Ha Ha Hack-A-Thon! 🐱‍💻

[led02]

No description provided.

[led02]

Okay, all new data model here.

• Plugins have been updated to use at least some common base. I.e., all are classes that derive from a common base class. The interface is still sparse but provides access to the all new provenance enhancements.
• Data is now expected to be expanded JSON-LD. However, there are nice wrappers in hermes.model.types that allow handling them like compact JSON-LD. Specialized derivatives for handling provenance data exist in hermes.model.prov.
• For merging data, there is an implementation of the JSON-LD containers in hermes.model.merge that can be configured using strategies on how to merge certain data.
• All those functionality depends on the pyld library for which some additional tools were built in.
  Especially the hermes.model.types.pyld_util.BundledLoader is from interest as it loads schemas from the package and also supports .ttl files with the help of RDLib.
• The caches are re-organized. The directory structure stays the same (.hermes//) but the files are different.
  • 'expanded.json' contains the expanded JSON-LD data. This is used for internal processing.
  • 'context.json' contains the JSON-LD context that should be used to get back to the compact form.
  • 'codemeta.json' contains the compact JSON-LD representation (which should be somewhat CodeMeta compatible).
  • 'prov.json' exists in a separate result directory and contains the PROV-O data as an compact JSON-LD graph.
• The cache is maintained by hermes.model.context_manager.HermesContext which is an awful name for a cache manager.

I think that's the most important stuff for now.

Of course, I did not do any proper testing (except for the obvious dog-feeding) and the documentation is rather sparse.
Some spots are still ugly and might need a good refactoring (e.g., provenance integration) but I tried to clean up the mess as good as possible.

Also, the code needs to be annotated so that REUSE is happy.

[led02]

Tests are still failing but I prefer not to deactivate them (yet) but hope for someone (maybe me) to fix them.

After a chat with @sdruskat I would propose to introduce this change step by step:

1. Import the new data model (i.e., hermes.model.types, hermes.model.merge and hermes.model.context_manager). This is also a good opportunity to rename some of the stuff. ;)
2. Adapt the plugin implementations to use the new data model. This requires adapting the command classes and the plugins itself. The implementations from this branch can be used as a template. However, I would not advise to directly import the implementations because the provenance recording is still very sloppy and might need a straight re-engineering.
3. Get rid of legacy stuff.
4. Having the plugins ported, decisions need to be made:

• The plugin class model is slightly improved in this branch. However, as the plugin model is meant to be refactored, this can be skipped.
• A fairly well tampered (little over-engineered) implementation for recording provenance is available in hermes.model.prov. This can be fitted to the plugins spitting valuable provenance information.
• Test should be adapted and added .... massively!

Happy to discuss this with you.