Update of "plugin meta data"
Artifact ID: | 478ea535493535266c5011e51f44a5cfe3298776 |
---|---|
Page Name: | plugin meta data |
Date: | 2015-05-04 21:33:33 |
Original User: | mario |
Mimetype: | text/x-markdown |
Parent: | cb4b8fe350077e3472fc349c56616937c6f3f858 (diff) |
Next | 63f0126ea8aed90fa07945efd4232a1f6d12b78d |
Plugin meta spec
Streamtuner2 now uses a proper plugin description scheme. Note that this is completely unrelated to the streamtuner2 or channels API, or how plugins are actually initialized (pkgutil, imp, implib, pluginbase/plugnplay).
Basically the top-level comment block is scanned on initialization. It simply lists a few key: values for documentation and technical descriptors. It's basically a subset of YAML embedded in a script comment.
# api: streamtuner2
# type: feature
# title: My Plugin
# description: Lists all the things
# version: 1.0
# category: music
# config: { name: myconf, value: 1, type: boolean, description: enable foo }
#
# And here goes a longer doc/comment.
The key names are case-insensitive of course. A few fields like title:, description: are required. Theoretically also the api: and type:. Though in ST2 they're currently just used for decorative purposes like the category: or version: field.
More interestingly the config: field can be used to describe plugin options. Default values will automatically end up in streamtuners conf.*
dict, and are presented to users in the config dialog.
- A value: represents the default.
- The type: select can be used for dropdown boxes, with a
select:
key=VAl|key2=VAL2.. attribute for options. - Other type: values are interpreted as strings, except type: boolean which becomes a checkbox.
- A per-option category: attribute is ignored at the moment.
- Params can be optionally quoted as in value: "Desc, desc, desc." in case they contain commas (which otherwise separate option attrs.)
And of course, the config: field can list multiple options. Simply make it multi-line (every plugin meta field can be) by indenting wrapped lines.
Why?
This scheme is designed to be language-agnostic. It originated in PHP way back (see libconfig), precisely to avoid a few common misdesigns.
Meta data does not belong in code.
Because that incurs always importing them, even when they may go uninitialized/unused.And it certainly should not reside in externalized ini/json data blobs.
(Or do you want ants? Because that's how you get ants.)Furthermore it encourages a useful documentation block atop.
(Instead of license blobs everywhere in lieu of proper comments.)
The meta description block incidentally can also serve as packaging control scheme. The pack: descriptor is used by fpm/xpm -s src. Which avoids manually doubling descriptions into separated setup.py/.cfg as the current Python packaging eco system does.
Note though that the plugin description scheme is primarily intended for in-application feature/plugin management. Reducing packaging effort is just a by-effect.
argparse config: struct
Similarly to internal plugin configuration settings, the config: list is now used for ArgumentParser definitions. Fields are slightly remapped to remain in line with regular options.
# config:
# { arg: -D, name: debug, type: bool, description: enable debugging }
# { arg: -e, name: enable[], type: str*, description: add plugin }
It's basically just the extra arg: field which discerns such entries from standard options. And the type: may carry an optional quantifier (mapped to argparses' nargs= flag). Or the name: can indicate a list property with []
attached (maps to argparses' append instead of store). Not every argparse peculiarity is provided for though. And using a docopt section would probably be a better option anyway.
Plugin list
The purpose of this scheme is feeding the plugin management in the settings window of course:
And this may now even become a Python package of its own, http://pypi.python.org/pypi/pluginconf/. Not quite