fix(#871): add injectable seams to the QfcQueue enqueue path

[drmoisan]

Make the QfcQueue enqueue path injectable and cover it headlessly

Summary

• Adds six internal seams to the QfcQueue enqueue path so it can be exercised without a live Outlook process or a real WPF dispatcher.
• Splits QuickFiler/Controllers/QfcQueue.cs into two new partial parts. The file stood at 507 lines, already over the repository's 500-line ceiling before any seam was added, so the split was a precondition rather than a cleanup.
• Adds a headless MSTest/Moq/FluentAssertions regression suite of 28 cases across two new partial test files.
• Raises line coverage of QuickFiler/Controllers/QfcQueue.Enqueue.cs from 0.152941 (13/85) to 1.000000 (85/85).
• Introduces no coverage-exclusion attribute and no assembly-level exclusion anywhere; the uncovered remainder is recorded as an explicit residual instead.
• Leaves a separately tracked out-of-scope defect deliberately unrepaired, and proves it was left untouched.

Why

The enqueue path could not be unit-tested. Every interesting branch reached out to a process-wide UI dispatcher, a static viewer queue, or an Outlook-bound move monitor, so no test could observe it without a live host. That is why the pre-change line rate on the enqueue part was 0.152941.

The fix is injection rather than rewriting: six narrow seams, each with a non-null production default and an ArgumentNullException setter guard, so production behaviour is unchanged while a test can substitute any single collaborator.

Two constraints shaped the design:

• The move-monitor interface is internal while QfcQueue is public, so an interface-typed constructor parameter or public property would be an inconsistent-accessibility error. MoveMonitor is therefore an internal property over the existing field.
• A new narrow IUiIdleDispatcher was introduced rather than reusing the existing dispatcher abstraction, because that abstraction expresses no priority for two of the three call shapes and would have silently promoted two call sites, changing when background page construction runs.

What Changed

Core production change

• QuickFiler/Controllers/QfcQueue.cs — seam S1 MoveMonitor; two regions moved out. 507 to 269 lines.
• QuickFiler/Controllers/QfcQueue.Tlp.cs (new) — the Tlp Manipulation region moved verbatim, plus seams S3 ItemViewerFactory, S4 ViewerRowPlacer, S5 ItemGroupFactory, S6 BackgroundTlpFactory. 329 lines.
• QuickFiler/Controllers/QfcQueue.UiIdle.cs (new) — the Helper Methods region moved verbatim, the UiThreadIdleDispatcher production adapter, and seam S2 UiIdleDispatcher. 108 lines.
• QuickFiler/Interfaces/IUiIdleDispatcher.cs (new) — three InvokeIdleAsync shapes. 35 lines.
• QuickFiler/Controllers/QfcQueue.Enqueue.cs — call sites routed through the seams. Unchanged at 200 lines.
• QuickFiler/QuickFiler.csproj — three <Compile Include> items. This is a legacy non-SDK project with no implicit source glob, so a missing item does not present as a missing-file error; it presents as the seam member not existing.

Tests

• QuickFiler.Test/Controllers/QfcQueueEnqueueTests.cs (new, 425 lines) and QfcQueueEnqueueTests.Harness.cs (new, 343 lines) — one partial class, harness separated from test methods.
• QuickFiler.Test/QuickFiler.Test.csproj — two <Compile Include> items.

No existing test file was modified.

Docs and evidence

• Feature spec.md and issue.md updated; 74 evidence artifacts under the feature folder's evidence/ tree.

Architecture / How It Fits Together

QfcQueue becomes a four-part partial class. Each seam is an internal property with a non-null default, so a caller that substitutes nothing observes exactly the previous behaviour.

Two seam shapes are used deliberately and are not an inconsistency:

• ItemViewerFactory and BackgroundTlpFactory initialize at their declaration, because their defaults are instance-free.
• ViewerRowPlacer and ItemGroupFactory use a lazy null-coalescing-assignment getter, because their defaults are instance methods and a C# field initializer cannot reference the instance.

UiIdleDispatcher is also lazy, so constructing a queue still performs no read of the process-wide dispatcher — which is what makes headless construction possible.

IUiIdleDispatcher declares InvokeIdleAsync<T>(Func<T>) and InvokeIdleAsync<T>(Func<Task<T>>). An argument of type Func<Task<T>> is applicable to both overloads with no better candidate, so every call passes an explicit type argument to avoid a CS0121 ambiguity.

Verification

Completed

Full C# toolchain in CLAUDE.md order, completed in a single clean pass with the formatter rewriting nothing:

Step	Result
dotnet tool run csharpier format .	exit 0, rewrote nothing
dotnet tool run csharpier check .	exit 0, 1632 files, none named
msbuild TaskMaster.sln /t:Rebuild ... /p:EnableNETAnalyzers=true /p:EnforceCodeStyleInBuild=true	exit 0, 0 errors / 0 warnings
msbuild TaskMaster.sln /t:Rebuild ... /p:TreatWarningsAsErrors=true	exit 0, 0 errors / 0 warnings
vstest.console.exe QuickFiler.Test	1423 / 1423 passed, 0 failed

Test population: a 1395-case baseline plus the 28 new cases equals the 1423 observed.

Coverage, measured by a single-assembly Cobertura run:

Scope	Before	After
QfcQueue.Enqueue.cs	0.152941 (13/85)	1.000000 (85/85)
QfcQueue.cs	0.496795 (155/312)	0.703226 (109/155)
QuickFiler package	0.810521	0.816886
Genuinely new lines	n/a	0.958333 (23/24)

New-code coverage clears the 90 percent floor in CLAUDE.md. The single uncovered new line is the BackgroundTlpFactory default lambda, which no test invokes by design and which is recorded as a residual.

