340 commits

Author	SHA1	Message	Date
arena-agent	7d1a7316d2	feat(doc): render multi-line cell text on grown rows ... Cell values holding newlines (legacy strings, or fresh ones the RFC-4180 quoting round-trip now produces) rendered collapsed inline; the text round-tripped but every display line squeezed onto one band. One shared line model now threads layout, renderer, caret, highlight, hit test, and the keyboard surface: - layout_projected_table grows a row by one 18px text line height per extra display line of its tallest visible cell over the 28px baseline; the table rect and the block flow below follow. Column widths stay fixed and single-line tables lay out byte-identical (control assertions pin both). Merge composition: a covered cell's hidden text never inflates its row, and a vertical merge anchor sums the grown heights of the rows it spans. - The renderer draws styled runs segment by segment: an embedded newline in a run resets x to the inset and advances one line, keeping the whole text block vertically centered so single-line cells draw exactly where they did. - table_cell_caret, cell_text_span_rects (one band per covered display line, replacing the single-rect helper), and the point-based cell_char_offset_at (y picks the band, x midpoint- splits within it) all resolve through one cell_text_line_col / cell_text_offset_at pair whose round-trip is unit-tested at every boundary, including empty lines and the newline's own offset. - ArrowUp/ArrowDown, previously dead in cell mode, step between display lines keeping the visual column (clamped per line), Shift extending the in-cell selection; they stay inert at the first and last line and on single-line cells, so no implicit row exit and no half-moved cell ranges. Defect fixed in-phase: an in-cell character span covering a newline copied as a raw slice, so a paste re-distributed it across cells. The in-cell copy branch now quotes through the same quote_tabular_field as every other tabular payload; the Shift+ArrowDown runtime test pins the quoted payload end to end. Tests: line-math boundaries, row growth with block flow and merge composition, multi-line caret rects, per-line selection bands, point hit-testing clamps, vertical-arrow step/inertness/collapse, a real tap parking on the tapped display line, and the quoted span copy via copyable_selection_text and the TextCopy hit.	2026-08-02 10:36:28 +00:00
Arena Agent	e4fbd71d78	fix(cad): finish 1.9 and 4.6, found by auditing the whole plan ... Asked whether every phase was complete, I checked each row against the code instead of against my own record. Phases 0-5 were done except two leftovers that had been reported as finished and were not. 1.9 -- the dead binding was still there: let rzyx = makepad_widgets::Mat4f::identity(); // Simplified — use transform directly let rzyx = mat4_mul(...); // immediately shadows it Harmless to execution, but it reads as though the rotation is being skipped, in the one function that builds the model matrix -- in a module where a rotation bug has already shipped four times. Deleted, with the real computation formatted so the Z*Y*X order is legible and the degrees contract stated. (The other half of 1.9, add_part's placement, was genuinely done: the slot comes from the monotonic id, not parts.len().) 4.6 -- 10 `v18b rev2:` prefixes survived the archaeology sweep, in arch_gltf, arch_pdf, viewport and workspace. Same treatment as the other 73: keep what the code does, drop which internal revision introduced it. Now zero. Also marked the 29 Phase 0/1/2/4 rows that were complete but never recorded as such, with the specifics rather than a bare "DONE" -- 0.2 notes the lockfile is at the workspace root (a per-crate one would be ignored, since nigig-build is a member); 1.1 notes the rotation contract settled on DEGREES, not the radians the plan proposed; 4.3 notes it was superseded by Phase 5.4 rather than done as written. Every numbered row in the plan is now DONE, or REJECTED with the measurement or counter-example that closed it. 783 lib + 154 integration tests pass. All 13 CI gates pass.	2026-08-02 10:27:45 +00:00
andodeki	24d90b6855	feat(spreadsheet-ui): add retained cell render cache	2026-08-02 10:20:49 +00:00
Arena Agent	583fbbd092	docs(cad): record Phase 2 and 3 outcomes, including three rejections ... Phase 2 is complete (2.6 was the last open item). Phase 3 is complete in the sense that every item has been either done or measured and closed with a reason. Done: 3.2 world-AABB cache (5.9x), 3.4 redraw_all removal (53 calls), 3.6 script regeneration on commit (425 us/frame at 500 parts), 3.7 async exports, 3.9 grid batching (5,400 -> 1 tessellation per frame at 1080p), 2.6 save dialogs. 3.1, 3.5 were already done in earlier phases. Measured and rejected, with the numbers in the table: 3.3 ParamHash memoisation. 50 ns/node for a Box. The polygon case is real (987 ns) but it is the vertex data, and bulk-hashing measured no faster; the fix would be a data-model change. 3.8 Cost-estimate parallel threshold. 1.01x on a warm cache, which is the common case. Two of the completed items were not what the plan described, and the table now says so rather than quietly claiming the original wording: 3.9 the plan blamed the "nice number" step computation. That is already a cheap if-else chain. The cost was stroke()-per-dash. 3.2 the plan said to key the AABB cache on ParamHash. Doing that would have served a stale box after every drag, because ParamHash deliberately excludes the transform. Also documents the export architecture in ARCHITECTURE.md 2d, including why the 3D viewer and Bake stay directory-based -- both write companion file pairs that reference each other by name.	2026-08-02 10:14:24 +00:00
Arena Agent	d61e215e9e	perf(cad): batch the 2D grid into one stroke (3.9); measure 3.3 and 3.8 ... Phase 3.9 -- the real cost was not the "nice number" step computation the plan named. That is already a cheap if-else chain, no log10/pow. It was draw_dashed_line calling stroke() after every 6px dash, and stroke() tessellates the entire accumulated path each time. A full-screen grid is ~50 lines of ~108 dashes: 5,400 tessellations per frame at 1080p, 14,688 at 4K. Split into queue_dashed_line (appends to the path) and draw_dashed_line (queues, then strokes) so single-line callers are unchanged. The grid queues every dash and strokes once. Bubble markers are collected and drawn after, not inside the loop -- they use draw_text and their own fills, which would otherwise land in the middle of the grid's path. Also one String allocation per grid line instead of two. the_2d_grid_strokes_once_not_once_per_dash pins it. My first version asserted exactly one stroke in the whole function and failed with 3: the work-plane cross below the grid is a separate feature with its own colour and correctly gets its own strokes. Scoped the assertion to the grid rather than weakening it. Negative test: swapping one queue_dashed_line back to draw_dashed_line fails it. --- Phase 3.3 (memoise ParamHash on CadNode): MEASURED, NOT DONE. Box: 50 ns/node -> 25 us/frame at 500 parts Extruded 64-gon: 987 ns/node -> 493 us/frame The Box case does not justify a cached field that every mutation would have to invalidate -- the exact hazard Phase 5.4 removed from part_geoms. The polygon case is the vertex data itself: I tried bulk-hashing the slice as raw bytes and measured 125 us vs 128 us for 500 x 64 verts, i.e. nothing. The only real fix is to give polygons an Arc identity the way Csg already has, which is a data-model change, not a cache. Benchmark kept so the next person starts from numbers. Phase 3.8 (cost estimate instead of node count): MEASURED, NOT DONE. 200 nodes, cold cache: 8.80ms seq / 6.21ms par -> 1.42x 200 nodes, warm cache: 5.81ms seq / 5.73ms par -> 1.01x On a warm cache -- the common case, since the preview renderer has already built every mesh -- parallel neither helps nor hurts. A cost estimate would have to hash every node to count cache misses, in order to choose between two paths that differ by 1% in the case it would most often face. The threshold comment now carries these numbers instead of "can be tuned based on real-world profiling". 783 lib + 154 integration tests pass. All 13 CI gates pass.	2026-08-02 10:13:13 +00:00
Arena Agent	bdf882b817	perf(cad): regenerate the parts script on drag commit, not per frame (3.6) ... `script_dirty` was set on every MouseMove and FingerMove of a part drag. That makes `sync_parts_from_any_dirty_viewport` call `generate_parts_script()` -- formatting every part into a String -- and the workspace then replaces the entire editor document via `set_editor_text_all`. Per motion event. Measured before changing it: 42 us at 50 parts, 170 us at 200, 425 us at 500, and that is the string formatting alone, before the code editor's own work. See bench_parts_script_regeneration_per_drag_frame. The reason it was set mid-drag no longer holds. The comment said it kept the split 2D/3D viewports in sync while dragging -- true when each viewport owned its own parts list, but since Phase 5.2 all three share one CadDocument. A move IS their state the moment it happens; they need a repaint, not a resync, and they get one. Both commit paths already set the flag: the MouseUp arm for mouse drags, and finish_part_drag for touch (reached from three places). So the script still regenerates exactly when it needs to -- once, when the edit is final. a_drag_regenerates_the_script_on_commit_not_per_frame pins it. It walks every arm that calls move_selected and asserts none of them set script_dirty, then asserts the commit paths still exist -- because the failure mode of this change is not "slow", it is "the script never updates at all", and a test that only checked the first half would miss it. Negative test: putting the assignment back fails it with the line number. 782 lib + 154 integration tests pass. All 13 CI gates pass.	2026-08-02 10:05:45 +00:00
andodeki	93faa96920	perf(spreadsheet-ui): invalidate grid after sheet switches	2026-08-02 10:03:16 +00:00
