Add literal to default matching styles?

[oantolin]

Add literal to default matching styles?

[oantolin]

It happens because with those matching styles (with any list of matching styles that includes orderless-regexp), the incomplete [0-9 gets compiled to a malformed regexp. Mmh.

[oantolin]

I'm looking into the best way to do it. Maybe there is something better, but matching against anything (an empty string, for example) should work: if it returns it's valid, if it errors it's not.

[oantolin]

I thought the (defalias 'orderless-regexp #'identity) implementation was cute, I'm kind of sad to see it go. 😛

[minad]

You could introduce (defalias 'orderless-regexp-I-know-what-I-am-doing #'identity).

[oantolin]

Sure, why not? The only thing that gives me pause is that you said elsewhere it didn't work for you. Let's figure out why not before doing anything.

[minad]

Yes, sure. I wonder why it didn't work like this. Otherwise I don't see big downsides of doing this. And one could use style dispatchers to mark certain parts of the input as explicitly literal/regexp. However style dispatchers will also introduce problems regarding to the matching of exact inputs.

[minad]

I set (setq orderless-matching-styles '(orderless-regexp orderless-literal orderless-initialism)).

If I invoke consult-line and enter [0-9, then it does not match anything. As soon as I close [0-9] it matches both 1329817321 and [0-9], assuming that [0-9] occurs literally in the buffer. This is the bug.

[oantolin]

Mmh. Does changing the order of orderless-regexp and orderless-literal make any difference (it shouldn't, but if it does that might be a clue)?

[minad]

No difference.

[oantolin]

I'll look into it. I thought my tests covered this case...

[minad]

Hmm, I see. Is it hard to detect if a regexp is valid and only include it in that case?

[oantolin]

It looks like re-builder uses that strategy, so I will too.

[minad]

I wonder why the current approach didn't lead to errors. Is this because the matching happens deep inside all-completions in the c code?

[oantolin]

The reason is that I already have a condition-case to catch errors in matching and just report no matches.

[oantolin]

I think I'll have the orderless-regexp style match invalid regexps literally. This seems to be the least surprising behavior.

[oantolin]

OK, done. a609004

Now I'll add orderless-literal to the default matching styles.

[minad]

Great, thanks! Nice that we also uncovered a bug through this discussion. cc @jaor

[jaor]

On Tue, Mar 23 2021, Daniel Mendler wrote: Great, thanks! Nice that we also uncovered a bug through this discussion. cc @jaor

Yes, nice outcome. Thanks.

[jaor]

BTW, another place where this bit me is when running an embark action that asks for confirmation, which inserts the candidate in the prompt. For instance (without orderless-literal), if i consult-buffer, make a buffer called `*<w3m>*<1>` current, embark-action and k to kill it, one cannot then RET to kill it.

[oantolin]

This why I always bind M-RET to exit-minibuffer, I get sick of any completion UI telling me what I can and cannot submit as input. All of them do it in different circumstances, I think.