test(ui): raise UI-command logic coverage to >=95% per file by azchohfi · Pull Request #645 · microsoft/winappCli

Alexandre Zollinger Chohfi (azchohfi) · 2026-07-15T18:18:40Z

PR 7 — Test coverage: UI-command logic

Raises test coverage on the winapp ui command surface (the UIA-driven inspect/click/type/screenshot/wait-for commands and the session resolver) toward the ≥95% per-file bar, part of the coverage-to-95% effort (issue #630). This slice covers command logic and the session resolver via injectable seams — the real in-process UIA COM/GPU/input layer (UiAutomationService and the Real* adapters) is deliberately left for a dedicated follow-up PR that drives a live window.

Real-workflow tests first (resolver disambiguation, auto-select, JSON envelopes, error paths), then targeted tests for the retry/warning branches — all deterministic, no live desktop required.

Per-file Debug line coverage (authoritative `coverage-report.ps1 -Configuration Debug`, full run)

All 32 UI logic files are ≥95% (min 97.6%). Highlights:

File	Before	After
Services/UiSessionService.cs	27.0%	100%
Commands/UiScreenshotCommand.cs	36.1%	100%
Commands/UiWaitForCommand.cs	50.0%	100%
Commands/UiSendKeysCommand.cs	91.7%	100%
Helpers/FrameworkHint.cs	—	100%
Commands/UiInspectCommand.cs	64.1%	99.4%
Commands/UiDragCommand.cs	89.2%	99.1%
Commands/UiSearchCommand.cs	49.5%	98.9%
Commands/UiListWindowsCommand.cs	65.1%	97.6%

(The remaining commands/helpers/models were already ≥95% or are now 100%.)

Product changes — test seams only, all behavior-preserving

No runtime-behavior change, no coverage exclusions, no visibility widened beyond internal. Production wires the real path in every case:

ISystemUiQuery window-metadata extension — GetWindowClassName / GetWindowSize / GetWindowOwner added to the seam (native P/Invoke bodies moved verbatim into SystemUiQuery). UiSessionService.PickLargestWindow is now an instance method that reads window sizes through the injected seam, so the "largest window" auto-select is unit-testable with fabricated sizes. Because UiListWindowsCommand / UiScreenshotCommand / UiAutomationService still call UiSessionService.GetWindowClassName/GetWindowInfo statically, a stateless s_sharedQuery bridges those static shims to the same real implementation (both point at the identical OS boundary; DI registers the real SystemUiQuery).
ISystemUiQuery into UiScreenshotCommand — the direct-HWND --window PID/title reads route through the seam (real handles resolve identically).
FrameworkHint.IsXamlClassName(string?) — the old IsLikelyXaml(long) (which read a class name off a live HWND) is replaced by this pure classifier; the command reads the class name through the seam and passes it in. Now unit-testable without a live XAML window.
IPollDelay / RealPollDelay — the winapp ui wait-for inter-poll delay is behind an interface (RealPollDelay is a straight Task.Delay passthrough), so the retry-loop "condition not met — keep polling" continuations are testable deterministically without a wall-clock wait.

Seam precedent: IOwnedWindowFinder/RealOwnedWindowFinder and the other Real* adapters in this same command surface.

Dead code deleted (per policy, not excluded): FrameworkHint.IsLikelyXaml(long) became unused once the command reads the class name through the seam (only a test + a doc cref referenced it; both updated).

Meaningful (non-vacuous) tests

Auto-select "largest window" now supplies distinct fabricated areas (0x222 300×300 vs 0x111 100×100; 0xC1 400×400 vs 0xC2 50×50) and asserts the largest handle wins — not merely "one of the candidates" (the prior assertion accepted either handle and proved nothing).
WM_CHAR-on-XAML warning is tested in both polarities: a XAML class (WinUIDesktopWin32WindowClass) fires the advisory warning (keys still posted), a Win32 class (Notepad) does not. Warnings route through the static ambient AnsiConsole (TextWriterLogger), so these two tests swap it to a capturing console under [DoNotParallelize] with a try/finally restore — the established pattern in this repo.
wait-for retry continuations are driven by stateful fakes (transient find-failure, present-then-gone, value-changes-across-polls) through the IPollDelay seam; each asserts the real outcome (exit code / JSON "found": true) and that a keep-polling iteration actually ran (FakePollDelay.CallCount). No wall-clock timing.

Honest ceiling

Services/SystemUiQuery.cs (native Win32/Process adapter) sits at ~27% and RealPollDelay/RealOwnedWindowFinder at ~0% — these are thin OS-boundary wrappers with no branching logic, the same honest-ceiling category as the existing RealForegroundGuard/RealMouseInput/RealKeyboardInput adapters, documented in their XML comments. They are not hand-written logic and carry no [ExcludeFromCodeCoverage].

