Support _auto_node_jsondata_hint feature and _node_indexer.

[khoale88]

• auto generate a hint from node json data.
• use node indexer for hint generator when indexing available.

[codspeed-hq]

CodSpeed Performance Report

Merging #82 will not alter performance

_{Comparing khoale88:issue-63 (01c6255) with master (9a12b88)}

Summary

✅ 81 untouched
🆕 16 new

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
🆕	test_no_hints[DiGraph]	N/A	15.6 ms	N/A
🆕	test_no_hints[MultiDiGraph]	N/A	15.7 ms	N/A
🆕	test_node_double_jsondata_hints[DiGraph]	N/A	36.7 ms	N/A
🆕	test_node_double_jsondata_hints[MultiDiGraph]	N/A	36.7 ms	N/A
🆕	test_node_jsondata_and_explicit_hints[DiGraph]	N/A	17 ms	N/A
🆕	test_node_jsondata_and_explicit_hints[MultiDiGraph]	N/A	17 ms	N/A
🆕	test_node_jsondata_hints[DiGraph]	N/A	17 ms	N/A
🆕	test_node_jsondata_hints[MultiDiGraph]	N/A	17.1 ms	N/A
🆕	test_no_hints[DiGraph]	N/A	15.6 ms	N/A
🆕	test_no_hints[MultiDiGraph]	N/A	15.7 ms	N/A
🆕	test_node_double_jsondata_hints[DiGraph]	N/A	35.5 ms	N/A
🆕	test_node_double_jsondata_hints[MultiDiGraph]	N/A	35.5 ms	N/A
🆕	test_node_jsondata_and_explicit_hints[DiGraph]	N/A	17 ms	N/A
🆕	test_node_jsondata_and_explicit_hints[MultiDiGraph]	N/A	17 ms	N/A
🆕	test_node_jsondata_hints[DiGraph]	N/A	17 ms	N/A
🆕	test_node_jsondata_hints[MultiDiGraph]	N/A	17.1 ms	N/A

[j6k4m8]

still reading through this but I love it so far! want to be sure I fully understand Indexers before merging, they seem like an amazing opportunity for some performance improvements to this codebase :)

[khoale88]

Sorry for the lack of documentation. Instead of complex Btree support fastter write, ArrayIndexer will have everything sorted in ascending order, and any searching would be logN. However, it hasn't deal with None value for now.

[khoale88]

I will handle null value with indexter in the #69

[khoale88]

@j6k4m8 I rebase this to the subquery PR, so you see there're a lot. But if you merge the PR and rebase this once more, change will be small.

I changed the implementation a little bit. I don't compute the node hints before hand. Instead, When the match happens, I convert the motif to indexer condition ast and then AND it with the where ast. This is much cleaner.

I also enable node auto hint by default.

I think it's worth to do 3 other indexing feature to complete this one.

• labels indexing. This is used in node label match.
• null attributes indexing. This is to handle null value faster.
• in operators in index queriers.

[j6k4m8]

Once the benchmarks finish, can I merge this in? :)

[khoale88]

Yes, please.

[j6k4m8]

Hooray! :D