Windows: inspect native context menus in the UIA live gate by coneilen · Pull Request #422 · scgopi/GraphCode

coneilen · 2026-09-23T00:42:06Z

Closes #418.

Tools/windows/uia-live-gate.ps1 had no way to open, read, or dismiss a native Win32
popup menu, so every context-menu behavior in the Windows shell was unverifiable live —
including the grayed Move Project... (unavailable: daemon support required) item that
GraphContextMenu.show() appends with MF_GRAYED.

What this adds

Gate-only fixture message MainWindow.wm_uia_context_menu (WM_APP + 44). App.showUiaContextMenu
calls the same GraphContextMenu.show() the mouse path calls, with the same target data;
only hit-test routing is bypassed. TrackPopupMenu blocks, but its own modal loop keeps
pumping, so the gate posts the request and inspects while the shell thread is blocked.
Anti-hang watchdog. A timer armed before show() fires inside the menu's modal loop and
calls EndMenu(), so an abandoned popup can never block the shell for the life of the process.
Gate popup capability: FindPopupMenuWindow (class #32768, scoped to the shell pid),
MN_GETHMENU + GetMenuItemCount/GetMenuStringW/GetMenuState, PostContextMenu, and
DismissPopupMenu; plus Wait-ForPopupMenu / Get-PopupMenuItems / Close-PopupMenu with
bounded waits, focus diagnostics, and hard failure — no skip path.
Live assertions: the project menu's ordered items; Move is command 5149 with its exact
unavailable text and a disabled state; a remote project's menu omits Move, Recycle Bin, and
Explorer; the popup dismisses and the shell's UIA tree still answers afterwards.
Missing coverage fixed: GraphContextMenu.zig and MainWindow.zig unit tests were not
executed by any harness. PR Make unavailable project relocation explicit on Windows #408's Move Project test has never actually run in CI. They now run
in WindowsShell.Tests.ps1.

Honest scoping of the evidence

A popup menu surfaces in the UIA tree only as an empty Pane with zero MenuItem children —
I verified this three times with a two-process spike before building anything. Item identity, text,
and the MF_GRAYED state are therefore read through MN_GETHMENU against the HMENU the shell
itself handed to TrackPopupMenu. That is live evidence of what the shell renders, but it is not
UIA-tree evidence, and it is labelled that way in the gate comments, the structural tests, and the
parity ledger.

investigation/ui-parity-matrix.md: the "Project context menu" row was Validated on a stale
claim that Windows keeps Move as an Explorer /select handoff — rewritten to the actual
grayed-unavailable behavior with live evidence. "Loop context menu" and "Canvas context menu" stay
Partial, now stating exactly what would close them.

Verification

Ran locally against the real built shell (graphcode-windows.exe, pinned Zig 0.15.2, sha256
matching bootstrap.ps1) with a focused probe driving the new hook:

MENU target=1 (local project) items=12
  [8] id=5149 enabled=False sep=False state=0x1 'Move Project... (unavailable: daemon support required)'
MENU target=2 (remote project) items=10   # no 5149, no 5151, no 5144
DISMISSED closed=True ... SHELL_TREE_RESPONSIVE true
WATCHDOG_CLOSED=True

Also run locally: pinned-Zig zig build, zig test src\GraphContextMenu.zig (69 pass),
zig test src\MainWindow.zig (6 pass), ValidationRunner.Tests.ps1 (PASS), and
Parser::ParseFile on every modified PowerShell file.

Not run locally: the full uia-live-gate.ps1 end-to-end. It stops early here on a pre-existing,
unrelated workspace-toolbar assertion because the pinned zmx provider is unavailable on this
machine (its pinned commit does not build against the locally installed Zig 0.16.0). CI is the
authoritative run for the new live assertions.

