[ms609-agent]

Closes #123.

What

Trees with different leaf sets were compared by pruning each pair with KeepTip() in R — once for the score (.SplitDistanceAllPairs() etc.), again in .MaxValue(), and again in NormalizeInfo() — holding n(n−1) pruned phylo objects in memory at once.

New cpp_pruned_pairs() (in pairwise_distances.cpp) restricts each tree's splits to the leaves it shares with its partner, drops trivial splits, merges splits that coincide after pruning, then scores the pair with the existing batch kernels and reports each pruned tree's information content (clustering entropy / splitwise info / split count). It runs in OpenMP over pairs, with per-thread scratch. Wired into:

• the different-leaves branches of .SplitDistanceAllPairs(), .SplitDistanceManyMany() and .SplitDistanceOneMany() (MCI, IRF, MSD, MSI, SPI, Nye/Jaccard, RF);
• .MaxValue() and NormalizeInfo() when the information function is ClusteringEntropy, SplitwiseInfo or NSplits. NormalizeInfo() also no longer prunes trees when infoInBoth or a numeric how is given.

Other scoring and information functions keep the R path.

Benchmark (150 random trees, 40–70 of 80 leaves, 1 thread, optimised build)

	before	after
TreeDistance(trees)	15.6 s	1.0 s
PhylogeneticInfoDistance(trees, normalize = TRUE)	19.3 s	0.9 s
RobinsonFoulds(trees[1:60], trees[61:150])	2.8 s	0.13 s
peak memory (gc() max, MB)	104	67

Bug fixes found along the way (in NEWS)

1. All-pairs values were wrong when leaf sets differed: KeepTip() on a Splits object keeps that tree's own leaf order, so the two pruned trees' bits could refer to different leaves. Example: MCI 2.60 from all-pairs vs 3.56 from comparing the pruned pair directly. The new code matches the KeepTip()-on-phylo reference for every metric (see test-pruned_pairs.R).
2. RobinsonFoulds(trees) with differing leaf sets went through the ClusterTable path, which ignores the leaf differences, and returned wrong values.
3. f(list1, list2) was transposed when leaf sets differed (rows were tree2), unlike the same-leaves case. Rows now always correspond to tree1. .SharedOnly() now lists N × M pairs in column-major order to match. One existing expectation in test-different-tips.R encoded the transposed shape and is updated. This is a user-visible change.

Notes for review

• SplitList can only be built from an RawMatrix, so each thread builds one before the parallel region, sized to the largest input, and overwrites its public fields for each pair. That depends on the TreeTools SplitList layout (row stride ≥ pruned bin count); a capacity constructor in TreeTools would be tidier but would need a LinkingTo bump.
• Pairs with fewer than four shared leaves score 0 with 0 information, so normalized distances are NaN, which matches the existing 1-vs-1 behaviour.
• I synthesised a test set rather than downloading the reporter's attachment.
• spelling::spell_check_package() errors locally on a Windows UTF-8 issue that already exists in NEWS.md; CI's tests/spelling.R will cover it.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.65%. Comparing base (9f3455f) to head (c3b2464).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #221      +/-   ##
==========================================
+ Coverage   97.57%   97.65%   +0.07%     
==========================================
  Files          60       60              
  Lines        6056     6305     +249     
==========================================
+ Hits         5909     6157     +248     
- Misses        147      148       +1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[github-actions]

Performance benchmark results

Call	Status	Change	Time (ms)
ClusteringInfoDistance(tr280)	⚪ NSD	-1.32%	9.11, 9.23 → 9.39, 9.29
ClusteringInfoDistance(tr50)	⚪ NSD	-0.37%	6.56, 6.68 → 6.66, 6.66
LAPJV(test2000)	⚪ NSD	-1.64%	95.5, 93.1 → 97.9, 95.3
LAPJV(test40)	⚪ NSD	1.99%	0.0181, 0.0181 → 0.0177, 0.0177
LAPJV(test400)	⚪ NSD	0.43%	3.23, 3.23 → 3.21, 3.22
MutualClusteringInfo(tr280)	⚪ NSD	-0.56%	18.7, 19 → 18.6, 19.4
MutualClusteringInfo(tr50)	⚪ NSD	-0.8%	17.4, 17.7 → 17.6, 18.1
NNIMaxStep(2000, normalize = TRUE)	⚪ NSD	-0.01%	0.214, 0.213 → 0.214, 0.214
NNIMaxStep(300, normalize = TRUE)	⚪ NSD	0.06%	0.0316, 0.0316 → 0.0316, 0.0316
NNIMaxStep(300)	⚪ NSD	-0.56%	0.00905, 0.00892 → 0.00904, 0.00901
PathDist(postTrees)	⚪ NSD	0.84%	2.31, 2.29 → 2.29, 2.26
PhylogeneticInfoDistance(tr180)	⚪ NSD	-0.33%	58.1, 58.4 → 58.3, 58.5
PhylogeneticInfoDistance(tr50)	🟣 ~same	-2.73%	43.1, 43.2 → 44.4, 44.3
RobinsonFoulds(tr280)	⚪ NSD	0.47%	4.77, 4.85 → 4.82, 4.76
RobinsonFoulds(tr280)	⚪ NSD	1.24%	4.62, 4.72 → 4.64, 4.58
RobinsonFoulds(tr50)	⚪ NSD	-0.67%	4.58, 4.67 → 4.69, 4.61