Skip to content

Schedule bounded angle work between tactile stages - #27

Merged
YasenSS merged 10 commits into
linker-bot:mainfrom
YasenSS:feat/python-tactile-stage-scheduling
Nov 16, 2026
Merged

YasenSS merged 10 commits into
linker-bot:mainfrom
YasenSS:feat/python-tactile-stage-scheduling

Conversation

@YasenSS

@YasenSS YasenSS commented Nov 16, 2026

Copy link
Copy Markdown
Contributor

Problem

Under the explicit Python runtime policy, a complete tactile acquisition can delay angle commands and reads. Unified telemetry read completion also constructs a full health report just to obtain the connection generation.

Change

  • Service angle commands and queries at complete Classic finger boundaries and O30 HOP V25 stage boundaries. Keep reply assembly and multi-frame angle batches uninterrupted.
  • Align query-only, command-feedback-only and adaptive feedback. Limit each boundary to four logical angle transactions, alternate ready work and preserve finite deadlines.
  • Cover nested failures, polling changes, close/reopen, recovery and shared-Session isolation. Preserve existing C++ APIs, Native defaults and the existing Session writer; add no send queue. O20, Python bindings and hand/arm integration are excluded.
  • Read the runtime connection generation without building a full health report. Prepare bilingual v0.3.12 documentation and record hardware coverage.

Validation

  • Current local checks: clang-format 18.1.8 format check, documentation checks and Git whitespace checks passed. Python validation: 61 passed, one platform-specific skip.
  • Recorded implementation checks: 24 scheduling cases / 453 assertions; Release CTest 20/20; Debug ASan+UBSan 10/10. These were recorded before the documentation-only commit.
  • O6, L20 and O30 each passed query-only, command-feedback-only, adaptive, five-minute saturation and Native-control hardware groups on Linux SocketCAN. Release notes identify the tested revisions. Physical shared-Session and injected fault/lifecycle timing remain untested; offline regressions cover isolation and failure behavior.

Remaining checks

  • Linux/Windows CI, the full Clang 18 Report Card and the twelve-target package matrix require remote validation.
  • Local analysis reported an easily-swappable callback-parameter finding in RuntimeAccess::with_stage_boundaries; it remains unresolved. The local macOS TGZ audit failed on AppleDouble entries; ZIP validation passed.
  • Documentation targets v0.3.12. CMake and the release-workflow default still specify 0.3.11 and must be aligned before tagging. This PR does not publish a release.

Add scoped, nonrecursive callbacks between complete tactile stages.
Read Python L20 and L20Lite fingers sequentially while publishing only
complete whole-hand samples; preserve the Native aggregate path.
Accept only declared current-state feedback as angle sampling evidence.
Correct the O30 V25 capability and collect optional O6 angle feedback.
Fence read completion counters by polling plan revision and provide an
internal clock seam for exact deadline regression tests.
Reuse the active device lease for pending angle targets and merged reads.
Limit each boundary to four logical transactions, alternate reads and
writes, and preserve finite execution budgets and actual send receipts.
Propagate nested connection failures while retaining ordinary inner errors
and isolate poll ownership during reentrant reads.
Cover feedback policies, complete protocol boundaries, bounded mixed
loads, nested failures, polling replacement, close, recovery, and shared
Session isolation with FakeTransport and controlled synchronization.

Validation: 24 tactile scheduling cases and 450 assertions passed.
Full offline CTest passed all 10 groups on macOS arm64 with C++17 and
SOURCE transport v0.3.4. No new hardware validation was performed.
Read the device connection generation from runtime health instead of constructing a full diagnostic report after each telemetry read. Preserve explicit health reporting and existing generation handling.
Document tactile-stage scheduling, compatibility and recorded validation.
Summarize O6, L20 and O30 hardware scope without extending past evidence.
Align version references and correct transport and interface guidance.
Correlate both L20 Lite angle feedback banks before adaptive polling
resumes. Fault the device runtime synchronously after a nested transport
failure while preserving peers on the shared Session.

Validate polling against the boundary cap without requiring four pending
intents on every stage. Cover retained invalid caches, recovery and peer
isolation, and address the reported clang-tidy diagnostics.

Validated Release communication, Python policy and discovery regressions
on macOS and Linux GCC 13, plus macOS ASan/UBSan. Repeated all 25 tactile
scheduling cases twenty times. Windows and the full Report Card await CI.
@YasenSS
YasenSS merged commit b5c60ae into linker-bot:main Nov 16, 2026
34 of 35 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant