[angeloINTJ]

Closes F4 of v2.13.0's Known list: on the Ângulo, GET /api/keypad no longer describes the original panel's keypad. It sends the faces, the state and the PIN policy, and names /api/ui as the route that lists where the Ângulo's keys are. The two Ângulo images get 208 and 216 B smaller, and the other nine are byte for byte what they were.

The defect

On an image built with the Ângulo, WebManager::handleApiKeypad( ) published the original panel's keypad geometry. A client that taps by those numbers misses the Ângulo's keys.

field	depends on the panel?
faces, up, kb, policy	no: they are core state, and the Ângulo's key k is the original's key k
grid[0..3]: keys, slots, cols, rows	no
grid[4..9]: x0, y0, w, h, pitch	yes. With 4 cards the original's keys are 152 × 44 at (6, 72), pitch 156/50. The Ângulo's are 146 × 44 at (12, 96), pitch 150/48
pop (groups keyboard only)	yes. Its one key height cannot describe the Ângulo's popup: 48 px wide, 52 tall on one row and 36 on two

Nothing that ships reads the route: WebUI.h does not, and the Ângulo's bench script tools/tft_ui/roteiro_ui.py reads only faces and kb and taps by /api/ui. The two bench tools that drive the original panel did read it.

The change

The maintainer chose on 2026-10-31 among options (a) to (d) from the plan. His choice:

• The route, option (c). On the Ângulo it sends faces, up, kb, policy and "targets":"/api/ui", with no grid and no pop. On the original panel the answer is unchanged.
• The tools refuse an Ângulo answer, with a message that says where to tap instead.

How the route is switched:

• A UI hook, not an #if. The hook is uiKeypadTargets( ):
  • the original answers nullptr, in src/ui/UiDriverMembers.h;
  • the Ângulo answers "/api/ui", defined inside the class in src/ui/angulo/Members.h.
• The handler. It builds the Ângulo's answer when the hook answers. Otherwise today's statements run, textually unchanged.
• Each side is a constant, so the branch an image does not take is not in the image. The Ângulo images carry neither the grid's format string nor the popup's. The original images with the route carry no "targets".
• The sprawl count stays at 368. tools/ui_texts_allow.json names the route literal in Members.h as a route path, not text the panel draws.

The tools:

• tools/keypad_answer.py is one reader for tools/panel_users_hw_test.py and tools/panel_fulltable_test.py. It refuses an answer that has targets, and its message names /api/ui and roteiro_ui.py. An answer with neither grid nor targets, as a device older than v25 sends, still gets the v24 layout.
• Rig checks once at its start, right after the web login:
  • a run stops before it makes the test accounts or re-points the telemetry;
  • before it stops, it deletes the login's own account, smap.
• The keypad read inside a run refuses the same way. It deletes smap too, because this tool's own cleanup is a step that a failure skips.
• Documents: the CHANGELOG (English and Portuguese), the manual's REST chapter and the regenerated docs/MANUAL.pt-BR.html, AGENTS.md, docs/AUTHORIZATION.md, docs/MANUAL*.md, docs/analysis/UI_DRIVERS.md, tools/README.md and CLAUDE.md's script count.
  • The script count goes from 211 to 214: the two new files, plus one script main already had but miscounted. Its "Called by CI" heading said 73 over 74 rows.

Tests first

• tools/test_keypad_answer.py, in CI's gates job. It has four cases: an original answer, an Ângulo answer (refused), an answer older than v25, and the groups keyboard with its popup. It also checks two things in the tools' source:
  • each tool reads the route only through the reader;
  • each of the tool's two refusals deletes the web login's account before it stops.
• Red first: it was red before the reader existed, because the module was missing (eb192b74).
• The account check was red before each of its two fixes (both in 5b4d8001) and is green after. Run with serial and requests stubbed:
  • on an Ângulo answer, the real refuse_angulo( ) and keypad_faces( ) send user del smap before they stop;
  • on an original answer they send nothing.
• Mutation controls: with the refusal taken out, exactly the Ângulo case fails. Each of the other three lines the test holds fails exactly the cases it serves.
• The images' bytes, by a script outside the repository. On 6981fd93 it fails 6 checks, all on the two Ângulo images: no ,"targets":", and the "pop":[%d and "grid":[ format strings still there. On this branch all 49 of its checks pass. With the two hooks' values swapped, it fails again.

Images

All eleven built from zero on both sides through the pio shim. A reviewer rebuilt all eleven independently and got the same sha256 and sizes.

Image	6981fd93 .bin	this branch .bin	Δ
pico2_w_release	1,038,352	1,038,144	−208
pico_wiretft_angulo	829,200	828,984	−216
the other nine	—	—	0, same sha256

Signed, as the CHANGELOG gives them (+241 B): 1,038,385 and 829,225. handleApiKeypad goes from 508 to 348 B on the Pico 2 W. The two budget entries in tools/flash_budget.json and their rows in docs/analysis/PLANO_STABLE.md §3.1 come down in this change, and check_flash_budget.py --table agrees.

Gates

• ci_gates_local.py --raso at the tip: 76 of 77 steps passed at 5b4d8001. The fuzz step fails here only because this machine has no libclang, and the AddressSanitizer step runs only in CI.

• Also passed:

  • check_authz.py, with a --list identical to main's;
  • check_readme_numbers.py;
  • check_feature_sprawl.py at 368;
  • check_ui_boundary.py with 0 findings, and test_ui_boundary.py 316 of 316;
  • check_ui_texts.py with 0 findings;
  • check_ui_link.py with 0 findings on both Ângulo ELFs.

• The regenerated HTML. tools/build_manual.py is not run by CI, and this machine's pandoc drops the table row classes. A post-step restores them:

  • on main first: the generator's output plus the step equals the committed file byte for byte, sha256 ce4ef77f;
  • on this branch: the same holds for the committed file, sha256 0acd0772.

  The HTML's diff is 2 hunks, only the changed text.

Review

• First pass, COM RESSALVAS. The reviewer rebuilt all eleven images and found the firmware change correct. It found one important defect, in a bench tool: Rig.__init__( ) makes the web login's full-admin account before the Ângulo check, and the refusal's SystemExit came before the cleanup, so a refused run left smap on the device.
• Re-check of the fix: APROVADO, with one remnant: the keypad read inside the run had the same gap. This tool deletes smap only in a step that a failure skips, and its finally does not delete it. Both are fixed in 5b4d8001, test first: the check now reads every handler of the Ângulo's answer in the tool.
• Not repeated: the reviewer did not re-run the firmware build after the fix, because that commit touches no firmware input.

Not run on a board

No device was contacted. The answers are proved from the images' bytes and by the host tests. The bench check needs the maintainer's word to flash the Pico 2 W:

1. Before flashing, with the keypad on the glass, the current answer must still carry grid, and keypad_answer.py must accept it.
2. After flashing this branch's pico2_w_release, open the keypad by /api/ui's targets:
   • the answer must carry targets = /api/ui and no grid or pop;
   • a PIN typed by /api/ui's targets must reach Settings;
   • the reader must refuse the new answer.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Uy7K2mzEmmjBnpxpdYNDFu