Is it possible for a Highlighter to re-render a line to undo cursor-based highlights?

[luketpeterson]

Consider the HistoryHinter example here: https://github.com/kkawakam/rustyline/blob/master/examples/example.rs which uses the MatchingBracketHighlighter.

If you type ( x ), it will bold-blue the first '(', as it should, when the closing ')' is typed. But if you then hit Enter, the '(' remains bold-blued on the now-complete line.

I have tried to fix this with my own implementation of a Highligher object, but neither the highlight_char nor the highlight Highlighter methods are called again on that original line. (They are called on a new line with a len=0, pos=0)

Is there a way (perhaps with the Validator?) to get the editor to re-render the line one last time? I see people may not want this as the default behavior if their tty speed is limited, but is there a way to make it happen?

Thank you.

[luketpeterson]

I was able to get the behavior I want by adding an additional "needs_final_refresh" method into Highlighter, that could cause this code path to be taken:

https://github.com/kkawakam/rustyline/blob/1c4091a0b20d68dda50c902abb723a6a74739ab9/src/edit.rs#L237C1-L238C1

Is there a way to do this in a more appropriate way? Or what design would you like to see in a PR?

Thanks again.

[gwenn]

We already refresh the line here:

rustyline/src/command.rs

Lines 28 to 32 in 1c4091a

	if s.has_hint() || !s.is_default_prompt() {
	// Force a refresh without hints to leave the previous
	// line as the user typed it after a newline.
	s.refresh_line_with_msg(None)?;
	}

But indeed currently, we cannot check / know that the Highlighter needs a final refresh.

[luketpeterson]

If adding a method to the Highlighter trait is distasteful, perhaps highlight_char could be called on a line that has the final EOL character at the end?

[gwenn]

No worry, we can add a new method to Highlighter trait with a default implementation.

[luketpeterson]

I tried to tie up this logic nicely in a PR, and I added the below method to the Highlighter trait:

fn final_highlight(&self, line: &str) -> bool

But I realized there is a complication.

The internal-mutability pattern (ie Cell) used in the MatchingBracketHighlighter means that the highlight_char() method actually has the side-effect of updating the highlighter state.

I followed this pattern and used the final_highlight() method to similarly clear the cell. But then highlight_char() is called from refresh_line_with_msg() again, resetting the cell to the original value.

There are several solutions but I only see one that doesn't involve changing the client-facing API or contract in some way.

• The lowest risk change is to make the internals of MatchingBracketHighlighter be tri-state. So final_highlight would set a "final" bool variable in the Cell, and that flag tells highlight_char() not to make any updates. (or tells highlight to render without any cursor-based info). This is unlikely to affect the behavior of any other clients, but it feels a little ugly to me.

• The other solutions are to somehow signal to highlight_char() or highlight() that a final render is expected. For example, calling those methods with a line that includes an EOF at the end would accomplish this. Unfortunately it looks like plumbing the information that a given refresh is "final" through the Refresher interface looks like it might cause some serious code disruption.

I raised a PR (#727) with the "least risky" change, but it feel like a little hackish so I understand if you have a better idea for a fix.

Thank you.

[gwenn]

Thanks for your investigation.
I am not sure but I guess that we need:

• a new flag in State that indicates that we are in the "final" phase.
• Use State.highlight_char to know if we need a final refresh.
• don't call Highlighter::highlight_char in "final" phase (set State.highlight_char to false).

See draft #729

[luketpeterson]

See draft #729

Yeah. I think this is a better solution. I just didn't feel personally comfortable enough changing the behavior of the editor state and trait method definitions, but you have more knowledge and I like your solution better.

My one nit is that I was thinking of it as "final" instead of "forced" (or is_final since final is a reserved keyword). ie. with that bool set to false, highlight_char decides whether highlighter needs to run with a given cursor position. With it set to true, highlight_char decides if it needs to run to redraw the line without any cursor influence. But of course the choice of names is ultimately up to you, and I am happy either way.

[gwenn]

In fact, this is not necessarily the final refresh because depending on Validator (for example if the statement is incomplete) a continuation line will be inserted...

[luketpeterson]

In fact, this is not necessarily the final refresh because depending on Validator (for example if the statement is incomplete) a continuation line will be inserted...

That is a good point. really it's a matter of the cursor position being non-existent after the line has been accepted. So you could make the pos argument into an Option<>, but if you want to keep the "force" naming for a bool it will also be fine.

Thank you again.

[gwenn]

Version 13.0.0 released.

[luketpeterson]

Thank you very much!