Skip to content

1.5.0: Material 3 only, rebuilt settings and setup, high refresh rates - #140

Merged
j0nsh1n merged 206 commits into
mainfrom
feat/material-only
Nov 20, 2026
Merged

j0nsh1n merged 206 commits into
mainfrom
feat/material-only

Conversation

@j0nsh1n

@j0nsh1n j0nsh1n commented Nov 18, 2026

Copy link
Copy Markdown
Owner

Draft for 1.5.0. Not ready to merge: work continues on this branch and each part is pushed after its review.

What 1.5.0 does

  • Material 3 is the only interface. The Widgets styles, the experimental designs and the look editor go. Checking six designs on two UI systems every release was too much to keep correct.
  • Everything that was only in the older styles is rebuilt in Material: alarms and the ring screen, alerts and sounds, study and school hours, focus timer lengths, copies of this week, Copy day / Copy last week / Reload week / Open Spotify, and the keys Ctrl+N, Ctrl+C, Ctrl+D, Delete, F1 (no command bar). All pop-ups are Material sheets.
  • Looks: High Contrast, Paper, Ink and Terminal come across as Material presets.
  • New setup: short multiple-choice cards with a live week preview beside them. Setup decides how much of Look's advanced controls to show.
  • Look settings: a Simple / More / All switch.
  • Settings: every setting the older styles had is checked and in Material. Open on: Week, Day, Month or My day. Animations: Off / Reduce / Normal / More.
  • Pop-ups with three or more answers: full-width choice cards. The safe choice is picked first and buttons stack when they don't fit, so Cancel never sits outside the pop-up.
  • Pair a phone says how long the code lasts, the tries left, and why a code closed.
  • Fallback: when the graphics card fails to draw, the app restarts once with CPU drawing and says so. A plain error page shows only if that fails too.

Still to come on this branch

  • Remove the rest of the older interface: groups 3-6, ending with the main window cut down to a small shell.
  • Material in its own Qt Quick window, so it draws at the screen's refresh rate (measured 62.5 fps on a 360 Hz screen today; 359.7 in a bare Qt Quick window).
  • feat: weekly fixed times with an end date, and reminders that stay with Snooze #118 rebuilt on 1.4.0: weekly fixed times with an Until date, "Repeats every week", and reminders that stay with Snooze.
  • Docs: spec.md, DESKTOP.md, README, CHANGELOG.

Checks so far

  • The whole desktop suite at ea8bd6a: 5167 passed, 1 skipped.
  • ruff passes, and Material code has an import test that keeps it free of the older interface's code.
  • Each group was reviewed by the lead before it was pushed.

Review note

This PR is over CodeRabbit's 100-file limit (about 350 files, mostly deletions). The review plan for the final state is still to be decided.

🤖 Generated with Claude Code

j0nsh1n and others added 30 commits November 17, 2026 23:25
…nt point can still go

spec.md said the store is the only code that writes the database; the sync crate and the Android
bindings also write through their own connections (engine/sync net/mod.rs open_db, engine/ffi
core.rs set_reminder_lead). The MCP restore-point comments now match the spec and README: at 20
student points a save drops their oldest, as any save does. Comments and docs only.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…hase

1.4.0 is released (PR #138); the Hack Club README is merged (#139). The new
Release 1.5.0 phase records Jonathan's 2026-11-17 picks (Material 3 only, what is
rebuilt and what goes, the CPU fallback, Material in its own Qt Quick window)
and folds in the Material settings decisions of R5 from the docs/after-review
branch (2026-11-08). context.md's handoff entries for merged branches are pruned.

Docs only.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…g files

hours/scale.py holds Scale, OPENS, opening_minute (and the KEY pattern Scale checks);
hours/month_cells.py holds MonthChip, MonthCell, month_cells, HOMEWORK, LOAD_CAP, LOAD_WORDS.
Code is moved unchanged. zoom.py and month.py import the names back, so the Widgets side and its tests
work as before; Material imports from the new modules.

Verified: ruff check desktop; AST of each moved definition equals HEAD's; pytest on sheets, month, setup,
account, settings, sync and desktop/tests/test_*.py (1987 passed, 1 skipped, 457 s) with all step 1 changes.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…gs choices leave the Widgets files

New modules, code moved unchanged: sheet_rules.py (from widgets.py), account_copy.py (from window.py),
setup_steps.py (from setup.py), settings_copy.py (from settings.py). The Widgets files import the names back;
window.py and one test import the moved day-name helpers from sheet_rules, where widgets.py no longer
re-exports them. Material imports from the new modules, and PASSWORD_LENGTH_HINT from client, where it lives.
Style names, look and layouts imports in bridge_account.py stay for later steps.

