Refactor completion

[tompng]

Fixes #611 and #621

Split logic and state of autocompletion and tab completion

Before

Reline has tab complete and autocomplete.
Tab complete uses

@completion_state = NORMAL | COMPLETION | MENU | MENU_WITH_PERFECT_MATCH | PERFECT_MATCH

Autocomplete uses

@completion_state = NORMAL | JOURNEY
@completion_journey_data
retrieve_completion_block(not stored to ivar, calling every time, causing dialog disappearing bug)

It's a separate feature but uses same instance variable @completion_state that making hard to fix some bugs.

After

Tab complete uses

@completion_state = NORMAL | COMPLETION | MENU | MENU_WITH_PERFECT_MATCH | PERFECT_MATCH

Autocomplete uses

@completion_journey_state = CompletionJourneyState.new(line_index, pre, target, post, list, pointer) | nil

State is separated. Logic is also separated. Now we can remove redundant proc calls and fix #611 #621

completion_journey_data

completion_journey_data is an API for dialog proc, defined in DialogProcScope. We need to keep compatibility.
It was an instance variable but changed to a method that calculates from @completion_journey_state.

DEFAULT_DIALOG_PROC_AUTOCOMPLETE

Move completion candidate list up logic to LineEditor
All completion logic is in LineEditor. DEFAULT_DIALOG_PROC_AUTOCOMPLETE only needs to create dialog content from completion_journey_data

[ima1zumi]

For my learning: Why is force: true needed?

[tompng]

If you type "1." and then paste "0.\t\t" to the terminal, "0." is inserted to @continuous_insertion_buffer (because in_pasting? is true).
Without this process_insert(force: true), 0. part will be ignored.

irb(main):001> 1.floor
               1.abs
               1.floor[selected]
               1.ceil

We need to force insert "0." into the buffer before calculating completion list.

irb(main):001> 1.0.ceil
               1.0.floor
               1.0.ceil[selected]
               1.0.round

[ima1zumi]

Thank you for your information!
But I could not paste "0./t/t" in my environment. Can you paste it?

I got this:

irb(main):001> 1.__id__
               1.__send__[selected]
               1.equal?

[tompng]

It was broken in the last commit...

Fixed and I added test for it

[ima1zumi]

For my learning: I'm not sure why select is needed. Is it possible to create a list that does not start with target?

[tompng]

Is it possible to create a list that does not start with target?

Yes, it's possible by this code

require 'reline'
Reline.autocompletion = true
Reline.completion_proc = ->*{['abc','def','ghi','jkl']}
p Reline.readmultiline('>'){true}

I'll remove this select because older reline does not reject candidates that does not start with target.

> x
  abc
  def
  ghi
  jkl

[tompng]

The old code was doing this select in L886

private def move_completed_list(list, direction)
  ...
  @completion_journey_data = CompletionJourneyData.new(
    preposing, postposing,
    [target] + list.select{ |item| item.start_with?(target) }, 0)

The old behavior (renders candidates that does not start from target, but disappears after pressing tab) seems buggy.

[ima1zumi]

How about separating list[0] from list?
CompletionJourneyData only uses by completion_journey_data. I don't think it constitutes a breaking change. By separating this, there will be no need to drop it at the caller side.
Moving a constant may also be considered a breaking change, but it doesn't seem to be used by anyone.

https://github.com/search?q=CompletionJourneyData+language%3ARuby+path%3A**%2F*.rb+NOT+is%3Afork&type=code

[tompng]

I think it's reasonable. It will make DEFAULT_DIALOG_PROC_AUTOCOMPLETE simpler.
However, I think removing list[0] will make completion_journey_data not a journey_data thing.
completion-journey includes target as pointer=0. Renaming it to something like completion_list_for_completion_dialog will break IRB so this is not possible now.

[ima1zumi]

Thank you for the explanation. I understand.

[ima1zumi]

LGTM!