Release Notes — v0.6.11¶

Developer experience: formatter CI gate, GSDC smartphone benchmark, faster CI¶

Release date: 2026-05-25 Type: Tooling / developer experience — no positioning change (all binaries and outputs are bit-identical to v0.6.10) Branch: release/v0.6.11

Overview¶

v0.6.11 is a tooling and developer-experience release. It contains no changes to the positioning engines, decoders, or numerical code — every binary and every solution output is bit-identical to v0.6.10. It bundles three streams of work merged to develop since the last release:

1. Formatter CI gate + a one-time repo-wide format baseline (#166)
2. GSDC-2023 smartphone SPP benchmark (#165)
3. Faster, network-free regression CI (#175)

1. Formatter CI gate + format baseline (#166)¶

The project already documented clang-format / taplo / ruff as the style authorities, but enforcement was local-only and files had drifted. v0.6.11 establishes a reproducible baseline and gates it in CI.

Repo-wide format sweep (PR #177) — a one-time, no-functional-change sweep applied the pinned formatters across the tree:

• Vendored / upstream-derived code is excluded via a new .clang-format-ignore (src/core/tomlc99, util/), so reformatting never pollutes upstream-sync diffs.
• The three sweep commits are recorded in .git-blame-ignore-revs so git blame skips them (GitHub honors this automatically; locally git config blame.ignoreRevsFile .git-blame-ignore-revs).

Build-free CI gate (PR #179) — a new lightweight workflow .github/workflows/format.yaml runs clang-format --dry-run --Werror, taplo fmt --check, and ruff format --check on every push / PR. It is kept separate from the ~10-minute regression job so style feedback is fast, and is check-only — it never pushes formatting commits back (signing policy + fork-PR token constraints). Tool versions are pinned (clang-format / ruff via pip wheels; the taplo release binary is verified by SHA-256), and the same .clang-format-ignore exclude is reused, so a local run matches CI exactly. ruff check (lint) is intentionally not part of the gate.

CONTRIBUTING.md and CLAUDE.md now document all three formatters, their pinned versions, config files, exclusions, and local fix commands.

2. GSDC-2023 smartphone SPP benchmark (#165)¶

A new benchmark harness evaluates SPP on Google's GSDC-2023 smartphone dataset — the environment where the SPP accuracy work in v0.6.10 (#116) is actually meant to pay off (clock jumps, large jitter), unlike the geodetic PPC set.

• scripts/benchmark/run_gsdc_benchmark.py — runs MRTKLIB SPP on GSDC-2023 traces, with scripts/benchmark/gsdc_to_rinex.py converting device_gnss.csv to RINEX and scripts/benchmark/download_brdc.py fetching broadcast ephemeris.
• scripts/benchmark/compare_gsdc.py — scores mrtk NMEA output against the GSDC ground_truth.csv, reporting the official GSDC score (mean of the p50 and p95 horizontal errors) and a coverage (Cov%) column.
• conf/benchmark/gsdc_p0.toml and docs/reference/benchmark-gsdc.md document the workflow.
• A position-EKF (P6) re-attempt on this data was evaluated and reverted (no gain); the negative result is recorded in docs/design/ for future reference.

3. Faster, network-free regression CI (#175)¶

• ctest now runs in parallel (-j4, matching the runner) with BLAS pinned to one thread per process for deterministic PPP-AR results; the timeout was raised to 20 minutes.
• The IONEX TEC fixture is vendored (tests/cmake/download_tec.cmake), removing a per-run network download, with an integrity check on the file.

Tests¶

ctest: 77 / 79 on the maintainer's machine. The two failures — rtkrcv_rt (a headless-terminal / RT file-replay environmental issue; excluded from CI via -LE realtime) and madocalib_pppar_ion_check (a LAPACK-vs-reference numerical-noise difference, ~1.6 cm vs a 0.5 cm tolerance) — reproduce identically on a clean develop build and are pre-existing and unrelated to this release. Because v0.6.11 changes no numerical code, all positioning outputs remain bit-identical to v0.6.10.

The new Formatter Check workflow is green across the repository.

Upgrade notes¶

• No configuration, API, or behavioural changes — drop-in for v0.6.10.
• Contributors: install the pinned formatters (see CONTRIBUTING.md) and, optionally, enable blame-ignore once with git config blame.ignoreRevsFile .git-blame-ignore-revs.