Retain literal types in collection backend

[jesper-friis]

Description:

Closes #160

Also added fallback_backend option to Triplestore.parse() and Triplestore.serialize() to allow calling parse() and serialize() with the collection backend (using rdflib under the hood).

Note that this PR builds on top of PR #161

Note also that this PR utilises DLite PR SINTEF/dlite#755. It is probably a good idea to merge that PR and create a new release of DLite before merging this PR to master.

Type of change:

• Bug fix and code cleanup.
• New feature.
• Documentation update.
• Testing.

Checklist for the reviewer:

This checklist should be used as a help for the reviewer.

• Is the change limited to one issue?
• Does this PR close the issue?
• Is the code easy to read and understand?
• Do all new feature have an accompanying new test?
• Has the documentation been updated as necessary?

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (c8e2fbd) 75.94% compared to head (4f532a6) 76.82%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #165      +/-   ##
==========================================
+ Coverage   75.94%   76.82%   +0.87%     
==========================================
  Files          17       17              
  Lines        1534     1536       +2     
==========================================
+ Hits         1165     1180      +15     
+ Misses        369      356      -13     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[francescalb]

I do not see where these are tested. According to codecov, they are, so currently it should be OK. But it is important that this is tested on purpose and not "by accident", e.g. by using a backend that right now does not contain parse but might in the future.

[jesper-friis]

They are tested in tests/backends/test_collection.py

[francescalb]

OK, I prefer tests to be a bit more explicit.

[francescalb]

Same comment as for parse above

[jesper-friis]

Tested in tests/backends/test_collection.py

[francescalb]

Suggested change

	# Possible bug in mypy
	# parse_object() is declared to return "Union[str, Literal]".
	# Dispite of that complains mypy about that `v` has no
	# attribute "value"
	obj = v if isinstance(v, str) else str(v.value) # type: ignore
	obj = v if isinstance(v, str) else str(v.value)

I have thested locally and with a draft pull request, I do not get any errors with mypy. Do you have an updated version?

[jesper-friis]

The problem came when I updated mypy. I got annoyed with it and changed the return type of utils.parse_object() from Union[str, Literal] to Union[str, Any] to get rid of these errors...

The source of the problem is that mypy suddenly has got the idea that Literal has got a special meaning, since it was added to typing.

It even doesn't understand if I change the return type of utils.parse_object() to Union[str, tripper.Literal].

Using Any in the return works around the problem. I have moved and updated the comment about buggy mypy to parse_object() and removed the # type: ignore as you suggested.

[francescalb]

Since we didn't need to change backend/backend_kwargs to fallback_backend/fallback_backend_kwargs in the tests, it seems that we do not have adeguate tests.

[jesper-friis]

Well, the collection backend does not provide a parse() method, so it will use the fallback backend. That is tested when calling ts.parse() in tests/backends/test_collection.py.

[francescalb]

Seems OK, although I do not understand the mypy issue.