Verification

Release build (both WinApp.Cli.csproj + WinApp.Cli.Tests.csproj, -p:UseSharedCompilation=false -nodeReuse:false): 0 warnings / 0 errors (warnings-as-errors, incl. CA1859).
Debug coverage self-verified from the cobertura report (numbers above).
Full suite: 1779 passed / 5 failed / 0 skipped — the 5 failures are exactly the pre-existing EndToEndTests.E2E_Node* tests needing the un-built Node src/winapp-npm/dist/cli.js; none from this PR.

Review

This branch was reviewed with the repo's pr-review skill (7 specialists + a multi-model cross-check across the GPT and Gemini families) before opening. It surfaced four medium findings — a vacuous "largest window" assertion and three declared coverage "ceilings" that were actually reachable once a single seam gap was closed (GetWindowClassName/GetWindowSize/GetWindowOwner were still hardcoded P/Invokes bypassing ISystemUiQuery). All four were fixed as above (the three "ceilings" are now genuinely 100%), and the vacuous test now asserts the correct largest-area window.

Co-authored-by: Copilot App 223556219+Copilot@users.noreply.github.com

Add fakes-driven tests across the UI-command surface (commands, session resolution, value setting, key parsing, helpers, models) to bring every hand-written file in scope to >=95% Debug line coverage without any real desktop/UIA dependency. Test-only additions dominate; two behavior-preserving OS-boundary seams were introduced so runnable resolver/discovery logic becomes testable with fakes (production wires the real implementations, behavior unchanged): - IOwnedWindowFinder / RealOwnedWindowFinder: extracts UiScreenshotCommand's owned-window desktop enumeration behind an interface. - ISystemUiQuery / SystemUiQuery + UiProcessInfo: abstracts UiSessionService's process enumeration and Win32 window queries; resolver decision logic (PID/name/partial-name selection, auto-select, HWND resolution) is now fully unit-testable. Also removed dead code in UiSessionService (unused GetProcessNameSafe and an unreachable RefreshWindowTitle fallback) and made ClassifyWindow internal static so its branches are directly testable. Genuine native/OS-only guards are left uncovered with explaining source comments (honest ceiling) rather than excluded; the two native seam impls (SystemUiQuery, RealOwnedWindowFinder) are 0% by design, matching the existing RealForegroundGuard category. No [ExcludeFromCodeCoverage] added to hand-written logic. No product behavior changes. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

…us/ceiling gaps Extend ISystemUiQuery with the window-metadata reads (GetWindowClassName / GetWindowSize / GetWindowOwner) so UI-command selection logic is fully testable with fakes, and close the four review findings: M1: route UiSessionService.PickLargestWindow through the injected seam and give the two auto-select tests distinct window areas so they assert the explicitly largest handle wins (previously vacuous "0x111 or 0x222"). UiSessionService.cs 27% -> 100%. M2: inject ISystemUiQuery into UiScreenshotCommand and read the direct-HWND PID/title through the seam (GetProcessIdForWindow/GetWindowText), replacing the inline PInvokes; add fake valid-HWND + owned-dialog tests. 36% -> 100%. M3: read the target class name through the seam and classify via the pure FrameworkHint.IsXamlClassName; add XAML/non-XAML warning tests (ambient console captured under [DoNotParallelize]). Deletes now-dead FrameworkHint.IsLikelyXaml. 91.7% -> 100%. M4: introduce IPollDelay/RealPollDelay and inject it into UiWaitForCommand so the "condition not met yet - keep polling" continuations (appear-after-transient, present-then-gone, value-changes-across-polls) are deterministically testable. 50% -> 100%. Seams (behavior-preserving, internal): ISystemUiQuery window-metadata extension (a UiSessionService.s_sharedQuery bridge keeps the static shims that UiAutomation- Service relies on pointed at the real impl) and IPollDelay. Production wires the real path for both. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

Copilot

Pull request overview

Raises deterministic winapp ui command coverage toward the ≥95% per-file target from issue #630.

Changes:

Adds injectable process, Win32 window, owned-window, and polling seams.
Routes screenshot, send-keys, wait-for, and session resolution through those seams.
Expands command, resolver, gesture, helper, and error-path tests.

