[arp242]

For zsh I use this to setup $PATH, which seems to address most of the use cases of this tool:

  ### Setup PATH
  dirs=(
      /usr/lib/ccache/bin                   # ccache
      ~/.local/script ~/.local/bin          # My local scripts.
      ~/.local/gobin                        # GOBIN
      /usr/local/bin /usr/local/sbin        # local takes precedence
      /bin /sbin /usr/bin /usr/sbin         # Standard Unix
      /usr/pkg/bin   /usr/pkg/sbin          # NetBSD
      /usr/X11R6/bin /usr/X11R6/sbin        # OpenBSD
      /opt/ooce/bin                         # illumos
      ~/.local/share/gem/ruby/*/bin(N[-1])  # Ruby
      ~/.luarocks/bin(N[-1])                # Lua
      ~/.local/dotnet/root ~/.local/dotnet/tools  # .NET
      /usr/lib/plan9/bin                    # Plan 9 from userspace
      ~/code/Vim/gopher.vim/tools/bin       # gopher.vim
      ~/.nimble/bin                         # Nim
      ~/.cargo/bin                          # Rust
      ~/.ghcup/bin                          # Haskell
  )
  typeset -U path=()                        # No duplicates
  # Use full path so /bin and /usr/bin aren't duplicated if it's a symlink.
  for d in $dirs:A; [[ -d $d ]] && path+=($d)
  unset dirs d

typeset -U makes the $path array "unique", and :A ensures the full path is always used (in case of symlinks).

You can probably do something similar with bash, but idk.

$path is tied to $PATH and an array, so you can use it as such for some of the other things:

  print ${(F)path} | grep /bin  # (F) to print one array per line
  print ${(F)path} | sort       # can also use (o) to print ordered

[throwaway84846]

I use 'path=( ${(u)^path:A}(N-/) )' instead of that for loop

[mst]

To just reduce the current entries to those that exist, I'd probably do

    PATH=$(perl -e 'print join ":", grep -d, split ":", $ENV{PATH}')

but that's because as a 20 year perl hacker and thus connoiseur of line noise, I'd rather read that than the shell clever.

(this doesn't mean the shell clever is bad, it just means I don't find -that- dialect of line noise as skimmable)

[JNRowe]

General point about comfort noted, but they don't do the same thing.

$^path(N) is an equivalent to your perl expression. The snippet throwaway84846 posted also removes duplicates and collapses symlinks from /usr-merge for example.

[adriangrigore]

Plan 9 from User Space binaries are on /usr/local/plan9/bin/ on OpenBSD.

[epogrebnyak]

How does (N[-1]) get resolved?

[arp242]

(N) sets the NULL_GLOB option for that pattern:

    % print doesnt-exist*
    zsh: no matches found: doesnt-exist*
    % print doesnt-exist*(N)
    (doesn't print anything, pattern is just ignored)

([n]) is array indexing to get entry n; -1 gets the last entry:

    % print a*
    a-1 a-2 a-3
    % print a*([-1])
    a-3
    % print a*([1])
    a-1

Globbing is automatically sorted by name.

So putting that together, things like this:

      ~/.local/share/gem/ruby/*/bin(N[-1])  # Ruby

will use the latest Ruby version in ~/.local/share/gem/ruby, and it won't error out if it doesn't exist (so I can freely copy this around even to machines without Ruby, or remove Ruby, etc).

[epogrebnyak]

Thanks for explaining, looks very clever. Is this syntax part of bash?

[arp242]

No, bash has nothing like this. It's all very specific to zsh. They're called "glob qualifiers" if you want to find out more.