Skip to content

Component build verification

Updated 7 November 2026. Kernel, frozen graphics, and HTIT build results below were recorded on 4 November; the native frontend and integrated Pocket checks were recorded on 7 November.

Component Inputs checked Result
Native graphical frontend Included SDL/OpenGL adapter, font/mesh generators, and source-release build inputs. Fresh-clone Mac build and focused frontend checks passed. Linux release-container and unpacked source-archive builds passed; GUI checks used software OpenGL. Physical Nokia acceptance remains separate. Build guide.
Integrated Pocket Shared desktop and application sources, isolated test state. 761 tests completed with 14 skips on 7 November. The native desktop entry describes the integration and its limits.
Linux 5.15.224 Pinned upstream archive, 55 maintained source changes, configuration and export whitelist. Independent fresh-checkout ARM build; 199 modules and 35 clock-module symbol checks. Full kernel record.
MBX module checkpoint Imported GPU source and the completed kernel build. Module compilation passed; all 63 imported symbol versions matched.
Mesa checkpoint Mesa 22.3.6 and 34 checksum-pinned Debian sysroot packages. ARM libGL.so.1 build passed. Required versions reached GLIBC 2.34, GLIBCXX 3.4 and CXXABI 1.3.9, within the selected Bookworm limits.
Verified 30 FPS snapshot Sealed module/Mesa production source, standard-gears fixture and fifteen offline regression suites. ARM module built with 64 matching symbol versions; Mesa rebuilt to the exact physical-test library hash. All fifteen suites pass. Source and commands.
HTIT hub Imported ESP-IDF source and Espressif32 6.10.0. Independent-clone firmware compilation passed.
HTIT executor Imported microReticulum firmware, vendored executor libraries and Espressif32 6.12.0. Independent-clone firmware compilation passed. The build used a locally generated private pairing PIN.
Map conversion Builders and generated fictional OSM/height fixtures; public regional data is downloaded separately. Offline fixture checks cover OSM import, geography/water tiles, terminal raster, terrain pyramid, bands and walk layers. Historical regional builds are recorded in the journal.

These are compilation and local-test results. Physical installation and device acceptance have their own records in the journal. The two graphics checkpoints have separate source identities and build commands.

The kernel builder records the selected compiler, configuration, source hashes and output checks. The graphics builders record dependency hashes and symbol or library compatibility checks. The HTIT wrapper records the source tree, PlatformIO version, environment and output hashes in its private build directory. Generated firmware carries the locally selected PIN, so independently generated executor builds have distinct bytes.

The full Pocket and bundled radio suites use isolated temporary home directories. Packet fixtures contain synthetic traffic. Map and terrain tests generate fictional reference packs; separately identified regional integration checks use prepared public data. The HTIT tests cover C/Python framing, an emulated radio engine, the official MeshCore client over local pseudo-terminals, CCMP validation and in-memory RSN peers. See the validation guide for the repeatable commands.

HTIT build · Map build

Fresh-checkout walkthrough, 7 November

This walkthrough cloned public revision dc97355 into an isolated Linux ARM64 Docker container. It followed the reader commands with fresh build directories. The container had no host checkout, USB devices, or radio services mounted. Native display checks used Xvfb and Mesa llvmpipe.

The first-session walkthrough launched with make, opened Help and Calendar, searched Help for “pairing,” switched windows with F6 and Shift-F6, quit with Ctrl-] then q, and restored all three windows on the next launch. The native compositor built, and the graphical frontend passed its input, frame-border, resize, and direct Room checks. make gui also launched under Xvfb as a normal user. Map configuration, planning, and fictional elevation/geography pipeline tests passed; this walkthrough did not download a regional map.

The manual Mesa command initially failed during configuration because the selected Python/Meson environment lacked required compatibility dependencies. The corrected guide creates the same pinned virtual environment used by the container build. With that environment, the frozen 30 FPS and earlier Mesa sources compiled successfully. All fifteen offline checkpoint suites passed. Both frozen MBX modules compiled against the newly built kernel, with 65 checked symbol versions for the 30 FPS module and 63 for the earlier module.

The kernel build completed with 199 modules and 35 clock-module symbol checks. Its combined image was 2,090,132 bytes, below the 2,095,104-byte limit. This is a new compiler-run result; the earlier image sizes above remain historical results. The executor and hub firmware both built, and the RSN checks passed 28 Python tests and 11 compiled-host tests.

Failures exposed by the reader commands

The normal-user source run completed 198 test files: 178 passed, 19 failed, and one (test_station_terrain_tools.py) exceeded the runner’s 240-second limit. These are test-file counts, rather than individual test-case counts.

The initial source runner reached legacy regional-data tests. They expected files such as tahoma-3dep.json, walk-layers.n8w, and lakewood-relief.json.gz in historical paths that the source release does not populate. The new area-selected map pipeline has its own passing fixture tests. The corrected tests generate fictional raster, road, water and elevation packs in temporary directories. The packaging test adds those generated inputs to its temporary inventory, while the source inventory pins the shipped code. Legacy elevation readers now honor the same data-directory override as the current catalogue.

The HTIT wire-protocol host test initially failed with GCC 12: wire.c compares a size_t loop index with 9 + f->len, and its -Werror build rejects the signed/unsigned comparison. Converting the length operand to size_t fixes the comparison. The corrected wire protocol passed all six host tests, and the hub firmware rebuilt.

The follow-up container run passed all 203 Pocket and bundled radio test files, including the previously timed-out station terrain suite. The run used a normal user and disabled network access during tests. All 28 Python RSN tests passed. The added test file covers the published game mouse-capture fix. The compiled RSN gate requires its checksum-pinned WPA source archive during dependency preparation, so its subsequent test execution can also run offline.

Run the source suite as a normal user. Nokia session checks deliberately reject root. Linux radio-proxy tests also need D-Bus and GLib bindings for the Python interpreter actually running the suite; installing Debian bindings for a different Python version leaves those imports unavailable.

GitHub container build, 7 November

Container build 42434184966 completed successfully for source revision bcd177a. The hosted run passed all 205 source test files, six HTIT wire-protocol host tests, 28 Python RSN tests and 11 compiled RSN tests. Source tests ran as a normal user with networking disabled. Fictional map and terrain packs supplied the test inputs.

The release build also completed the kernel, frozen graphics, HTIT firmware, native frontend and software-OpenGL checks. Docker reused matching cached build stages. The workflow uploaded separate n810-source and n810-documentation archives. Readers compile native programs and firmware from the source archive using the component guides.

An earlier hosted run exposed a timing race in the launcher test: the child process could exit between the helper's predicate check and its process-status check. The helper now checks its final condition before reporting failure. The corrected test passed 60 consecutive single-CPU Docker runs before the successful hosted build. The test change retains the assertion that the launcher exits successfully.

These results establish compilation and automated checks in containers. The first-session interaction walkthrough above records the separate manual exercise. Physical installation, radio behavior and N810 GPU performance remain tied to their dated device observations.