I suspect there's a good chance it's not getting real console handles in CI systems either... maybe we should also print type(sys.stdout) (or possibly we have to dig into .buffer.raw)?

Tested it on my own repo with:

import sys
print(sys.stdin.encoding, sys.stdout.encoding)
print(sys.stdin.buffer.raw, sys.stdout.buffer.raw)

And the result is:

cp1252 cp1252
<_io.FileIO name='<stdin>' mode='rb' closefd=False> <_io.FileIO name='<stdout>' mode='wb' closefd=False>

Instead of what I've got on my local machine:

utf-8 utf-8
<_io._WindowsConsoleIO mode='rb' closefd=False> <_io._WindowsConsoleIO mode='wb' closefd=False>

So I think there is no real consoles on GHA, and we can't run this case on it. I guess we can just add @requires_resource('console') to skip it on GHA and just run it with other environments?

After reading #114565, I'm not sure whether @requires_resource('console')is the correct approach. It seems intended to be skipped during local development to avoid unnecessary output, and should be enabled in CI?

However, I found that it is not enabled or run in CI because the console resource is not defined in libregrtest (#133519).

If we intend to enable the console resource in CI, should we add another resource like windows-console to allow skipping this test in environments that don't have a Windows console?

Or we can just skip the test partway through (raise unittest.SkipTest or something like that) when we figure out that it's not a real console. If there's only a small handful of cases, then there's no real need to define a resource, and it's a legacy test, not a critical one.