Verified: ruff check desktop; AST of each moved definition equals HEAD's (one docstring gains a space from
ruff format); pytest on sheets, month, setup, account, settings, sync and desktop/tests/test_*.py:
1987 passed, 1 skipped.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Reads every desktop/native/material/**/*.py with ast (top-level and function-level, relative and plain
imports). The allow-list holds host.py's look import (plan step 3) and bridge_account.py's layouts.registry,
look and setup style names (step 7); an unlisted import, a new name from an allowed file, or a stale entry fails.

Break-it proof: a function-level `from desktop.native.widgets import Dialog` in bridge_week.py turned the
test red; a new `look` name in host.py (allowed file) turned it red; both reverted, test green again.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…essing hidden Widgets buttons

Each row used findChild + button.click() and read enabled/text from the Widgets button. The rows now
run the same window actions (_quick_focus, _plan_with_left_over(everything=True), _open_help,
_open_about, _log_out) and compute enabled themselves (Running late and Sign out grey while a save
runs; the rest stay on). The unused Widgets material_more QMenu and its helpers are removed.

The old test patched the removed _button; replaced by two tests that make findChild return None for
those buttons. Break-it: putting a findChild click back in moreAction turned the new test red; removed
again, green. Verified: ruff check, mypy backend, pytest on sheets (-k more etc.), more_count, material
imports/sheets/window/week, account, settings, layouts_window.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…choice and the device, not the Widgets look

appearance_of no longer asks the window for the account's theme pack, the Widgets
look or design_look. System follows the device palette; Light and Dark win; contrast
is Material's own saved choice; signed-out screens follow the device at standard
contrast. host.py leaves the step-1 import allow-list. Widgets keys in the look file
are read by nobody here and left as they are.

Verified: ruff check; settings, account, Material window/settings and import tests
(949 passed); new test_material_appearance_own_state.py (8 passed). Break-it: making
appearance_of read the Widgets pack again turned 3 of those 8 red (system follows the
device, a light pack on a dark device, signed-out); restored, green.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Step 3 made signed-out Material always standard contrast. Before, the sign-in
screens carried the High contrast of the last look worn on this computer; the
look file is the computer's, not the account's, so Material's saved contrast
is that same answer. Light or dark still follows the device when signed out.

Verified: the signed-out test now also checks the contrast; with the old
'std' put back it fails (1 failed, 9 passed), restored 10 passed; ruff clean.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…CPU, with a plain error page if that fails too

The Widgets week no longer appears as a fallback. The look file keeps cpu_drawing_version (replacing
material_failed); start-up sets QT_QUICK_BACKEND=software for that version. A newer version tries the
graphics card again. Failure on the CPU shows an error page with Quit and Try again.

Verified: ruff check desktop; pytest day/, account/, settings/, smoke and import tests; the break-it check
(showing the week on a CPU failure) turned test_no_failure_path_shows_the_widgets_week red.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… a packaged build from its executable

sys.orig_argv in an AppImage points into its mount, which goes away when the
failing copy quits. The new copy now starts from $APPIMAGE; a packaged build
from its executable with its arguments; a source run repeats the interpreter.

Verified: three new cases plus the fallback file, 13 passed; with the AppImage
rule taken out one fails, restored 13 passed; ruff clean.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Draws Alerts, Focus, Planning and Look from the top, so each after picture has a before beside it.
Verified: render.py drew 8 pictures, 0 failed.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…n accent over any preset and a font choice

The four old looks' colours from look.py become Material schemes (one mode each: High contrast,
Ink and Terminal dark, Paper light), laid out as theme.scheme() lays out the others. The accent
(Same as preset by default) and the category-chip switch use the saved accent and accent_chips
preferences; a font choice sits in the look file's material object and a preset's font and corners
are only defaults. Readability is checked per scheme (4.5 to 1, 7 to 1 for High contrast).

