[BUG] Rich is incompatible with pytest capsys fixture

Describe the bug Adopters of rich will eventually discover that their use of capsys in pytest unittest gets broken when they integrate the library.

While this is somehow expected we should document a recipe for adapting testing in such way that does not break when using rich.

Example test:

import pytest

def test_output(capsys):
 	foo() # expected to produce stdout content
	stdout, stderr = capsys.readouterr()
	assert stdout == "..."
	assert stderr == "..."

This has repercussions even for logging functions and I even encountered errors when parametrizing tests like ValueError: I/O operation on closed file. which are caused by Console() instance being closed after the first test.

As rich itself makes use of pytest it should not be a big deal to include an example.

[kat25230]

I’m having the same issue, I used click.echo and it worked, but now I tried using rich and when I’m looking at the: capsys.readouterr() I’m seeing partial log was captured, that is some of the output is missing. Any ideas what’s causing this? (I’m using rich version 11.1.0, with python=3.7.12).

[unparalleled-js]

I am experiencing this problem only with rich Tree output. Rich tables and other output is testable in capsys, but the tree doesn’t appear there. I can see the tree happening, it just isn’t captured.

Edit: I have tried using the example above to custom echo the tree. I have always tried the rich-click plugin. Edit 2: Maybe this is a different problem because even with -s, the tree doesn’t appear Edit 3: Found one solution: send the output to a separate file and then read the file and make assertions there. Also note you have to close the file first for this to work

EDIT 4: By explicitly setting the file to sys.stdout, everything worked 😃

Edit 5: Okay, clarification: I was setting the file to sys.stdout in the lib that made the call. HOWEVER, it only works when setting the file to sys.stdout in the actual test… This seems like a bug!

Edit 6: Ok, resorting back to edit 3 because when using sys.stdout, most of the output is lost wherever there is markup characters, so it is not accurate. However when using a separate file, I can see all the output.