Skip to content

Latest commit

 

History

History
1034 lines (881 loc) · 63.2 KB

testing.md

File metadata and controls

1034 lines (881 loc) · 63.2 KB

Testing and verification

CPU temperature availability wording (1.3.1)

On 2026-10-10, the interface-open failure and initial CPU temperature detail were changed to CPU temperature is not available. The native error code and unavailable/error classification are preserved. Regression coverage checks the missing-interface message and initial snapshot. Provider access, sampling and rendering behavior are unchanged.

Verification used an isolated checkout at out/release-source-1.3.1, containing only this patch, its version bump and release documentation. The local thermal-interface research and optional diagnostic were excluded. From that checkout, using the pinned developer shell:

pwsh -NoProfile -File scripts/package-release.ps1
. ./scripts/enter-dev-shell.ps1
foreach ($preset in @('windows-x64-debug', 'windows-x64-asan')) {
    cmake --preset $preset
    cmake --build --preset $preset
    ctest --preset $preset --output-on-failure
}
cmake --build --preset windows-x64-release --target format-check
cmake --build --preset windows-x64-release --target tidy
cmake --build --preset windows-x64-release --target analyze
dumpbin /dependents out/release/1.3.1/Loadbar.exe
dumpbin /headers out/release/1.3.1/Loadbar.exe
git -c core.safecrlf=false diff --check

Release packaging configured, built and passed all four CTest suites in 6.15 seconds. Debug and ASan configure/builds passed 4/4 suites in 13.52 and 32.40 seconds respectively. Format-check, the full clang-tidy scan and MSVC analysis passed. Logs remain under the checkout's ignored out/v131-*.log paths. Resource checks verified the embedded 1.3.1 version; archive and dependency evidence is in packaging.md.

The running application was not replaced or restarted. Live presentation of the wording, the full interactive matrix, hardware accuracy, controlled performance/soak and clean-machine execution remain pending. Existing 1.3.0 measurements are not measurements of this binary.

Always-visible readouts (1.3.0)

On 2026-10-09, after the typography/spacing correction, Debug, Release, RelWithDebInfo and ASan configure/builds passed all four CTest suites each (11.37 s, 6.72 s, 5.60 s and 31.99 s). Logs are out/typography-<preset>-{configure,build,test}.log. Format-check, the full clang-tidy scan and MSVC analysis passed (out/typography-{format-check,tidy,analyze}.log).

Final publication packaging reran Release configure/build and all four CTest suites with pwsh -NoProfile -File scripts/package-release.ps1: 4/4 passed in 5.89 seconds. The staged executable matches the live performance observation by hash. dumpbin /dependents and dumpbin /headers were repeated against out/release/1.3.0/Loadbar.exe, confirming Windows imports and the security flags recorded in packaging.md. Logs are out/v130-package.log and out/v130-{dependents,headers}.log.

The new setting defaults on for fresh and migrated settings. Regression coverage includes all 64 preference combinations, schema 13 validation, presentation-only collection policy, checkbox background, draft/Cancel, disabled-hover keyboard navigation and saved hover choice. Pure-model tests cover byte-unit boundaries, warm-up/retained/unavailable values, CPU aggregation, missing GPU memory, hidden device mapping, and 360 uniform/hybrid/unknown layout combinations over all edges, CPU shapes, short edges and 100–225% text scale. Checks require all CPU tiles, readouts and thermal bounds to fit without overlapping widget allocations. Component-specific readout widths are checked on every edge, with equal non-CPU graphic lengths, compact divider padding and identical geometry at 0% and 100%. Another 144 layouts cover hidden GPU/network/drives, partial rows/columns and all three content alignments. Embedded-font tests verify maximum percentages, capacity/rate strings, small-positive values, and direction symbols against their column bounds and explicit baseline band at four scales.

The existing graphics-only/hover fixtures explicitly select readouts off, preserving those regressions. New production-renderer fixtures cover all edges, 96/144/192 DPI, normal/high contrast, 1400/1600-DIP edge lengths, unchanged hover/rest cache reuse, capacity-only changes without layout movement, readout off/on, and text-resource failure/retry. A memory-only text change creates exactly one new DirectWrite layout; these offscreen counts are not whole-process performance data.

Checks found and fixed a secondary-text cache-width dependency, short-edge vertical stacking, and device-only horizontal rows inheriting a taller hybrid CPU band's height. Clang-tidy found signed shift counts in the new test; those now use unsigned literals. MSVC analysis flagged the RAM suffix lookup; it now uses checked array access. The typography pass also caught vertical thermal-label collisions with tighter spacing; groups retain that clearance, and wrapped columns budget the upright label width. No diagnostics were suppressed.