Verified: settings/test_material_new_presets, test_material_tokens, sheets/test_material_scheme_values:
passed; the same-as-preset rule broken on purpose went red and was restored.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… Look settings it was missing

Alerts: reminder sound, tone picker with a test play, Spotify link, do-not-disturb override, volume,
end chime, tray, mute while a session runs. Focus: Short, Standard, Long or Custom with a slider per
length, auto-split, mark homework finished. Planning: Study hours (windows, protected times, cutoff)
and School hours, drafts saved the way the Widgets dialogs save them. Look: accent, chips, font.
Group titles in every section. The single 25/5 switch is replaced; a saved 25/5 reads as Custom.

Verified: settings/ account/ and the import allow-list ran green but for one test that already
fails on HEAD (test_settings_look_choices late failure); the timer presets, the accent default and
the study-hours rules broken on purpose went red and were restored.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…relaunch replaced

test_material_is_not_current_after_a_late_failure_or_a_rebuilt_settings_page
checked the Widgets Settings card 'while the Widgets week is up' after Material
failed. Since step 6 a failure relaunches Material on the CPU and the Widgets
week never shows, so the scenario cannot happen; it failed from 62e49aa on.
The new flow is covered by test_material_shell_fallback.py (a late graphics
error asks for one CPU copy; no failure path shows the Widgets week).

Verified: test_settings_look_choices.py + test_material_shell_fallback.py,
30 passed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…py day, Copy last week's fixed times, Reload and Spotify tools

Keys: Ctrl+N (new homework) and F1 (Help) are window shortcuts, off while typing, in a sheet, Focus or
sign-in. Ctrl+C, Ctrl+D and Delete act on the block that has the keyboard focus and end in the same
calls as its menu (Delete asks the same question and gives the same Undo). Help lists them.
Tools: four rows under Planning in More; Spotify only with a link set; all grey while a save runs.
Verified: new key and tool tests (18), sheets/, week/ (2390), Material shell, import and static audit
tests, ruff. Broke Delete and the typing guard, and the Spotify and busy rules on purpose: red, then green.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The Timer subtitle was cut off; it now reads 'Pick a set of lengths, or Custom to set each one.'
Verified by before and after pictures at 1280x800, light and dark.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…ing screen

Alarms group in Settings > Alerts (list, add, edit, delete), AlarmSheet, and a RingScreen the window
opens instead of AlarmRingDialog while Material is shown (Dismiss, Snooze, Spotify, Esc/Enter).
ALARM_NOTE and ring_days moved to settings_copy.py so Material imports no Widgets module.
Verified: 16 new tests (4 mutations went red and back to green), settings/ and alarm tests pass;
regress184 time-box failures are identical at the parent commit.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…es or Material gives up

Material's ring screen has no dialog to stop the sound as it closes, so closeEvent and the give-up
paths now end the ring as a dismiss. Verified by two new tests, red with the calls removed.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…terial can use it

The Widgets dialogs import it back; behaviour is unchanged. Verified by the settings and sheets suites.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…sheets

While Material is shown none of the Widgets Help, About, Update or sign-out dialogs open. An update found
while another sheet is open waits for it. Verified by 24 new tests (each sheet mutated red, restored green),
the sheets, settings, account, alarm and updater suites.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…unt_copy so Material can use it

The Account notes, recovery-code status, restore point rows and preview summary, and the backup-import summary
move out of settings.py (which imports them back); the account file and delete wording is added to account_copy.
A restore point's week count now reads the backend's own 'weeks' key as well, so the list no longer says 0 weeks.
Verified: ruff, account/, settings/ and sheets/ suites.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Account (More, and Settings > This computer) and Copies of this week (More > Planning) open Material sheets
with the Widgets dialogs' words and actions. Delete account, import backup, replace week and Restore each ask
first in a Material question with the safe answer first. While Material is shown the Widgets AccountDialog,
RestoreDialog and TransferPreviewDialog never open. 26 new tests through the real backend; each guard was
mutated red and restored green. Two old tests changed to the Material behaviour (More rows, Esc from Account).

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The shared row text now picks week or weeks from the count, for both looks.
Verified: new test (4 cases) passes.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Used for the step 7 before and after pictures.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Removes every way to the Widgets styles: the Look page's Style row and the
Widgets styles button, MaterialBridge.useWidgets and the widgetsRequested
signal, the window's switch back and forth, the "Back to Material" More row,
the Widgets design card and the Look grid's Material card, and the setup
style, look and colours pages (setup now opens on Your week). Every look
file opens Material, a saved "widgets" included. The smoke test now walks
Material's own sign-up and setup. The step 1 import allow-list is empty.

