--output like Ack (or: rethink --replace)

[mernen]

First, a question: what exactly is the use case for --replace? I don't think I've ever seen any other tool offering this option, and I can't think of many uses.

I do, though, find Ack's --output useful every now and then. It's a similar feature, but it replaces the contents of the entire line rather than only the matched part. When coupled with -h (hiding line numbers and file names), it allows one to pipe the results to other tools to get statistics about the matches in ways I'm not sure how I'd achieve with ripgrep (or other competitors, for that matter).

$ # what are the std modules in use?
$ ack '\bstd::\w+' --output '$&' -h | sort | uniq -c | sort -nr
  29 std::path
  28 std::io
  18 std::os
  16 std::sync
  14 std::ffi
  10 std::error
   9 std::fs
   9 std::fmt
   7 std::str
   6 std::env
   5 std::cmp
   4 std::thread
   4 std::process
   4 std::mem
   4 std::collections
   3 std::borrow
   2 std::time
   2 std::result
   2 std::ops
   2 std::cell
   1 std::vec
   1 std::slice
   1 std::marker
   1 std::iter
   1 std::hash
   1 std::ascii

(ripgrep's --replace can be fully emulated in Ack using $` and $' in the output expression, but those perlisms are rather cryptic, and also pretty awkward to use on the CLI)

[BurntSushi]

Ack's --output is isomorphic to ripgrep's --replace. Apparently, it works slightly differently. As you noticed, it enables one to replace each match if they want to. At the same time, it's pretty easy to replace the entire line as well. All you need to do is make the regex match the full line. So \bstd::\w+ would be written as ^.*\bstd::\w+.*$. To use --replace, you'd then want to create a capturing group as well. So, ripgrep's version of your ack command is:

$ rg '^.*(\bstd::\w+).*$' --replace '$1' --no-filename | sort | uniq -c | sort -nr

Which has the same output. The --no-filename is like ack's -h flag.

(ripgrep's --replace can be fully emulated in Ack using $` and $' in the output pattern, but those perlisms are rather cryptic, and also pretty awkward to use on the CLI)

Hmm, yes, I'm actually not familiar with those perlisms. :-)

[mernen]

Huh, indeed, it didn't occur to me at all to use ^.* and .*$ to replace the full line.

Since they are fully isomorphic, if there's no intention to change how --replace works, I suppose this issue should be closed — even though I still find --output more useful (i.e. simpler on the more common use case), I don't think it's a good idea to have both, and I can certainly live just with --replace.

[BurntSushi]

Yeah, I can see how --output might be more convenient. Tricky issue.

I think we can at least improve the docs of --replace to include an example that replaces the entire line.

[mernen]

Just saw #34, about the grep feature --only-matching. I think it'd be the best of all worlds:

• As far as I can see, -or would be equivalent to Ack's --output
• The alternative use case (today's --replace) would still be available, and significantly less cryptic and annoying to type than Ack's --output '$`expr'"$'"
• -o is a grep standard and works with other matches. I did consider adding a flag, but it seemed quite ridiculous to have one just for --replace. It didn't occur to me this same flag could have other (natural) uses.

[BurntSushi]

@mernen Good point. It actually didn't occur to me that --only-matching would be combined with --replace like that, but it makes perfect sense!