Bump click minimum version to 8.3.1

[stratakis]

Click 8.1.8 through 8.3.0 have broken pager invocation for multi-argument PAGER values, which causes behave test failures.

Fixed on pallets/click@7db1f20

• I've added my name to the AUTHORS file (or it's already there).

[meeuw]

Maybe try this?

Suggested change

	"click >= 8.3.1",
	"click >= 4.1,<8.3.2",

[stratakis]

This will "allow" some configs with broken click versions to be used (8.1.8 to 8.3.0). I could exclude all the problematic ranges, which would be ugly to put in pyproject toml but doable. I could also explore fixing the compatibility issue in pgcli.

Would something like click >= 4.1, != 8.1.8, != 8.2.*, != 8.3.0, < 9 work? Noting that the next click version would be 9, not 8.3.2

[meeuw]

Sounds great! Please commit the following so we can try.

Suggested change

	"click >= 8.3.1",
	# Click 8.1.8 through 8.3.0 have broken pager invocation for multi-argument PAGER values, which causes behave test failures.
	"click >= 4.1, != 8.1.8, != 8.2.*, != 8.3.0, < 9",

[meeuw]

@stratakis thanks for your PR, I approved the CI tests.

For anyone reading this, rationale can be found here:

Required for click's rebase to 8.3.1.

https://src.fedoraproject.org/rpms/pgcli/pull-request/14

[DiegoDAF]

Confirmed working. We've been running with click 8.3.1 on our fork with the full test suite passing (2759 tests).

The old <8.1.8 upper bound was a workaround for the broken pager in 8.1.8-8.2.x. This is the proper fix.

[j-bennet]

Looks like this fails in the 3.10 build.

[stratakis]

Looks like this fails in the 3.10 build.

This is due to click 8.3.1 requiring python >= 3.10, as uv checks that in the CI run. In essence pgcli should move to requiring python >=3.10.

[meeuw]

Python 3.10 hasn't reached EOL:
https://devguide.python.org/versions/

Maybe remove support end of this year?

[stratakis]

Python 3.10 hasn't reached EOL: https://devguide.python.org/versions/

Maybe remove support end of this year?

The build that is failing is not the 3.10 one but uv checking the whole dependency matrix, including Python 3.9. So Python 3.10 should be fine. Let's see with the latest changes with the exclusion of the bad click versions it should hopefully not use the incompatible 3.9.

[stratakis]

And as I'm not familiar with pgcli's codebase I fed the relevant data into claude to see what it can come out with. Gave it pgcli's codebase including my latest commit, the integrations tests issue, click's codebase and asked for approaches for making it compatible with lower click versions as well. Maybe it's helpful, maybe not:

LLM output

Problem: Click 8.1.8 through 8.3.0 have a bug in _pipepager() where multi-argument PAGER strings are
not handled correctly. In 8.1.8, which() is called on the entire PAGER string instead of just the
command name. In 8.2.0-8.3.0, shlex.split() was added but Popen is called with shell=True + a list
argument, so only the first element is executed. Fixed in 8.3.1 with shell=False.

The behave tests set PAGER to python3 /path/to/wrappager.py ---boundary---, which triggers this bug.
The pager silently falls through to less, boundary markers are never emitted, and all pager-dependent
tests fail.

Approaches to make pgcli work with all click versions:

1. Make wrappager.py a standalone executable with no arguments. Hardcode the boundary string (it's
   always ---boundary---), rely on the existing shebang, chmod +x, and set PAGER to just the absolute path
   of the script. A single-path PAGER works with all click versions. Smallest change.
2. Pass the boundary via environment variable instead of argv. Set PAGER_BOUNDARY in environment.py,
   read it with os.environ in wrappager.py, and set PAGER to just the script path. Same single-path
   benefit, but keeps the boundary configurable from the test harness.
3. Use the exclusion-based version constraint. "click >= 4.1, != 8.1.8, != 8.2.*, != 8.3.0" allows
   older click versions (which work) and 8.3.1+ (which is fixed), while excluding the broken range. No
   code changes needed but doesn't protect users who have PAGER="less -R" or similar multi-argument
   values.
4. Bypass click.echo_via_pager in pgcli. Replace the click.echo_via_pager(text, color) call in output()
   with pgcli's own pager invocation using subprocess.Popen(shlex.split(...), shell=False). Fixes the
   issue for real users too (not just tests), but is a larger change.
5. Monkey-patch click.echo_via_pager in the test harness. Override it in environment.py's before_all()
   with a corrected implementation. Contained to tests, but ugly.

Approach 1 or 2 combined with approach 3 covers both the test suite and real-world usage with minimal
changes.