[tompng]

First step of #708

This will make :matched :matching check simple and fast, especially after we merge MAPPING(size==256 array) and key_bindings(small hash)

# Before O(N)
key_mapping.keys.select { |lhs|
  start_with?(lhs, input)
}
...

# After O(1)
:matched if @key_bindings[key]
:matching if @matching_bytes[key]

Structure

@key_bindings = {
  [27, 91, 65] => :ed_prev_history,
  [27, 91, 50, 48, 48, 126] => :bracketed_paste_start,
}
@matching_bytes = {
  [27] => true,
  [27, 91] => true,
  [27, 91, 50] => true,
  [27, 91, 50, 48] => true,
  [27, 91, 50, 48, 48] => true,
}

Composite key binding

# Before
kb = @key_actors[@editing_mode_label].default_key_bindings.dup
kb.merge!(@additional_key_bindings[@editing_mode_label])
kb.merge!(@oneshot_key_bindings)

# After
Reline::KeyActor::Composite.new([@oneshot_key_bindings, @additional_key_bindings[@editing_mode_label], @key_actors[@editing_mode_label]])

KeyStroke#expand

# Before
expanded = key_stroke.expand(buffer).map{ |expanded_c|
  Reline::Key.new(expanded_c, expanded_c, false)
}

# After
expanded, rest_bytes = key_stroke.expand(buffer)

When key bindings is { 'abc' => :foo, 'ab' => :bar, 'xy' => :baz }, abxy should be [:bar, :baz].
Old implementation will expand abx to [Key(:bar), Key('x')] but it should be expanded: [Key(:bar)], rest_bytes: ['x']. The rest bytes x could be part of xy.

[st0012]

nit: can we also convert this into a normal class?

[st0012]

Have we removed that from IRB?

[tompng]

Not yet.
Even if we remove it, we need to support [Reline::Key.new(nil, 0xE4, true)] for a while, not to break an environment that new Reline is installed but IRB is not updated.

[st0012]

Yes I'm aware of that and I'm not arguing we need to change the approach here 🙂 I just hadn't see any related PR on IRB so want to confirm.

[tompng]

I opened ruby/irb#963

[st0012]

I didn't test the changes manually, but they look good to me 👍