RED: gate had zero popup-menu capability -> new assertion "project context menu never opened a native popup window" fails on main because PostContextMenu, FindPopupMenuWindow, and the shell hook do not exist
GREEN: wm_uia_context_menu opens the real TrackPopupMenu and MN_GETHMENU reads it -> live probe returned id=5149 state=0x1 disabled with the exact unavailable text, and the remote menu omitted 5149/5151/5144
REGRESSION: ValidationRunner structural tests plus newly wired zig tests -> deleting the popup helpers, the disabled-Move assertion, the EndMenu watchdog, or the GraphContextMenu/MainWindow test wiring now fails validation even on a runner with no live desktop

CI evidence (authoritative)

The windows-shell job is green and the live gate now publishes what the native popup
actually contained, from run on 48cfa6e:

contextMenuItemCount          : 12
contextMenuMoveProjectText    : Move Project... (unavailable: daemon support required)
contextMenuMoveProjectEnabled : False
contextMenuMoveProjectState   : 0x1
contextMenuDismissed          : True
remoteContextMenuItemCount    : 10

The job log also shows the previously-unrun unit tests executing:
8/69 GraphContextMenu.test.the real Move Project menu item is disabled with its explicit reason inline...OK.
So the live-gate blocker noted above is resolved by CI: the assertions did run, on a real
desktop, against the real TrackPopupMenu.

…vocation Moves the structural guard added for #424 to the end of WindowsShell.Tests.ps1, after all zig test invocations, instead of right after Resolve-TestZig. Placed first, the guard's expected failure (reserving GraphContextMenu.zig/MainWindow.zig for #418/#422) short-circuited the whole script in CI before any of the newly-wired tests ever ran on the actual runner, leaving only local verification as evidence. Placed last, CI now executes and reports every invocation for real before the guard's contract check runs, while the guard still fails the job overall until #418/#422 lands. RED: with the guard first, CI failed at the guard on the first push and never exercised a single newly-wired zig test -> no real CI signal existed for the wiring itself, only local runs. GREEN: relocated the guard after every Invoke-Native/Invoke-NativeQuarantined call and reran the full harness locally with pinned Zig 0.15.2 -> all 37 invocations execute (90/93 Sidebar.zig, 228/231 App.zig, everything else 100%), quarantines are tolerated correctly, and the guard still throws last naming exactly GraphContextMenu.zig and MainWindow.zig. REGRESSION: reran the same real-mutation guard check (removing a wired file name) after relocating the block -> guard still throws the identical not-wired message immediately, confirming the guard's detection logic is unchanged, only its position in the script moved. Signed-off-by: Colin Neilens <coneilen@microsoft.com> Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

The live gate had no way to open, read, or dismiss a native Win32 popup menu, so every context-menu behavior in the Windows shell -- including the grayed Move Project item that GraphContextMenu.show() appends with MF_GRAYED -- was unverifiable live. Add a gate-only fixture message (MainWindow.wm_uia_context_menu) that asks the shell to open a real context menu through the same GraphContextMenu.show() the mouse path calls, a watchdog timer that calls EndMenu() so an abandoned popup can never block the message loop, and gate-side popup discovery, introspection, and dismissal. A popup surfaces in the UIA tree only as an empty Pane with no MenuItem children, so the menu is read through MN_GETHMENU and the Win32 menu API against the HMENU the shell handed to TrackPopupMenu. That is live evidence of what the shell renders, not UIA-tree evidence, and is labelled as such in the gate, the tests, and the parity ledger. The gate now asserts the live project menu's ordered items, that Move is command 5149 with its exact unavailable text and a disabled state, that a remote project's menu omits Move/Recycle Bin/Explorer, and that the popup dismisses without wedging the shell. GraphContextMenu.zig and MainWindow.zig unit tests were never executed by any harness; they now run in the Windows shell validation. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> Signed-off-by: Colin Neilens <coneilen@microsoft.com>

The context-menu assertions were silent on success, so a passing CI run left no positive record of what the native popup actually contained. Emit the observed item count, the Move Project item's exact text, enabled flag and raw MF_ state bits, and the dismissal result into the gate's summary JSON so the live evidence is auditable from the job log. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> Signed-off-by: Colin Neilens <coneilen@microsoft.com>

