`**` glob does not match files in the root directory

[zx8]

$ fish --version
fish, version 3.1.2-1083-ga30630a1d

---

bash

$ mkdir foo
$ touch bar foo/bar
$ echo **/bar
bar foo/bar

fish

$ mkdir foo
$ touch bar foo/bar
$ echo **/bar
foo/bar

Is this intentional or an oversight? The documentation says:

** matches any files and directories in the current directory and all of its subdirectories.

[sudormrfbin]

A lone ** matches any files and directories in the current directory; the description you saw was an example. A few paragraphs above it is mentioned:

** matches any string of characters. This includes matching an empty string. The matched string may include the / character; that is, it recurses into subdirectories.

And of particular interest in the specific example you mentioned is this line in the docs (emphasis mine):

Note that augmenting this wildcard with other strings will not match files in the current working directory ($PWD) if you separate the strings with a slash ("/").

So to match files in the current directory also, you specify the glob without the / like so:

$ echo **bar
bar foo/bar

[ridiculousfish]

Right, the forward slash is part of the glob and must be matched. I think fish shell is pretty committed to this decision. Closing, but thanks for filing!

[zx8]

@gokulsoumya So to match files in the current directory also, you specify the glob without the / like so:

$ echo **bar
bar foo/bar

The only problem with this is that it will also match e.g. foobar.

---

@ridiculousfish Right, the forward slash is part of the glob and must be matched. I think fish shell is pretty committed to this decision. Closing, but thanks for filing!

That's fair enough, but it also makes sense to me if ** matched . (i.e. the working directory) because then **/bar would expand and pick up ./bar while still being semantically correct, right?

[zx8]

Also, fish seems like the odd one out here:

fish

$ echo **/bar
foo/bar

bash

$ echo **/bar
bar foo/bar

zsh

$ echo **/bar
bar foo/bar

tcsh

$ set globstar
$ echo **/bar
bar foo/bar

python

Python 3.8.5 (default, Jul 21 2020, 10:48:26)
[Clang 11.0.3 (clang-1103.0.32.62)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import glob
>>> glob.glob("**/bar", recursive=True)
['bar', 'foo/bar']

ruby

$ irb
irb(main):001:0> Dir.glob('**/bar')
=> ["foo/bar", "bar"]

go (via https://github.com/bmatcuk/doublestar)

package main

import (
	"fmt"

	"github.com/bmatcuk/doublestar"
)

func main() {
	g, _ := doublestar.Glob("**/bar")
	fmt.Println(g)
}

[foo/bar bar]

rust (via https://docs.rs/glob/0.3.0/glob/)

use glob::glob;

fn main() {
    for entry in glob("**/bar").expect("Failed to read glob pattern") {
        match entry {
            Ok(path) => println!("{:?}", path.display()),
            Err(e) => println!("{:?}", e),
        }
    }
}

"bar"
"foo/bar"

node (via https://www.npmjs.com/package/glob)

$ node -e 'var glob = require("glob"); glob("**/bar", function (er, files) { console.log(files); })'
[ 'bar', 'foo/bar' ]

[ridiculousfish]

Ok, maybe we should change it.

[mqudsi]

I've been thrown by this before, tbh.

I think the current behavior is fishy (not in a good way), and since fish only uses globs for matching against files, it seems like it's not a stretch to incorporate path semantics into the glob matching.

That said, git behaves similar to fish:

mqudsi@Blitzkrieg /m/d/r/fish> git ls-files '*.in' | grep -v /
config_cmake.h.in
fish.pc.in
fish.spec.in
mqudsi@Blitzkrieg /m/d/r/fish> git ls-files '**/*.in' | grep -v /

[zx8]

@ridiculousfish Hope I'm not being too cheeky here, but I noticed your recent commits to master are poking around the ** glob code - any chance you might be able to address this issue while you're in the thick of it? 🙏

[ridiculousfish]

I found a way to add this behavior without too much extra work. Most other glob implementations require that ** appear in a segment by itself; fish does not which makes this trickier. I added a special case to incorporate that behavior; that is if ** appears in a segment by itself we also match in the same directory as if that segment were missing. Updated docs too, tried to make them simpler.