Starting in fish 3.6.0, abbreviations no longer trigger with following semicolon

[lwolfsonkin]

Previously, starting in Fish 3.1.0, I reported #6970, which was that abbreviations stopped expanding with a following semicolon. At the time, I was given the suggestion to add bind ';' self-insert expand-abbr to my config.fish, which fixed my problem.

In my usual workflow, I use a large set of abbr abbreviations for all the hg subcommands, and often chain them together by typing e.g. the literal amend; and upload which would expand to hg amend; and hg upload and each of amend and upload respectively would expand once I typed a following space, enter, or semicolon (among others, I presume).

Now, I'm having the same issue in Fish 3.6.0 (and 3.6.1), both with and without the bind ';' self-insert expand-abbr workaround in my config, on two different machines of mine.

I see that Fish 3.6.0 overhauled the abbreviation handling, and I'm assuming that somehow broke my workflow. Was that intentional? Might it be possible for me to work around this in the existing version?

Thank you!

[faho]

Try the other way around:

bind ';' expand-abbr self-insert

In fact that's how our bindings are supposed to work already, so lemme just fix that.

[lwolfsonkin]

Ah, that workaround solved my immediate issue. Thanks @faho! I appreciate it!

[zanchey]

@faho, did you have a fix in mind here?

[faho]

@faho, did you have a fix in mind here?

I was gonna switch it from "self-insert expand-abbr" to "expand-abbr self-insert", but that has some awkward side effects.

The main issue: if you have an abbr with --set-cursor, now triggering it inserts the space.

Other changes are that it changes what part "undo" undoes and so you might have a space less. Which seems not terrible but I'm not sure we want those repercussions here.

I think we need to look at the cursor_backtrack idea again.

[zanchey]

OK, pushing this out of 3.7.0 for now

[g0t4]

FYI flipping the order messes up the new abbr with --set-cursor, you end up getting the self-insert at the cursor position after its moved... for example:

abbr --set-cursor='!' gcmsg 'git commit -m "!"'

Which exands to git commit -m "<CURSOR>" and so if self-insert comes after then I get git commit -m " <CURSOR>" which is definitely not my intention :)

@faho this may require changes in expand-abbr, or perhaps a new expand-abbr-and-self-insert that correctly places the self-insert and then backtracks accordingly to perform the expand without breaking ;|&<>) from triggering expansion.

This is my current workaround (note not a fix for all issues, just the ones that affect me most of the time):

# most of the time I use space to trigger --set-cursor abbr's and it just so happens expand-abbr works with self-insert before & after for " " only
bind --preset $argv " " self-insert expand-abbr
# rest of these I assume I am only looking for vanilla abbrs to work (so --set-cursor abbrs are broken still):
bind --preset $argv ";" expand-abbr self-insert
bind --preset $argv "|" expand-abbr self-insert
bind --preset $argv "&" expand-abbr self-insert
bind --preset $argv ">" expand-abbr self-insert
bind --preset $argv "<" expand-abbr self-insert
bind --preset $argv ")" expand-abbr self-insert

[g0t4]

As for cursor_backtrack, IIAC this goes back a char before trying expansion and ' ' won't be treated as a token so it correctly finds and expands the previous "command" positioned token (if any). Whereas with ;|&<>) I would assume those are all treated as tokens and so expansion doesn't work unless the backtrack is 2...

fish-shell/src/reader.rs

Lines 2919 to 2925 in 05ac1b7

	rl::ExpandAbbr => {
	if self.expand_abbreviation_at_cursor(1) {
	self.inputter.function_set_status(true);
	} else {
	self.inputter.function_set_status(false);
	}
	}

fish-shell/src/reader.rs

Lines 4404 to 4420 in 05ac1b7

	fn expand_abbreviation_at_cursor(&mut self, cursor_backtrack: usize) -> bool {
	let (elt, el) = self.active_edit_line();
	if self.conf.expand_abbrev_ok && elt == EditableLineTag::Commandline {
	// Try expanding abbreviations.
	self.update_commandline_state();
	let cursor_pos = el.position().saturating_sub(cursor_backtrack);
	if let Some(replacement) =
	reader_expand_abbreviation_at_cursor(el.text(), cursor_pos, self.parser())
	{
	self.push_edit(elt, Edit::new(replacement.range.into(), replacement.text));
	self.update_buff_pos(elt, replacement.cursor);
	return true;
	}
	}
	false
	}
	}

All of this feels like the fish shell input functions need to more intimately understand abbr, and what is being inserted, to find a solid workaround?