SignalGraph authoring extraction and D1 TSan refresh (2026-10-15)

This packet moves control-thread node and connection authoring methods from signal_graph.cpp to the private signal_graph_authoring.cpp translation unit. It does not change the public headers, serialization, prepare/live-swap policy, optimized executor, or independent reference walk. The moved methods retain their GraphMutationLock and invalidate_live_locked_() ordering.

Provenance and ownership

• Original Release and TSan validation base: origin/main at bcde1146d8d82c932d71b47037cfc4e7fe83a019.
• PR rebase base: origin/main at 666157bda2acaee99fa7c667c20af774040ad0a3.
• Validation used an isolated worktree at the protected base.
• signal_graph.cpp shrank from 2,413 to 1,737 lines; the new private TU is 725 lines after repository formatting.
• The existing signal_graph_prepare.cpp, signal_graph_live_swap.cpp, signal_graph_executor_routing.cpp, and signal_graph_reference_walk.cpp remain separate and unchanged.

Release validation

The focused host, parity, and grouped graph targets built through tools/ci/governed-build.sh cmake --build build --target ... in Release. pulp-test-host-signal-graph passed 11,977 assertions in 136 cases; pulp-test-signal-graph-executor-parity passed 48,742 assertions in 63 cases. Focused CTest discovery and execution for SignalGraph|Translated routing|Parallel routing|Standalone translated routing|Lowerable Custom|Routed PDC both found 207 cases; all 207 passed. This includes the prepared-edit and live-swap cases registered in the grouped graph binary.

After rebasing onto the PR base above, the three focused targets rebuilt in Release through the governed wrapper. The full host and parity binaries passed the same 11,977/136 and 48,742/63 assertion/case totals; the sample-region sequencer binary passed 250 assertions in 10 cases, including its source authority check against the moved authoring method.

Dedicated Debug/TSan refresh

The TSan tree used Debug, PULP_SANITIZER=thread, tests ON, examples OFF, and GPU OFF. Both binaries built through the governed wrapper. Each run used TSAN_OPTIONS=halt_on_error=1:history_size=7:suppressions=$PWD/test/tsan.supp. No ThreadSanitizer report was emitted.

Selection	Result
Full host graph binary	11,977 assertions / 136 cases passed
Full executor parity binary	48,742 assertions / 63 cases passed
Host [race]	65 assertions / 9 cases passed
Host [live-swap]	129 assertions / 4 cases passed
Parity [threads]	416 assertions / 3 cases passed

The refreshed suites include the later scalar-status, pinned routed-status, and standalone routing-publication regression tests added after the earlier D1 receipt. This is deterministic sanitizer evidence on the listed base and diff, not a claim that all possible thread schedules were explored.

Binary	Size	SHA-256
build-tsan-d1-current/test/pulp-test-host-signal-graph	33,446,584 bytes	f46d9234bab4cce2c2245d88eb711e93d1022af6103c29e7cffaec3ef259fbb8
build-tsan-d1-current/test/pulp-test-signal-graph-executor-parity	27,465,432 bytes	3e1aa08a187a7b347405bc891bf3145b0d447ac9fe77fdd76a4ddf2811dabf84