Tests: tests that drove the removed switch are removed or rewritten (listed
in the step 7 report); tests of Widgets screens keep their Widgets start
through a conftest rule that steps 8 and 9 remove with them. Verified with
the account, setup, settings, e2e, day, sheets and focus suites.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
SheetFrame's footer and the Settings prompt kept their buttons in one right-anchored row, so four
answers ran out of the card on the left. They now stack, right-aligned, Cancel on top.

Verified: test_prompts_stay_in_their_card (walk of every prompt, Normal/Large, 1280x800/1024x640);
forcing stacking off turns the walk red, restoring turns it green.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Full-width tonal cards with the safe, undoable choice picked first, a 2 px primary outline, tint and
a check badge on the picked one; Cancel + Delete / Save changes in the footer. Up/Down move the pick,
Enter confirms, Esc cancels. The answers and what each does are unchanged.

Verified: sheets/, settings/ and the dialog week tests, 1876 passed; ruff check desktop clean.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…er to stack them

The footer works out stacking a turn after its buttons exist (Qt.callLater), so a
four-button dialog could draw one frame side by side and spilling out of the card
before it stacked. Both frames (SheetFrame, settings Prompt) now keep the footer
at opacity 0 until the first fit has run.

Not covered by a test: a test that steps the event loop could not catch the
unstacked frame (with the guard removed it still passed), because the deferred
fit runs inside the same processEvents round. Kept as a guard; the button-in-card
walk (test_prompts_stay_in_their_card.py) still passes.

Verified: sheets/, settings/ and week/test_edited_weeks_are_asked.py, 1866 passed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
j0nsh1n and others added 27 commits November 19, 2026 19:15
…e draw-error window (Grok review)

Grok's review of step 11 (FAIL): in Material's own window the shell's Esc shortcut now matches, and it did not
stand aside for Month's open date bar, so Esc with the keyboard on a date moved focus to the rail and left the bar
up. It now stands aside while the bar is open, as it does for Week and Day menus. The alarm dialog shown when
Material cannot draw gets the draw-error window as its parent. The first-run setup test no longer takes focus
itself before asserting the window has it. Checked the other Esc handlers against the shell shortcut: Focus's end
question is already covered (ending), Setup sits in the account layer, sheets and prompts are excluded.

Verified: new test_esc_with_the_keyboard_on_a_date_closes_its_bar_and_stays_on_the_month fails without the
shortcut condition and passes with it; test_material_first_run_setup.py passes (9) without the focus call.
Not done: file dialogs keep no transient parent (their tests patch Qt's static calls); the real-screen check on
Wayland decides whether they need one.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ig on it

Sonnet 5.5 (guarded): MaterialPage is a QQuickView, NativeWindow a QObject controller (close, Ctrl+S, file
dialogs, draw-error window, instance raise, tray). Opus 5.5 investigator (Jonathan's OK): Esc during a Week range
drag was lost to two matching shortcuts; WeekView's extra one goes. Lead: the rig finds Material in its own
window; a guard that no key has two enabled shortcuts; the shell's Esc stands aside for Month's date bar; the
alarm dialog opens over the draw-error window. Grok: FAIL then PASS. Whole suite on this tree: 2524 passed,
1 skipped; real plays material-light and material-month-light clean. Real-screen checks (focus, Alt+Tab, tray,
dialogs on Wayland, 360 Hz) still to do with Jonathan.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Jonathan, 2026-11-20: "The Ctrl+Scroll to zoom in and out is gone again". On his KDE Plasma Wayland session Qt lists
the only pointer as a touchpad, and the hours pane's WheelHandler took the default devices, which leave a touchpad
out, so Ctrl+wheel reached the week and changed nothing (the 1.3.0 notes' "Ctrl+wheel zoom on some Linux
desktops"). The handler now takes every pointing device. Plain scrolling never used it and is unchanged.