All eight no-behaviour-change properties were checked against the merge base, including that every relocated member moved verbatim, that no nullable pragma was added to relocated code, and that no public member of the queue class was added, removed, retyped or re-signed.

Recommended

• dotnet tool run csharpier check .
• msbuild TaskMaster.sln /t:Rebuild /m /p:Configuration=Debug "/p:Platform=Any CPU" /p:EnableNETAnalyzers=true /p:EnforceCodeStyleInBuild=true
• msbuild TaskMaster.sln /t:Rebuild /m /p:Configuration=Debug "/p:Platform=Any CPU" /p:TreatWarningsAsErrors=true
• vstest.console.exe QuickFiler.Test\bin\Debug\QuickFiler.Test.dll /InIsolation /TestCaseFilter:"TestCategory!=LiveOutlook"

Backward Compatibility / Migration Notes

No breaking change. All six seams are internal; the public surface of QfcQueue is unchanged. Every seam has a non-null production default, so untouched callers behave exactly as before.

QuickFiler/Controllers/QfcQueue.cs lost its UTF-8 byte-order mark when the repository-wide CSharpier pass normalized it. No other Write Set file carried one. This is the formatter's own output under the mandated command, not a hand edit, and both rebuild gates are clean afterwards.

Risks and Mitigations

• Relocated code could have drifted during the split. Mitigated by mechanical comparison against the merge base: 224 of 224 lines identical for the Tlp region and 34 of 34 for the Helper Methods region.
• A seam could make tests tautological, passing even if production regressed. Mitigated by asserting values that flow through production code to an observable outcome, and by one test that leaves ItemGroupFactory at its default so the production body is exercised rather than displaced. Review flagged two of nine argument assertions in a single test as comparing null against null; see Follow-ups.
• Rollback is a straight revert; nothing here is stateful and no data shape changed.

Review Guide

Suggested order:

1. QuickFiler/Interfaces/IUiIdleDispatcher.cs — smallest file, defines the new abstraction.
2. QuickFiler/Controllers/QfcQueue.cs and QfcQueue.Enqueue.cs — the real edits.
3. QuickFiler/Controllers/QfcQueue.Tlp.cs and QfcQueue.UiIdle.cs — largely mechanical moves; the new content is the seam declarations and the adapter class.
4. The two test files.
5. Evidence artifacts only if a specific figure needs corroboration.

Two things that look like defects but are deliberate:

• A commented-out ContextIdle line is retained in QfcQueue.UiIdle.cs. It travelled with a relocated method body; deleting it would break the verbatim-move property.
• In QfcQueue.Enqueue.cs the running-jobs increment sits outside the try whose finally decrements it. That is a real pre-existing defect, it is separately tracked, and this PR deliberately does not repair it so the change stays scoped.

Follow-ups

• One test calls Dequeue() without first asserting the queue count. Because Dequeue blocks and the enqueue path swallows exceptions, a future regression could hang the assembly rather than fail cleanly. A one-line count assertion would make it fail fast.
• In the nine-argument pass-through test, the home-controller and viewer assertions both compare a harness-supplied null against null, so an argument-position swap of those two slots would not be caught.
• [TestCleanup] does not restore the SynchronizationContext that [TestInitialize] clears.
• The spec.md acceptance criterion AC19 quotes a pre-change rate of 0.503205 where the measurement is 0.496795. The two sum to exactly 1.000000 — the document quoted the miss rate. No gate outcome changes, since the comparison passed against the stricter figure.
• Acceptance criterion AC22 is the one criterion not met as literally worded, and is left unchecked. Its substantive requirement holds: no untouched production or test file was modified. It fails only because the branch carries three tracked agent-memory files that the declared Write Set does not enumerate. The remedy is a one-line documentation amendment, not a code change.

GitHub Auto-close

None — GitHub CLI validation was unavailable when this body was generated, so no closing keyword is emitted.

This PR implements issue #871. Please close it manually after merge. The context bundle's author-asserted list also harvested several unrelated issue numbers from prose inside the feature documents; emitting closing keywords from that list would have closed issues this PR does not address.

🤖 Generated with Claude Code

https://claude.ai/code/session_01GSsDVPgS66HpWg9Qroc427

---

Disclosure: AC22 is unchecked, and this pull request is merged at 21 of 22

AC22 is UNCHECKED and is not being marked as passing. It is disclosed here rather than
checked off, so that the gap stays visible to anyone reading this pull request later.

Its substantive requirement is met, and that was verified. No untouched production file and no
untouched test file was modified by this delivery. The change footprint is the one the plan
declared.

What fails is the Write Set enumeration, not the work. The branch carries tracked
.claude/agent-memory/ files that the specification's declared Write Set does not enumerate.

AC22 as written is unsatisfiable by any agent-executed change in this repository.
.claude/agent-memory/ is a tracked tree and every agent writes to it during a run, so every
agent-executed delivery carries files a Write Set cannot enumerate in advance. A criterion that
cannot be satisfied in principle is a defect in the criterion rather than in the work, and it will
recur on every item until the template wording is corrected. That correction is tracked as
issue #885.

The item's own run declined to amend AC22's wording or the Write Set to make the criterion pass,
on the grounds that doing so would convert a criterion's own violation into a pass by editing the
criterion. That judgment was endorsed, and the acceptance decision was taken separately and
explicitly instead: what was measured is recorded in the checkbox, and the acceptance of the item
despite it is recorded here.