Skip to content

omamobile-shell 0.2.38: touch batch 5 (onboarding and PIN as sheets, a Reminder tile, contrast, landscape follow-ups, polish) - #123

Closed
derekross wants to merge 7 commits into
mainfrom
touch-batch-5
Closed

derekross wants to merge 7 commits into
mainfrom
touch-batch-5

Conversation

@derekross

Copy link
Copy Markdown
Owner

Draft: built and tested offline, not yet run on the Pixel 7 Pro. Opened so its items are visible and nobody picks them up twice (see #122). It merges after a session on the phone, and one of its steps changes the real account password, which sudo also uses.

Touch plan batch 5 (docs/TOUCH-PLAN.md, "Batch 5"):

  • Landscape follow-ups from batch 4's phone check: a portrait row of app tiles stays a row under the widgets in landscape, and the landscape view never shows more pages than the layout has.
  • A Reminder tile in quick settings over Omarchy's own omarchy-reminder: minute presets, a message field, pending reminders with a two-tap Clear all. (Nightlight was already a tile; it now has a test.)
  • Onboarding and PIN setup as shell sheets instead of gum prompts (1.11), in a new omamobile.onboarding plugin. The scripts still decide what every answer does: the sheet calls new entry points (omamobile-onboarding --status --json | --identity NAME EMAIL | --invite-agent, omamobile-set-pin --status | --apply TYPE), and both scripts fall back to the terminal when the shell doesn't have the sheet (OMAMOBILE_PROMPTS=terminal forces it).
  • Contrast (2.7): secondary text dims only as far as 4.5:1 allows (4.64:1 on Tokyo Night, 4.55:1 on Catppuccin Latte), and unavailable tiles dim only their glyph. Checked by sampling the rendered images.
  • Polish (3.7): Wi-Fi and Bluetooth Forget take two taps, a drawer long-press says "Added to home", a missing app's tile says "Not installed", recents show the app's name instead of its id, and the shell README gains a Gestures table.

Version: omamobile-shell 0.2.37. Main is at 0.2.32, and 0.2.33 to 0.2.36 are held by #99, #115, #117 and the low-battery branch; the number is set again at merge if the order changes. The commit subjects still say 0.2.31, from before the rebase; the squash title takes the real one.

Tests: packages/omamobile-shell/test/all 230 ok on this head (rebased onto main 2026-10-28); offscreen renders of every new or changed surface, upright and in landscape; tools/pr-check clean.

On the phone (announce first: the shell restarts, the screen rotates in step 1): rotate to landscape and check the tile row and page count; set a 1-minute reminder; System > PIN opens the sheet (the PIN change is real); omamobile-onboarding shows the sheet; the contrast and polish items by eye.

🤖 Generated with Claude Code

derekross and others added 6 commits October 28, 2026 11:56
The batch 4 phone check found two things in the landscape view of the
home (TOUCH-PLAN 2.4): a portrait row of app icons turned into a column
at the right edge, since every item's x and y swapped, and the view
showed one page more than the layout, since an item with no room on its
page was pushed to a new one by the same re-placement a loaded file gets.

HomeModel.transposeLayout now turns only the widgets. The app tiles are
laid out afresh in rows under the widgets, in portrait reading order with
a new portrait row starting a new landscape row, so a row stays a row. A
widget with no room where it lands takes the first free spot at its
size, else the largest smaller size down to its minimum; every page
keeps to itself, so the view has exactly the layout's pages, and an item
with no room at all on its page is left out of the view rather than
moved. Nothing is saved either way.

test/home-model.js covers both (two 2x2 widgets and a row of four tiles
on page 1 keep one page with the tiles in a row; two portrait rows stay
two rows; a full page re-places, shrinks and then leaves out, never adds
a page), and test/home-render checks that the two app tiles share a row
under the widgets and that the battery stays on the first page.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
TOUCH-PLAN 1.12's last keyboard-only feature with a touch path. Omarchy
sets a reminder with Super+Ctrl+R and a keyboard flow (the omarchy.reminders
overlay types into a key catcher), which a phone can't drive. The new tile
closes quick settings' third page; a tap opens a page behind it, like the
Wi-Fi picker: a row of minute presets or a typed number, a message, and
Set, which runs `omarchy-reminder <minutes> [message]`. The timer, its
toast and the confirmation stay Omarchy's own; the page starts nothing of
its own. Above the form, what is pending from `omarchy-reminder show
--json`, with a two-tap Clear all (`omarchy-reminder clear`). The form
keeps to the bottom of the card, so it stays above the on-screen keyboard
while a field has focus. A long-press on the tile shows Omarchy's own
list (`omarchy-reminder show`), and the tile's status counts the pending
timers (tile-state.sh), lighting while one is.