Commands, from the pinned x64 developer environment:

. ./scripts/enter-dev-shell.ps1
foreach ($preset in @('windows-x64-debug', 'windows-x64-release', 'windows-x64-relwithdebinfo', 'windows-x64-asan')) {
    cmake --preset $preset
    cmake --build --preset $preset
    ctest --preset $preset --output-on-failure
}
cmake --build --preset windows-x64-debug --target format-check
cmake --build --preset windows-x64-debug --target tidy
cmake --build --preset windows-x64-debug --target analyze
dumpbin /dependents out/build/windows-x64-release/Loadbar.exe
dumpbin /headers out/build/windows-x64-release/Loadbar.exe
git -c core.safecrlf=false diff --check

Preview PNGs are generated by CTest in out/build/windows-x64-release/render-fixtures/: readout-2-normal.png / readout-2-heavy.png are horizontal at 1400 × 60; readout-1-normal.png and readout-3-normal.png are Left and Right at 60 × 1400. readout-2-4-core.png shows four uniform cores/threads at 1400 × 60, with matching vertical fixtures named readout-1-4-core.png and readout-3-4-core.png. Each has a 1600-DIP counterpart with -1600 before the scenario, for example readout-2-1600-normal.png. The supplied reference's 3200 × 120-pixel strip corresponds to the 1600 × 60 comparison. They use explicit synthetic fixtures through the production renderer. The supplied reference is preserved in docs/media/loadbar-always-readout.png. The paired README GIFs compare readouts off/on at the same scale; see animation.md.

For manual validation, exit the current Loadbar and launch the new Release executable. Readouts should appear without hover. In Settings, toggle Always show readout and Apply; confirm the bar reflows, the hover-info checkbox enables/disables without losing its choice, and tooltips remain independent. Test both vertical edges and a shortened edge. Cancel/Close must discard uncommitted changes. Exit Loadbar or tray/bar Exit releases the AppBar. No live AppBar was launched/reconfigured by the agent. Interactive Shell/hardware accuracy, controlled performance/soak and clean-machine execution remain pending. Release packaging and publication evidence is recorded in packaging.md.

Temperature control and resource follow-up (1.2.0)

On 2026-10-09, final Debug, Release, RelWithDebInfo and ASan configure/builds passed all four CTest suites each (10.80 s, 4.15 s, 4.25 s and 26.16 s). Logs are out/v120-final-<preset>-{configure,build,test}.log. These include the review fix for coalesced temperature toggles, not only the earlier provider/rendering changes.

Coverage includes schema 12 migration/defaults/validation; the fifth native checkbox's background, keyboard order, draft/Cancel and responsive layout; disabled cold-start and zero new sensor calls; pending cancellation and prompt re-enable; completed-buffer release; last-valid ages; and no unrelated counter/peak reset. A real worker started paused verifies that off/on transitions survive configuration coalescing without performing hardware reads. Production-renderer tests cover all edges, hover/high contrast, no-temperature pixel equivalence, cache reuse, font invalidation and failed-resource recovery. Disabled temperatures do not schedule freshness/retention deadlines or trigger layout/paint changes.

Commands run from the pinned developer environment:

. ./scripts/enter-dev-shell.ps1
foreach ($preset in @('windows-x64-debug', 'windows-x64-release', 'windows-x64-relwithdebinfo', 'windows-x64-asan')) {
    cmake --preset $preset
    cmake --build --preset $preset
    ctest --preset $preset --output-on-failure
}
cmake --build --preset windows-x64-debug --target format-check
cmake --build --preset windows-x64-debug --target tidy
cmake --build --preset windows-x64-debug --target analyze
pwsh -NoProfile -File scripts/package-release.ps1
dumpbin /dependents out/release/1.2.0/Loadbar.exe
dumpbin /headers out/release/1.2.0/Loadbar.exe
git -c core.safecrlf=false diff --check

The first full clang-tidy run flagged an optional access in a new test. Binding the optional to a const reference before its checked use resolved it without suppressions or weakened assertions. Final format-check, full clang-tidy and MSVC analysis passed with no actionable project diagnostics. Logs are out/v120-final-{format-check,tidy,analyze}.log. Packaging repeated Release tests successfully (4/4, 4.02 s), verified the single-entry ZIP, and staged the exact inspected executable. Hashes and import/security evidence are in packaging.md.

The first adversarial pass had no actionable findings and independently probed failed cancellation, path rebinding, late completion after disable and cleanup. The performance pass confirmed all three original resource improvements but found that coalesced off/on settings could lose the temperature reset. The added temperature generation fixes that path independently of the GPU/network/global counter generations; review passes restart after this fix. Fake/offscreen counts are in the performance review.

