Recommended way to detect interactive terminal?

[hugovk]

I'm maintaining the termcolor colour library and recently added tty detection, so colour is disabled for things like pipes:

def _can_do_colour() -> bool:
    """Check env vars and for tty/dumb terminal"""
    if "ANSI_COLORS_DISABLED" in os.environ:
        return False
    if "NO_COLOR" in os.environ:
        return False
    if "FORCE_COLOR" in os.environ:
        return True
    return (
        hasattr(sys.stdout, "isatty")
        and sys.stdout.isatty()
        and os.environ.get("TERM") != "dumb"
    )

https://github.com/termcolor/termcolor/blob/ec269e608a08b33beb739b6ab5f17b9217a5692a/src/termcolor/termcolor.py#L106-L118

Unfortunately this means colour is disabled in Jupyter/JupyterLab: termcolor/termcolor#28

I see there was a plan to set isatty to True for ipykernel.iostream.OutStream, but instead it was made configurable to make sure things like pagers (less, more) didn't get confused:

• Make the isatty method of OutStream return True #683
• Make the isatty method of OutStream return True #683 (comment)

So at the moment, if a Jupyter user wants colour from termcolor, they need to either set an env var FORCE_COLOR=1, or configure ipykernel.iostream.OutStream so isatty=True.

But if we wanted to add auto-detection to termcolor's _can_do_colour, what sort of thing should we query to know Jupyter is being used in an interactive way?

Thank you!

[blink1073]

Hi @hugovk, we inject some environment variables here, we could add an environment variable that you could put in your _can_do_color test, or perhaps set FORCE_COLOR=1 if that makes more sense.

[hugovk]

An env var would be great.

Or setting FORCE_COLOR=1 is an interesting idea, it's also supported by a number of other tools, such as Rich, pytest, nox, tox, pypa/build (see links in Textualize/rich#2425).

Having said that, I see you're already setting env["CLICOLOR"] = "1" there, so I should probably add support for that into termcolor:

• https://bixense.com/clicolors/

[blink1073]

Based on that link, it seems like CLICOLOR_FORCE is what we should be using, since we are not a tty.

[blink1073]

But we could set both since rich already expects CLICOLOR.