Feature #17745: `IO#close_on_exec=` returns different value when called with `send, __send__, public_send` or not

• Status: Closed
• Assignee: -
• Target version: -

[kachick (Kenichi Kamiya)]

$ ruby -v -e 'p(STDIN.close_on_exec = 42)'
ruby 3.0.0p0 (2020-12-25 revision 95aff21468) [x86_64-darwin20]
42

$ ruby -v -e 'p(STDIN.__send__ :close_on_exec=, 42)'
ruby 3.0.0p0 (2020-12-25 revision 95aff21468) [x86_64-darwin20]
nil

Is this an intentional behavior?
ruby/spec has the test case, But I can't think any benefit this different returning value 🤔

PR: https://github.com/ruby/ruby/pull/4321

[Eregon (Benoit Daloze)]

The PR looks good to me.

As you may know recv.foo= value always return value, no matter what the foo= method returns.
So in general the return value of assignment methods might differ and is probably best to be ignored.

[nobu (Nobuyoshi Nakada)]

I also think the "returns nil" example should be removed simply.

[jeremyevans0 (Jeremy Evans)]

• Tracker changed from Bug to Feature
• ruby -v deleted (ruby 3.0.0p0 (2020-12-25 revision 95aff21468) [x86_64-darwin20])
• Backport deleted (2.5: UNKNOWN, 2.6: UNKNOWN, 2.7: UNKNOWN, 3.0: UNKNOWN)

[Eregon (Benoit Daloze)]

Is there any reason to call close_on_exec= with send/__send__, or to use its return value?

I would be on the side of "changing this should not hurt", but it seems other committers do not necessarily agree (I guess their point is code should not rely on that anyway).

[nobu (Nobuyoshi Nakada)]

• Status changed from Open to Closed

Applied in changeset git|42745ecae3ce95774f2fcc66467320471ca0460a.

---

Remove meaningless IO#close_on_exec= spec [Feature #17745]