Night light was already a tile (omarchy-toggle-nightlight through
tile-action.sh); test/all now pins that too.

Tests: the minutes rule, argv and JSON in test/tiles-model.js, the probe
in test/tile-scripts.sh, the wiring in test/all, and two shots of the
page (upright and on its side) in test/quicksettings-render, whose
harness now sizes the pages' window stand-ins.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
TOUCH-PLAN 2.7. Secondary text (the drawer's count, its placeholder and
"No apps match", the tiles' status lines, the panels' detail rows, the
sheets' hints) was the foreground darkened by a fixed factor or drawn at
60-75% opacity: on Tokyo Night that came to 3.5-4.3:1 against the
surface, under WCAG's 4.5:1 for text, and a factor that suits one theme
fails another. An unavailable tile dimmed its name as well as its glyph,
and a slider row without a backend dimmed itself whole, reason included.

Readable.js (the same file in every plugin) gains contrast(), flatten()
and secondary(fg, surface): the foreground mixed toward the surface only
as far as 4.5:1 allows, on any theme, and the foreground itself when
nothing dimmer would do. Every such text and every placeholder (against
the field's own fill, through Style.controlFill, since a focused field is
lighter) goes through it; a tile dims only its glyph and chevron, a
slider row only its glyph. The 1.3 darkenings left are glyphs.

Checked from the pictures, not by eye: test/launcher-render and
test/quicksettings-render crop each named text from the PNG with magick,
take the histogram's commonest colour as the surface and the farthest in
luminance as the text, and require 4.5:1 (every tile's name and status,
the sliders' reasons, the drawer's count, placeholder and "No apps match"
in a new no-match shot, the reminder page's placeholders), in both
themes and both orientations; parts the sheet's scroll clips are marked
and skipped. test/all pins the helper's numbers and the greps.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…cents

TOUCH-PLAN 3.7's small parts:

- Wi-Fi and Bluetooth Forget take two taps, like the other destructive
  actions: the first turns the button into "Forget <network>?" (or the
  device's name) for a few seconds, and cancelling the sheet disarms it.
- A long-press in the drawer is answered: the home page shows the new
  tile with "Added to home" (the drawer closes as before, so the notice
  is on the page the tile landed on).
- A home tile whose app isn't installed says "Not installed" rather than
  its raw desktop id; the widgets sheet's Missing list still has the id
  for removing it.
- Recent apps' cards and the Show with… rows are titled by the app's name
  from its desktop entry (by id, then a heuristic lookup when Quickshell
  has one), falling back to the window class.
- The README gains a Gestures table: every swipe, tap and hold on the
  shell, by surface, and the two-tap rule.

Tests: test/switcher-touch titles cards by name under qmltestrunner with
a DesktopEntries stub; test/all pins the two-tap Forget in both panels,
the tile's wording, the home's notice, the lookup and the table.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…mpts

TOUCH-PLAN 1.11. The first run (omamobile-onboarding) and System > PIN
(omamobile-set-pin) asked their questions with gum in a floating
terminal: arrow keys and Enter, which a touch screen can't give, and the
terminal keyboard's arrow row is 33 px tall.

A new panel plugin, omamobile.onboarding, asks the same questions in the
same order: name and email (Continue or Skip), then a PIN or password
unless there is one (then the note the prompts printed, with Keep it or
Change it), the choice between a PIN and a password (the current one
first, as gum listed them), the entry one step at a time as passwd asks
(the current one when there is one, the new one, the new one again, with
the ABC/123 switch and the fields kept above the keyboard), and done,
with Omarchy's agent invitation on the first run. System > PIN is the
same sheet from the choice on.