Verified: the existing Ctrl+wheel test failed on the real Wayland display (56 == 72) and passes with the fix; a new
test sends the wheel from a touchpad-type device so CI's offscreen runs catch it (red without, green with); mutation
case RED; Week and Day zoom files pass (22).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ture renders read Material's own window (steps 11.7 + 11.8)

Material draws in its own Qt Quick window and Qt Quick paces its animations (350-360 frames a second measured on a
360 Hz screen), so the Widgets page motion goes: motion.py's Clock, Shift and Dim fades, switch_page and helpers
(536 -> 31 lines), motion_level.frame_interval_ms, their tests (5 files deleted) and three page-fade mutation
cases. apply_ui_effects stays (Qt's menu, tooltip and combo fades for the remaining widget windows) with a new
test and mutation case. NativeSmoke's Material check also requires the window visible and the shell built;
render.py's QML labels and focus ring read the QQuickView through the test helpers, and the seam guard's last
known exception is gone.

Builder: Sonnet 5.5 (guarded T3 slice, brief step11/brief-11-7-8.md; report step11/11-7-8/step-11-7-8.md).
Verified in the box: ruff clean, 458 passed with -n 1; breaking the painted-colours check turned its new test red,
restored green.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Since group 5b3 (20bad69) week_zoom.py imported the removed Widgets helpers (a_day, a_hand, a_lane_week, ...), so
render.py --list reported screens_week.py as a problem and dropped every Week picture. The seven Widgets zoom
screens go; the Material week's one-zoom-step screen stays.

Verified: render.py --list now lists 107 screens with no problems (was 88 and one); ruff clean.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…red weeks

Since the week can start on any day (3b598a1), session.revision is a list when the open week spans two stored
weeks, and the smoke test compared it with 1: a TypeError on every tick, then "timed out waiting for saved week"
(found while checking step 11.8; it failed the same way before the window move). It now waits until any stored
week of the open week is saved, since the save bumps the one that holds the new blocks.

Verified: --smoke-test offscreen was ok false (timed out) before, now ok true, "week shown after setup Solve",
Material painted 161 colours.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…pictures read Material's own window

Sonnet 5.5 (guarded): the Widgets page motion and frame clock go, apply_ui_effects stays; NativeSmoke and render.py
read the QQuickView. Lead: the week zoom picture module (broken since 5b3) keeps its Material screen; the smoke
test's saved-week check handles a week that spans two stored weeks. Grok: PASS. Suite on the step's branch 2511
passed, 1 skipped; render.py --list 107 screens, no problems; --smoke-test ok, Material painted 161 colours.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
CI's Python lint (verify-python on a5f7231) failed on desktop/main.py:264 at 111 characters; the comment was reworded
in 77e1390 after the last local lint run. Comment only.

Verified: ruff check . passes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…back from Snooze (#118, U7)

Jonathan, 2026-11-20 (#118 plan): a reminder stays on screen as a card when no pop-ups are shown or "Leave
reminders on screen" is on, with Snooze and Close. The engine says when: the block's start minute from its reminder
key, that the card stays until that minute, and that a snooze is the alarm's five minutes but never past the start.
Everything is in minutes of the local day, as the reminder poll reads the clock, so a day the clocks change does not
move it (the old #118 branch worked in milliseconds from midnight, which a review found an hour off on that day).
Python gets thin wrappers in desktop/native/remind.py.

Verified: cargo fmt, clippy -D warnings, desk_remind 20 passed; making the card stay through the start minute
turns its test red, restored green; maturin develop --release, and the wrappers answer 960 / True / 960 / 945.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… U1)

Jonathan approved #118 (2026-11-20): a student's own fixed time can repeat every week. One rule says what stands
in every week: is_standing_block = School, Setup's activities, or a block whose id starts with weekly-. Repeat
intents, split-off days, deleting a whole repeat and the MCP every_week rule use it; Setup keeps is_setup_block,
so it still makes and stops only its own. A one-day change of a weekly time says "open it and choose This week
and later". Python gets calendar.is_standing_block. No storage or sync change: a weekly- repeat is an ordinary
standing row.