Arena Agent	84ed7d2e43	perf(cad): drop 53 redundant cx.redraw_all() calls (Phase 3.4) ... cx.redraw_all() sets a flag that repaints every widget in the application. Every one of the 52 calls in viewport.rs, and the 1 in viewport_2d.rs, sat DIRECTLY after `self.area.redraw(cx)` -- the targeted redraw was already there and the full-app repaint added nothing. Verified mechanically before deleting: a scan for `cx.redraw_all()` not preceded by `area.redraw(cx)` returns zero hits in both files. On a drag this ran per motion event: a whole-application relayout to move one part. What I did NOT touch, and why: workspace.rs (15) cross-widget coordination. Both viewport sync paths end in a redraw of the OTHER viewports, and that is what makes removing the viewport's own calls safe. Removing these would be a different change with a different argument. viewport_input.rs (28) event paths; 13 are not paired with an area.redraw at all, so each needs reading on its own terms rather than a bulk edit. cad_editor_sheet.rs (2) not the viewport. The risk here is a missed repaint, which no test can see, so I checked the mechanism rather than relying on the suite staying green: cross- viewport repaint runs through sync_parts_from_any_dirty_viewport (which calls vp.redraw on each destination) and sync_view_from_any_dirty_viewport (which ends in its own redraw_all). Both live in workspace.rs and are untouched. the_viewport_does_not_ask_the_whole_app_to_repaint pins it as a source check, because asserting on repaints needs a live Cx the suite does not have. Negative test: reintroducing one pairing in viewport_2d.rs fails it with the file and line named. 781 lib + 154 integration tests pass. All 13 CI gates pass.	2026-08-02 10:02:20 +00:00
Arena Agent	73c4c49cb9	perf(cad): cache the world AABB per placement -- 230us -> 39us (Phase 3.2) ... Measured before building, because I had previously dismissed this item as "smaller" without checking. It is 5.9x at 500 parts, and hover picking runs on mouse-move, so it is a per-frame cost. pick_part's broad phase transformed 8 local corners by the model matrix for every part on every pick. Phase 5.3 had already removed the expensive half (it no longer re-meshes to read bounds), leaving 8 matrix multiplies per part -- cheap individually, 230 us/frame at 500 parts. SceneCache::world_aabb_for now memoises the result. The key is a NEW type, PlacedHash, not the existing ParamHash. This is the whole subtlety of the change: ParamHash deliberately excludes the transform, because a local-space mesh cannot change when a part moves (Phase 5.4). A world-space AABB is exactly the opposite -- moving the part is the entire point. Reusing ParamHash here would serve a stale box after every drag and make parts unpickable at their new position, which is the picking equivalent of the stale part_geoms bug. Making it a distinct type rather than "ParamHash plus a flag" means the two cannot be confused at a call site. a_move_invalidates_the_world_aabb_even_though_it_keeps_the_mesh pins the asymmetry directly: the same move that rebuilds the AABB must still hit the mesh cache. Negative test: making PlacedHash ignore the transform -- i.e. reverting it to ParamHash -- fails that test. Restored and green. retain_world_aabbs is paired with every retain_meshes call site, for the same reason that one exists: the map is keyed by NodeId and nothing drops an entry when its node is deleted, so without it the map grows for the session. 775 lib + 154 integration tests pass. All 13 CI gates pass.	2026-08-02 09:58:23 +00:00
arena-agent	54e1148b39	feat(doc): round-trip tabular clipboard payloads with RFC-4180 quoting ... Cells holding tabs, newlines, CRs, or quotes re-distributed across the grid on a copy/paste cycle — the documented caveat of every tabular clipboard milestone. This closes it on both sides: - table_grid_tsv quotes such fields on the way out (wrapped in double quotes, inner quotes doubled) via a new quote_tabular_field helper; plain and empty fields stay raw, so payloads stay byte-compatible with spreadsheets and plain text editors. The cell-range payload and the block-span document payload share the one builder, so both inherit the quoting at once. - paste_table_payload replaces its split('\n')/split('\t') walk with split_tabular_payload, an RFC-4180-style tokenizer: quotes open only at field start (mid-field quotes are literal), doubled quotes read as one, tabs/newlines/CRs inside quotes are literal field text, CRLF rows outside quotes keep their tolerance, an unterminated quote reads to the end as best effort, and a single trailing newline adds no phantom row (a deliberate empty row survives). - Caret parking, no-op skipping, empty-field clears, and the one-undo grouped write semantics of the raw paste milestone are unchanged; the caret offset in a multi-line value counts the newline too. Tests: unit coverage for the writer and the tokenizer (every quoting rule plus the quote/split round-trip property), runtime coverage of copy quoting, paste restoring embedded tab/newline values verbatim with one-undo and redo, and an end-to-end copy-cut-paste cycle; a doc-engine materialize test pins special-character cell text surviving peer sync and undo/redo verbatim. README caveats updated and the milestone documented.	2026-08-02 09:54:40 +00:00
Arena Agent	07cef07dfb	feat(cad): async exports with a save dialog (Phase 2.6 + 3.7) ... Both plan items are the same call, so doing them separately would mean writing the plumbing twice: the file picker's `save_data` takes owned bytes, and serialising off the UI thread is what produces owned bytes. Before: every export serialised on the UI thread straight into a File at a hardcoded path -- generated/stl/model.stl, generated/3d/model.glb, generated/floor_plan.pdf. A large scene froze the editor for the whole serialise, and the second export of a session silently destroyed the first. Now: PDF, STL spawn_export_to_target -> worker thread -> save dialog SVG stays sync (the preview widget needs the bytes on the UI thread, so there is nothing to move) but gains the dialog CLI already a String in memory; gains the dialog 3D viewer async, but stays directory-based ON PURPOSE: it writes TWO files and viewer.html references model.glb by relative name, so renaming the GLB through a picker would break it Bake stays directory-based: writes the parts.obj/parts.cad pair, a workspace artefact rather than a document, and Solid::write_obj takes a path not a writer Results come back through an mpsc channel drained on NextFrame, the same mechanism the rebuild worker already uses -- not a second bespoke one. The status says "exporting…" while in flight, and the completion message for a dialog export says "ready — choose where to save" rather than claiming the file is written, because at that point it is not. Reporting success before the write is the exact bug fixed in the Save buttons earlier. The 3D companion HTML is now only written if the GLB actually landed. Previously a truncated GLB still got a viewer.html beside it, which is how "export succeeded" turned into a blank page. ExportTarget::suggest gives each export a distinct default name (<stem>-<project>-<counter>.<ext>) so successive exports do not propose to overwrite each other. sanitize_file_stem exists because a project name is user text that ends up in a save dialog. It can contain a path separator, "..", a NUL, a leading dash, or 300 characters of emoji. The test found a real bug in my first version: replacing "/" with "_" turns "../../etc/passwd" into "_.._.._etc_passwd", so trimming leading dots BEFORE the replacement leaves "_.." behind. Trim after, and include "_". Deleted in the same commit as their cause: write_floor_plan_pdf and export_to_file, both now unreachable. Verified no callers remain anywhere in crates/. New tests: bytes land at the requested path; a failed write is reported rather than swallowed (parent is a regular file, so create_dir_all cannot succeed); an empty export still produces a file; the hostile project-name corpus; successive suggestions do not collide. 772 lib + 154 integration tests pass. All 13 CI gates pass.	2026-08-02 09:52:27 +00:00
andodeki	52eff6167d	perf(spreadsheet-ui): invalidate full grid after scrolling	2026-08-02 09:51:13 +00:00
andodeki	7fdf436510	test(pay): Phase 5 lifecycle matrix as domain tests (R2.3) ... The exit criterion names five scenarios: permission denial, cancellation, backgrounding, app restart, out-of-order callbacks. Four of the five are state questions, not hardware questions. A device adds confidence that Android really emits a given callback sequence; it cannot tell you how the domain reacts, because the state machine decides that. So the matrix runs against the real coordinator on every commit instead of when a phone is free, and a regression names the invariant it broke. 13 tests in crates/nigig-pay-domain/tests/lifecycle_matrix.rs, including the cases that only exist as races: a success arriving after a cancellation; backgrounding before a grant (must refuse) versus after one (must be preserved — the user did authorise); restart before dispatch versus after; a foreign grant; a replayed grant. Plus a clean-path test so the matrix cannot pass by refusing everything. ## A coverage hole the matrix found a_restart_after_dispatch_cannot_redispatch passed with the duplicate-dispatch budget removed. The state machine refuses Submitted -> Dispatching first, so the budget was never reached. That is good defence in depth and bad coverage — nothing proved the budget still worked. the_dispatch_budget_survives_a_state_machine_walk_back forces the intent back to Dispatching, exactly as a faulty recovery path would, leaving the budget as the only guard. It fails when the budget is removed. The forcing hook is behind a `test-hooks` feature, not #[cfg(test)]: an integration test is a separate crate and does not see cfg(test), so the method was simply missing. The isolated runner enables it explicitly, otherwise that test is silently filtered out and proves nothing. ## Verified by injection authorization gate removed -> 7 of 13 fail dispatch budget removed -> 1 fails (the new one) ## What still needs hardware That Android actually produces these sequences: permission dialogs, process-death timing, callback ordering under memory pressure. This file asserts the response is correct for each sequence; a device confirms the sequences are the real ones. Different claims, both needed. Tracked as R2.3b. ## Validation domain 148 unit + 13 matrix, fmt, clippy -D warnings, bench pass storage 46 / platform 64 / mpesa 29 / pay-ui 78 pass clippy -p nigig-pay-ui --no-deps -D warnings 0 errors	2026-08-02 09:47:52 +00:00
andodeki	703bba3652	perf(spreadsheet-ui): distinguish dirty cells and full invalidation	2026-08-02 09:38:43 +00:00
andodeki	d567978119	feat(pay): key rotation for the encrypted ledger (R2.2) ... Examined R2.2 the way R2.1 turned out to need, rather than assuming the whole item was device-blocked. Most of 3.1 was already present: DatabaseKeyProvider, open_encrypted, wrong-key rejection distinct from corruption, keystore-unavailable failing closed, and a test asserting no PII appears in the raw file. ## The real gap was rotation, and it is pure logic A StaticTestKeyProvider key lives forever. An Android Keystore key does not: KeyPermanentlyInvalidatedException is thrown after fingerprint re-enrolment, adding or removing a screen lock, or a device restore. That is ordinary, not exceptional. With no rotation path the only responses were "lose the ledger" or "keep using a key that no longer exists" — and the second is not available, because the key is gone. Deleting the ledger is not an option either. It destroys the record of money that may have left the account, which is the same reasoning that makes retention.rs refuse to sweep unreconciled rows. rotate_key uses PRAGMA rekey, which re-encrypts every page inside SQLCipher's own transaction, then proves the new key reads the data before returning — a rekey that reported success but left the file unreadable would otherwise only surface on the next launch, by which time the old key may be gone. 5 tests: records survive rotation, the superseded key stops working, an empty key is refused without damaging the file, rotation is repeatable, and the schema version is untouched. Verified by neutering rotate_key: 3 fail. Storage tests 41 -> 46. ## Ordering, documented at the trait The caller persists the new key only after rotate_key returns Ok. The reverse order leaves a stored key that does not open the file. This order leaves, at worst, a re-keyed file whose new key was not saved — recoverable by rotating again from the old key still in the keystore. ## What remains The JNI call itself: KeyGenParameterSpec with user authentication required, the AndroidKeyStore provider, and catching KeyPermanentlyInvalidatedException. The full contract is written on DatabaseKeyProvider so it is not rediscovered from scratch. Tracked as R2.2b. Everything except the platform call is already exercised by the sqlcipher suite. ## Validation storage 46 (was 41) with --features sqlcipher pass domain 148 / platform 64 / mpesa 29 / pay-ui 78 pass clippy -p nigig-pay-ui --no-deps -D warnings 0 errors builds: pay, mpesa, core pass rotation injection: 3 tests fail when rotate_key is neutered pass	2026-08-02 09:37:00 +00:00
arena-agent	a66d7d2198	feat(doc): carry table grids in document-level clipboard payloads ... A block-span selection — Shift+Arrow across a table, select-all, any multi-block drag — used to copy blank lines where tables sat, so copying a document lost every table's content. Table blocks now contribute their whole grid at their block position as tab/newline lines, built by the same table_grid_tsv helper the cell-range payload uses (extracted from cell_range_clipboard_text): one builder, one convention, no drift between "copy a range" and "copy across a table". Stored cell text exports verbatim — a merge's covered cells keep their hidden values — and empty tables still contribute nothing. Cutting such a span was already structurally correct through replace_block_range / CancelBlockRange, so the milestone is payload-only: the payload now matches what actually disappears — verified by a cut over [paragraph, 2x2 table, paragraph] draining the document with "lead\na\tbc\nd\te\ntail" in the payload and one undo restoring blocks AND every cell value. Also covered: select-all through the TextCopy hit, and a partial mid-paragraph span splicing the grid between its text fragments in order. The stale "tables are skipped" select-all bullet is retired.	2026-08-02 09:35:13 +00:00
andodeki	85f21a6f35	perf(spreadsheet-ui): consume dirty cells during intent dispatch	2026-08-02 09:34:16 +00:00
Arena Agent	c6e7635c5a	test(cad): fuzz the coordinate parser; audit all 84 indexing sites ... I dismissed indexing_slicing as "mechanical churn" without checking. That was an unchecked claim about 105 reported panic sites, which is exactly what I have been objecting to elsewhere in this codebase. Audited all of them. 105 clippy hits are 84 unique lines. Every one is bounded: 43 fixed-size arrays indexed by a literal or a 0..N loop whose bound matches the array -- mat4_mul, mat4_inverse, ray_aabb_intersect, the 8-corner projected box. Unindexable by construction. ~30 behind an explicit length check in the same function: verts.len() >= 12 (I-beam web), >= 8 (HSS inner wall), pts.len() == 4, chamfer_rect's `< 4` early return, polygon_area's `n < 3`, pick_part's per-triangle bounds test. ~11 `% len()` on a non-empty slice, or selection[i] over 0..selection.len().min(3). Converting these to .get() adds ~84 `else { continue }` arms guarding conditions the compiler or an adjacent check already rules out, each one a place to get the fallback subtly wrong. Not gated; the audit is recorded in ARCHITECTURE.md 2c so the next reader gets the evidence rather than the dismissal. The one genuinely input-facing site is fuzzed instead. parse_coord_input takes raw text from the coordinate box on every keystroke and indexes parts[0..2] after a split. Two new tests: an adversarial corpus (multi-byte leading characters, lone separators, RTL and combining marks, 500-char inputs, inf/NaN/1e400) and every char-boundary prefix of a valid input, because the box parses as you type. Building the corpus is where the actual work was. My first version -- garbage strings -- passed even with the length guards deleted, because a first component that fails to parse makes `?` return before the second index is evaluated. It looked like a strong test and tested nothing. The cases that reach the guards are the ones whose FIRST component is valid: "@5", "5<", "@5<45". Negative test, per guard: spherical parts.len() == 3 removed -> FAILS, index out of bounds polar parts.len() == 2 removed -> FAILS, index out of bounds relative parts.len() >= 2 removed -> still passes, and that is correct: the branch is gated on contains(','), and a string containing a comma always splits into at least two parts, so the check is redundant. Verified rather than assumed; left in place because it states intent locally. 763 lib + 154 integration tests pass. All 13 CI gates pass.	2026-08-02 09:33:31 +00:00
andodeki	874a8481fa	perf(spreadsheet-ui): track dirty grid cells	2026-08-02 09:23:13 +00:00
andodeki	1d3e6ab72a	fix(pay): correlate USSD callbacks to the payment that asked for them ... R2.1. Review items 2.7 and 5.3. ## SessionRegistry was built in Phase 5 and never wired The pump still read: while let Some(ev) = robius_ussd::next_event() { ... if let Some(id) = h.current.take() { ... } } next_event() drains a process-wide queue and its entries carry no session id, so every event was applied to whatever `current` happened to be. Reproduced before changing anything: payment A is dispatched then abandoned with events still queued; payment B starts; the pump drains A's ResultText and SessionEnded and applies both to B. An abandoned payment settles the one that replaced it. ## Now - Dispatch claims the single in-flight slot. The USSD backend returns no session handle, so the intent id is the correlation id — enough, because the registry only has to tell this payment from the previous one. - Every event is admitted against the live operation before it can touch an intent. Foreign and stale events are logged and dropped. - Terminal events are de-duplicated; progress chatter still repeats freely. ussd_duplicate_key mirrors ProviderSignal::duplicate_key in the platform crate, and a test pins the two together. - All six terminal and teardown paths retire the session id, so a late duplicate cannot revive a closed operation. 6 tests, including the abandoned-payment scenario by name. Verified by removing the close call: that test goes red. CI gate asserts the pump still admits, dispatch still claims, and at least six paths still close — matching the method rather than a receiver literal, because rustfmt wraps the call. ## What this does not do The pump still lives in PayFlowHandler, which still owns the pending-store writes and the bulk queue. Moving *ownership* to PaymentCoordinator changes who cancels on teardown and who observes an out-of-order callback, which is what ADR 0007's device matrix exists to check. That is now tracked as R2.1b. The correlation defect — the one that could settle the wrong payment — is closed, and it did not need a device. I had previously filed the whole of R2.1 as device-blocked; that was too coarse. ## Validation nigig-pay-ui 78 (was 72) / nigig-mpesa 20 pass domain 148 / storage 41 / platform 64 / mpesa 29 pass clippy -p nigig-pay-ui --no-deps -D warnings 0 errors builds: pay-ui, pay, mpesa, core; default and --no-default pass correlation injection: abandoned-session test fails without it pass pin-capture guard pass	2026-08-02 09:21:45 +00:00
Arena Agent	072dcde979	docs(cad): record Phase 5/6 outcomes, including the two rejections ... Updates the file table in ARCHITECTURE.md for the three new files and corrects the viewport.rs/workspace.rs descriptions, which still claimed to hold the rendering and input code that moved out. A doc that lies is worse than no doc. Marks the plan items honestly: 5.5 DONE viewport.rs 8,023 -> 4,796. The original six-file target was not met and is not being pursued: the remaining 4,796 lines have no seam comparable to the two that were taken. 5.6 DONE workspace.rs 3,991 -> 3,273. 5.7 REJECTED after counting. "Branched on in 40 methods" was 14, and 21 of the 37 branches were in one function. 5.8 REJECTED. The premise is false -- kind is not derivable from the solid, because six PartKinds share CadSolid::Box. Phase 6 marked partially done, with what is left stated plainly: the indexing_slicing ratchet is 105 sites in CAD, mechanical churn rather than defect-finding, and the panic family it was really aimed at is now at zero and gated. Recording the rejections in the plan matters as much as the completions: both items would have destroyed something real, and the next reader should find the counter-evidence rather than the instruction.	2026-08-02 09:21:22 +00:00
