When I migrated to fish from bash, I found glob patterns like "tmp[0-9][0-9].pdf" aren't supported. Then, recently fish has stopped to support the ? wildcard:

$ fish --version
fish, version 4.0.0
$ ls tmp00.pdf
tmp00.pdf
$ ls tmp??.pdf
ls: tmp??.pdf: No such file or directory
$

.0/fish_for_bash_users.html says, "Fish only supports the * and ** glob (and the deprecated ? glob) as syntax."

So, has somebody written a fish function that can simulate the good-old glob? which might be used like

for file in (fmatch 'img..\.jpg') # to get img00.jpg, img01.jpg, . . .
   # do something on $file

With some help from ChatGPT, I've been able to write something that filters the output from '*' with string match -r -- $regex -- $file, but the handling of paths is incredibly hard to me:

# bash
for file in ../../*/img??.jpg # How to implement this?

and also the constant need of escaping . in regex is tedious, especially when your path includes ../.

I tried fd -g <glob pattern> but this glob doesn't seem to allow for paths.

I need the good old Glob. Perhaps I should admit the defeat :-) and do this

function glob --description 'Good old glob'
    if test (count $argv) -ne 1
        echo "Usage: glob <glob_pattern>"
        return 1
    end
    bash -c 'printf "%s\n" $*' _ $argv[1] # call bash!
end

Edit: Initially, I wrote

bash -c "ls $argv[1]"

But a better version was proposed by @CharlesDuffy, which I've replaced mine with, in the above fish function. [Please refer to @CharlesDuffy's discussion below for why printf and what problems exist in $*. But, given that the above solution still has some problems, it's probably better to install this glob command instead of writing a fish function like the above.]

I'm already using the glob.fish function defined as above. So far so good. glob tmp[0-9][0-9].pdf works!

When I migrated to fish from bash, I found glob patterns like "tmp[0-9][0-9].pdf" aren't supported. Then, recently fish has stopped to support the ? wildcard:

$ fish --version
fish, version 4.0.0
$ ls tmp00.pdf
tmp00.pdf
$ ls tmp??.pdf
ls: tmp??.pdf: No such file or directory
$

https://fishshell/docs/4.0/fish_for_bash_users.html says, "Fish only supports the * and ** glob (and the deprecated ? glob) as syntax."

So, has somebody written a fish function that can simulate the good-old glob? which might be used like

for file in (fmatch 'img..\.jpg') # to get img00.jpg, img01.jpg, . . .
   # do something on $file

With some help from ChatGPT, I've been able to write something that filters the output from '*' with string match -r -- $regex -- $file, but the handling of paths is incredibly hard to me:

# bash
for file in ../../*/img??.jpg # How to implement this?

and also the constant need of escaping . in regex is tedious, especially when your path includes ../.

I tried fd -g <glob pattern> but this glob doesn't seem to allow for paths.

I need the good old Glob. Perhaps I should admit the defeat :-) and do this

function glob --description 'Good old glob'
    if test (count $argv) -ne 1
        echo "Usage: glob <glob_pattern>"
        return 1
    end
    bash -c 'printf "%s\n" $*' _ $argv[1] # call bash!
end

Edit: Initially, I wrote

bash -c "ls $argv[1]"

But a better version was proposed by @CharlesDuffy, which I've replaced mine with, in the above fish function. [Please refer to @CharlesDuffy's discussion below for why printf and what problems exist in $*. But, given that the above solution still has some problems, it's probably better to install this glob command instead of writing a fish function like the above.]

I'm already using the glob.fish function defined as above. So far so good. glob tmp[0-9][0-9].pdf works!

• shell
• glob
• fish

• 1 fish supports ? just fine. It's not clear why you are conflating regular expressions and glob patterns, though. . has no special meaning in a glob. Neither * nor ? will match the path separator, though, so to search in an arbitrarily deep path, you need to use **, which matches zero or more path components: **/foo would match /foo, /bar/foo, /xyz/bar/foo, etc. – chepner Commented Mar 5 at 15:56
• 1 @chepner "fish supports ? just fine" . . . It seems to have been removed recently. I've edited my question above. I understand the rest of what you say. – Ryo Commented Mar 6 at 18:40
• 1 BTW, the recommendation to use printf instead of ls was very deliberate. (Alternate source, written in light of objections to that FAQ). Admittedly, a better solution would use printf '%s\0' instead of printf '%s\n', but I don't know how to make the fish end of that work off the top of my head (specifically, I don't know how to tell fish to expect NUL delimiters when reading a subprocess's output into an array). – Charles Duffy Commented Mar 6 at 23:02
• 1 @xpusostomos, $* unquoted in bash concatenates all your arguments into a single string, then word-splits that string and expands each element in it as a separate glob -- so if your original argv has items with spaces, those spaces are instead treated as separating individual glob expressions. – Charles Duffy Commented Mar 8 at 17:17
• 1 @xpusostomos, ...additionally, with echo $* you can't tell if a space in the output is intended to separate two distinct items, or is a case where a glob matched a filename with a space as part of its literal content (where that entire filename needs to be understood by the caller as a single object). – Charles Duffy Commented Mar 8 at 17:18

2 Answers 2

I happened to be looking for a solution to this exact problem today...

Standing on the shoulders of everyone in the discussion above (special thanks to Ryo and Charles Duffy), I came up with this improved but somewhat "uglified" and more complex version of OP’s function...

function bash_glob --description 'Expand bash-style glob string into an array'
    [ (count $argv) -eq 0 ] && echo 'Usage: bash_glob <pattern>' && return 1

    bash -c '
        shopt -s nullglob;
        IFS=;
        eval "printf \"%s\\\\\\0\" $*"
    ' _ $argv | string split0
end

Example:

me@my-computer ~> bash_glob '.config/[e-n]??[cm]*/README.{md,txt}'

.config/emacs/README.md
.config/nvim/README.md

me@my-computer ~> ▇

Main Changes:

• \0 as an array separator: Avoids wrong processing of paths containing whitespaces.
• IFS= as a security measure: Prevents word splitting and potential code injection in Bash.
• eval: Ensures Bash fully expands $*, and that the glob patterns are processed entirely within Bash.
• shopt -s nullglob: Prevents unexpanded glob patterns from appearing as extra arguments - this way, only matched files are included in the result.

Only with these changes was I able to get the results I needed, since I did have to handle filenames with spaces, glob patterns that wouldn’t always expand, and other edge cases. This approach avoids most issues, at least.

I myself use the solution provided by @rsenna (Thank you!) et al, but I post this just to suggest an alternative, which people might be interested in.

This project

https://github/p-ranav/glob

provides a glob library. As an example, it also provides a very simple stand-alone command for globbing. Out of the box, it doesn't work like glob <pattern> but it's a very short simple C++ program that can be easily modified.

So, here I report I just modified the source code in a trivial way, compiled everything using the provided make script, and successfully obtained a command line tool that can be an alternative to the fish-shell function shown in other messages.

If I'm very much bored (which wouldn't happen in the near future), I might turn this glob command into a Homebrew package . . .