GH-4950 LMDB: extensible ID scheme

[kenwenzel]

Implement an extensible ID scheme for the LmdbStore that also allows to embed values into IDs.

GitHub issue resolved: #4950

Briefly describe the changes proposed in this PR:

Use a binary encoding for IDs that allows to represent the value type, if the value itself is embedded within the id and the embedded value itself.

---

PR Author Checklist (see the contributor guidelines for more details):

• my pull request is self-contained
• I've added tests for the changes I made
• I've applied code formatting (you can use mvn process-resources to format from the command line)
• I've squashed my commits where necessary
• every commit message starts with the issue number (GH-xxxx) followed by a meaningful description of the change

[hmottestad]

The failing integration tests: It's a test that is marked as disabled, but has somehow started to run again. Not sure how that happened.

[kenwenzel]

@hmottestad It would be good to have this in 5.0.0 as it is a breaking change but I am not sure if I can finish the whole logic.

[hmottestad]

Sorry. I've already published the M3 milestone build and we now have a feature freeze until the final release. The LMDB Store is experimental, so you should be able to make breaking changes for 5.1.0. Might be better too, since then users can upgrade to 5.0.0 and get all the fixes and enhancements without any major breaking changes.

[hmottestad]

PS: The develop branch is now 5.1.0 so you can merge into the develop branch without affecting the 5.0.0 release.

[nguyenm100]

rdf 1.2 is largely done and jena has an implementation. apache/jena#2805

might be a good idea to brush the dust off of this pr. how much is left to do @kenwenzel ?

[kenwenzel]

@nguyenm100 Currently, only the ID scheme in the value store is implemented.
What is missing:

• embedding of (numeric) values into IDs
• handling of triple terms
• indexes for triple terms as part of the triple store

@hmottestad Do we already have an issue for RDF 1.2 and/or can we just adapt the current RDF-star support?

[nguyenm100]

Hey @kenwenzel is it possible to just do (1) embedding of values into IDs and merge this pr and then (2) and (3) with rdf 1.2 implementation? or is there some dependency from 1.2 for this to be usable?

[kenwenzel]

@nguyenm100 Yes, that's perfectly possible.

[nguyenm100]

Thanks @kenwenzel are you able to finish (1) and merge this pr? If so, and at that point does it make sense to pull the experimental designation from lmdbstore? (or does that force a major version upgrade?)

[kenwenzel]

@nguyenm100 (1) is a bit of work - I think at least 2 or 3 days, but there is no direct dependency to RDF 1.2.

[kenwenzel]

@odysa @nguyenm100 @hmottestad Do we need to make this backwards compatible with the current ID scheme or do you think a breaking change is OK?

[nguyenm100]

Hi Ken, what's the loe for backward compat? we have one client keeping a rolling weekly snapshot that I would need to chat with.. alternatively, we can introduce it an lmdb2 sail (is that what tdb did?)

[kenwenzel]

@nguyenm100 It should be possible by extracting some constants and considering them in ValueStore and TripleStore. It might have a small performance impact.

[nguyenm100]

Hey Ken, all good on our side. changes don't need to be backward compat for us. tx

[kenwenzel]

@hmottestad Can I merge this?