Reviewed changes

Copilot reviewed 26 out of 26 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
`Services/UiSessionService.cs`	Uses injectable OS queries.
`Services/SystemUiQuery.cs`	Implements real process/window queries.
`Services/ISystemUiQuery.cs`	Defines the OS-query seam.
`Helpers/RealPollDelay.cs`	Implements production polling delay.
`Helpers/RealOwnedWindowFinder.cs`	Extracts owned-window enumeration.
`Helpers/IPollDelay.cs`	Defines polling abstraction.
`Helpers/IOwnedWindowFinder.cs`	Defines owned-window abstraction.
`Helpers/HostBuilderExtensions.cs`	Registers new services.
`Helpers/FrameworkHint.cs`	Makes XAML detection pure.
`Commands/UiWaitForCommand.cs`	Injects polling delay.
`Commands/UiSendKeysCommand.cs`	Injects window-class lookup.
`Commands/UiScreenshotCommand.cs`	Injects window metadata/discovery.
`Tests/UiSessionServiceTests.cs`	Expands resolver coverage.
`Tests/UiHelpersTests.cs`	Covers UI helper behavior.
`Tests/UiCommandTests.WaitFor.cs`	Covers wait and retry paths.
`Tests/UiCommandTests.SimpleVerbs.cs`	Covers simple UI commands.
`Tests/UiCommandTests.SendKeys.cs`	Covers errors and XAML warnings.
`Tests/UiCommandTests.Search.cs`	Covers search rendering/errors.
`Tests/UiCommandTests.Screenshot.cs`	Covers screenshot workflows.
`Tests/UiCommandTests.ListWindows.cs`	Covers window listing branches.
`Tests/UiCommandTests.Inspect.cs`	Covers inspect rendering/filtering.
`Tests/UiCommandTests.Gestures.cs`	Covers gesture guards and errors.
`Tests/UiCommandTests.cs`	Registers new test fakes.
`Tests/FrameworkHintTests.cs`	Updates classifier tests.
`Tests/FakeUiServices.cs`	Adds configurable UI test seams.
`Tests/BaseCommandTests.cs`	Supports custom cancellation tokens.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Alexandre Zollinger Chohfi (azchohfi) · 2026-07-15T20:13:43Z

+        try
+        {
+            var handle = new Windows.Win32.Foundation.HWND((nint)hwnd);
+            var buffer = new char[256];


Fixed in f83031c. Bumped the GetWindowText buffer back to char[512] to match the prior inline direct-HWND path this seam replaced (origin/main used 512 in two places). char[256] silently truncated window titles of 256-511 chars, so the seam is now genuinely behavior-preserving as its doc-comment claims.

Alexandre Zollinger Chohfi (azchohfi) · 2026-07-15T20:13:45Z

+        Assert.AreEqual(0, exitCode);
+        Assert.IsTrue(File.Exists(path));


Fixed in f83031c. Verified real behavior first: --window 0 with no --app clears the missing-app guard, DiscoverAllWindows returns null, then ResolveSessionAsync throws "Specify --app..." (proven by UiSessionServiceTests.ResolveSession_WhitespaceApp_ZeroHwnd_Throws) -> exit 1, no file written. Rewrote the test to model that real rejection and assert exit 1 + no screenshot, instead of the vacuous exit-0 the fake previously allowed. Added a genuine positive single-window test so the discovery-null single-capture success path stays covered.

Alexandre Zollinger Chohfi (azchohfi) · 2026-07-15T20:13:47Z

+        Assert.AreEqual(0, exitCode);
+        Assert.IsTrue(File.Exists(path));


Fixed in f83031c. --window 999999 -> GetProcessIdForWindow yields pid 0 -> DiscoverAllWindows bails -> ResolveByHwnd throws "Window HWND 999999 not found or not accessible" (proven by UiSessionServiceTests.ResolveByHwnd_WindowNotFound_Throws) -> exit 1, no file. Rewrote the test to assert the real exit-1 + no-screenshot outcome rather than the fake accepting any handle.

github-actions · 2026-07-15T18:38:11Z

Build Metrics Report

Binary Sizes

Artifact	Baseline	Current	Delta
CLI (ARM64)	36.31 MB	36.34 MB	📈 +28.0 KB (+0.08%)
CLI (x64)	36.50 MB	36.52 MB	📈 +25.5 KB (+0.07%)
MSIX (ARM64)	15.19 MB	15.21 MB	📈 +16.8 KB (+0.11%)
MSIX (x64)	16.12 MB	16.13 MB	📈 +10.4 KB (+0.06%)
NPM Package	31.70 MB	31.72 MB	📈 +24.0 KB (+0.07%)
NuGet Package	31.74 MB	31.75 MB	📈 +14.8 KB (+0.05%)