Cherry-picked #420's NativeForms/UpdateOfferPresentation harness wiring (commit 775d77b) onto this branch: my branch point (68eabe5) predated that merge, so the reentrancy-guard Assert-Contract, the source-list entry, and the zig test invocation for UpdateOfferPresentation.zig were all absent here even though they exist on main. Restored via cherry-pick rather than a full rebase, per instruction to hold on rebasing until #422 lands. That cherry-pick alone was not sufficient: this PR's own anti-drift guard maintains a second, independent file list (\) that #420 never touched (the guard did not exist on main). Newly restoring the UpdateOfferPresentation.zig invocation without adding it to that list would have made the guard itself flag it as unwired. RED: after cherry-picking 775d77b, the guard's \ array still lacked "UpdateOfferPresentation.zig" -> a manual simulation of the guard's detection logic reported it as missing alongside the two entries correctly reserved for #422. GREEN: added "UpdateOfferPresentation.zig" to \ next to "NativeForms.zig" -> the same simulation now reports exactly and only GraphContextMenu.zig and MainWindow.zig as missing, matching the #422 reservation. REGRESSION: reran the full harness end-to-end locally with pinned Zig 0.15.2 and GRAPHCODE_WINGHOSTTY_ROOT set -> all 37 invocations execute (including the restored NativeForms 96/96 and UpdateOfferPresentation 1/1), the same 3 pre-existing Sidebar.zig failures are tolerated by name, and the guard still throws on exactly GraphContextMenu.zig, MainWindow.zig. Signed-off-by: Colin Neilens <coneilen@microsoft.com> Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

…nti-drift guard Fixes #424. WindowsShell.Tests.ps1 hand-maintained a list of zig test invocations that missed 17 files (Accessibility.zig was fixed separately by #421; GraphContextMenu.zig/MainWindow.zig are reserved for in-flight #418/ #422). Wires all 17 remaining orphaned files with per-file link flags verified against pinned Zig 0.15.2, and adds a structural guard that enumerates graphcode-windows\src\*.zig, detects files containing a est " block, and throws if any is missing from the wired-file list -- so this drift cannot recur silently. First-run triage: - WorktreeDialog.zig: one test used a stale fixture (dirty instead of locked) to exercise armConfirmation()'s fail-closed path; sweepSelectable() intentionally permits dirty rows. Fixed the test fixture, not the code. - App.zig: one test's App struct literal predated three fields (sidebar_state, declared_entry_ids, kept_worktree_paths) added since it last compiled. Fixed the test to match App.init()'s initialization. - Sidebar.zig: 3 real, pre-existing test failures traced to one root cause (layoutFor()/projectSectionHeight() count a recent_projects entry that is also the open project, but appendRows() correctly excludes it from rendered rows, desyncing row/scroll y-math). Per explicit instruction not to modify Sidebar.zig source, these are quarantined at the harness level with an explicit reason string, not fixed or deleted. Reported as a real product bug for separate follow-up. - App.zig transitively reruns the same 3 Sidebar failures (it imports Sidebar.zig); quarantined identically with a note explaining why. RED: WorktreeDialog and App.zig tests failed on first run -> both had stale fixtures, not product bugs; fixed the tests to match current code. GREEN: all 17 newly-wired files now execute; 90/93 Sidebar.zig and 228/231 App.zig tests pass -> the 3 residual failures are quarantined with cause. REGRESSION: full harness run end-to-end with pinned Zig 0.15.2 exits 0 and the anti-drift guard fails loudly on a real mutation removing a wired file -> guard verified functional, not merely asserted. Signed-off-by: Colin Neilens <coneilen@microsoft.com> Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

…vocation Moves the structural guard added for #424 to the end of WindowsShell.Tests.ps1, after all zig test invocations, instead of right after Resolve-TestZig. Placed first, the guard's expected failure (reserving GraphContextMenu.zig/MainWindow.zig for #418/#422) short-circuited the whole script in CI before any of the newly-wired tests ever ran on the actual runner, leaving only local verification as evidence. Placed last, CI now executes and reports every invocation for real before the guard's contract check runs, while the guard still fails the job overall until #418/#422 lands. RED: with the guard first, CI failed at the guard on the first push and never exercised a single newly-wired zig test -> no real CI signal existed for the wiring itself, only local runs. GREEN: relocated the guard after every Invoke-Native/Invoke-NativeQuarantined call and reran the full harness locally with pinned Zig 0.15.2 -> all 37 invocations execute (90/93 Sidebar.zig, 228/231 App.zig, everything else 100%), quarantines are tolerated correctly, and the guard still throws last naming exactly GraphContextMenu.zig and MainWindow.zig. REGRESSION: reran the same real-mutation guard check (removing a wired file name) after relocating the block -> guard still throws the identical not-wired message immediately, confirming the guard's detection logic is unchanged, only its position in the script moved. Signed-off-by: Colin Neilens <coneilen@microsoft.com> Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