Arena Agent	015cf44422	fix(cad): remove the reachable panics; gate unwrap/expect at 5 ... Phase 6, scoped to what is provable rather than a blanket -D warnings. Measured the CAD module first: ~200 clippy warnings, but the panic family -- the part the plan actually cared about, copying the pay crates' ratchet -- was only 8: 2 unwrap, 6 expect, 0 panic!. Three were real, five are genuine constructor invariants. Fixed: - code_editor.rs x2. `lazy_init_session(); self.session.as_mut().unwrap()` in both draw_walk and handle_event. Correct today, but the guarantee lived across a function boundary the compiler cannot see, so an unwrap sat on a widget draw path waiting for a third caller to forget the prologue -- and a panic there kills the editor with unsaved work in it. Added `editor_and_session()`, which splits the borrow and returns Option, so both sites take an early return instead. I first tried folding init into `get_or_insert_with`. That silently dropped the `keep_cursor_in_view = Once` side effect, which only happens on the create path. Caught it by grepping for the field rather than trusting the refactor; reverted. - cad_scene.rs x1. MeshCache::get_or_build did `.write().expect("mesh cache poisoned")` while every other method on the type already degraded with `if let Ok(..)`. Reachable: the export path calls get_or_build on a spawned thread, so one panicking worker poisoned the lock and the next draw took the UI thread down with it. The cache is pure derived data -- every entry rebuilds from its node -- so a poisoned lock now costs memoisation, not correctness. The mesh is built before the lock is taken, and the double-check still prefers a racing thread's entry so Arc::ptr_eq comparisons stay consistent. Left alone, with reasons: 4 x cad_scene "default material/layer always exists" (SceneBuilder::new inserts both; verified) and 1 x arch_gltf serde_json::to_vec over a Value built in that file. New gate: "No new unwrap/expect in CAD production code", allowlist of 5. A bare count drifts upward quietly and a blanket ban just gets #[allow]-ed, so the count is pinned and each exemption is named in the comment. The gate skips #[cfg(test)] by BRACE DEPTH rather than stopping at the first one. That matters: arch_gltf.rs has production code after two test modules, so the existing panicking-macro gate's "stop at first #[cfg(test)]" awk cannot see line 850 at all. My first attempt used the same awk idiom and reported 4 of 5 -- I only noticed because the number disagreed with clippy. Verified the older macro gate is not currently hiding anything, but it is hiding it by luck. Both negative tests pass: an unwrap added to viewport.rs is caught, and one added to arch_gltf.rs *after* its test modules -- the exact blind spot -- is also caught, named with file and line. 13 gates now, all green. 761 lib + 154 integration tests pass.	2026-08-02 09:20:12 +00:00
arena-agent	4df2193859	docs(doc): retire stale in-cell paste and migration-status claims ... Tabular paste superseded the clipboard section's "in-cell pastes always stay a single whole-cell write, newlines included", and the doc-engine README still framed the controller migration as "next" even though the workspace navigation already runs on CrdtDocEditor (the classic editor stays registered as a fallback). Found during the phase audit.	2026-08-02 09:17:57 +00:00
Arena Agent	b26f4c8b91	refactor(cad): split handle_event into pointer and touch dispatch ... Phase 5.7. handle_event was 952 lines holding two independent dispatch loops -- one over raw Event (mouse buttons, motion, wheel), one over hit-tested Hit (finger down/move/up/scroll). They share no locals, so the seam was already there. The plan item asked to "consider making 2D/3D two widgets" because the view mode was "branched on in 40 methods". I counted before acting: 14 methods, 37 branches, and 21 of those branches are inside handle_event alone. The remainder are mostly one-liners picking a pan axis or a zoom helper. Two widgets would have duplicated the entire input layer to delete a handful of matches! guards, and doubled the surface the shared CadDocument must keep coherent -- which is precisely what Phase 5.2 was spent removing. The size problem was the event handler, not the enum. Split the handler; keep the enum. Reasoning recorded in the module doc so the plan item is not re-attempted from its original framing. Verified as a move rather than a rewrite: the sequence of Event::/Hit:: match arms across viewport.rs + viewport_input.rs is identical to HEAD's -- 41 arms, same order -- and the 3D-only desktop camera fall-through appears exactly once, as before. Order between the two loops is preserved and load-bearing: pointer arms set drag state the hit arms read. viewport.rs: 5,702 -> 4,796 lines (8,023 at the start of Phase 5.5). 761 lib + 154 integration tests pass. All 12 CI gates pass.	2026-08-02 09:14:22 +00:00
Arena Agent	f02de5645f	refactor(cad): split handle_actions into three per-panel handlers ... Phase 5.6. `CadWorkspace::handle_actions` was a single 1,133-line method: one flat sequence of `if button.clicked(actions)` arms covering every control in the editor. Adding a control meant reading all of it to find where the related ones lived. Split along the seams the comments already marked -- the panel groupings existed, they just were not expressed in the code: handle_view_and_plane_actions work/inclined planes, per-axis grids, reference and clip planes, construction toggles, coordinate input handle_export_and_file_actions CLI/OBJ/PDF/3D/STL/SVG export, the AI attach-image and model controls, open/save/save-as handle_properties_panel_actions colour swatches, layer/material dropdowns, per-axis position/rotation/ size inputs, DOF toggles Order is the risk in splitting a sequential handler: several arms depend on running after an earlier one has updated state. Verified mechanically rather than by reading -- the full sequence of `ids!(..)` widget references through handle_actions plus the three extracted bodies is byte-identical to HEAD's, 147 references in the same order. 19 helpers became pub(super). That is not a widening: they were already reachable from anywhere in the cad module, and pub(super) keeps them exactly there. Nothing was made `pub`. workspace.rs: 3,991 -> 3,273 lines. 761 lib + 154 integration tests pass. All 12 CI gates pass.	2026-08-02 09:11:20 +00:00
andodeki	648c662a1e	perf(spreadsheet-ui): cache visible text measurements	2026-08-02 09:10:46 +00:00
Arena Agent	1bc5d792a8	refactor(cad): extract the 25 draw_* methods into viewport_render.rs ... Phase 5.5, first cut. viewport.rs was 8,023 lines with a single 5,930-line `impl CadViewport`. The drawing methods are the largest coherent seam in it: 25 methods, 2,321 lines, that only read editor state and emit draw calls. Rust merges inherent impl blocks for the same type across files, so this is a pure move -- no signature changed and no caller was touched. Verified mechanically rather than by eye: the set of method names across viewport.rs + viewport_render.rs is byte-identical to the set in viewport.rs at HEAD, 187 before and 187 after, none lost, none gained. Two items needed pub(super) -- `part_model_matrix_cadnode` and `DrawCadMesh::draw`. That is not a widening: both were already reachable from anywhere in the cad module, and pub(super) keeps them there. No item was made `pub`. The module doc states the rule for what belongs in the file: a method lives here if it takes &mut Cx2d/&mut Cx3d and emits drawing commands. Anything that DECIDES what to draw -- picking, snapping, hit-testing, tool state -- stays put. Deliberately mechanical, because the previous attempts to split this file were judgement calls and did not hold. viewport.rs: 8,023 -> 5,702 lines. --- Phase 5.8 (remove kind_hint): REJECTED, with the counter-example recorded as a test rather than a comment. The plan item says to "derive kind from the solid variant". That is not possible. Six PartKinds -- Cube, Wall, Slab, Door, Window, Beam -- all build CadSolid::Box, and Cylinder/Column both build CadSolid::Cylinder. kind_hint is not duplicated state; it is the only record of which one the user asked for. Removing it would silently downgrade every Wall, Slab, Door, Window and Beam to a Cube, taking the properties panel's thickness controls and the Extrude tool's kind filter with it. several_part_kinds_share_one_solid_variant_so_kind_is_not_derivable asserts the collision directly, so the next person to read that plan item finds the evidence instead of executing it. I also built a script_tag/from_script_tag round-trip so the kind could survive a save, then deleted it before committing: nothing reads parts.cad back, so it would have been a write-only annotation -- the exact "add an abstraction, declare the migration done" pattern this phase exists to stop. The real gap is that the parts list has no serialisation at all, which is a feature, not a cleanup. 756 lib + 154 integration tests pass. All 12 CI gates pass.	2026-08-02 09:08:27 +00:00
arena-agent	c5c1e0a3e9	feat(doc): paste tabular payloads across table cells ... A TextInput carrying tabs or newlines while the caret lives in a table now pastes the spreadsheet way — one cell per tab stop, one row per line — starting at the caret cell, or at the armed range's normalized top-left (consuming the range like any paste-over- selection). The distribution rides the grouped set_table_cells from the previous milestone so the rectangle un-pastes in one undo step, and the caret parks at the last cell the payload touched. Rows or columns past the table edge clip, CRLF strips per line like text-block pastes, empty fields clear their targets, and unchanged cells are skipped so a paste lands neither redundant LWW ops nor dead group members. Plain payloads keep the whole-cell char splice, and cells still never spawn blocks. The pre-existing in-cell paste test asserting the old "newline stays embedded in the cell" behavior is rewritten to the distribution semantics; new runtime tests cover the 2x2 distribution with caret parking and the one-undo/redo round trip, edge clipping without wrap-around, backward-spanned ranges pasting from their top-left, CRLF with empty-field clears, and the mobile menu's Paste action distributing from the pressed cell. Engine integration coverage replays a grouped multi-cell write into a peer and asserts the projections converge.	2026-08-02 09:02:05 +00:00
