fix(merge-tree sequence): rb tree crash

[connorskees]

Resolves a crash in our red-black tree implementation that occurs when multiple references slide off the tree at once.

ADO#4477

[msfluid-bot]

⯆ @fluid-example/bundle-size-tests: -723 Bytes

Metric Name	Baseline Size	Compare Size	Size Diff
aqueduct.js	442.37 KB	442.13 KB	⯆ -241 Bytes
connectionState.js	680 Bytes	680 Bytes	■ No change
containerRuntime.js	238.34 KB	238.34 KB	■ No change
loader.js	178.18 KB	178.18 KB	■ No change
map.js	48.05 KB	48.05 KB	■ No change
matrix.js	139.12 KB	138.89 KB	⯆ -241 Bytes
odspDriver.js	90.2 KB	90.2 KB	■ No change
odspPrefetchSnapshot.js	41.75 KB	41.75 KB	■ No change
sharedString.js	159.86 KB	159.62 KB	⯆ -241 Bytes
sharedTree2.js	266.29 KB	266.29 KB	■ No change
Total Size	1.71 MB	1.71 MB	⯆ -723 Bytes

---

Baseline commit: 1031b62

Generated by 🚫 dangerJS against aaea07a

[Abe27342]

FWIW I think the work item here wasn't talking about op reentrancy. I interpreted it as issues with LocalReferencePosition.link and LocalReferenceCollection.removeLocalRef mutually calling each other, leading to issues.

Not really against these tests though, I guess they do add some coverage for actions taken as part of a sequence delta callback, which is a reasonable thing applications might do. I think we have some coverage of this type of thing at the merge-tree level but not really at the sequence level.

[Abe27342]

this looks good to me. It'd be good to have Tony look at it before merging as he might recall some more of the subtleties with why we didn't originally link to undefined.

[Abe27342]

this looks good to me. It'd be good to have Tony look at it before merging as he might recall some more of the subtleties with why we didn't originally link to undefined.