fix(model): replace O(N²) prefix sum in checkDuplicateTransactionsInSubtree (regression from #198)#904
Closed
oskarszoon wants to merge 2 commits into
Closed
Conversation
…ubtree (regression from bsv-blockchain#198) The loop summing b.SubtreeSlices[i].Size() for i < subIdx was O(N) per subtree, making the full checkDuplicateTransactions pass O(N²). Each Size() acquires a read lock, so at N=450k subtrees this saturates all cores with lock contention. The loop comment claiming (subIdx * subtreeSize) is invalid was wrong: baseIdx is the sum of subtrees BEFORE subIdx. Per Block.go:1293, every subtree except the last must be full-size, so all subtrees before subIdx are full-size regardless of whether subIdx itself is the smaller last subtree. The O(1) formula is always correct. Adds TestCheckDuplicateTransactions_LargeBlockNotQuadratic (N=20k subtrees, completes in <1s vs tens of seconds with the old code) and TestCheckDuplicateTransactionsInSubtree_BaseIndexCorrectness covering subIdx=0, smaller last subtree, subtreeSize=1, and cross-subtree dup detection.
Contributor
|
🤖 Claude Code Review Status: Complete Review SummaryThis PR correctly fixes a critical O(N²) performance regression in duplicate transaction checking. The change is mathematically sound, well-tested, and follows project conventions. Findings: No issues found. Analysis: The fix replaces an O(N) loop that summed all predecessor subtree sizes with a closed-form O(1) calculation (baseIdx = subIdx * subtreeSize). This is correct because:
The inline comment correctly quotes the invariant and explains the optimization, satisfying the "Philosophy" requirement from AGENTS.md to make tradeoffs explicit. |
icellan
approved these changes
May 19, 2026
|
Contributor
Benchmark Comparison ReportBaseline: Current: Summary
All benchmark results (sec/op)
Threshold: >10% with p < 0.05 | Generated: 2026-05-19 19:07 UTC |
Contributor
Author
|
Closing as duplicate of #910 — Simon's PR lands the same A subset of edge-case subtests from this PR is worth carrying forward — |
5 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #900.
Summary
Block.checkDuplicateTransactionsInSubtreewalked all preceding subtrees on every invocation, callingSubtree.Size()per subtree — an O(N) prefix sum executed undercheckDuplicateTransactionsConcurrencyworkers in parallel, producing O(N²) atomic RWMutex traffic. On a 450k-subtree block this saturated 11.6 cores inSize()lock churn with effectively zero useful work.Regression introduced in
58a68d376(PR #198).Fix
Replace the O(N) loop with
baseIdx := subIdx * subtreeSize. The "all subtrees need to be the same size as the first tree, except the last one" invariant (model/Block.go:1289) means every subtree beforesubIdxis full-size — the smaller-last-subtree never appears in the sum. Closed-form computation is O(1) per call.Comment quotes the invariant inline so a future drift in the invariant's location doesn't decouple from the formula.
Test plan
TestCheckDuplicateTransactions_LargeBlockNotQuadratic— synthetic 20k-subtree block; verifies wall-time well under a 30s budget. (Local: ~18ms.)TestCheckDuplicateTransactionsInSubtree_BaseIndexCorrectness— 4 subtests coveringsubIdx=0, smaller-last-subtree,subtreeSize=1degenerate, cross-subtree duplicate detection.go test -race ./model/...clean.make lint-newclean.