Arena Agent	5ffc515f91	perf(cad): build the command scene snapshot lazily -- 254us -> 0.8us ... CadCommandCtx::new eagerly called scene_cache.scene_for(parts), which is O(nodes): every node cloned, every material re-registered. No production command reads that scene -- MoveNode, ResizeNode, RotateNode, YawNode, ModifyNode, CreateNode and DeleteNode all address a node by id. Only tests and the benchmark call ctx.scene(). Every command mutation bumps the PartsStore generation, so the next context construction was a guaranteed cache miss. move_selected issues one command per selected part per frame, making a drag O(commands x nodes) for work that is O(1) per command. This arrived with Phase 5.1, when the generation counter turned a previously accidental cache hit into a guaranteed miss; the 0.4us baseline predated it. The snapshot is now built on the first scene() call and memoised. CommandContext::scene() returns Arc<CadScene> rather than &CadScene so an implementation can build on demand instead of keeping one alive for the context's lifetime. This was also a latent CORRECTNESS bug, not only waste. The eager snapshot was captured BEFORE the command ran, so a command that mutated and then read scene() saw its own edit missing. Every mutating method now drops the memo; those invalidate_snapshot() calls are placed directly beneath the existing scene_cache.mark_dirty() calls so a new mutator cannot silently miss one. Two tests, both verified to fail against the old code: - a_command_that_never_reads_the_scene_does_not_build_one asserts on SceneCache::rebuild_count (a new #[cfg(test)] counter) rather than wall-clock, so it is deterministic rather than machine-dependent. Fails 20-vs-0 when construction is made eager again. - a_lazily_built_scene_reflects_edits_made_earlier_in_the_same_context reads the scene BEFORE mutating, then again after. Reading only after the mutation passes even with invalidate_snapshot() gutted -- the lazy build simply happens later -- so the first read is what gives the test teeth. I checked: the obvious version of this test was vacuous. The benchmark was also measuring the wrong thing. It called ctx.scene() each iteration to read the start position, which no production path does: move_selected and finish_part_drag both read p.pos() off the parts list. Fixing the lazy build alone moved undo 228us -> 0.5us but left execute at 231us, because the benchmark was timing its own scene read. It now mirrors the production callers. Measured: execute 254us -> 0.8us, undo 246us -> 0.5us (1000 commands over a 1000-node scene), ~300x. No other benchmark regressed. 755 lib + 154 integration tests pass. Test-name list diffed: +2, nothing dropped. All 12 CI gates pass.	2026-08-02 08:57:10 +00:00
andodeki	f33e999991	perf(spreadsheet-ui): improve unicode text width estimate	2026-08-02 08:53:26 +00:00
andodeki	5800beb552	fix(pay): close the R1 gaps against the completion standard ... You are right that my first pass at R1 fell short. It deferred an item on a judgement call, and it fixed three defects without regression tests naming them. Four gaps, all closed here. ## 1. S8 was deferred; it is now done as far as the platform allows I skipped certificate pinning as "wasted work if the endpoints get dropped". That was my call to make about effort, not an external blocker. Investigated properly: Makepad's HttpRequest exposes no pinning API. Its only TLS control is set_ignore_ssl_cert, which weakens verification. Pinning is not implementable at this layer without patching the platform crate. What *is* enforceable is the property pinning mostly buys — that a mistyped, injected or attacker-supplied URL cannot be dialled. check_transport gates every request on HTTPS plus a four-host allowlist, at all three dial sites in both copies of the client. 7 tests: lookalike hosts (api.coingecko.com.evil.example), embedded credentials (https://evil@real/), explicit ports, plain HTTP, malformed URLs, and an assertion that TLS is never disabled. Verified by disabling the allowlist: 3 tests fail. ## 2. The 13-digit phone defect had no test naming it I fixed it and moved on. It now has a regression test quoting the original duplicated branches, plus a property test that normalisation output is either empty or exactly a valid 10-digit 07/01 number — no third outcome. ## 3. The fee-policy UI wiring was untested The domain guard had 11 tests; the wiring that connects it to the pay sheet had none, so nothing proved the sheet actually consults it. Four tests now cover the shipped policy: it identifies the bundled tariff, refuses once stale, still quotes while current, and keeps "unknown band" distinct from "stale table". ## 4. The exchange client had no tests at all It does now, via the transport module above. ## A test that failed against itself tls_verification_is_never_disabled_in_this_module asserts the module never calls set_ignore_ssl_cert — and the literal in the assertion put the string in the file, so it failed on first run. The needle is now assembled at runtime. Recorded because it is exactly the kind of thing that gets "fixed" by deleting the test. ## Completion standard, now written into the plan A phase is done when: no item is deferred on a judgement call; no capability is removed to satisfy a review item; defects found while implementing are fixed in the same phase even if absent from the review; every fix carries a test that fails without it; and CI enforces it. ## Validation domain 148 / storage 41 / platform 64 / mpesa 29 pass nigig-pay-ui 72 (was 66) / nigig-mpesa 20 pass clippy -p nigig-pay-ui --no-deps -D warnings 0 errors builds: pay-ui, pay, mpesa, core pass allowlist injection: 3 tests fail when disabled pass pin-capture guard pass Pre-existing and untouched: `cargo test -p nigig-pay --lib` fails to build on clean HEAD (ClassifiedTransaction not in scope in transact.rs). Verified by stashing. The transport tests are exercised through the nigig-mpesa copy.	2026-08-02 08:52:32 +00:00
andodeki	3fda46b1eb	fix(spreadsheet-ui): flush toolbar mutation intents per event	2026-08-02 08:46:05 +00:00
Arena Agent	11ef0fbf67	fix(cad): GPU buffers self-invalidate; stop re-meshing on every drag frame ... Phase 5.4. Two defects with one root cause: part_geoms was keyed on the raw node id, so staleness was invisible to the type and correctness rested on seventeen scattered `part_geoms.remove(&id)` calls at the edit sites. All seventeen are deleted; the map is now keyed on (id, ParamHash), the same key MeshCache uses, and an entry whose hash no longer matches its node is simply never read. 1. subdivide_selected drew the wrong geometry. It replaced each selected part's solid with a fresh Csg and invalidated the MESH cache, but never removed the part's part_geoms entry -- so the stale uploaded buffer stayed a hit and the viewport kept drawing the un-subdivided shape. An audit of every mutation path found this was the only edit site missing its manual eviction, which is exactly the failure mode a manual protocol produces. 2. ParamHash covered the transform and the material, so a pure move invalidated the mesh. It should not: build_mesh reads neither. The cached TriMesh is in the node's LOCAL space, and all four consumers -- the draw loop, pick_part, the STL and glTF exporters -- apply the model matrix themselves. Dragging a part therefore re-triangulated it on every frame to produce byte-identical triangles, at up to 886 ns per extruded part per frame, per selected part. The hash now covers the solid parameters and nothing else. Two existing tests asserted the old behaviour and were INVERTED, not deleted -- they described what the code did rather than what it needed to do: cache_detects_transform_edits_via_param_hash -> a_transform_edit_reuses_the_cached_local_space_mesh mesh_cache_self_invalidates_on_parameter_and_transform_edits -> mesh_cache_self_invalidates_on_solid_parameter_edits New: build_mesh_output_does_not_depend_on_the_transform guards the assumption the key now rests on -- if anyone makes build_mesh bake the transform in, it fails and the key must grow it back. Plus mesh_cache_hits_on_a_pure_transform_edit, mesh_cache_hits_on_a_material_edit and an_uploaded_buffer_is_not_reused_after_the_solid_changes. Negative test: restoring the transform to the hash makes the first two fail; restored and green. Deletion is still explicit (`retain` on the live id set) because it is the one case a content hash cannot express -- there is no node left to hash. ParamHash is pub(crate), not pub: it is how the caches agree on staleness, not a consumer contract. Also recorded in BENCH_BASELINE.md, not fixed here: bench_command_execute _overhead is 227 us/cmd against a stale recorded 0.4 us. Verified pre-existing -- 254 us on the commit before this branch, so this work slightly improves it. CadCommandCtx::new eagerly builds a scene snapshot that most commands never read, and every command bumps the generation, so it is O(commands x nodes). The fix is to make that snapshot lazy; it is a separate change and gets its own commit. 753 lib + 154 integration tests pass. Test-name list diffed, not just the count. All 12 CI gates pass.	2026-08-02 08:44:42 +00:00
andodeki	9a14c0ccad	refactor(spreadsheet-ui): dispatch formula edits through intents	2026-08-02 08:43:50 +00:00
