Make counsel-ag respect ivy-re-builder-alist

[CeleritasCelery]

closes #1342. counsel-ag nor its derivatives support out of order matching, even though the underlying programs do. This means builders like ivy--regex-ignore-order or ivy--regex-plus are essentially ignored. I took the liberty of enabling out of order matching for the searchers that support it. This is one of the killer features of helm-ag IMHO. This will respect ivy-re-builder-alist.

The only corner case for this is rg which did not add support for look-around's till 0.10.0 which was released Sep 2018. I implemented a solution to automatically handle the version checking, but if that seems too heavy handed I can remove it. Instead we could just define a custom variable like counsel-rg-look-around-enabled-p. However that would mean it would not work out of the box anymore.

Last thing I changed that may be up for debate is I removed the :caller options from counsel-ag. See #1934 for more details. I think this is the right call, because now you can set a different re-builder for each type of searcher, which was impossible before. If you still want to match all counsel-ag related functions, you can set the key to the collection (counsel-ag-function).

[basil-conto]

Thanks for working on this!

Last thing I changed that may be up for debate is I removed the :caller options from counsel-ag. See #1934 for more details. I think this is the right call, because now you can set a different re-builder for each type of searcher, which was impossible before. If you still want to match all counsel-ag related functions, you can set the key to the collection (counsel-ag-function).

As I mentioned in the linked issue, I don't think we should remove :caller from counsel-ag. It makes counsel-ag less customisable, it breaks existing user configurations, and it doesn't solve the real problem, which is that other search functions piggy-back counsel-ag in a very ad-hoc and messy way.

[CeleritasCelery]

Thank you for your review @basil-conto. I have pushed another commit based on your feedback. I did not change the wrappers around counsel-ag because that feels like a it should be tackled in a separate issue. Let me know what you think.

[CeleritasCelery]

I was thinking about this some more, and I feel that I made it more complicated then it needed to be. I removed the custom variable and instead just check if rg doesn't produce an error when we use that flag. If accepts that flag, we assume that it works. Let me know what you think.

[basil-conto]

LGTM.

[CeleritasCelery]

@abo-abo I now have a CA from gnu.org. If that was the hold up for merging this.

[abo-abo]

Merged, thanks.

Please review my follow up commit with new added tests. I had to research what negative lookaheads were, since I didn't use them before. With the added tests, hopefully it should be more clear.

[dsedivec]

Just a note for anyone using ivy--regex-ignore-order and ripgrep: this transformed searches such as foo to (?=.*foo) for me, which made ripgrep run about 20x slower—so slow that counsel-rg never finished on my work code base, even after several minutes.

If you have this problem, I suggest switching to another regexp builder for counsel-ag, such as ivy--regex-plus. Here's how I set this in my init.el:

(setf (alist-get 'counsel-ag ivy-re-builders-alist)
      #'ivy--regex-plus)

[CeleritasCelery]

@dsedivec, note that ivy--regex-plus Will use look arounds if you attempt to add negative matching (e.g. foo ! bar). I would recommend using ivy--regex instead.