Cherry-picked #420's NativeForms/UpdateOfferPresentation harness wiring (commit 775d77b) onto this branch: my branch point (68eabe5) predated that merge, so the reentrancy-guard Assert-Contract, the source-list entry, and the zig test invocation for UpdateOfferPresentation.zig were all absent here even though they exist on main. Restored via cherry-pick rather than a full rebase, per instruction to hold on rebasing until #422 lands. That cherry-pick alone was not sufficient: this PR's own anti-drift guard maintains a second, independent file list (\) that #420 never touched (the guard did not exist on main). Newly restoring the UpdateOfferPresentation.zig invocation without adding it to that list would have made the guard itself flag it as unwired. RED: after cherry-picking 775d77b, the guard's \ array still lacked "UpdateOfferPresentation.zig" -> a manual simulation of the guard's detection logic reported it as missing alongside the two entries correctly reserved for #422. GREEN: added "UpdateOfferPresentation.zig" to \ next to "NativeForms.zig" -> the same simulation now reports exactly and only GraphContextMenu.zig and MainWindow.zig as missing, matching the #422 reservation. REGRESSION: reran the full harness end-to-end locally with pinned Zig 0.15.2 and GRAPHCODE_WINGHOSTTY_ROOT set -> all 37 invocations execute (including the restored NativeForms 96/96 and UpdateOfferPresentation 1/1), the same 3 pre-existing Sidebar.zig failures are tolerated by name, and the guard still throws on exactly GraphContextMenu.zig, MainWindow.zig. Signed-off-by: Colin Neilens <coneilen@microsoft.com> Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

…422; correct Sidebar quarantine attribution Rebased onto origin/main after #422 (issue #418) merged as 06e092e, which wired GraphContextMenu.zig and MainWindow.zig into the harness's zig test invocations and source-list array. #422 never touched this PR's own \ guard array, because that array did not exist on main when #422 was authored -- it is this PR's own addition. Added both filenames to the list and updated the guard's explanatory comment, which was otherwise now stale (still described the two files as reserved and not-yet-landed). Also corrected the Sidebar.zig quarantine reason: an independent review determined that only 2 of the 3 quarantined failures are the real layoutFor()/appendRows() product bug (#428, fix in flight as #430); the third (sidebar scroll clamps overflow, shrink, and resize, expected 334 found 410) is a separate, stale test expectation -- the 76px delta is the Activity block height that contentBottom/paint() correctly account for and the test's oracle omitted. Left it quarantined (not fixed) since this PR must not modify Sidebar.zig source, but the reason string now attributes each failure accurately instead of lumping all three under one root cause. RED: mechanically diffed this branch's wired-file set against origin/main after rebasing -> main had 22 entries (20 plus #422's 2), this branch still reported only 38 in its own \ guard list, and running the guard's detection logic directly showed it still flagging GraphContextMenu.zig and MainWindow.zig as unwired despite their zig test invocations now existing in the script. GREEN: added both names to \ and reran the same detection logic -> zero missing files reported. REGRESSION: ran the complete harness end-to-end locally with pinned Zig 0.15.2 and GRAPHCODE_WINGHOSTTY_ROOT set -> all 39 invocations execute including the newly-landed GraphContextMenu.zig (8/8) and MainWindow.zig (6/6, appearing twice via App.zig's transitive import), the same 3 Sidebar.zig failures are tolerated by name with the corrected attribution text rendering intact, and the full script now exits 0 -- the guard no longer fires at all. Signed-off-by: Colin Neilens <coneilen@microsoft.com> Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