Test Results

✅ 2220 passed, 1 skipped out of 2221 tests in 700.3s (+147 tests, -3.8s vs. baseline)

Test Coverage

⚠️ 72.8% line coverage, 66.4% branch coverage · ✅ +4.3% vs. baseline

CLI Startup Time

46ms median (x64, winapp --version) · ✅ no change vs. baseline

Updated 2026-07-15 21:33:46 UTC · commit c761e29 · workflow run

… tests Addresses three auto-review findings on the UI-command coverage PR: - SystemUiQuery.GetWindowText: restore the char[512] buffer the seam replaced (the prior inline direct-HWND path used 512). char[256] silently truncated window titles 256-511 chars, making the seam's "behavior unchanged" claim inaccurate. - UiCommandTests.Screenshot: the --window 0 and invalid-HWND tests asserted success (exit 0 + file written), which the fake allowed but production rejects -- ResolveSessionAsync throws for a blank --app with a non-positive HWND, and ResolveByHwnd throws for a dead handle (both proven directly by UiSessionServiceTests). Rewrote both to model the real rejection and assert exit 1 with no screenshot written, and added a genuine positive single-window test (-a) so the discovery-null -> single-capture path stays covered. All four affected files remain at 100% Debug line coverage. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

Reconcile PR7 (UI-command logic coverage) with two changes that landed on main while it was in review: - #629 (NuGet migration): auto-merged. HostBuilderExtensions.cs keeps both #629's NugetSourceProvider/NugetPackageDownloader registrations and my IOwnedWindowFinder/IPollDelay/ISystemUiQuery seam registrations. BaseCommandTests.cs: #629 and this branch independently added the same 3-arg ParseAndInvokeWithCaptureAsync overload; collapsed to one (kept #629's 2-arg delegator + my doc-comment) to avoid a duplicate method. - #603 (--allow-system-keys on ui send-keys): reconciled both product and tests so the feature AND my IPollDelay/ISystemUiQuery seam both survive. UiSendKeysCommand.cs: kept #603's AllowSystemKeysOption, warnings list, never-bypassable/soft-combo guard rewrite and JSON Warnings, plus my ISystemUiQuery systemQuery ctor param and IsXamlClassName seam read. UiCommandTests.SendKeys.cs: manual conflict resolution keeps the exact union of both sides' test methods (41 total, no duplicates). Debug 0W/0E; Release both projects 0W/0E. UiCommandTests + UiSessionServiceTests + SystemKeyGuard: 310 passed / 0 failed / 0 skipped. All 32 UI-scope files remain >=95% (UiSendKeysCommand 232/232=100%). Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

…629)" This reverts the NuGet-client migration (#629) to unblock ADO main builds, which fail on the prerelease NuGet.* 7.9.0-rc packages sourced from the dnceng dotnet-tools feed (not nuget.org). See #653. Conflict resolution for coverage PRs that landed on top of #629: - FakeNugetService.cs restored to its pre-#629 form. - BaseCommandTests.cs: reverted #629 changes but kept the generic ParseAndInvokeWithCaptureAsync(..., CancellationToken) overload, which is unrelated test infra used by UI-command cancellation tests (#645). - Deleted PackageInstallationServiceTests.cs and UpdateCommandTests.cs (new in #629; later expanded by #640/#641 for the migrated code). - Deleted PackageInstallationServiceCacheMarkerTests.cs (added by #641; exercised the migrated NugetService via now-removed NugetFeedTestHelpers and INugetService.IsPackageInstalled). Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> Copilot-Session: d55408ba-4e7d-41aa-95a0-f7c801632eaf

