feat(spark): Support multi-label in Spark

[yangxk1]

Describe the enhancement requested

Multi-label is currently supported in C++, but not yet implemented in Spark.
Can reference #650 #654

Component(s)

Spark

[unical1988]

@yangxk1 can you say more about the feature, i am willing to take on this pr

[yangxk1]

Hi @unical1988 ! thanks for your interest in this issue!
Multi-label support was proposed in #96. In short: a vertex/edge can have multiple labels. This feature has already been implemented in C++ (test_multi_label.cc), where a bloom_filter is used to enable fast filtering.
You can refer to the issues and PRs mentioned in this issue, as well as the corresponding C++ code.
Looking forward to your contribution!

[unical1988]

Is it supported in java?

[yangxk1]

Is it supported in java?

Not yet, it’s only supported in C++.

[unical1988]

So then i will start a pr for the its java version

[yangxk1]

Does "java" refer to the project under maven-projects/java/?

[unical1988]

Yes

[yangxk1]

This project under maven-projects/java/ is implemented using FastFFI, which tightly couples it with C++ and makes it hard to maintain. You can take a look at the discussion here #704.
We are planning to stop adding new features to this library.
If you're interested, you can contribute to the /info/ module instead, which is a pure Java implementation.

[unical1988]

Ok, shall we sync about that? How to start?

[unical1988]

Are there graph properties or operations yet to implement? Let me know.
@yangxk1 Something like adding two graphs or subtracting them? or obtaining sub-graphs where a constraint is satisfied (a condition on the vertex size or other...)?

[yangxk1]

@Thespica might know what the next steps are.
In my opinion, all the functionality in the C++ core should be implemented in java-info.

[unical1988]

@yangxk1 are you referring to those implemented in cpp/src/graphar/, for example the version_parser or the chunk_info_reader)? could you name a few (from what i see LookupKeyIndex AddVectorElement ConcatEdgeTriple createAdjacentList are among the features that are not implemented) ?

[yangxk1]

Yes! And for example, Parquet/csv file read/write and property filtering.

[unical1988]

OK, I will do a PR on those, can I you tell me what are the maven and java best version to use for the latest graphAr version? On windows only