homeassistant_espscreen/docs/TEST_RESULTS_0259.md

Test results app 0.2.59 / firmware 0.2.51 (2026-09-16)

Max found the history card unusable ("no y axis, no real x axis") and chose option A of two designs: a line with axes for numbers, with the touch readout of option B (a finger on the graph shows the value and time of that moment). History for on/off, people and numbers too, fetched when the card opens, for 1 hour, 24 hours or a week with the same number of points. During the work he asked for three changes, all in this release: the graph stays as it is under a finger (only the words at the top change), the rings of the highest and lowest moment sit on the line's peak and valley, and a sensor that only exists for two hours shows no data before that. See CHANGELOG 0.2.59 and the compatibility note in docs/RELEASING.md.

Automated

• Python (.venv-portal): 282 tests OK. New tests/test_history_card.py (28): which entities get a line or a timeline; time-weighted averages with gaps; statistics rows (milliseconds, gaps, min and max); the highest and lowest moment; axes in round steps for °C, °F, %, W, kWh, negative values, lux, m³/h, ppm, V and a flat line; decimals from the display precision; time ticks every quarter hour, six hours or midnight, across a daylight saving change; a door's timeline with Home Assistant's words, runs with the real times of their state (an opening of 5 minutes in a 15-minute slot reads 01:00-01:05, 300 s), people with zones and their own colours, statuses beyond six states joining "Other"; a sensor two hours old (averages only for its own hours, also from statistics with one compiled hour, and a sensor with one state) and a binary sensor two hours old (no data before it); a motion sensor that changes every 30 s for a week (20,160 changes: 13 ms, 354 bytes); the largest possible timeline (96 runs, 32-character states) at 3.6 KB of the 4096 a screen takes. The manager: a number and a door each get one message through the screen's action, a day from hourly statistics and an hour from the changes; screens asking at once share one fetch and a reopened card uses the cache; a failed fetch sends and keeps nothing and the next question is answered; entities that are not on the screen, other inboxes, other ranges, a light and an offline screen get nothing; the event loop survives a failure. Firmware source checks: the event, the answer it waits for, the entities with a history card, range keys kept out of the disabled keys, the press lock, and small_font in both profiles and packages.
• C++ (clang++ -std=c++17 -Wall -Wextra -Werror -I. -I components/smart_display): 15/15 PASS. New test_history_view: clock times in 24 and 12 hours with weekdays and negative offsets, axis times ("4 PM", "4:15 PM"), durations, numbers with thousands and glued units, the part and the run under a finger, the monotone line (a setpoint stepping 55-45-60 stays between 45 and 60, flat stays flat, every point on the line, a small buffer is not overrun) and through (a moment takes its part's point, a second moment in the same part joins beside it, a full buffer refuses).
• generate_packages.py --check green.

Builds (ESPHome 2026.6.2)

Check profiles with the generated packages and local components, the same as for 0.2.58.

Profile	RAM	Flash	vs 0.2.58
CYD	23.5% (77,060 B)	77.8% (1,427,015 B)	+512 B RAM, +15,192 B flash
Guition	24.5% (80,436 B)	18.8% (1,530,331 B)	+512 B RAM, +15,068 B flash

No warnings from runtime_tiles.h or history_view.h (the switch's part/state selector, moved from the old switch card, now casts to lv_style_selector_t).

Firmware on the Mac (host builds, driven over the ESPHome API)

A demo home with a week of history (door, coffee machine, a person with zones, a washing machine status, a boiler setpoint, and a plug added two hours ago) answers with the real Manager.card_history; a virtual LVGL pointer drives the card. Guition 40/40, CYD 40/40:

• Opening a card fires one esphome.screen_history event (inbox, entity, 24); the card shows the value now and "Loading history..." until the answer, then the line with High and Low and their times and the axis labels the message carries.
• A finger reads the average of the hour under it with its time range and "average of 1 h" in the accent colour; dragging moves the readout, past the edge it holds the last hour, lifting restores the header. The objects of the graph (count, positions, visibility) are identical before and under the finger, on the line and the timeline.
• "1 week" asks for 168 hours and waits with the week selected; a late answer for the day is ignored; the week's High and Low name the weekday and a week part reads as a weekday time range averaged over 7 h. "1 hour" asks for 1 hour and a part reads "average"; tapping the range shown asks nothing; a new state updates the header without a new question; reopening starts on 24 hours and asks again.
• The door: the heading follows the state ("Off" before the answer, "Closed" with it, "Open" when the door opens while the card is open), "Open · 22 min" and "6 times"; a finger on the opening of 12:30 reads "12:30 – 12:35" and "5 min", the unavailable stretch "14:00 – 14:20" and "20 min".
• A person with zones (home, "once"), the week; a washing machine status without counts; the coffee machine with its toggle, which sends switch.turn_off; the boiler setpoint with its slider.
• The plug added two hours ago: its line only over its last three parts, High 1,850 W and Low; a finger before it existed reads "No data" with the time, on its last hour its average.
• Without an answer "No history available" after 8 s, and a late answer is still drawn; an empty hour says "No history in this period"; malformed history messages change nothing; no errors in the log.
• Full redraws on the host: the line card about 70 ms like the cover, weather and vacuum cards (the fill about 6 ms, the line about 12 ms). The host's LVGL loop stalls for 1-2 s at random, on any card; the checks wait for the state they expect instead of fixed pauses.
• Renders of every card on both boards; the README shows the new history cards. All README screens were rendered again from the standard demo home: the pages and the alert are unchanged, the light card and both weather cards changed since 0.2.41 and were replaced. The Guition weather card's round back button lay half under the card (since 0.2.43); it now starts below the button, checked in a new render. On the CYD moving it squeezed the days too much, so the CYD keeps its layout. A playing media player's volume slider is grey on the tile since 0.2.43 (Home Assistant colours it); left for a separate change.

Not tested / for the owner

• A live read-only check against Home Assistant could not run: both local tokens were refused. The card uses the same two calls the tile graphs already use (recorder/statistics_during_period, now with min and max, and /api/history/period).
• Real hardware: on Studio 1 after the update, open a temperature, a door or a person, switch between 1 hour, 24 hours and 1 week, and slide a finger over the graph. Free heap on a CYD with a week's timeline open.