…#654) Reverts the NuGet-client migration (#629) and its stacked coverage PR (#647). Tracks #653. ## Why After migrating `NugetService` to the official NuGet client libraries, `main` builds fail on ADO due to a security issue: #629 pins the `NuGet.*` packages to the prerelease **`7.9.0-rc.36120`** sourced from the dnceng **dotnet-tools** public feed (not nuget.org). The security gate flags consuming a prerelease from a secondary public feed. This was always intended as temporary — the `NuGet.UseSystemTextJsonDeserialization` switch needed to keep the Native AOT publish clean isn't in a nuget.org stable yet. ## What this reverts - **#647** first (reverse-merge order, since it's stacked on #629): removes the added coverage tests and the `NugetPackageDownloader.cs` test seams. - **#629** next: restores the hand-rolled `NugetService`, and removes the `NuGet.*` RC pins, the `dotnet-tools` source + `packageSourceMapping` from `nuget.config`, the `.pipelines/release-nuget.config` upstream, and the `Directory.Packages.props` pins. ## Conflict resolution notes Several coverage PRs merged after #629 built on top of its surface, so a plain revert left conflicts. Resolved as follows: - `FakeNugetService.cs` — restored to its pre-#629 form. - `BaseCommandTests.cs` — reverted #629's NuGet-specific changes but **kept** the generic `ParseAndInvokeWithCaptureAsync(..., CancellationToken)` overload, which is unrelated test infrastructure used by UI-command cancellation tests from #645. - Deleted `PackageInstallationServiceTests.cs` and `UpdateCommandTests.cs` (both new in #629; later expanded by #640/#641 for the migrated code, which no longer exists). - Deleted `PackageInstallationServiceCacheMarkerTests.cs` (added by #641; exercised the migrated `NugetService` against a local folder feed via `NugetFeedTestHelpers` + `INugetService.IsPackageInstalled`, all of which are removed here). ## Validation - `dotnet build winapp.sln -c Debug` — **0 warnings / 0 errors** - Test project builds clean; UI-command and NuGet-service test suites pass (447 passed / 0 failed / 1 skipped). - No remaining references to `dotnet-tools`, `7.9.0`, or `packageSourceMapping` in `nuget.config` / `Directory.Packages.props` / `.pipelines/release-nuget.config`. ## Re-landing This migration should be reapplied **only once stable `NuGet.Protocol 7.9.0` and all related `NuGet.*` packages are published on nuget.org** (not on dnceng dotnet-tools or any other secondary public feed). At that point: repoint the pins to the nuget.org stable, and drop the `dotnet-tools` source, `packageSourceMapping`, and the release-nuget.config upstream. See #653. --------- Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

Alexandre Zollinger Chohfi (azchohfi) and others added 2 commits July 15, 2026 09:33

Copilot AI review requested due to automatic review settings July 15, 2026 18:18

Copilot started reviewing on behalf of Alexandre Zollinger Chohfi (azchohfi) July 15, 2026 18:19 View session

Copilot AI reviewed Jul 15, 2026

View reviewed changes

Alexandre Zollinger Chohfi (azchohfi) and others added 2 commits July 15, 2026 13:08

Alexandre Zollinger Chohfi (azchohfi) added the docs-not-needed label Jul 15, 2026

Alexandre Zollinger Chohfi (azchohfi) merged commit b0e02a2 into main Jul 15, 2026
22 of 24 checks passed

Alexandre Zollinger Chohfi (azchohfi) deleted the azchohfi-tests-ui-commands branch July 15, 2026 21:37

Alexandre Zollinger Chohfi (azchohfi) mentioned this pull request Jul 16, 2026

Revert NuGet-client migration (#629, #647) to unblock ADO main builds #654

Merged

Navigation Menu

Repository navigation

test(ui): raise UI-command logic coverage to >=95% per file - #645

test(ui): raise UI-command logic coverage to >=95% per file#645
Alexandre Zollinger Chohfi (azchohfi) merged 4 commits into
mainfrom
azchohfi-tests-ui-commands

Alexandre Zollinger Chohfi (azchohfi) commented Jul 15, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Alexandre Zollinger Chohfi (azchohfi) Jul 15, 2026

Uh oh!

Alexandre Zollinger Chohfi (azchohfi) Jul 15, 2026

Uh oh!

Alexandre Zollinger Chohfi (azchohfi) Jul 15, 2026

Uh oh!

github-actions Bot commented Jul 15, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

		Assert.AreEqual(0, exitCode);
		Assert.IsTrue(File.Exists(path));

Conversation

Alexandre Zollinger Chohfi (azchohfi) commented Jul 15, 2026

PR 7 — Test coverage: UI-command logic

Per-file Debug line coverage (authoritative coverage-report.ps1 -Configuration Debug, full run)

Product changes — test seams only, all behavior-preserving

Meaningful (non-vacuous) tests

Honest ceiling

Verification

Review

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Alexandre Zollinger Chohfi (azchohfi) Jul 15, 2026

Choose a reason for hiding this comment

Uh oh!

Alexandre Zollinger Chohfi (azchohfi) Jul 15, 2026

Choose a reason for hiding this comment

Uh oh!

Alexandre Zollinger Chohfi (azchohfi) Jul 15, 2026

Choose a reason for hiding this comment

Uh oh!

github-actions Bot commented Jul 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Build Metrics Report

Binary Sizes

Test Results

Test Coverage

CLI Startup Time

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Per-file Debug line coverage (authoritative `coverage-report.ps1 -Configuration Debug`, full run)

github-actions Bot commented Jul 15, 2026 •

edited

Loading