Add entriesTail method to RBTree

[jzxchiang1]

Inspired by Java's TreeMap#tailMap method but this one automatically gives you an iterator.

The use case is cursor pagination through a long list.

I've been trying to implement cursor pagination in the IC. Think a long list of items that a client needs to fetch, e.g. comments on a post, stories on a feed, etc. The client should fetch a small number of items at a time, say 20, in order to minimize server load as well as meet IC message requirements (<2 MB for requests, I think).

Offset pagination is one option, but it's not really viable at all for real-time, dynamic data. If items get removed or added while the user is scrolling through a partial list, it can result in skipped or duplicate data rendered on the client. At least there isn't really a performance hit on the IC, since it can be implemented with array indexing thanks to orthogonal persistence.

As a result, cursor pagination is generally preferred. In web2 land, this is generally done with a SQL query WHERE id > ... LIMIT .... This requires id be sorted in some database index. On the IC, the only data structure that can emulate a sorted map like this is RBTree.

So in this PR, I add a new method to the RBTree.RBTree class that allows a user to pass in some cursor (i.e. x) to continue iterating where they left off, crucially in ~O(log n) time.

[jzxchiang1]

@matthewhammer might be a good candidate to review, given that he wrote the RBTree.mo code. If someone could approve the workflow run (for tests), that would also be great. Thanks!

[jzxchiang1]

Can you please run it again?

[ggreif]

My (minor) concern is that having such a method will add the need to implement similar methods for other container also. Not necessarily a bad thing, but a library designer should make a statement about this.

[ggreif]

nit: semicolons before closing braces are redundant.

[jzxchiang1]

My (minor) concern is that having such a method will add the need to implement similar methods for other container also. Not necessarily a bad thing, but a library designer should make a statement about this.

Fair enough, but I think this method is somewhat unique to RBTree, since it's the only ordered container available in motoko-base right now.

[ghost]

Dear @jzxchiang1,

In order to potentially merge your code in this open-source repository and therefore proceed with your contribution, we need to have your approval on DFINITY's CLA¹.

If you decide to agree with it, please visit this issue and read the instructions there.

— The DFINITY Foundation

Footnotes

1. Contributor License Agreement ↩