Ripgrep: Have a --count option that counts occurrences instead of matched lines:

[santagada]

Right now -c/--count counts line matches, it would be great to have an option to count matches and not lines matched. Looking at the example:

$ echo "foo foo" > example.txt
$ rg -c foo example.txt
1
$ rg -o foo example.txt
1:foo
1:foo

would be interesting to have a way to get the total matches (2) in this case.

[BurntSushi]

Seems reasonable I guess. Maybe a --count-all flag?

[santagada]

I personally would prefer a --count-matches to be clear on what is the difference between the two

[okdana]

Would it make sense to just tie the behaviour to -o?

If i wasn't already familiar with grep i think that's how i would have expected it to work tbh. Because why else would someone want to use -o with -c?

[BurntSushi]

Hmm. I guess I feel like -o/--only-matching is more of a flag for output control, but I agree it is a possibly reasonable path forward.

[santagada]

-o is totally unrelated to counting. I just used it to make the example simpler. My current problem is searching inside binary files where we see many matches on each "line" because counting lines on binary files seems very wrong (but the material of another bug request).... but I have use for both behaviors even on text files. for me --count and --count-matches seems the way to go.

[okdana]

-o is totally unrelated to counting

Yeah, but it doesn't have to be. It literally tells rg to return 'only the matches' instead of whole lines. Why it doesn't change the behaviour of -c to count 'only the matches' instead of whole lines, i can't think of a super compelling reason, besides the fact that that's how grep happens to work. (And grep -o isn't part of POSIX — i think it was a GNU extension added in the early 2000s — so there's not really even a standard AFAIK.)

idk, i don't feel super strongly about it, so that's all i'll say. Just thought i'd mention that that's how my intuition would've told me it should behave.

[balajisivaraman]

I had a look at this (while looking a the code for the --stats feature) and can send a PR for this. But first of all, do we still want this feature implemented?

Another thing to note about the behaviour of grep on my machine, with GNU Grep 3.1, -o -c doesn't behave as @okdana says. (ag, with version 2.1.0, prints no. of matches and not no. of lines with -c.)

$  echo "foo foo" > grep.txt
$ grep -c 'foo' grep.txt
1
$ grep -o -c 'foo' grep.txt
1
$ grep -o 'foo' grep.txt
foo
foo
$ ag -c 'foo' grep.txt
2

If we do want this, shall I put it behind --count-matches as suggested or as a combination of -o -c?

[BurntSushi]

@balajisivaraman I don't know the right answer here. Part of me leans toward --count-matches so that it is explicit. If grep -o -c still reports a line count instead of match count, then there may be value in keeping ripgrep's current behavior there (which is the same as grep).

Also, I think @okdana was pointing out that grep -o -c does report number of lines matched instead of number of total matches, and that that is one possible reason why ripgrep might want to do the same (to be consistent with grep).

[balajisivaraman]

Ah, that does make a lot of sense. My bad for misunderstanding. I'll begin work under the assumption that we want --count-matches for this. 👍

[okdana]

I was actually suggesting that rg do something different from everybody else and have -o modify the behaviour of -c, since it seems like an intuitive result of the combination to me (and using them together for any other reason is essentially pointless). I still like the idea, personally, but given how much emphasis there's been on matching grep's behaviour lately i can see why the other approach would be chosen.

[kbknapp]

I don't see why it can't be both.

• I like --count-matches because it's explicit and easy to search for in the manpages/help. It's also super intuitive.
• I like -c -o because it's short and makes conceptual sense (rg -co 'foo' test.txt would be awesome) . When I look at the grep output, I actually stop and think, "Huh? Oh well, that's strange but I guess so." I'd have a hard time people relying on this behavior

Adding a blurb in the --count-matches help saying, analogous to -c -o is easy too.

[BurntSushi]

@okdana Yeah I know, but I definitely feel a compelling attraction to match grep's behavior in subtle cases like this. But you and @kbknapp present strong arguments, and I like @kbknapp's suggestion for having both.

@balajisivaraman If you want to start with --count-matches and continue on your current trajectory, then that sounds great. That can be introduced in the next minor version release. Changing the behavior of -c -o will have to wait for the next semver bump, so it should be in a separate PR. :-)

[balajisivaraman]

One question I have is how should --count-matches behave when -v is passed in. Once again, the default ag behaviour is befuddling to me. (Or maybe I'm missing something?)

$ echo -e "foo foo\nbar bar\nbaz baz" > grep.txt
$ ag -c 'foo' grep.txt
2
$ ag -c 'baz' grep.txt
2
$ ag -c 'bar' grep.txt
2
$ ag -v -c 'foo' grep.txt (If we're counting non-matching lines, shouldn't this be 2?)
1
$ ag -v -c 'bar' grep.txt (This returns nothing, but should again be 2.)
$ ag -v -c 'baz' grep.txt (Same as above.)
1

Do we want rg to invert matches in the same way it would if -c were passed in, or should we lean towards no interaction between --count-matches and -v? I'm not sure which way to go.

[BurntSushi]

@balajisivaraman The -v flag basically causes any ripgrep option that deals with individual matches to not work because the -v is a line oriented "match" or "no match." I don't think we should copy ag here, but instead, --count-matches -v should probably behave the same as -c -v.