Ripgrep: Provide a `--directories` option to list directories

[williamsmj]

I'm using ripgrep with fzf to fly around my filesystem much faster than I could by using find with fzf.

This approach uses rg --files, which is not only a nice terse, sensible alternative to a long, pedantic find command. It's also orders of magnitude faster!

It would be great if I could do rg --directories and have it work just like rg --files, but give me all the directories, rather than all the files.

find obviously supports a huge variety of options that are out of scope for a tool focussed on searching within files. This particular use case seems (to me at least! ymmv!) common enough that it might be worth supporting with ripgrep.

[bag-man]

+1!

[BurntSushi]

This is a dupe of #169, which I closed on the grounds that it is out of scope. Please read that issue though, because it includes ideas for workarounds. Do they work for you?

My strong preference here is for someone to write a Rust tool that does this for you. It would be relatively simple to do today, since all of the directory traversal and gitignore code has been factored out into separate libraries. The --files listing is a secondary feature of ripgrep. It is intended to tell you what files ripgrep will search.

[BurntSushi]

@bag-man Please avoid +1 comments.

[williamsmj]

Thanks! Sorry I didn't search old isuses!

Piping rg --hidden --files --null | xargs -0 dirname | uniq to fzf works perfectly for me. Thank you.

(rg --hidden --files --null | xargs -0 dirname | sort -u works less well because nothing comes out of the pipeline until rg is finished, so I can't start searching in fzf immediately)

[BurntSushi]

@williamsmj Since #169, ripgrep has grown a --sort-files flag which will do what you want. It will force ripgrep to use a single thread, but it will emit results incrementally instead of blocking until the entire traversal is complete.

[williamsmj]

So rg --hidden --sort-files --files --null 2> /dev/null | xargs -0 dirname | uniq?

Or can I drop some part of the pipeline?

Note the important thing about piping through sort for this use case is the -u not the sort.

[BurntSushi]

@williamsmj Ah I see. If you don't care about the sort, then it'd be fine to drop --sort-files so long as uniq does what you want. Otherwise, it looks good to me!

[ddickstein]

I still got duplicates when I tried this command. Removing --sort-files and replacing uniq with sort -u seemed to work more reliably. Perhaps there's a bug in the sort?

[BurntSushi]

@ddickstein can you open a new issue with a reproducible example? ripgrep should never be showing the same file path twice in the output of --files.

[ddickstein]

Well the duplicates would appear with the dirname call, right? The idea here is that we have a list of sorted files, then we call dirname on all of them (introducing duplicates), but then we call uniq which resolves duplicate directory names. Since there were duplicates in output and I'm assuming there isn't a bug in uniq though it's limited by the need for its input to be sorted, I would guess that the problem is because of a sorting failure with --sort-files rather than a deduping failure with --files. The problem disappeared when I instead did sort+unique at the end by omitting --sort-files and replacing uniq with sort -u. I can try to reproduce at some point and file an issue, but with that change my setup is working the way I'd like.

[BurntSushi]

@ddickstein Oh I see. --sort-files is correct as far as I can see. --sort-files sorts the complete file paths, which happens before dirname is applied. For example, this is valid sorted output of rg --files --sorted-files:

a/b.txt
a/c/foo.txt
a/d.txt

But if you apply dirname to each, then you get

a/c
a

So basically, you need to sort/dedup after applying dirname.

[ddickstein]

Oh interesting. So if it sorted by file type (file v directory) and then alphabetically instead of just alphabetizing file paths then it would have worked? I wonder what the other use cases would be for one kind of sort vs the other.

[matematikaadit]

Relevant project, https://github.com/sharkdp/fd. A simple, fast and user-friendly alternative to find.
It also uses regex and ignore crates.