Error: 'GroupPlugin' object has no attribute 'is_serve' when deploying with mike

[shamoon]

Hi there, really confused as to why I'm getting the above error when trying to deploy via ci. We were previously using mike 1.1.2 and had no issues but have to upgrade to 2.0 for reasons. I realize this isnt the mike repository but I think the solution is here (or LMK if Im wrong about that).

Now, when executing
run: MKINSIDERS=true mike deploy --update --push ${{github.ref_name}} we hit the above error,

error: 'GroupPlugin' object has no attribute 'is_serve'
Error: Process completed with exit code 1.

mkdocs.yml looks like:

...
plugins:
  - group:
      enabled: !ENV MKINSIDERS
      plugins:
        - optimize
        - typeset
        - social

Appreciate any help!

[kamilkrzyskow]

Is this the whole traceback of the error @shamoon? If possible I would like to see the full output.

[kamilkrzyskow]

Perhaps mike swallows the output to make it more "user friendly".
Please add the --debug flag to the command @shamoon
I see it was added in 2.0.0
https://github.com/jimporter/mike/blob/master/CHANGES.md

If that doesn't work I will debug it a bit later myself.

[kamilkrzyskow]

I managed to reproduce the error message locally

Log Output

$ mike --debug serve
Traceback (most recent call last):
  File "<frozen runpy>", line 198, in _run_module_as_main
  File "<frozen runpy>", line 88, in _run_code
  File "C:\MyFiles\_git\homepage-help\venv\Scripts\mike.exe\__main__.py", line 7, in <module>
  File "C:\MyFiles\_git\homepage-help\venv\Lib\site-packages\mike\driver.py", line 460, in main
    return args.func(parser, args)
           ^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\MyFiles\_git\homepage-help\venv\Lib\site-packages\mike\driver.py", line 310, in serve
    load_mkdocs_config(args)
  File "C:\MyFiles\_git\homepage-help\venv\Lib\site-packages\mike\driver.py", line 142, in load_mkdocs_config
    cfg = mkdocs_utils.load_config(args.config_file)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\MyFiles\_git\homepage-help\venv\Lib\site-packages\mike\mkdocs_utils.py", line 33, in load_config
    return cfg['plugins'].run_event('config', cfg)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\MyFiles\_git\homepage-help\venv\Lib\site-packages\mkdocs\plugins.py", line 507, in run_event
    result = method(item, **kwargs)
             ^^^^^^^^^^^^^^^^^^^^^^
  File "C:\MyFiles\_git\homepage-help\venv\Lib\site-packages\material\plugins\group\plugin.py", line 85, in on_config
    command = "serve" if self.is_serve else "build"
                         ^^^^^^^^^^^^^
AttributeError: 'GroupPlugin' object has no attribute 'is_serve'

There are several things contributing to this error. However, I think that mike does a nasty shortcut, which should be addressed from their end, and mkdocs-material does another shortcut that should be addressed here as well 😄.

Firstly mike, when loading the MkDocs config it also triggers the on_config event for the plugins, it skips the on_startup event.
https://github.com/jimporter/mike/blob/533e996066579124a8c406c4d70e54fb78ac8dd6/mike/driver.py#L136-L143
https://github.com/jimporter/mike/blob/533e996066579124a8c406c4d70e54fb78ac8dd6/mike/mkdocs_utils.py#L30-L33

Secondly mkdocs-material, it sets the self.is_serve attribute in the on_startup event (which runs before on_config) and then checks the value in on_config expecting the value to be set in the previous event.

mkdocs-material/src/plugins/group/plugin.py

Lines 41 to 44 in 6e0b497

	# Determine whether we're serving the site
	def on_startup(self, *, command, dirty):
	self.is_serve = command == "serve"
	self.is_dirty = dirty

This event is omitted by mike, as written above. The issue here is that on_startup is required to set the self.is_serve attribute as it's the only event having access to this information. However this isn't the recommended way to set variables overall, as https://www.mkdocs.org/dev-guide/plugins/#on_startup describes that they should be set in __init__ and/or on_config.

So it's hard to pinpoint who is responsible for the fix here, probably both parties. I think there is a PEP code style guide for Python that recommends that every class object attribute should be set in __init__ even if that value is just None.
I also didn't dive deep to check if the on_config event is triggered anywhere else after mike loads the MkDocs config, but mkdocs-material could also safeguard against the on_config run when self.is_serve is None (from __init__), because on_serve didn't run.

[shamoon]

Ha youre faster than me! Really appreciate the deep dive, Im definitely not familiar-enough with the codebase to have much to add much less open a PR at either repo. So please let me know how best to proceed, I'd be happy to open an issue in the mike repo (and/or here), or I can hang tight.

Thanks again

[kamilkrzyskow]

I don't really understand why mike triggers the on_config event for every plugin during load, probably mike assumes that every plugin makes modifications in on_config "as per the standard" and wants to manage a "finished" config object 🤔
Therefore, I don't really know the best way forward when it comes to mike, but, ensuring that on_startup runs before on_config would be a good start :P You could create an issue upstream about it @shamoon ✌️

As for the PR here in mkdocs-material, I can make one later today if squidfunk won't be faster. Most plugins don't use the on_startup event, and when they use it it's typically to track if MkDocs is served, at least I think so. Therefore, it is safe to assume a default state of build mode, and that's what I will implement.

EDIT:
I just checked that both of the Insiders social and optimize plugins already assume is_serve as False in their __init__:
https://github.com/squidfunk/mkdocs-material-insiders/blob/afbb2303e76b10fd2a2b8652a0251cd3eaa87701/src/plugins/social/plugin.py#L79-L83
https://github.com/squidfunk/mkdocs-material-insiders/blob/afbb2303e76b10fd2a2b8652a0251cd3eaa87701/src/plugins/optimize/plugin.py#L60-L64

So it seems that only the group plugin got unlucky and didn't get the same treatment 😅

[squidfunk]

Thanks for investigating! This is a bug on mike's end and a bug on our end. The bug on our end can be fixed as in the optimize and social plugin, as you discovered, @kamilkrzyskow – PR appreciated!

[shamoon]

Thanks so much. For reference: jimporter/mike#193

[squidfunk]

The issue was resolved in 55fe1cc, which is already in master, thanks for @kamilkrzyskow.

[shamoon]

Thank you to you both! Do these changes get synced to insiders occasionally?