Use latest pluggy

[goodboy]

Latest version is now 0.5.1.

The list of benefits includes:

• speedier _MultiCall loop using for instead of while
• warnings for the deprecation of __multicall__
• warnings for hook impls which mismatch specs
• more comprehensible error messages on incorrect usage
• important fix for callbacks where historic hooks would not be called for already registered plugins

[nicoddemus]

Took a quick stab at this but it seems there's a problem (or misinterpretation on my part) on how firstresult works:

Traceback (most recent call last):
  File "C:\pytest\.env27\Scripts\pytest-script.py", line 11, in <module>
    load_entry_point('pytest', 'console_scripts', 'pytest')()
  File "c:\pytest\_pytest\config.py", line 50, in main
    config = _prepareconfig(args, plugins)
  File "c:\pytest\_pytest\config.py", line 160, in _prepareconfig
    pluginmanager=pluginmanager, args=args)
  File "c:\pytest\.env27\lib\site-packages\pluggy\__init__.py", line 680, in __call__
    return self._hookexec(self, self._nonwrappers + self._wrappers, kwargs)
  File "c:\pytest\.env27\lib\site-packages\pluggy\__init__.py", line 240, in _hookexec
    return self._inner_hookexec(hook, methods, kwargs)
  File "c:\pytest\.env27\lib\site-packages\pluggy\__init__.py", line 234, in <lambda>
    methods, kwargs, specopts=hook.spec_opts, hook=hook
  File "c:\pytest\.env27\lib\site-packages\pluggy\callers.py", line 104, in execute
    gen.send(outcome)
  File "c:\pytest\_pytest\helpconfig.py", line 70, in pytest_cmdline_parse
    if config.option.debug:
AttributeError: 'list' object has no attribute 'option'

Here's the code for pytest_cmdline_parse() impl:

pytest/_pytest/helpconfig.py

Lines 65 to 70 in 0824076

	@pytest.hookimpl(hookwrapper=True)
	def pytest_cmdline_parse():
	outcome = yield
	config = outcome.get_result()
	if config.option.debug:
	path = os.path.abspath("pytestdebug.log")

outcome.get_result() is returning a list, while we expected it to return a single Config object.

That's using pluggy-0.5.1.

[RonnyPfannschmidt]

@tgoodlet yay - this means we missed and important core detail of pluggy that @hpk42 never had a correct test for ^^ and 0.5.1 is not actually good, just not completely broken

[goodboy]

@nicoddemus hah! Nice catch.
This firstresult stuff seems to be quite subtle (and untested) hehe.

I think the difference here is that the way the original _LegacyMultiCall terminates its while loop when firstresult is used. The recursion unwinds returning a single value in a very sneaky way.

I can see 2 ways to handle this:

• make _Result aware of the firstresult notion
• handle and reset the outcome based on the firstresult condition before sending to hookwrappers

I personally vote for the latter.

I'll suggest a test and fix first tomorrow later this morning.
It's a good thing we're catching this stuff now so we can add tests for it 👍

[goodboy]

@RonnyPfannschmidt yes you're totally right.
I guess it's not all bad because at least we're figuring out the tests we need.
This is why I was partial a while back to running pytest unit tests as part of pluggy CI.

Either way sounds like we'll have 2 patch releases in 2 days...yey

[RonnyPfannschmidt]

@tgoodlet well its not too bad because there is no pytest release yet that we did break, but its a pain that we cause tox/devpi users trouble

[goodboy]

@RonnyPfannschmidt yeah definitely not great.
At least it seems like none of those projects are using firstresult hook wrappers.

Alright looking at a fix.

[goodboy]

@RonnyPfannschmidt @nicoddemus please see pytest-dev/pluggy#72

Sorry about the lack of thoroughness with this. I should really at least run the pytest suite manually before making a pluggy release.

Would you guys be ok if I did run the latest version of pytest as part of the CI job from now on?

[RonnyPfannschmidt]

@tgoodlet we also need a job on pytest to test regulary with latest pluggy and in future attrs as well

[goodboy]

@RonnyPfannschmidt sounds like a good plan.