Orderless completion doesn’t work in comments?

[kings2u]

I use the Orderless package for completion-styles, and the author recommends at https://github.com/oantolin/orderless/#company to use (setq orderless-component-separator "[ &]") in order to sort Company results with Orderless instead of Company. This works great while typing code in an Emacs-Lisp buffer using the built-in Company backends, but if I’m trying to get completion when typing in a code comment, using & as a separator between search strings no longer works. This is the case despite setting company-dabbrev-code-everywhere to t, though I’m not sure that’s relevant.

Does anyone have advice on how to use Orderless for searching Company in all possible contexts?

[dgutov]

Unfortunately, Orderless only works through completion-styles, and when we're in a comment, that usually means using the company-dabbrev backend, which is a classic Company backend which doesn't work through CAPF API.

You can take a look https://github.com/PythonNut/company-flx, but it provides a different matching method than Orderless.

[dgutov]

Good find. Does it work better with the change below?

diff --git a/company.el b/company.el
index 1fffe96..2e36d52 100644
--- a/company.el
+++ b/company.el
@@ -2256,7 +2256,7 @@ each one wraps a part of the input string."
   (let ((re (funcall company-search-regexp-function text))
         (i 0))
     (cl-dolist (line lines)
-      (when (string-match-p re line (length company-prefix))
+      (when (string-match-p re line)
         (cl-return i))
       (cl-incf i))))
 

[kings2u]

Indeed it does! The only thing it doesn’t fix is that the search string bo in bookmark-edit-annotation-text-function and bookmark-read-annotation-text-function is not highlighted with the company-tooltip-search face like the other candidates containing tex and bo are. Not a huge deal, but it’s something to note if there’s an easy fix for that too. I look forward to this package update :)

[dgutov]

There was a similar problem there, should be fixed now too. 👍

And it was actually a leftover from the old design where we only did prefix matching -- so the code made sure to skip the prefix when doing the search, assuming the user would want to only search in the remaining part of the completions (since they begin the same). This assumption no longer holds, so it's easier to remove it than to do some kind of adaptation.

[dgutov]

If you like how it works now, you can write your own reply and mark is as the answer ;)

[kings2u]

Everything works swimmingly now, thank you again :) I will of course write a summary and mark it as the answer.

[kings2u]

The above reply from @dgutov solved the issue for me. For those who stumble onto this issue, I recommend changing the user option company-search-regexp-function to company-search-words-in-any-order-regexp, then commence a search using company-filter-candidates (bound to C-s). This accomplishes the most common use case for the Orderless package of narrowing based on out-of-order strings.