Binding to `exit` command causes `TypeError`

[13k]

inputrc:

Control-d: exit

❱ ./bin/console
irb(main):001> ^D
RELINE_SRCDIR/lib/reline/line_editor.rb:975:in 'Kernel#exit': no implicit conversion of String into Integer (TypeError)
	from RELINE_SRCDIR/lib/reline/line_editor.rb:975:in 'Method#call'
	from RELINE_SRCDIR/lib/reline/line_editor.rb:975:in 'Reline::LineEditor#wrap_method_call'
	from RELINE_SRCDIR/lib/reline/line_editor.rb:941:in 'Reline::LineEditor#run_for_operators'
	from RELINE_SRCDIR/lib/reline/line_editor.rb:1001:in 'Reline::LineEditor#process_key'
	from RELINE_SRCDIR/lib/reline/line_editor.rb:1027:in 'Reline::LineEditor#input_key'
	from RELINE_SRCDIR/lib/reline/line_editor.rb:1005:in 'Reline::LineEditor#update'
        <omitted>

I didn't follow the whole call chain, but I'm guessing key.method_symbol == :exit and wrap_method_call simply grabs the method from top-level Kernel and calls it with a string parameter, causing the TypeError exception.

[t-mangoe]

I briefly looked into this issue. As expected, it appears that the Kernel's exit method is being called.
Because the exit method is inherited, the following respond_to? method returns true. (then, the exit method is executed.)

reline/lib/reline/line_editor.rb

Line 955 in 4fe1ff9

return unless respond_to?(method_symbol, true)

I think only methods defined in the LineEditor class should be called. How about that?

[tompng]

exit is not an existing GNU-Readline functon name. GitHub code search of Readline's config shows that "[key binding here]": exit doesn't seem to be used in real world, so I don't see a use case. Reline just doesn't support exit functionality just like GNU Readline.

The problem is not only for Kernel's method.

"\C-d": initialize
"\C-e": move_history

Will call LineEditor#initialize or LineEditor#move_history (internal method) with wrong argument.

Reline maintainer team, including me, thinks there is a maintainability issue in this part

private def ed_end_of_history(key)
  move_history(Reline::HISTORY.size, line: :end, cursor: :end)
end
alias_method :end_of_history, :ed_end_of_history

• It registers end-of-history as a function. This part is OK (except it just look like normal method aliasing)
• Unknown weather ed-end-of-history is intended to register as a function, or it's just a meaningful method name that expresses it originates from Editline
• end-of_history and end_of-history etc are also registered
• Hard to distinguish valid function from unused method that need to be removed as a refactoring.
• And problems described in this issue: No error handling, No warning, etc.

To fix this root cause, we need a large-diff refactoring that introduce explicit function-name register mechanism.

Workaround pull requests are welcome (If the workaround is simple enough)