for what it's worth, I'm not sure git bash is a necessary use-case for us. There are many alternatives like shims, powershell, and wsl for windows users.

I'm generally not a Windows user but I agree that shims and powershell seem to work well enough. Windows is a PITA anyways, in my last project Git Bash was my only option (was not allowed to use WSL or install anything else). While probably rare there might be some users restrained to such a setup.

right but even then shims will work. What won't work is you won't see env vars unless you wrap things in tasks but as I'm sure you've heard me explain, that's kind of how I think projects should be setup anyways since it doesn't rely on shell extensions.

mise en would be another alternative

I noticed that when using shims in context of cygwin/mysys a .cmd suffix has to be added though.

isn't that already happening?

What i mean is that i would expect for example that i can use java -version in Git Bash but instead i have to use java.cmd -version.

is this something that can be configured with PATHEXT? or is there a git bash equivalent for that?

PATHEXT is a Windows environment variable and i am not aware of an equivalent in Git Bash/Cygwin. Emulating PATHEXT in Bash does also not seem to work.

export PATHEXT=.com:.exe:.bat:.cmd

if declare -f command_not_found_handle >/dev/null; then 
    eval "original_command_not_found_handle() $(declare -f command_not_found_handle|tail -n +2)"
fi
command_not_found_handle(){
    local PATHEXT_EXPANDED i
    IFS=: read -a PATHEXT_EXPANDED<<<"$PATHEXT"
    for i in "${PATHEXT_EXPANDED[@]}"; do
        if type "$1$i" &>/dev/null; then
            "$1$i" "${@:2}"
            return $?
        fi
    done
    if declare -f original_command_not_found_handle >/dev/null; then
        original_command_not_found_handle "$@"
    else
        return 127
    fi
}

While this PR makes activate work for Git Bash and Cygwin detection seems a bit hacky. I wonder if we should drop this for now. Having at least the shims work would be great though.

The following hook might be a workaround for the shims .cmd ending issue. We might simply document this as possible workaround in the troubleshooting docs "Windows problems" section.

command_not_found_handle()
{
   cmd=$1
   shift
   args=( "$@" )

   IFS=:
   for dir in $PATH; do
      for executable in $dir/$cmd.bat $dir/$cmd.cmd $dir/$cmd.exe; do
         if [ -f "$executable" ]; then
            "$executable" "${args[@]}"
            return
         fi
      done
   done

   orig_command_not_found_handle "$cmd" "${args[@]}"
} 
if declare -f command_not_found_handle >/dev/null; then 
    eval "orig_command_not_found_handle() $(declare -f command_not_found_handle|tail -n +2)"
fi

One possible fix for this is to add two shims on windows, one for cmd and one for git bash:

Tested with two shims for node

node.cmd (unchanged)

@echo off
setlocal
mise x -- node %*

node (no extension)

#!/usr/bin/env bash

mise x -- node "$@"

This works totally fine on Git Bash, Cmd, and Powershell (as cmd and powershell will ignore the file with no ext)

Unsure if this works well with the current implementation of reshim() as at first glance it expects a one-to-one shim to app? (My primary language isn't rust so happy to be wrong!)

@ffaen Was also thinking about something in that direction which should be doable.