live grep AND operator

[rieje]

In Swiper (Emacs package equivalent), whitespaces act as wildcards so searching word1 word2 searches as word1.*word2. It is possible to replicate this very convenient behavior in Telescope.

Is it possible to do the same with fzf-lua? It's might be the only thing keeping me using Telescope.

And optionally, is it possible to disregard order of words so that word1 word2 also searches word2 word1 (word1.*word2 and word2.*word1, respectively)?

Much appreciated.

[ibhagwan]

Live grep uses a regex expression sent directly to rg, the example you mention should work out of the box, more so if you use live_grep_glob (or set grep.rg_glob=true) you an also use iglob filtering, for example, excluding spec files:

Notice the regex highlight of the foo.*bar

And optionally, is it possible to disregard order of words so that word1 word2 also searches word2 word1 (word1.*word2 and word2.*word1, respectively)?

There’s probably a regex you can craft for that kind of scenario or use grep_project and use fzf’s exact matching with 'word2 'word1:

[ibhagwan]

To be clear, the prompt should simply be foo bar and in the Telescope example it was configured to replace whitespaces with .* to achieve this, i.e. whitespace implicitly means .* for rg.

You can customize your own rg_glob_fn and replace spaces with .* but this isn't something I'd add as builtin as it makes no sense for most users:
https://github.com/ibhagwan/fzf-lua/wiki#how-can-i-send-custom-flags-to-ripgrep-with-live_grep

[rieje]

Oh yea, I totally don't expect this to be a builtin since it's not a standard behavior. I'm not a programmer and I tried implementing this, but it only seems to work (not tested extensively) if glob separator is used, any ideas?

Attempt at modifying the query to do the following:1) a single space between words should be replaced with .*. E.g. querying word word2, it searches for word.*word2. 2) To search for an actual space, add an additional space. E.g. to search for word<ONE SPACE>word, query word<TWO SPACES>word2, to search for word<TWO SPACES>word2, query word<THREE SPACES>word2. -- should also be supported where everything that comes after in the query should be treated as rg flags.

grep = {
  rg_glob = true,

 rg_glob_fn = function(query, opts)
    ---@type string, string

    -- modify query. for a single space between words, replace with wildcard
    local mod_query = query:gsub("([^%s])%s([^%s%-])", "%1.*%2")

    -- to avoid wildcard when wanting to search for space, add an
    additional space, i.e. use 1 space for wildcard, 2 spaces for 1 space,
    3 spaces represent 2 spaces
    mod_query = mod_query:gsub("  +", function(match)
      local length = #match
      if length > 2 then
        return string.rep(" ", length - 1) -- reduce the number of spaces by 1
      else
        return " " -- for exactly two spaces, return one space
      end
    end)

    -- everything that comes after ' --' are interpreted as rg_flags
    local regex, rg_flags = mod_query:match("^(.*)%s%-%-(.*)$")

    return (regex or mod_query), rg_flags
  end,
},

If I include -- then it looks like it works, but not without it. I would think return (regex or mod_query) should take care of that? It doesn't seem straightforward to debug this.

---

Side note, I was a bit confused by rg_glob meaning, wouldn't rg_flags be more appropriate since it can take rg arguments such as -w (--word-regexp)? I was thinking glob to mean * as in wildcard.

[ibhagwan]

If I include -- then it looks like it works, but not without it. I would think return (regex or mod_query) should take care of that? It doesn't seem straightforward to debug this.

It is pretty straight forwards, use io.stdout:write() to print debug info into fzf.

Side note, I was a bit confused by rg_glob meaning, wouldn't rg_flags be more appropriate since it can take rg arguments such as -w (--word-regexp)? I was thinking glob to mean * as in wildcard.

In hindsight yes, but this was originally (and still used to parse just --iglob flags.

[rieje]

glob_separator = "" solved the issue for me, though it doesn't sound like it would work or make sense in the context of the following code. It's because rg_glob_fn requires the query to contain glob_separator. Not sure what to think of this since it works but what it implements was not the intention of the rg_glob_fn. The final code:

grep = {
  -- for convenience, regex for query is rebuilt using rules inspired by
  -- https://oremacs.com/swiper/#ivy--regex-plus:
  -- 1) single spaces are replaced with `.*` (consecutive spaces are not
  --    affected)
  -- 2) to include literal spaces, use an extra space (e,g, 'word1  word2'
  --    for 'word1 word2')
  rg_glob = true,
  glob_separator = "", -- needed to make this work
  -- @return string, string?
  rg_glob_fn = function(query, opts)
    local regex, flags = query:match("^(.-)%s%-%-(.*)$")

    local function rebuild_regex(str)
      local result = str:gsub("%s+", function(r)
        if #r == 1 then
          return ".*"
        else
          return r:match("%s(.*)")
        end
      end)
      return result
    end

    if not regex then
      return rebuild_regex(query), ""
    end

    return rebuild_regex(regex), flags
  end,
},

[ibhagwan]

Good job @rieje.