal?(*args).

[Eregon (Benoit Daloze)]

Indeed:

h = {a: 1}
h = Hash.ruby2_keywords_hash(h)
args = [h]

p h.equal?(h) # => true
p h.equal?(*args) # => false
p h.equal?(*args, **{}) # => true

What was very confusing is in ruby/spec, I had code like:

after_usage.should_not.equal?(marked)

to check whether a copy was made for #18625.
And this didn't fail, even though after_usage and marked are the same object.
The reason is this behavior, and MSpec internally dispatching through __send__.
I worked around it by calling equal? more directly and avoiding __send__ for the case of equal? but obviously it's kind of a hack.

What this means is ruby2_keywords delegation doesn't preserve identity of the last argument if it is a marked Hash.
And an unintentionally marked Hash does not seem so unlikely to happen with #18625.

I updated the description of #18625 to this, there is no copy currently on master for that case (I misunderstood that originally due to this issue):

after_usage.equal?(marked) # => true, BUG, should be false