nigig-ci	25f32f7870	test(sms): build a real test suite (Phase G) ... 50 tests -> 102, and the two that were there at the start of this work are deleted. Where this started: robius-sms had ZERO tests, and nigig-sms had two -- bulk_sub_tab_default_is_contacts and bulk_sub_tab_variants_distinct. Both asserted a derived Default and a derived PartialEq. Neither mentioned SMS. Neither could fail short of the compiler breaking. That is the defect that produced every other defect in this plan: nothing could prove a change was safe, so nothing was ever deleted and every bug survived contact with review. Property tests (proptest, new dev-dependency) Seven over truncate_preview, format_timestamp, badge_text, and five more over segment_count, the rate limiter and ScheduleRequest. These are the ones that matter, because the hand-written cases in this repo all encode a bug someone had ALREADY found. proptest searches the space instead. I verified that by reinstating the original byte-slicing truncate_preview and confirming prop_truncate_preview_survives_mixed_scripts and prop_truncate_preview_respects_the_char_limit both fail against it -- they would have caught A3 before it shipped. prop_rate_limiter_respects_capacity models the window independently and asserts the invariant across random clock sequences, rather than re-implementing the limiter's own arithmetic in the assertion. Integration tests (2 new files, public API only) robius-sms/tests/sms_pipeline.rs and nigig-core/tests/sms_store.rs go through the public surface the application actually uses. The unit tests inside src/ can see private helpers; these cannot, which is the point -- they catch a refactor that keeps every unit test green while breaking the caller-visible contract. Two of them are privacy canaries. e1_message_bodies_are_never_persisted and e1_no_body_text_reaches_the_serialised_store fail if anyone removes #[serde(skip)] from OfflineSmsMessage.body. Verified by removing it: both fail, the other six pass. Nothing else in the tree would have noticed the inbox silently going back to plaintext on disk. Named regression tests One per defect, named for it -- c1_*, d1_*, d3_*, e1_*, e7_*, a4_*, c3_*, c7_* -- so a future reader goes from a failing test straight to the bug it guards rather than to a git archaeology session. New coverage for logic that had none - build_timeline_items / build_filtered_timeline_items: date-divider placement and the message indices the draw loop uses to index conv_data.messages. An off-by-one there renders the wrong body in the wrong bubble; it had no test at all. - kind_to_offline / kind_from_offline round-trip: the only thing stopping a cached Sent message reappearing as Inbox after a restart, which would flip the bubble to the wrong side of the screen. - normalize_number: what C1 groups on, across five formatting variants plus short codes and alphanumeric senders. MessageKind::from_android_type / to_android_type were hoisted out of sys/android/inbox.rs onto the type, the same way ScheduleRequest::validate was in A4, so the provider mapping is testable off-device. An unrecognised TYPE value is preserved verbatim in Unknown rather than defaulted, and there is a property test asserting the round trip is total over every i32. CI: a test-count FLOOR at 100. A floor rather than a ratchet -- unlike the clippy count, there is no reason to ever want this number to fall. Deliberately NOT faked: the JNI cursor loop, the keystore round-trip and broadcast delivery still need an emulator. A mock returning what I expect would test my expectations, not Android. Those remain called out in the Phase A and E commit messages. Verified: 11/11 checks. 48 robius-sms + 46 nigig-sms + 8 sms_store = 102. clippy -D warnings clean on host and aarch64-linux-android; nigig-sms ratchet holds at 32 (my first draft added an orphaned `use super::*`, caught by the ratchet and removed rather than baselined).	2026-08-02 08:43:16 +00:00
andodeki	79950bbba6	refactor(spreadsheet-ui): dispatch grid intents through workspace	2026-08-02 08:34:27 +00:00
andodeki	a264f53eb7	feat(pay): complete phase R1 of the remaining-work plan ... All four R1 items. Two of them uncovered defects that were not in the review, and R1.4's corpus found a live bug. ## R1.1 versioned fee policy (U9) The band table is a static "effective Jan 2024" snapshot. When Safaricom revises a tariff, nothing notices: the old number is quoted and the user authorises a total they are not charged. FeePolicy attaches provenance and a 400-day trust horizon. Past it, fee_for returns FeeError::PolicyOutOfDate rather than a number, and the sheet refuses to quote exactly as it already does for an unknown band — "no band for this amount" and "our table is old" stay distinguishable because they need different messages. Verified by disabling the check: 4 tests fail. Domain tests 137 -> 148. ## R1.2 quality gate for nigig-pay-ui (Q2) Correcting my own earlier count: 9 of the 10 unwraps were in tests. The one production case, on the dispatch path inside the biometric branch, is now a fail-closed path — no request, no prompt, no dispatch. nigig-pay-ui now denies unwrap_used/expect_used outside tests and CI runs clippy --no-deps -D warnings. Scoped with --no-deps because matrix_client and robius-ussd carry pre-existing warnings that are not this crate's to fix, and a gate that fails on someone else's code gets disabled. Turning the lint on surfaced 13 more issues, one a real defect: normalise_phone had two identical branches, and the 13-digit "254…" arm produced an 11-digit result — not a valid MSISDN, but non-empty, so it flowed on as a recipient. The duplication was hiding it. ## R1.3 exchange API (S7/S8/S10) The client forged origin/referer for api2.bybit.com and p2p.binance.com, impersonating those exchanges' own web clients against internal endpoints. Removed from both copies (nigig-pay and nigig-mpesa — item A5 again), along with the framework-identifying User-Agent. CI rejects either regrowing. Requests are still made, now honestly identified. If those endpoints reject an honest client the P2P panes fall back to their offline cache, which is the true state of the integration rather than a disguised one. Not done, deliberately: certificate pinning. Pinning an endpoint the product may drop is wasted work, and whether to keep these endpoints is a product call recorded in the plan. ## R1.4 adversarial CSV corpus (7.5) parse_csv turns an untrusted file into a payment list. Corpus covers empty input, injection-shaped fields, overflow, NUL, RTL override, full-width digits, a 5,000-row file and malformed numbers. The bar is not "parses correctly" but "never silently produces a payment nobody intended". Verified it can fail. UI tests 61 -> 66. ## Validation domain 148 / storage 41 / platform 64 / mpesa 29 / pay-ui 66 pass clippy -p nigig-pay-ui --no-deps -D warnings 0 errors builds: pay-ui, pay, mpesa, core; default and --no-default pass pin-capture guard pass fee-policy injection: 4 tests fail with the check removed pass corpus injection: catches a fabricating normaliser pass	2026-08-02 08:33:16 +00:00
Arena Agent	139f6de25a	refactor(cad): one CadDocument shared by all three viewports ... The three CadViewport instances each owned a private PartsStore, kept in step by copying the whole list every frame (parts_snapshot / replace_parts_snapshot). Both are now deleted. They share one Rc<RefCell<CadDocument>>: an edit in any viewport IS the edit in all of them, so there is nothing to reconcile. Copying was not an implementation detail, it was the source of a class of bugs that were individually fixable and collectively unfixable: - Only one viewport wins a sync frame, so the loser's id allocations were absent from the winner's snapshot and the assignment moved the counter BACKWARDS. Two parts drawn in different views between syncs got the same NodeId -- which keys MeshCache, part_geoms and every command. (Bounded earlier by sharing the allocator; now moot.) - Breaking on the first dirty viewport left later ones dirty, so the loser overwrote the winner's edit on the next frame. - Each sync deep-copied a Vec<CadNode> per destination and forced a full scene rebuild in each: ~98.5 us/frame at 100 parts, sustained for the whole of a drag. sync_parts_from_any_dirty_viewport survives but no longer moves data. It clears every script_dirty flag in one pass, returns the reporting viewport's script, and asks the others to repaint. Reads go through vp.parts() -> Ref<PartsStore>, writes through vp.parts_mut() -> RefMut. Both borrow self, so "never hold a read guard across a write" is checked by the compiler rather than left to reviewer discipline. Where a read loop's body needs &mut self (a draw call), the handle is cloned first via document() + read_parts() so the guard is tied to the local Rc instead of to self -- restoring the disjoint-field borrows the plain struct member used to give for free. Deleted in the same commit as their cause: - parts_snapshot / replace_parts_snapshot (40 lines) - split_for_command + CommandBorrows, which existed only to prove to the borrow checker that parts, scene_cache and command_stack were distinct fields. The parts list is no longer a field at all; with_command_ctx replaces them. - PartsStore::as_mut_vec, the last migration escape hatch. Its two remaining callers only wanted to edit fields, so they use a new iter_mut(); structural changes go through the specific methods, which is what makes "the generation moved" mean something. share_document runs from request_rebuild as well as initialize_split_viewports: the split viewports are created lazily, and one that missed the join would quietly edit a document nobody can see. Selection, part_geoms, scene_cache and camera state stay per-viewport by design -- what is highlighted is a property of a view, not the model. Test swap, verified by diffing the full test-name list rather than the count (both are 761): command_borrows_fields_accessible is gone -- it constructed a struct literal and read its fields back, asserting something the compiler already guarantees, and would have passed against every sync bug this removes. In its place, an_edit_through_one_document_handle_is_visible_through_the_others asserts the actual property. Negative test: replacing the shared handles with three independent documents makes it fail (left: 0, right: 1); restored and green. 746 lib + 154 integration tests pass. All 12 CI gates pass.	2026-08-02 08:30:09 +00:00
andodeki	59dd29d1ce	feat(spreadsheet-ui): queue grid mutation intents	2026-08-02 08:27:51 +00:00