Verified: new engine tests (the rule, a new weekly repeat, its one-day note) and store tests (made from the open
week it shows in every later week and no earlier one; moved this week and later it keeps earlier weeks and a
later week's own change; stopped it leaves earlier weeks); making is_standing_block Setup-only turns 2 engine and
3 store tests red, restored green; cargo clippy --workspace -D warnings clean; cargo test --workspace: 97 ok.
Not yet: the .venv engine rebuild waits for the U7 helper's run (shared venv).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… are read (#118, U2)

A repeat may carry until (YYYY-MM-DD): a change may set it, an empty value clears it, and a change whose Until is
before the day it starts is refused. Weeks are cut at read (store week_with_standing -> shown_with_standing): a
block shows only its days up to its Until, and goes once none is left; nothing is stored cut. Before a change to
the repeat merges a later week's copy, the days the repeat's old Until cut from that week are put back
(uncut_copy): a copy saved while cut holds only what the week showed, and the old #118 branch read those missing
days as the week's own removals, so moving Until later never brought them back (its review failure). The sheet's
sentence "Until can't be before the first day, Tue 24 Nov." is until_problem, with a Python wrapper.

Verified: store tests by what weeks show: Until inside a week shows the days up to it and none after; moving Until
later shows the cut days again even in a week saved while cut, from that week or an earlier one; a day taken out
by hand in a cut week stays out when Until moves; engine tests for setting, clearing and refusing Until and the
sentence. Merging without the put-back turns the two moved-Until tests red, restored green. cargo clippy
--workspace -D warnings clean; cargo test --workspace: 97 ok. The desktop's instant open-week view
(apply_to_week) is not cut-aware; the store's answer is what the app shows after a save (U4 tests that).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…e weeks that hold it by hand (#118, U3)

Jonathan, 2026-11-20: up to 200 repeats (School, Setup's activities and weekly fixed times together), and a new
weekly time that a later week already holds by hand is named for that week with nothing removed. The engine holds
MAX_REPEATS (exported to Python for the backend's export and change limits, which stopped at 9 and would have
failed an export at the tenth) and REPEATS_FULL; the store refuses a repeat new to the account past the limit.
twin_in_week finds a later week's hand-made copy (same name, start and length on a day the repeat, cut at its
Until, also has); the save's answer carries doubled_weeks only when some week shows it twice, so every other
answer keeps its shape.

Verified: store tests (200 repeats stand, the 201st is refused with the sentence, an edit to one that stands is not
new; a hand-made Guitar in the week of 23 Nov is named and both stay); letting the limit allow one more and
dropping the named weeks each turn their test red, restored green; cargo fmt, clippy --workspace -D warnings
clean, cargo test --workspace 97 ok. The backend's use of MAX_REPEATS is committed after the engine is rebuilt
into the venv and the backend tests run.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…lose (#118, U7)

When the computer shows no pop-ups (no tray, or a tray that cannot show messages) or "Leave
reminders on screen" is on, a block's reminder becomes a Material card with Snooze and Close
instead of the Got it strip. Snooze brings it back five minutes later, never after the block
starts; at the start the card goes and "starts now" takes over. A moved or deleted block drops
its card; a renamed one shows the new name. Several cards stack, newest on top. A screen reader
is told without the keyboard moving. Focus notices keep the strip. All timing comes from the
engine (remind.py).

Verified by the helper (guarded Sonnet 5.5): ruff clean; test_reminder_cards.py (12 new) and
test_block_reminders.py 22 passed; settings, sheets, focus and reminder/alarm tests exit 0;
three mutation cases each red when broken by hand, green restored.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…nd a save names its restore point (#118, U3)

- backend/app.py: MAX_STANDING comes from the engine's MAX_REPEATS (200), so exports, imports and
  restore points hold more repeats than School and eight activities; STANDING_RULE names weekly
  fixed times. A save with snapshot_label now answers with the restore point it made, so the
  "this week and later" notice's Undo can put it back.
- backend/models.py: TimeBlock keeps `until` (YYYY-MM-DD); without it the desktop and the server
  dropped a weekly time's end date from a week's own copy.
- engine repeats.rs edit_intent: a form sent without Until for a repeat that has one clears it
  (every form that edits a weekly time shows Until; School and Setup never carry one).

