[BUG] Progress bar data is not piped when a script using rich is run in a subprocess

[joeyballentine]

Describe the bug

Progress bar updates are not piped via stdout when running any script that uses rich in a subprocess. I specifically noticed this when trying to parse the progress bar of a pip install (pip uses rich for its progress bar), when I noticed that it only pipes the data of the final (finished) bar, and none of the updates.

This stack overflow post I found describes the issue in slightly more detail and provides a minimal repro. They mention using python's Popen, but the same thing happens in nodejs's child_process.spawn function.

Here's an example of what it outputs:

Collecting onnx\r\n
  Downloading onnx-1.11.0-cp39-cp39-win_amd64.whl (11.2 MB)\r\n
---------------------------------------- 11.2/11.2 MB 6.7 MB/s eta 0:00:00\r\n

And what I would expect it to output:

Collecting onnx\r\n
  Downloading onnx-1.11.0-cp39-cp39-win_amd64.whl (11.2 MB)\r\n
     -                                        1.0/11.2 MB 6.7 MB/s eta 0:01:00\r\n
     ----                                     2.2/11.2 MB 6.7 MB/s eta 0:00:50\r\n
     ------------                             6.8/11.2 MB 6.7 MB/s eta 0:00:30\r\n
     ---------------------------              8.2/11.2 MB 6.7 MB/s eta 0:00:20\r\n
     ------------------------------------     10.0/11.2 MB 6.7 MB/s eta 0:00:10\r\n
     ---------------------------------------- 11.2/11.2 MB 6.7 MB/s eta 0:00:00\r\n

Platform
Windows

P.S. Please don't tell me not to run pip install commands this way. I have a specific use case and it will be a waste of time to argue about this. This is a general issue that extends far beyond pip, and I'm sure being able to parse progress bars from scripts is helpful to more people than just me.

[willmcgugan]

Rich will strip anything dynamic when it detects it is not writing to a console.

If you don't want this you can force it to behave as a terminal with force_terminal=True. However, since you don't control that, you should consider using a psuedo-terminal rather than a subprocess.

[joeyballentine]

Thanks for the response. For now, I will use a pseudo-terminal, and I will see if the pip team is open to making force_terminal an option.

[github-actions]

Did I solve your problem?

Why not buy the devs a coffee to say thanks?

[joeyballentine]

@willmcgugan sorry to revive this issue, but is there any chance you'd consider adding the ability to use an environment variable to force this instead of needing to change every codebase that uses rich?

[willmcgugan]

Potentially. See pypa/pip#10909 for background