perf(ymake): pre-size TUpdIter::Nodes hash map — Δ=-0.26s (-1.9%), p=0.0029, r=0.504

[KorsarOfficial]

Pre-size TUpdIter::Nodes hash map for graph traversal

Summary

Call reserve(graph.Size()) on the TUpdIter::Nodes hash map in the
TUpdIter constructor to pre-allocate capacity before DFS graph traversal.
This eliminates incremental rehashing as nodes are discovered, reducing
configure graph time by ~1.9%.

8-line change in devtools/ymake/add_iter.h.

Evidence

Warm configure graph benchmark (single ymake invocation, n=15 each):

Metric	Before	After	Delta
Mean	13.587s	13.327s	-0.260s (-1.91%)
Median	13.480s	13.227s	-0.253s
Std	0.359s	0.377s
95% CI	[13.451s, 13.822s]	[13.202s, 13.614s]

Statistical validation:

• Mann-Whitney U: U1 = 179, p = 0.0029 (one-sided), significant at α = 0.05
• Effect size r = 0.504 (large by Cohen's convention)
• Cohen's d = 0.705
• BCa 95% CI for mean difference: [-0.009s, +0.493s]
  (CI straddles zero due to n=15 sample size, but MWU confirms directional
  effect; confidence level: MEDIUM)

Note: Benchmark measures a single-config ymake run (~13.5s), not the full
two-config run (tools-pic + main, ~60.6s). Percentage improvement applies
identically to each invocation.

Amdahl contribution (combined with Phase 17):
S_(17+19) = 1.0353, contributing to S_cold = 1.157 total cold-build speedup.

Supplementary: data/19-statistical-validation.json, data/19-benchmark-before.json,
data/19-benchmark-after.json.

Changes

devtools/ymake/add_iter.h
  - Add Nodes.reserve(graph.Size()) in TUpdIter constructor
  - 8-line change; no behavioral modification

Patch

patches/19-combined.patch

This patch includes three files:

• devtools/ymake/add_iter.h -- the graph optimization (this PR's change)
• devtools/ymake/asio_extern_templates.cpp -- correctness fix for Phase 17
• devtools/ymake/ymake_async.h -- companion to asio fix

The asio portion of the patch requires Phase 17 (patches/17-combined.patch)
to be pre-applied; asio_extern_templates.cpp must exist before this patch
can modify it. The asio fix itself corrects a discovery made during Phase 19
work: strand<any_io_executor> cannot be explicitly instantiated under the
Unified Executors TS (not Networking TS). It is included here to keep the
fix co-located with its discovery context.

To apply only the graph optimization without Phase 17:

git apply --include='devtools/ymake/add_iter.h' patches/19-combined.patch

Testing

# Build only (header change in ymake)
ya make devtools/ymake/bin

Build success verifies correctness. reserve() does not change hash map
semantics -- only pre-allocates capacity. Verify identical build output:

ya dump build-plan <target> | jq -S . > after.json
diff <(jq -S . baseline.json) after.json  # expected: empty

See upstream/test-results.log for test execution status and environmental
constraints. Historical validation: patch applied in yatool Docker container
during Phase 19 implementation; build-plan diff confirmed empty.

CLA

I hereby agree to the terms of the CLA available at:
https://yandex.ru/legal/cla/?lang=en

[KorsarOfficial]

Evidence Report

Full statistical analysis for this optimization:
https://github.com/KorsarOfficial/yatool/releases/download/v1.0-perf-analysis/07-optimization-evidence.pdf — Section 6: Graph Construction (Δ=-0.26s, -1.9%, p=0.0029)

Also available:

• 03-graph-analysis.pdf — Build graph analysis (|V|=4053, strace profiling)
• findings.pdf — Top 5 bottlenecks summary

All reports: https://github.com/KorsarOfficial/yatool/releases/tag/v1.0-perf-analysis