Verified: backend tests for standing weeks, restore points, saving, repeat intents, export and
import 58 passed; fwtest gate --backend-only passed; cargo test engine + store 886 passed. Each
new rule broken by hand (MAX_STANDING 9, no restore_point in the answer, no until field, no
Until clear in edit_intent): its test went red, restored green.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Helper branch t3/reminder-cards-snooze-close (guarded Sonnet 5.5), checked with check-slice
(10 paths, all allowed by the brief or noted: controller.py and qmldir were needed).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…t, imports, and the save answer

- apply_to_week puts back the days a repeat's Until hid before it compares (as the store does), and
  takes the week start the days count from; a week on screen from another day than Monday uncuts in
  its own dates. Moving Until later or clearing it from a Wednesday week brought Saturday back nowhere
  before (review finding 1). MCP callers pass their stored Monday.
- The limit counts every repeat a starting one would stand beside, in its week or a later one, and a
  stopped repeat started again counts (finding 2: stop/add/restart, and adding from an earlier week,
  both passed 200).
- An import whose standing rows hold more than 200 at once is refused (engine
  most_standing_at_once), and a row's place is at most 199 (finding 4).
- The Rust save answer names its restore point as the Python one does (finding 6); RULE names weekly
  fixed times and until (finding 5).
Finding 3 (Setup's re-run stops weekly- repeats) is U6, in the running helper's brief.

Verified: cargo test --workspace 1209 passed, clippy -D warnings clean, fmt clean; backend tests 418
passed; fwtest gate --backend-only passed. Red/green by hand: no uncut in apply_to_week (midweek
test red), old limit condition and no later-week count (each limit case red), no import count and
place le=200 (import test red), no restore_point in the Rust answer (test_changes red).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…s them alone (#118, U4-U6)

The block sheet gets "Repeats every week" and an Until date (with No end) for a student's own fixed
time. Add with it on makes a weekly- repeat; turning it on for a one-off makes that block the
repeat; an edit asks "This week only" or "This week and later" (no question when only the switch or
Until changed, or for one day); turning it off keeps this week's as a one-off. An edit is built
from the repeat the session holds, never the week's copy. "This week and later" saves (and deleting
a weekly or Setup block from this week on) make a restore point first; the notice's Undo puts it
back and says later changes go too. A new weekly time that later weeks already hold by hand names
those weeks and removes nothing. Setup's re-run keeps every weekly- repeat as it stands.

Verified by the helper (guarded Sonnet 5.5): ruff clean; 15 sheet-flow tests and 3 on-screen tests
added; sheets + host seam green after two fixes; week/settings/Setup/account repeat tests 187
passed, 1 xfailed, 4 strict-xfail now passing (regress184 item1, Setup only); three mutation cases
red when broken by hand, green restored.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…the days Until hid (#118)

The edited repeat's days were the open week's ticked days. A week that Until had cut (or that had
its own edit) lacks some of the repeat's days, so ticking one more dropped those from every later
week: the old #118 failure. Only the days the student ticked or unticked now change.

Verified: new test_ticking_a_day_keeps_the_days_until_hid_from_this_week red with the old rule,
green with the fix; sheets, host seam and the 38 repeat/Setup/reminder test files 737 passed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ys hint and switch row line up (#118)

Jonathan (2026-11-20): a sheet opened for one day changes that day only, so "Repeats every week" and
Until are not shown there (a change to them was ignored before); the days hint becomes "Pick every
day it happens." so it does not read against the new switch. The switch row now lines up with the
fields around it (the settings-row component added its own 16 px inside).

Verified: sheets, test_weekly_fixed_times, host seam, shortcut guard 497 passed; the one-day rule
broken by hand turned its test red, restored green.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…6, Grok re-check)

Finishing Setup's week page went through window._setup_step_left -> _apply_standing with only
School and the activities, so every weekly- repeat was stopped; only set_setup_blocks had the filter.
Both now use session.setup_repeats(blocks): Setup's blocks plus every weekly fixed time as it stands.

Verified: new test_finishing_setups_week_page_again_leaves_a_weekly_time_alone drives the Setup page;
red with the page path on answer["blocks"], green with the fix. fwtest mutate on
weekly_fixed_times.json and reminder_cards.json: all six cases red (Setup case now points at
setup_repeats).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…es (#118, Grok re-check)

apply_to_week puts back the days the old Until hid so the compare does not read them as removed,
but then returned those copies as the week shown: a new title on Guitar (Thu and Sat, Until Fri)
showed Saturday until the next read cut it. The changed copies are now cut at their Until again,
in the week's own dates (a Monday's, or the shown week's). The store still writes every day.