The scripts stay the one definition of what each answer does, factored
into entry points the sheet calls and the gum flow shares:
omamobile-onboarding --status --json, --identity NAME EMAIL (Omarchy's
git.sh and xcompose.sh), --invite-agent; omamobile-set-pin --status and
--apply TYPE, which reads the current and the new secret from stdin
(never argv, a file or the log), drives passwd on a pty the way the
terminal answered its prompts, and records the choice and lock-on-start.
Its exit code says what happened (3 wrong current, 5 refused with the
reason on stdout) and the sheet goes back to the step that needs another
answer. Run with no arguments, both scripts summon the sheet when the
shell has it and fall back to the terminal prompts otherwise (or with
OMAMOBILE_PROMPTS=terminal); the login nudge's tap lands on the sheet.

Tests: test/onboarding.sh drives every entry point against a fake passwd
that prompts as pam_unix does (the first PIN, a wrong current one, a
refusal with its reason, a change to a password, the nudge, the
invitation, the sheet-or-terminal choice), and checks no secret reaches
the output; test/onboarding-model.js the flows, steps and outcomes;
test/onboarding-render every page upright and on its side in both
themes; test/all the wiring, the lint, and that the sheet repeats none of
the scripts' logic.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The Batch 5 section: the two landscape follow-ups from the batch 4 phone
check, the Reminder tile (1.12), onboarding and the PIN as sheets (1.11),
contrast sampled from the renders (2.7) and 3.7's small parts, with the
tests, what was not touched, the phone checklist and the list of what
remains (2.1's font seeds, 3.1, 3.3, 3.5 About, 3.6 haptics, 3.7's rest,
zoom, the Omarchy patches 3-6). The status line and the rows 1.11, 1.12,
2.7 and 3.7 say what is done.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
0.2.33 to 0.2.37 are held by #99, #115, #117, #124 and the low-battery
branch.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@derekross derekross changed the title omamobile-shell 0.2.37: touch batch 5 (onboarding and PIN as sheets, a Reminder tile, contrast, landscape follow-ups, polish) omamobile-shell 0.2.38: touch batch 5 (onboarding and PIN as sheets, a Reminder tile, contrast, landscape follow-ups, polish) Oct 28, 2026
derekross added a commit that referenced this pull request Oct 28, 2026
L-279 to L-293: the 9a's slot B hold and what made devinfo writable
(#125), per-device flags checked by building both command lines
(#101), env defaults in the shared build library and CI waiting for
approval (#79), GLib retry paths and radio stops checked at the
controller (#99), silent re-pushes and stacks with several tops (the
tegu stack), and two of our own: merge only after every check has
finished (#126), and an empty diff is checked for empty inputs
(kernel #123).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
derekross added a commit that referenced this pull request Oct 28, 2026
…wenty-sixth rounds (#127)

* docs/review: 19 lessons and 3 corrections from the twenty-fifth round

The confirmed candidates from round 25 (L-260 to L-278): the 9a slot B
marking as a devinfo question, cross-repository links, squash-merged
stacks and version files, stale hand-built prebuilts, and before-merge
Pixel 7 Pro runs on omamobile-int. C-16 to C-18 correct our own earlier
comments on kernel #98, kernel #80 and #74; L-241 is corrected in place.

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

* docs/review: 15 lessons from the twenty-sixth round

L-279 to L-293: the 9a's slot B hold and what made devinfo writable
(#125), per-device flags checked by building both command lines
(#101), env defaults in the shared build library and CI waiting for
approval (#79), GLib retry paths and radio stops checked at the
controller (#99), silent re-pushes and stacks with several tops (the
tegu stack), and two of our own: merge only after every check has
finished (#126), and an empty diff is checked for empty inputs
(kernel #123).

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

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
@derekross

Copy link
Copy Markdown
Owner Author

Closing my own draft from 10-28: it never ran on the 7 Pro, and the shell has moved on about 60 versions since. I'll re-cut whatever's still wanted from it on today's main.

@derekross derekross closed this Nov 7, 2026
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