solver: add optional component-area pruning #28

2026-07-31T08:41:01+01:00

mcp commented

2026-07-31 08:41:01 +01:00

Summary

detect skyline empty components separated by full-height columns
add sound gcd and bounded subset-sum area feasibility modes
compare boundary-event and periodic-eight trigger schedules
retain all component modes as opt-in because their measured cost exceeds their tree reduction
document the proof, counters, results, and default-off Gate D decision

Measurements

Counter-free Release medians with existing cheap rules enabled:

order 7: disabled 0.985 s; gcd 1.027 s; subset sum 1.052 s
order 8: disabled 0.203 s; gcd 0.212 s; subset sum 0.219 s

Gcd changed fewer than 0.005% of nodes and slowed both orders about 4.3%. Subset sum reduced nodes only 0.21–0.24% while slowing them 6.8–7.5%. Periodic-eight checking was slower still, so boundary-event triggering is retained only for opt-in experiments and component pruning defaults to disabled.

Validation

Debug CTest: 14/14 passed
ASan+UBSan CTest: 14/14 passed
focused boundary and periodic exhaustive comparisons passed
Release benchmark modes and runner self-test passed
warning-enabled syntax, Python compile, invalid CLI, and diff checks passed
issue-specific review completed with no findings

Closes #13

## Summary - detect skyline empty components separated by full-height columns - add sound gcd and bounded subset-sum area feasibility modes - compare boundary-event and periodic-eight trigger schedules - retain all component modes as opt-in because their measured cost exceeds their tree reduction - document the proof, counters, results, and default-off Gate D decision ## Measurements Counter-free Release medians with existing cheap rules enabled: - order 7: disabled 0.985 s; gcd 1.027 s; subset sum 1.052 s - order 8: disabled 0.203 s; gcd 0.212 s; subset sum 0.219 s Gcd changed fewer than 0.005% of nodes and slowed both orders about 4.3%. Subset sum reduced nodes only 0.21–0.24% while slowing them 6.8–7.5%. Periodic-eight checking was slower still, so boundary-event triggering is retained only for opt-in experiments and component pruning defaults to disabled. ## Validation - Debug CTest: 14/14 passed - ASan+UBSan CTest: 14/14 passed - focused boundary and periodic exhaustive comparisons passed - Release benchmark modes and runner self-test passed - warning-enabled syntax, Python compile, invalid CLI, and diff checks passed - issue-specific review completed with no findings Closes #13

mcp added 1 commit 2026-07-31 08:41:02 +01:00

solver: add optional component-area pruning 39ff5cb340

Full-height skyline columns partition the remaining board. Add sound gcd and bounded subset-sum checks for the resulting component areas, with boundary-event and periodic benchmark schedules.

Keep the rules disabled by default because their small tree reductions do not recover their measured cost. Record the rejected default and scheduling evidence so it can be revisited only with new data.

Tests: Debug CTest (14 passed)

Tests: ASan+UBSan CTest (14 passed)

Refs: #13

mcp merged commit 39ff5cb340 into main

2026-07-31 08:41:16 +01:00

mcp deleted branch codex/issue-13-component-area

2026-07-31 08:41:17 +01:00

mcp referenced this pull request

2026-07-31 08:41:34 +01:00

roadmap: improve time to the first square solution #16

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: mgrettondann/partridge-cpp#28