Verified: midweek and Monday rename cases added; with the cut off the test goes red, restored
green. cargo test --workspace 1209 passed, clippy -D warnings and fmt clean; engine rebuilt into
.venv; backend 418 passed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…them

Helper branch t3/implement-u4-u6-u5 (guarded Sonnet 5.5, check-slice ok 10 paths) plus lead fixes on it: ticking a day
changes only that day (not days Until hid), one-day sheets hide the switch and Until, the days hint,
the switch row lined up, and Setup's own page keeping weekly- repeats (Grok re-check finding 1).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…y CI, not locally)

Look (Gemini 4 Neon, FAIL): reminder cards sit in the bottom right corner instead of over the
week's hours, on a light surface with an outline edge, Snooze a text button (the tonal fill matched
the card); a question shows above the save notice and dims it; "is already in a later week" for one
week; the delete card says "Undo also takes back anything changed after it."; the repeat switch row
is as tall as the switch, like "I missed it"; Next stops on an Until problem.
Grok U4-U6 (PASS, minors): Copy last week's fixed times leaves weekly times out; a save refused at
the 200 limit is taken back (drop_refused_save) so Save sends a corrected one; a new weekly time on
the open week is cut at its Until; turning the switch on is one Undo step named "changing".

Verified locally: ruff, rustfmt, cargo clippy -D warnings, engine test_repeats (new-repeat cut red
when broken, green restored). Desktop tests NOT run locally: Jonathan stopped local runs (RAM and
swap while he tests games); CI runs them on this push.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
CI's mutation-pattern check (fwtest gate --backend-only) found the case's old text "Pick more days to
repeat it." 0 times after the hint became "Pick every day it happens." (Jonathan, 2026-11-20).

Verified: every scripts/mutations/*.json case's old text found exactly once (static count, no tests
run locally).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…screens for #118's new states

The cards' conditional anchors left them bottom-left over the rail's Settings; they are now placed
with x (16 px in from the right, centred on a phone) and 288 px wide, inside the 320 px side panel,
clear of the week's hours. New picture screens (desktop/tests/pictures/screens_sheets_118.py) draw the
weekly switch with Until, the Until error, "Change Guitar?", the notice with Undo, the doubled-weeks
prompt and two reminder cards, light and dark with motion off, for the look review.

Verified: reminder cards, block reminders, weekly fixed times, sheets and host-seam tests 521 passed
(4 workers, capped, memory watched); pictures drawn by fwtest pictures render (145, 0 failed) and the
cards redrawn after the fix.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…118, U8)

The phone writes no repeats; it reads weeks through week_with_standing, which cuts a weekly- row at its
Until. today() for a weekly Guitar on Thursdays with Until Thu 15 Oct shows it on the 8th and 15th, not
the 22nd; with Until Wed 14 Oct the 15th has none.

Verified: red with cut_at_until disabled, green restored; cargo test -p flexweek-ffi 97 passed, clippy
-D warnings and fmt clean. Older phones (v1.4.0 engine) read with with_standing (no cut) and rebuild a
saved block from TimeBlock's known fields, dropping until: noted for the release notes draft.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Reminder cards' Close is tonal, not filled: with two cards up the screen had two filled buttons
  (one filled button per screen); Snooze stays a text button.
- The notice after a weekly-time save names what happened: "Guitar repeats every week from this week
  on." (add, or a one-off turned into the repeat), "Guitar stops repeating after this week." (switch
  off), "Changed Guitar from this week on." (an edit), each with the Undo line.
- The card is 16 px inside on every side (height content + 32; it was 12 px under the buttons).

Verified: weekly fixed times (new notice asserts), reminder cards, block reminders and the button-kind
rule 45 passed (capped, memory watched); reminder and doubled-weeks pictures redrawn, light and dark.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@j0nsh1n
j0nsh1n merged commit 5b19a22 into main Nov 20, 2026
33 checks passed
j0nsh1n added a commit that referenced this pull request Nov 20, 2026
Conflicts:
- desktop/native/settings.py: main removed the Widgets settings page, so its
  Report a problem button goes with it; main's file is kept.
- bridge_settings.py: both import lists kept (feedback plus main's new ones).
- tools/fwtest/areas.toml: main's shorter list plus desktop/native/feedback.py.
- CHANGELOG.md: the phone entries go into main's Added/Changed/Fixed lists;
  Report a problem now says Settings > Computer only.
- test_report_problem.py: the Widgets settings page checks are removed; the
  Material checks stay.

Verified: test_report_problem.py and test_get_developer_builds_switch.py,
8 passed; ruff check on the touched files passed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
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