…nti-drift guard Fixes #424. WindowsShell.Tests.ps1 hand-maintained a list of zig test invocations that missed 17 files (Accessibility.zig was fixed separately by #421; GraphContextMenu.zig/MainWindow.zig are reserved for in-flight #418/ #422). Wires all 17 remaining orphaned files with per-file link flags verified against pinned Zig 0.15.2, and adds a structural guard that enumerates graphcode-windows\src\*.zig, detects files containing a est " block, and throws if any is missing from the wired-file list -- so this drift cannot recur silently. First-run triage: - WorktreeDialog.zig: one test used a stale fixture (dirty instead of locked) to exercise armConfirmation()'s fail-closed path; sweepSelectable() intentionally permits dirty rows. Fixed the test fixture, not the code. - App.zig: one test's App struct literal predated three fields (sidebar_state, declared_entry_ids, kept_worktree_paths) added since it last compiled. Fixed the test to match App.init()'s initialization. - Sidebar.zig: 3 real, pre-existing test failures traced to one root cause (layoutFor()/projectSectionHeight() count a recent_projects entry that is also the open project, but appendRows() correctly excludes it from rendered rows, desyncing row/scroll y-math). Per explicit instruction not to modify Sidebar.zig source, these are quarantined at the harness level with an explicit reason string, not fixed or deleted. Reported as a real product bug for separate follow-up. - App.zig transitively reruns the same 3 Sidebar failures (it imports Sidebar.zig); quarantined identically with a note explaining why. RED: WorktreeDialog and App.zig tests failed on first run -> both had stale fixtures, not product bugs; fixed the tests to match current code. GREEN: all 17 newly-wired files now execute; 90/93 Sidebar.zig and 228/231 App.zig tests pass -> the 3 residual failures are quarantined with cause. REGRESSION: full harness run end-to-end with pinned Zig 0.15.2 exits 0 and the anti-drift guard fails loudly on a real mutation removing a wired file -> guard verified functional, not merely asserted. Signed-off-by: Colin Neilens <coneilen@microsoft.com> Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

…vocation Moves the structural guard added for #424 to the end of WindowsShell.Tests.ps1, after all zig test invocations, instead of right after Resolve-TestZig. Placed first, the guard's expected failure (reserving GraphContextMenu.zig/MainWindow.zig for #418/#422) short-circuited the whole script in CI before any of the newly-wired tests ever ran on the actual runner, leaving only local verification as evidence. Placed last, CI now executes and reports every invocation for real before the guard's contract check runs, while the guard still fails the job overall until #418/#422 lands. RED: with the guard first, CI failed at the guard on the first push and never exercised a single newly-wired zig test -> no real CI signal existed for the wiring itself, only local runs. GREEN: relocated the guard after every Invoke-Native/Invoke-NativeQuarantined call and reran the full harness locally with pinned Zig 0.15.2 -> all 37 invocations execute (90/93 Sidebar.zig, 228/231 App.zig, everything else 100%), quarantines are tolerated correctly, and the guard still throws last naming exactly GraphContextMenu.zig and MainWindow.zig. REGRESSION: reran the same real-mutation guard check (removing a wired file name) after relocating the block -> guard still throws the identical not-wired message immediately, confirming the guard's detection logic is unchanged, only its position in the script moved. Signed-off-by: Colin Neilens <coneilen@microsoft.com> Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

Cherry-picked #420's NativeForms/UpdateOfferPresentation harness wiring (commit 775d77b) onto this branch: my branch point (68eabe5) predated that merge, so the reentrancy-guard Assert-Contract, the source-list entry, and the zig test invocation for UpdateOfferPresentation.zig were all absent here even though they exist on main. Restored via cherry-pick rather than a full rebase, per instruction to hold on rebasing until #422 lands. That cherry-pick alone was not sufficient: this PR's own anti-drift guard maintains a second, independent file list (\) that #420 never touched (the guard did not exist on main). Newly restoring the UpdateOfferPresentation.zig invocation without adding it to that list would have made the guard itself flag it as unwired. RED: after cherry-picking 775d77b, the guard's \ array still lacked "UpdateOfferPresentation.zig" -> a manual simulation of the guard's detection logic reported it as missing alongside the two entries correctly reserved for #422. GREEN: added "UpdateOfferPresentation.zig" to \ next to "NativeForms.zig" -> the same simulation now reports exactly and only GraphContextMenu.zig and MainWindow.zig as missing, matching the #422 reservation. REGRESSION: reran the full harness end-to-end locally with pinned Zig 0.15.2 and GRAPHCODE_WINGHOSTTY_ROOT set -> all 37 invocations execute (including the restored NativeForms 96/96 and UpdateOfferPresentation 1/1), the same 3 pre-existing Sidebar.zig failures are tolerated by name, and the guard still throws on exactly GraphContextMenu.zig, MainWindow.zig. Signed-off-by: Colin Neilens <coneilen@microsoft.com> Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

