Files

T

Max Gramser 1a6308aba7 Fix what a full test round of today's releases found: a ticking single clock, a switch that works in standby, no taps from a swipe (0.2.78)

A test and code-analysis round of 0.2.72 to 0.2.76 (suites at every tag, sanitizers, ESP32 and GitHub builds, host
harnesses on both boards, the editor in a browser, best practice checked against ESPHome, LVGL, Home Assistant and
Vue) found these, fixed here with Max's choices for the product questions:

- Firmware 0.2.65: the second hand of a single analog clock moves again; on a page that is one full on/off tile the
  waking push also switches it; on the Guition a flick is never a tap and a finger that slides off a tile doesn't tap
  it; Previous/Next react to a firm press; the same Go to page tile may sit on several pages; a layout the screen has
  no memory for is refused instead of crashing; 64-bit redraw mask for 48 tiles; Montserrat 14 no longer built.
- App: Save refuses a layout too long for one message or too big for eight pages; an offline screen keeps its
  firmware version from the device registry; HA refusals reach the editor; 128 KB requests; changelog out of the
  live stream; cached hashed assets; tini as init and a clean stop; changelog parsing without asterisks or crashes.
- Editor: clicking the open screen keeps unsaved edits, edits during a save stay unsaved, a Page row and phone
  dragging to other pages, repeated page tiles, limits from the server, HA's error messages, a type-checked build.
- tools/check.sh + GitHub Actions + a CYD flash budget; docs match 48 tiles and the new editor.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

2026-09-18 17:25:23 +02:00

4.4 KiB

Raw Blame History

Test results app 0.2.78 / firmware 0.2.65 (2026-09-18)

Max: "ga eens je ultra best doen om de commits van vandaag lokaal 100% te testen", then "gedegen code analyse en check op internet of dit best practise is", and after the report "can you address them?". This release fixes what that test round of 0.2.72 to 0.2.76 found, with Max's choices for the product questions: a swipe is never a tap, the waking push also switches a full-page switch, and the same page tile may sit on several pages. See CHANGELOG 0.2.78 and the compatibility note in docs/RELEASING.md.

The test round of 0.2.72 to 0.2.76 (before the fixes)

Python suite at every release point: 386 (2cb3479, 9f5694b), 398 (1ed42dc), 409 (bb25353, 261fa82), 0 skips, no order dependence over 43 runs (reverse, shuffled, -X dev). Only the older test_camera.Feed timing flake failed.
C++: 17/17 at every tag with the project's flags, ASan + UBSan, libc++ hardening, g++ 16 with _GLIBCXX_DEBUG, plus 594 new runtime_model checks, 20,022 comparisons with the add-on's packing and 200,000 fuzz rounds, all clean.
Editor: Vitest 46/46, vue-tsc clean, and the committed bundle rebuilt byte for byte at 1ed42dc, bb25353 and 261fa82.
ESP32 builds of both boards, also fetched from GitHub the way users get them; the add-on image built and started.
Host harnesses: full page 37/37 per board, page buttons 12/12 per board (the 0.2.71 rule fails 3/12 and 4/12, as it should), wake/sleep, settings, cover and history on the Guition without regressions; 4 bpp icons measured (2 grey levels became 15 to 16).
Found and fixed here: the frozen second hand of a single analog clock (since 0.2.74), unsaved edits lost when clicking the open screen, dragging to page 2+ on a phone, Identify and Try it answering 500, a 48-tile layout too long for one message saving anyway, packing past slot 47, an offline screen counted as very old firmware, the 32-bit redraw mask, a crash instead of a refusal when memory runs out, a firm press on Previous/Next doing nothing, python as PID 1 (every stop took 10 s), the 16 KB request limit, asterisks in What's new, docs that still said twenty tiles.

Automated (this release)

tools/check.sh (new; the same script runs in GitHub Actions), on this release rebased onto 0.2.77 (the media card): Python 454 tests OK, C++ 18/18 with -Wall -Wextra -Werror, packages and icons current, Vitest green, vue-tsc clean, build, and the committed editor bundle equal to a fresh build.
tools/check.sh --firmware with ESPHome 2026.6.2, on copies of both board profiles with the fallback access point and captive portal users get:

Board	firmware.bin	of the app slot	vs 0.2.75
CYD	1,652,384 B	90.0 % of 1,835,008 B (183 KB free)	-3,200 B (Montserrat 14 no longer built: -13,936 B; the 0.2.77 media card adds the rest)
Guition	2,026,608 B	24.9 % of 8,126,464 B

The CYD sits at the edge of the new budget's "tight" band (90-93 %): the next release states its flash delta.

Firmware on the Mac (host builds, virtual finger)

Full-page harness (fullpage_host.py, 37 checks: taps, holds, drags, keys, page jumps, 48 tiles, six forecast/clock flips), on the fixes before the rebase onto 0.2.77: Guition 37/37, CYD 37/37. The CYD's first run gave 36/37: the harness repeats a tap when the host loop stretches the contact past 350 ms, and the firmware had already counted the first one (two light.toggle calls 0.7 s apart); the rerun without that stall passed 37/37.

Not tested

On real hardware: nothing of this release has run on a bench screen yet. To check on the glass: a single analog clock ticks; a flick across a full-page light on the Guition does nothing while a tap and a hold still work; one push on a dimmed screen with a full-page light switches it; a firm press on Next turns the page.
Unit tests cover the repeated page tiles (tests/test_runtime_model.cpp and the Python suite), the memory refusal (a stand-in for the free block size on the host), a consumed contact after a flick and the 300 ms page-bar case. The gesture handler, the slide-off check and the waking push run inside LVGL and have code review only, no host harness run with a finger.
Build machine note: Homebrew's Python 3.14.7 cannot load pyexpat on macOS 26.2 (system libexpat lacks XML_SetAllocTrackerActivationThreshold), which breaks PlatformIO's penv; the builds here ran with ESPHome 2026.6.2 in a Python 3.11 venv.

4.4 KiB Raw Blame History