bherbruck 5b4709a6f8 geom: raise MAX_SOLVE_BRANCHES to 2^23, measured on the tiny-scale torus booleans ...

The port's own global-solve work bound (not upstream) was 2^20. The tiny-scale box/torus
booleans (major 1e-4, minor 3e-5) need up to 3,408,795 branches in one solve and C++
finishes them; 2^23 gives 2.46x headroom while keeping a runaway (fuzz) solve to about a
minute. unculled_solves_stop_at_the_branch_bound now keeps 139 solutions at the bound
(20 under 2^20; ~14 s in the difftest profile). New #[ignore] regression
max_solve_branches_tiny_torus (both booleans finish; ~3 min release), run nightly
(owner-approved step). Finding: the tiny-torus cases can't be compared with C++, which
reaches UB-175 on them.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJzQKZ9sjU1DhgS253e8J5

2026-10-26 15:58:17 -04:00

..

actions/cpp-reference

ci: build without sccache when its cache storage is down

2026-10-26 12:39:03 -04:00

workflows

geom: raise MAX_SOLVE_BRANCHES to 2^23, measured on the tiny-scale torus booleans

2026-10-26 15:58:17 -04:00