There is a project argonaut I'm working on; this tool is used to simplify defining command-line arguments in the shell, similar to what cobra does in Go. Here is an example to help illustrate the issue.

This is a script the user wrote: cool.sh

#!/bin/bash set -e EVAL_CONTENT=`argonaut bind \  --flag=flag1 \  --flag-flag1-default=hello \\  --flag-flag1-choices=hello,world \  --flag=flags2 \  --flag-flags2-multi=true \  --flag-flags2-choices=one,two,three  -- "$0" "$@" ` eval "$EVAL_CONTENT"

Then run

./cool.sh --flags2 one --flags2 three

EVAL_CONTENT will be assigned

FLAG1='hello' FLAG2='one,three' 

So after executing eval, the user can directly use these two environment variables in their own scripts.
If you enter the wrong parameters, for example

./cool.sh --flags2 one --flags2 four

will output

Error: value four for flag flags2 is not in allowed choices [one two three] Usage: cool.sh [flags] Flags: --flag1 string (default "hello") --flags2 stringArray -h, --help help for cool.sh 

So far, everything has worked perfectly, until I executed

./cool.sh --help

It's a complete disaster. Because EVAL_CONTENT will become the help message, and eval will obviously throw an error. The workaround I can think of right now is

1. Redirect the help message output to stderr
2. Return a nonzero exit code

As for redirecting the help output, I've already found the API, but when dealing with the latter I discovered that cobra doesn't support customizing the help's return value, because help never reaches the Run method and is handled by cobra beforehand, even if I define the help flag myself. I even think this is a bug: if the user defines help themselves, it indicates they want to handle help requests themselves, not have the framework do it.