arena-agent	08c3eed18d	feat(doc): copy and clear cell ranges through the clipboard ... An armed table cell range now copies as tab/newline text (rows top to bottom, cells left to right, the spreadsheet convention) through the same copyable_selection_text the TextCopy/TextCut hits answer with, so the long-press cell menu floats the full action set instead of a paste-only one. Cut and Backspace/Delete clear every non-empty spanned cell via the engine's new set_table_cells, which pops each sub-write's compensation into one Compensation::Group — the span un-clears in a single undo step where Backspace previously dropped the range and edited only the caret cell. Single writes keep the leaf compensation and empty write lists record nothing. Grouping cell writes surfaced a real undo asymmetry: an undo after a redo re-applied the redone cell text, because redo pushed the write's own text as its undo compensation (inverse() only swaps the verb, keeping the stale text) while undo's leaf special-case captured live cell state only in one direction. undo()/redo() now resolve cell-text inverses per compensation member against the projected text at both transition boundaries (redo_compensation/ undo_compensation), so leaves and groups round-trip through arbitrary cycles and the "cells stay out of groups" restriction is gone.	2026-08-02 08:21:31 +00:00
nigig-ci	1670ddf49c	refactor(sms): delete the dead code and the duplication (Phase F) ... Net -596 lines. No behaviour change except F10, which replaces a label that was lying. F2 -- four copies of one stub backend. apple.rs, linux.rs and windows.rs were BYTE-IDENTICAL 66-line files, and unsupported.rs was the same again. That duplication is what let them drift: Phase C3 had to fix `Error::Unknown` in exactly one of the four, because only one had it wrong. Collapsed into sys/stub.rs, which each platform module invokes. 268 lines become 35 plus one shared definition. The module is cfg'd out on Android, which has a real implementation and would otherwise report the macro as unused under -D warnings. F3 -- TWO dead compose implementations. SmsComposePage (189 lines) was registered in the VM and instantiated nowhere. Separately, the FAB and its compose overlay were left in the DSL as `visible: false` with a comment saying "FAB removed: SMS compose/inbox navigation now lives in SmsActionBar" -- but 102 lines of DSL and 53 lines of handler stayed behind, wired to a button no user can reach. Deleted both, and send_reply() with them: it existed only to serve the unreachable overlay. Compose navigation is SmsActionBar's, as the comment already said. F4 -- a whole second contact subsystem, unreachable. sms_screen.rs carried its own CONTACTS_CACHE, contacts_loaded(), load_contacts_into_cache(), display_name_for_number(), normalize_number(), try_load_contacts() and a contacts_load_attempted field. Nothing called any of it -- the live implementation is in conversations_list.rs. Worth noting the dead copy was also the WRONG one: its display_name_for_number did an O(n) linear scan of the whole phone book per lookup, where the live version is O(1) because cache_contact_number inserts under both the raw and normalised key. F5 -- the page tree was written out twice. sms_bulk_page, sms_schedule_page and sms_more_page were each declared under Desktop AND under Mobile, byte-identical apart from indentation. Any change to a page header had to be made in both places or the layouts silently diverged. Now three named widgets plus a shared SmsPageHeader, referenced from both variants. F8 -- serde, serde_json and robius-location were declared by nigig-sms and referenced nowhere in its sources. F9 -- was_scrolling was read twice per frame from the same portal list; the copy in handle_event was bound and never used. F10 -- the character counter was a hardcoded lie. The old compose page rendered "0 / 160 characters" and never updated it. It died with F3, but the bulk composer -- where the money actually goes -- had no cost indication at all. It now shows live segment count as you type, using segment_count() from Phase A5, because segments are the billing unit and "160" is only right for GSM-7: one emoji forces UCS-2 and drops the limit to 70. This is the only user-visible change in the commit. F1 and F7 were already done, in Phase A (shared cursor.rs) and Phase D1 (I/O out of draw_walk). The deletions orphaned eight imports, which are also removed. Together that takes the nigig-sms clippy ratchet from 49 to 32 -- these were not suppressed, the code they reported on is gone. Verified: 10/10 checks. clippy -D warnings clean on host AND aarch64-linux-android, 28 robius-sms tests, 22 nigig-sms tests, nigig-build still builds, metadata --locked clean.	2026-08-02 08:21:07 +00:00
andodeki	f3c2aa7a83	feat(spreadsheet-ui): define grid mutation intent type	2026-08-02 08:11:18 +00:00
andodeki	23fce675de	feat(pay): USSD automation on by default; containment moves to packaging ... Option A, as requested. Plus an audit of the whole review against the code. ## The default flips nigig-pay-ui default = [] (leaf stays off; see below) nigig-pay default = ["demo"] nigig-mpesa default = ["demo"] pageflipnav default = ["native", "demo"] `cargo run -p pageflipnav` now drives *334#, shows the PIN field and dispatches. That is the app's primary function and it works out of the box. A release build opts out: cargo build -p pageflipnav --no-default-features --features native This was not one line. A first attempt flipped the four `default =` lines and the opt-out still leaked: the app crates depended on nigig-pay-ui with its own defaults, so `--no-default-features` on pageflipnav was silently re-enabled one level down. Verified with a compile probe rather than cargo tree, which truncates. The inner deps now carry `default-features = false`, and the probe confirms both directions: default -> demo ON, --no-default-features -> demo OFF. ADR 0007's Play-policy note is untouched. The containment requirement of review item 0.1 is not dropped — the flag exists, CI exercises both directions, and a shipped build still cannot dispatch. What changed is which way it points by default, so development and device testing are not fighting it. CI guards inverted to match: they now assert automation is on by default *and* that the packaging opt-out still works. check-no-pin-capture.sh now probes the packaging build, since the default legitimately captures a PIN. ## REVIEWS/IMPLEMENTATION_AUDIT.md Every phase checked against the code, not against the tranche notes. Where they disagreed the code won. Summary: phases 0-5 and 7 done bar 5.2 and Keystore provisioning; phase 6 substantially done with the thread_local session ownership outstanding; phase 8 partly. ## B7 found live while auditing Month navigation had never been examined. Both copies of the transactions widget still stepped months with Duration::days(31) and years with Duration::days(366). Reproduced before touching it: 2025-12-28 -1 month => 2025-11-27 (drifts a day) 2026-03-30 -1 month => 2026-02-27 (drifts; repeated steps skip a month) 2027-06-15 +1 year => 2028-06-15 (366d wrong on a non-leap year) Now uses checked_add_months/checked_sub_months, which clamp to the end of the target month, and checked_add_signed on the day path. An unrepresentable date leaves the view where it was. Neither claimed done nor flagged open — simply never looked at. That is the argument for auditing code rather than notes. ## Validation domain 137 / storage 41 / platform 64 / mpesa 29 / pay-ui 61 pass cargo check: pay-ui, pay, mpesa, core (default and opt-out) pass demo-on-by-default probe, both directions pass no-PIN-capture guard against the packaging build pass Unrelated and still blocking a full APK: nigig-map fails to compile on clean HEAD (12 errors, no field center_lat on ViewportState).	2026-08-02 08:06:43 +00:00
Arena Agent	c83f8e083f	refactor(cad): commands mutate through PartsStore, not a bare Vec ... First migration step of Phase 5.2, and it closes a real hazard rather than only moving types around. `CadCommandCtx` held `&mut Vec<CadNode>`, taken out of the store by `split_for_command` via `as_mut_vec()`. Every command therefore edited the parts list *underneath* `PartsStore`, so none of their mutations bumped its generation. `SceneCache::scene_for` compares generations, so the only thing preventing a permanently stale scene was each command remembering to call `mark_dirty()` by hand -- the exact discipline-based invalidation the generation counter was introduced to replace. Every command happens to call it today. One omission in a new command would have produced a scene that never rebuilds, with no failing test. `CadCommandCtx` and `CommandBorrows` now carry `&mut PartsStore`. move_node and update_node go through `get_mut_by_raw_id`; create, delete and insert already used store methods once the type changed. The generation now moves on its own and the `mark_dirty()` calls become belt-and-braces instead of load-bearing. `CadCommandCtx::new` also builds its scene with `scene_for`, so it shares the cache rather than rebuilding unconditionally. Four tests. Three exercise the real `CadCommandCtx` against a real `PartsStore` -- the mock keeps its own `Vec` and structurally cannot observe the generation, which is why the gap survived this long. The fourth pins the hazard itself: a mutation that skips the generation still needs `mark_dirty()`, demonstrated through a `#[cfg(test)]` `as_mut_vec_without_bump` that models the old path. Negative-tested by restoring the bypass: "a command edit must move the generation, not rely on a manual mark_dirty()". Scope kept deliberately narrow: commands.rs only, ~93 sites left in viewport.rs and workspace.rs. ARCHITECTURE.md updated with what is done and what remains. 746 lib + 154 integration, 0 failed. All twelve gates pass, benchmarks still build and run.	2026-08-02 08:01:19 +00:00
