🐛 Make sure rich.markup is imported when rendering help text

[g-arjones]

rich.markup is used to escape strings when rendering help texts.

This fixes a regression introduced by #1128

[svlandeg]

So just to clarify some edits I just made: The original content of the PR imported rich.markup before using escape, because due to the lazy imports from #1128 this module wasn't available.

I've rewritten this and moved the escaping functionality to rich_utils. My proposal is to never use rich directly in the core library, to isolate all this functionality within rich_utils and avoid any other future regressions.

There's a few more edits and maybe an additional test I'd like to add to this PR, so I'll put this in draft while I continue to work on it.

And thanks for the report & PR, @g-arjones ! 🙏

[svlandeg]

@g-arjones: do you have a minimal example that crashes on master?

[g-arjones]

@svlandeg Sure, here you go:

from typing import Annotated

import typer


app = typer.Typer(rich_markup_mode=None)


@app.command()
def foo(bar: Annotated[str, typer.Argument(help='foobar')]) -> None:
    """foobar."""
    pass


app()

Then run with --help

[svlandeg]

Thanks! This has been a bit tricky to get into the test suite, because other tests do import rich.markup at some point, which is why we didn't catch this regression earlier. I've now added a test that catches your originally reported regression on master, and that should go green on this PR.

I will add one more edit to this PR to fix a similar issue with accessing rich.traceback directly in main.py.

[g-arjones]

Thanks for taking the time!

[tiangolo]

Thanks for the report and fix! 🍰

And thanks @svlandeg for the tweaks, confirmation, test... 🙌

This will be available in Typer 0.17.4 released in the next few minutes. 🎉