…422; correct Sidebar quarantine attribution Rebased onto origin/main after #422 (issue #418) merged as 06e092e, which wired GraphContextMenu.zig and MainWindow.zig into the harness's zig test invocations and source-list array. #422 never touched this PR's own \ guard array, because that array did not exist on main when #422 was authored -- it is this PR's own addition. Added both filenames to the list and updated the guard's explanatory comment, which was otherwise now stale (still described the two files as reserved and not-yet-landed). Also corrected the Sidebar.zig quarantine reason: an independent review determined that only 2 of the 3 quarantined failures are the real layoutFor()/appendRows() product bug (#428, fix in flight as #430); the third (sidebar scroll clamps overflow, shrink, and resize, expected 334 found 410) is a separate, stale test expectation -- the 76px delta is the Activity block height that contentBottom/paint() correctly account for and the test's oracle omitted. Left it quarantined (not fixed) since this PR must not modify Sidebar.zig source, but the reason string now attributes each failure accurately instead of lumping all three under one root cause. RED: mechanically diffed this branch's wired-file set against origin/main after rebasing -> main had 22 entries (20 plus #422's 2), this branch still reported only 38 in its own \ guard list, and running the guard's detection logic directly showed it still flagging GraphContextMenu.zig and MainWindow.zig as unwired despite their zig test invocations now existing in the script. GREEN: added both names to \ and reran the same detection logic -> zero missing files reported. REGRESSION: ran the complete harness end-to-end locally with pinned Zig 0.15.2 and GRAPHCODE_WINGHOSTTY_ROOT set -> all 39 invocations execute including the newly-landed GraphContextMenu.zig (8/8) and MainWindow.zig (6/6, appearing twice via App.zig's transitive import), the same 3 Sidebar.zig failures are tolerated by name with the corrected attribution text rendering intact, and the full script now exits 0 -- the guard no longer fires at all. Signed-off-by: Colin Neilens <coneilen@microsoft.com> Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