nigig-ci	737a3e5d5d	security(sms): encrypt scheduled payloads, report real send status (E3, E7) ... Closes the two Phase E items I had left open and documented as open. E3 -- scheduled message bodies were plaintext on disk. Pending schedules must outlive the process so SmsAlarmReceiver can send them when the alarm fires and so they survive a reboot, so recipient and body go to SharedPreferences. MODE_PRIVATE is the right primitive -- the file is UID-scoped -- but the contents were in the clear, readable by anything running as the same UID and swept into cloud backup by default. Same asset class as the inbox (THREAT_MODEL.md T-I4). Adds SmsScheduleCrypto: AES-256-GCM, fresh IV per value, key generated inside the platform AndroidKeyStore and non-exportable. An attacker with the prefs file but not the keystore gets ciphertext. Deliberately NOT androidx.security.EncryptedSharedPreferences: that is a Gradle dependency, and this crate compiles its Java with bare javac against android.jar (see build.rs), so using it would mean a Gradle build or a vendored jar. AndroidKeyStore and javax.crypto are both in android.jar and give the property that matters. The key is deliberately NOT user-authentication-bound: an alarm fires while the device may be locked and the receiver must decrypt with no user present. This protects against another app and against an extracted backup, which is the threat in scope -- not against someone holding an unlocked handset. Fails CLOSED. If the keystore is unavailable, encrypt returns null and schedule_sms errors rather than writing plaintext. A row that cannot be decrypted -- wrong key after a reinstall, tampering, or written by an older build -- is treated exactly like a missing row and skipped; sending a garbled body would be worse than not sending. E7 -- "sent" was a guess. Both the sentIntent and deliveryIntent arguments were null, so nothing could report back. send_sms returning Ok meant "the JNI call returned", not that the radio accepted the message and certainly not that it arrived -- and the UI rendered that as a tick. A send rejected for no service, no SIM or a throttled radio was indistinguishable from a delivered one. Adds send_sms_tracked, which attaches real PendingIntents and returns a correlating token, plus SmsSentReceiver to collect the platform result and SendOutcome/SendReport to express it: Sent (radio accepted) is now a different value from Delivered (handset acknowledged), and failures carry the RESULT_ERROR_* code. Three details worth recording: - multipart takes ArrayList<PendingIntent>, one entry per part, so the intent is repeated part_count times. Passing null here, as before, meant no status for exactly the messages most likely to fail: the long ones. - the request code is derived from (token, kind), or the two intents collide and the delivery report overwrites the send report. - the broadcast is package-scoped and the receiver registered NOT_EXPORTED, so another app cannot forge a delivery report. If the receiver class is unavailable the send still goes out with null intents: losing the status report is much better than losing the message. Also fixes A5 on the scheduled path. SmsAlarmReceiver still called sendTextMessage directly, so a scheduled message over 160 GSM-7 characters -- or 70 with any emoji -- was silently truncated by the carrier. It now divides and sends multipart, as the Rust send path has since Phase A. CI: two gates, both negative-tested by reverting the fix and confirming they fail. One asserts schedule_sms never writes request.recipient or request.body directly; the other asserts the send path still passes sent/delivery intents in both the single-part and multipart calls. THREAT_MODEL.md T-I4 and the delivery-confirmation row move from open to fixed, with the residual risk stated: callers may still use the untracked send_sms, which remains honest about meaning only "handed to the platform". Tests: robius-sms 25 -> 28. Verified: 13/13 CI checks, clippy -D warnings clean on host and aarch64-linux-android, both new Java classes javac-compile and dex. NOT verified on a device. The keystore round-trip, the broadcast delivery and the token correlation all need an emulator or handset; there is still no CI runner on this repo.	2026-08-02 07:58:44 +00:00
andodeki	7566374237	refactor(spreadsheet-ui): make shared model handle mandatory	2026-08-02 07:51:41 +00:00
andodeki	9f0e133c4b	fix(pay): the APK crate could not reach the demo flag at all ... Reported: building the pageflipnav APK, the Pay sheet shows no PIN field and reports that dispatch is unavailable, with no way to enable the *334# automation. That is a real defect and it is worse than the earlier feature-forwarding gap. pageflipnav is the crate that produces the APK. It depends on nigig-pay and nigig-mpesa but declared no `demo` feature, so: grep -c demo crates/pageflipnav/Cargo.toml -> 0 The flag was unreachable from the only build that matters. Adding `--features demo` to nigig-pay does not change what the APK contains, so every instruction I gave for enabling the automation was useless to anyone building the real app. pageflipnav now forwards it: demo = ["nigig-pay/demo", "nigig-mpesa/demo"] Verified with a compile probe rather than cargo tree, which truncated its output and initially suggested the wiring had failed: a `#[cfg(feature = "demo")] compile_error!` in nigig-pay-ui fires twice under `cargo check -p pageflipnav --features demo` and zero times without it. A CI guard asserts pageflipnav keeps forwarding the flag, and the README now leads with the pageflipnav command and states plainly that building nigig-pay alone does not affect the APK. Unrelated: nigig-map fails to compile on clean HEAD (12 errors, `no field center_lat on ViewportState`), so a full pageflipnav build is currently blocked by that regardless of this change.	2026-08-02 07:51:25 +00:00
arena-agent	00b3eb6e98	feat(doc): select the whole editing context with Ctrl/Cmd+A ... Keyboard select-all lands in the CRDT editor, mirroring makepad's text_input. A parked cell cursor selects its whole cell text (the in-cell character span, collapsing any armed merge range); otherwise the selection spans the first layout glyph to the last with the caret following the focus, so Copy, Cut, style toggles and Backspace-over-span all treat it like a maximal Shift+Arrow selection. Table blocks ride the range like any middle block: the clipboard payload keeps skipping their cells while a cut drains them through replace_block_range, so one undo restores the document. Touch sessions in Edit mode float the platform clipboard menu on the fresh selection (the keyboard-select-all pattern from text_input), mirrored through clipboard_menu like the long-press request; the cell-rect lookup it shares is factored into cursor_cell_rect. A desktop Ctrl+A never makes a menu request, and a document without glyphs leaves the caret put. copyable_selection_text goes crate-visible so hosts rendering their own copy affordances read the same payload the hits answer with.	2026-08-02 07:47:06 +00:00
Arena Agent	0d05a5e62f	feat(cad): add the shared CadDocument (Phase 5.2 foundation) ... Introduces `CadDocument` and `SharedCadDocument` (`Rc<RefCell<CadDocument>>`) -- one owner for the parts list, its generation counter and the id allocator -- with the semantics the viewports will move onto. Deliberately NOT wiring `CadViewport` onto it in this commit. That is 103 call sites (73 viewport.rs, 20 workspace.rs, 8 commands.rs), each needing its borrow scope checked individually, and bundling it with the type definition would produce a diff nobody can review and a bisect target nobody can isolate. This lands the destination, proven by test; the migration follows one file per commit. Four tests, each negative-tested: - an edit through one handle is visible through the other -- the whole point of the phase; - the generation counter is shared, so SceneCache::scene_for cannot serve a stale scene to a viewport that has not synced yet; - two separately constructed documents do NOT share, so the first two cannot pass for the wrong reason; - sequential borrows are fine and overlapping ones are rejected. That one states the borrow discipline as an executable rule rather than a comment, using try_borrow_mut so it documents the failure without aborting. Verified before designing, rather than assumed: - Every `self.parts` read in viewport.rs is a short-lived expression or a loop, and only two loops mutate `self` at all (~6231, ~6265) -- both touching `selection`, a different field, so neither becomes a BorrowMutError. - `split_for_command` hands out `&mut Vec<CadNode>` tied to `&mut self`, which is the one construct genuinely awkward under RefCell. Every caller already scopes it in a block and drops it before touching `self` again, so it maps onto a RefMut guard -- but that holds because of how the callers are written, not because the signature enforces it, and ARCHITECTURE.md now says to re-check it rather than assume it. The migration rule -- never hold a `parts()` guard across a call that can reach `parts_mut()` -- is recorded with the two loops to re-check first if a BorrowMutError ever appears. 738 lib + 154 integration, 0 failed. All twelve gates pass.	2026-08-02 07:43:52 +00:00
andodeki	f021bdc0f5	fix(pay): the gate message hid the way to turn automation back on ... Reported: the *334# automation used to work and now cannot be reached, and the status message reads as though the capability was removed. The capability was not removed. Verified with git: - `demo` and the USSD dispatch gate are both present in cc05abd, the initial commit, before any of my work. - 67abe2b "phase zero payment containment" (2026-07-27) predates my first commit d0f5e74 (2026-07-29); `git merge-base --is-ancestor` confirms it. - `git diff d0f5e74..HEAD` on pay_flow_handler.rs shows I added and removed no `cfg!(feature = "demo")` gate at all. - robius-ussd, including the 534-line AccessibilityService that drives the menu, is untouched by my commits. What I did get wrong is the wording. The original message was: "USSD dispatch disabled: compile with `--features demo` to enable" Blunt, but it named the flag. I replaced it with "This app tracks payments — it doesn't send them", which is accurate about the default build and says nothing about how to change it. Read on a device, that sounds like the feature was deleted. That is a real regression in usability and it is mine. The message now leads with the fact that it is off *in this build* and names the flag again, while keeping the honest framing about the PIN: "Automated *334# payment is off in this build. Rebuild with `--features demo` on Android to enable it, or send in the M-Pesa app and it will appear here automatically. No M-Pesa PIN was requested or stored. [tracker-3]" Marker bumped to tracker-3 so the running build is identifiable.	2026-08-02 07:43:45 +00:00