…-drift guard (#426) * Wire 17 orphaned Zig test files into WindowsShell.Tests.ps1 and add anti-drift guard Fixes #424. WindowsShell.Tests.ps1 hand-maintained a list of zig test invocations that missed 17 files (Accessibility.zig was fixed separately by #421; GraphContextMenu.zig/MainWindow.zig are reserved for in-flight #418/ #422). Wires all 17 remaining orphaned files with per-file link flags verified against pinned Zig 0.15.2, and adds a structural guard that enumerates graphcode-windows\src\*.zig, detects files containing a est " block, and throws if any is missing from the wired-file list -- so this drift cannot recur silently. First-run triage: - WorktreeDialog.zig: one test used a stale fixture (dirty instead of locked) to exercise armConfirmation()'s fail-closed path; sweepSelectable() intentionally permits dirty rows. Fixed the test fixture, not the code. - App.zig: one test's App struct literal predated three fields (sidebar_state, declared_entry_ids, kept_worktree_paths) added since it last compiled. Fixed the test to match App.init()'s initialization. - Sidebar.zig: 3 real, pre-existing test failures traced to one root cause (layoutFor()/projectSectionHeight() count a recent_projects entry that is also the open project, but appendRows() correctly excludes it from rendered rows, desyncing row/scroll y-math). Per explicit instruction not to modify Sidebar.zig source, these are quarantined at the harness level with an explicit reason string, not fixed or deleted. Reported as a real product bug for separate follow-up. - App.zig transitively reruns the same 3 Sidebar failures (it imports Sidebar.zig); quarantined identically with a note explaining why. RED: WorktreeDialog and App.zig tests failed on first run -> both had stale fixtures, not product bugs; fixed the tests to match current code. GREEN: all 17 newly-wired files now execute; 90/93 Sidebar.zig and 228/231 App.zig tests pass -> the 3 residual failures are quarantined with cause. REGRESSION: full harness run end-to-end with pinned Zig 0.15.2 exits 0 and the anti-drift guard fails loudly on a real mutation removing a wired file -> guard verified functional, not merely asserted. Signed-off-by: Colin Neilens <coneilen@microsoft.com> Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> * Run the anti-drift guard last so CI still validates every zig test invocation Moves the structural guard added for #424 to the end of WindowsShell.Tests.ps1, after all zig test invocations, instead of right after Resolve-TestZig. Placed first, the guard's expected failure (reserving GraphContextMenu.zig/MainWindow.zig for #418/#422) short-circuited the whole script in CI before any of the newly-wired tests ever ran on the actual runner, leaving only local verification as evidence. Placed last, CI now executes and reports every invocation for real before the guard's contract check runs, while the guard still fails the job overall until #418/#422 lands. RED: with the guard first, CI failed at the guard on the first push and never exercised a single newly-wired zig test -> no real CI signal existed for the wiring itself, only local runs. GREEN: relocated the guard after every Invoke-Native/Invoke-NativeQuarantined call and reran the full harness locally with pinned Zig 0.15.2 -> all 37 invocations execute (90/93 Sidebar.zig, 228/231 App.zig, everything else 100%), quarantines are tolerated correctly, and the guard still throws last naming exactly GraphContextMenu.zig and MainWindow.zig. REGRESSION: reran the same real-mutation guard check (removing a wired file name) after relocating the block -> guard still throws the identical not-wired message immediately, confirming the guard's detection logic is unchanged, only its position in the script moved. Signed-off-by: Colin Neilens <coneilen@microsoft.com> Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> * Point quarantine reason at issue #428 instead of #424 The three Sidebar.zig quarantine entries pointed at #424 (this PR's own tracking issue) with the phrase "first-run finding" as a stand-in for a real bug report, since no dedicated issue existed yet for the Sidebar layout bug when it was first quarantined. The coordinating session filed #428 with the full root-cause writeup (layoutFor()/projectSectionHeight() vs appendRows()'s isProjectOpen skip, the 76px delta, and the hit-testing failure). Repointing the quarantine reason string at #428 so it resolves to the actual bug report instead of this wiring PR. RED: quarantine reason string referenced #424, which is this very PR and not a bug report -> anyone reading the quarantine message would have to guess where the real Sidebar.zig fix should land. GREEN: repointed both Sidebar.zig and App.zig quarantine reason strings to #428 and reran the full harness locally with pinned Zig 0.15.2 -> same 90/93 Sidebar.zig and 228/231 App.zig pass counts, guard still fails last on exactly GraphContextMenu.zig and MainWindow.zig, exit code unchanged. REGRESSION: confirmed Invoke-NativeQuarantined's known-failure matching is by test name string only, independent of the reason text -> the wording change cannot affect which failures are tolerated. Signed-off-by: Colin Neilens <coneilen@microsoft.com> Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> * Add UpdateOfferPresentation.zig to the guard's wired-file list Cherry-picked #420's NativeForms/UpdateOfferPresentation harness wiring (commit 775d77b) onto this branch: my branch point (68eabe5) predated that merge, so the reentrancy-guard Assert-Contract, the source-list entry, and the zig test invocation for UpdateOfferPresentation.zig were all absent here even though they exist on main. Restored via cherry-pick rather than a full rebase, per instruction to hold on rebasing until #422 lands. That cherry-pick alone was not sufficient: this PR's own anti-drift guard maintains a second, independent file list (\) that #420 never touched (the guard did not exist on main). Newly restoring the UpdateOfferPresentation.zig invocation without adding it to that list would have made the guard itself flag it as unwired. RED: after cherry-picking 775d77b, the guard's \ array still lacked "UpdateOfferPresentation.zig" -> a manual simulation of the guard's detection logic reported it as missing alongside the two entries correctly reserved for #422. GREEN: added "UpdateOfferPresentation.zig" to \ next to "NativeForms.zig" -> the same simulation now reports exactly and only GraphContextMenu.zig and MainWindow.zig as missing, matching the #422 reservation. REGRESSION: reran the full harness end-to-end locally with pinned Zig 0.15.2 and GRAPHCODE_WINGHOSTTY_ROOT set -> all 37 invocations execute (including the restored NativeForms 96/96 and UpdateOfferPresentation 1/1), the same 3 pre-existing Sidebar.zig failures are tolerated by name, and the guard still throws on exactly GraphContextMenu.zig, MainWindow.zig. Signed-off-by: Colin Neilens <coneilen@microsoft.com> Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> * Wire GraphContextMenu.zig/MainWindow.zig into the guard's list after #422; correct Sidebar quarantine attribution Rebased onto origin/main after #422 (issue #418) merged as 06e092e, which wired GraphContextMenu.zig and MainWindow.zig into the harness's zig test invocations and source-list array. #422 never touched this PR's own \ guard array, because that array did not exist on main when #422 was authored -- it is this PR's own addition. Added both filenames to the list and updated the guard's explanatory comment, which was otherwise now stale (still described the two files as reserved and not-yet-landed). Also corrected the Sidebar.zig quarantine reason: an independent review determined that only 2 of the 3 quarantined failures are the real layoutFor()/appendRows() product bug (#428, fix in flight as #430); the third (sidebar scroll clamps overflow, shrink, and resize, expected 334 found 410) is a separate, stale test expectation -- the 76px delta is the Activity block height that contentBottom/paint() correctly account for and the test's oracle omitted. Left it quarantined (not fixed) since this PR must not modify Sidebar.zig source, but the reason string now attributes each failure accurately instead of lumping all three under one root cause. RED: mechanically diffed this branch's wired-file set against origin/main after rebasing -> main had 22 entries (20 plus #422's 2), this branch still reported only 38 in its own \ guard list, and running the guard's detection logic directly showed it still flagging GraphContextMenu.zig and MainWindow.zig as unwired despite their zig test invocations now existing in the script. GREEN: added both names to \ and reran the same detection logic -> zero missing files reported. REGRESSION: ran the complete harness end-to-end locally with pinned Zig 0.15.2 and GRAPHCODE_WINGHOSTTY_ROOT set -> all 39 invocations execute including the newly-landed GraphContextMenu.zig (8/8) and MainWindow.zig (6/6, appearing twice via App.zig's transitive import), the same 3 Sidebar.zig failures are tolerated by name with the corrected attribution text rendering intact, and the full script now exits 0 -- the guard no longer fires at all. Signed-off-by: Colin Neilens <coneilen@microsoft.com> Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> * Remove Sidebar/App.zig quarantine now that #430 fixed the underlying bug PR #430 (merged as ece5935) fixed the real Sidebar.zig layout/row-count desync structurally via a single shared predicate, projectIsVisibleInSection, so all three previously-quarantined tests now pass unconditionally. Landing the quarantine would have shipped a misleading tolerate-list for an already-fixed bug, so remove it entirely instead: - Delete Invoke-NativeQuarantined and its Get-FailingZigTestNames helper (Sidebar.zig/App.zig were their only consumers). - Delete the sidebarLayoutOpenProjectKnownFailures/-Reason variables. - Convert both call sites to plain Invoke-Native. Verified end-to-end locally with pinned Zig 0.15.2: exit 0, all 93 Sidebar.zig tests and the full 232-test App.zig suite pass unconditionally, zero quarantine output. Re-ran the guard's set-difference check against origin/main (ece5935): empty, with UpdateOfferPresentation.zig and the NativeForms Assert-Contract block both still intact. Signed-off-by: Colin Neilens <coneilen@microsoft.com> Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> --------- Signed-off-by: Colin Neilens <coneilen@microsoft.com> Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

coneilen mentioned this pull request Sep 23, 2026

Wire orphaned Zig test files into WindowsShell.Tests.ps1 and add anti-drift guard #426

Merged

coneilen and others added 2 commits September 22, 2026 19:23

coneilen force-pushed the coneilen-microsoft-context-menu-uia-automation branch from 48cfa6e to 846b663 Compare September 23, 2026 02:23

coneilen merged commit 06e092e into main Sep 23, 2026
10 checks passed

Navigation Menu

Windows: inspect native context menus in the UIA live gate - #422

Windows: inspect native context menus in the UIA live gate#422
coneilen merged 2 commits into
mainfrom
coneilen-microsoft-context-menu-uia-automation

coneilen commented Sep 23, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

coneilen commented Sep 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What this adds

Honest scoping of the evidence

Verification

CI evidence (authoritative)

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

coneilen commented Sep 23, 2026 •

edited

Loading