The follow-up performance review found no actionable issues. Its isolated /O2 /MT /W4 /WX production-code probe (out/v120-performance-one/build-final.ps1) passed the actual-worker coalescing, Collector reset isolation, provider and renderer regressions, preserving the previously measured reductions in vendor calls and hover-layout creation.

A different final reviewer then found no actionable correctness, lifecycle, settings, rendering-cache or resource issues. Its independent three-drive probe exercised delayed successful completion after failed cancellation and rapid off/on, alongside the production temperature/vendor regressions. The production Release offscreen renderer passed 24 continuity/resize, 49 scaling and 65 fixture checks in its isolated output directory. Evidence is under out/review-v120-final-independent/. This establishes two independent, consecutive clean reviews after the coalescing fix, not completion of the manual release gates.

Independent probes (fake providers/offscreen rendering only) completed with exit code 0:

./out/vendor-capability/run-tests.ps1
./out/v120-performance-one/build-final.ps1
pwsh -NoProfile -File out/review-v120-final-independent/run.ps1
./out/build/windows-x64-release/loadbar_render_tests.exe out/review-v120-final-independent/render

Manual follow-up: exit the current application, launch the candidate, open Settings and apply Show temperatures off/on. Confirm readouts/effects disappear, ordinary metrics continue, and available temperatures return; repeat on a vertical edge and after reopening Settings. Close dismisses Settings; Exit Loadbar or tray/bar Exit ends monitoring. Live Shell/display changes, hardware accuracy, controlled 1-Hz performance, one-hour soak and clean-machine execution have not been performed by this change. The owner initially kept v1.2.0 staged, then reported successful manual use and requested publication with the remaining validation gaps disclosed. The agent did not launch or reconfigure the live AppBar. The owner's report does not establish completion of the full manual validation matrix or the performance and clean-machine gates.

Before publication, cmake --preset windows-x64-release, cmake --build --preset windows-x64-release and ctest --preset windows-x64-release --output-on-failure passed again (4/4, 4.03 s). SHA-256 verification confirmed the rebuilt executable, staged executable and sole ZIP entry were identical; both assets matched SHA256SUMS.txt. The staged full diff check reported whitespace in the byte-preserved upstream headers, notices and third-party checksum list. The project-owned diff check passed with git diff --cached --check -- . ':(exclude)third_party/**' ':(exclude)resources/fonts/OFL.txt'; all vendor files matched their pinned SHA-256 values. Upstream files were not reformatted.

Temperature tint and alignment refinement (1.2.0, unpublished)

On 2026-10-09, Debug, Release, RelWithDebInfo and ASan configure/builds succeeded and each passed all four CTest suites (10.91 s, 5.77 s, 5.88 s and 26.07 s respectively). The non-mutating format check, full clang-tidy and MSVC analysis passed with no actionable project diagnostics. Analysis logs are out/temperature-tint-tidy.log and out/temperature-tint-analyze.log.

New tests cover original icon colors through 60°C, a half blend at 75°C, final red at 90°C, the 90–95°C glow ramp for all four thermal component types, and saturation above 95°C. Production-renderer tests distinguish fractional tint/glow changes at the same displayed integer, retain high-contrast behavior and verify that network is unaffected. Resource creation counters confirm that fractional changes reuse text layouts/ink measurements and glow masks. Existing retained/unavailable and resource-failure recovery tests also pass.

Geometry and embedded-font ink checks cover all edges, both CPU shape settings, 40/60/120-DIP requests (with the readable minimum), Windows text scales 1/1.5/2.25 and 96/120/144/192 DPI. Signed and three-digit readouts clear their icons and stay within hit/paint bounds. Horizontal ink bottoms share the graphic bottom after pixel snapping. Initial test failures identified two old expectations (16-DIP icons and unchanged fractional tint); these were updated to the new behavior with separate coverage for unchanged text outside the ramps.

Commands run from the pinned developer environment:

. ./scripts/enter-dev-shell.ps1
foreach ($preset in @('windows-x64-debug', 'windows-x64-release', 'windows-x64-relwithdebinfo', 'windows-x64-asan')) {
    cmake --preset $preset
    cmake --build --preset $preset
    ctest --preset $preset --output-on-failure
}
cmake --build --preset windows-x64-debug --target format-check
cmake --build --preset windows-x64-debug --target tidy
cmake --build --preset windows-x64-debug --target analyze
dumpbin /dependents out/build/windows-x64-release/Loadbar.exe
git -c core.safecrlf=false diff --check

Release Loadbar.exe is 1,360,384 bytes, reports 1.2.0, and imports only Windows components. SHA-256: A812BA480BDAFCF7E6DDBA00CEBBC0FCCB5B4D0D82F9CF4BE5B2E6D50F3AB665. Installed-driver optional loads remain unchanged. Fresh reference previews and five temperature-ramp-top-1400x60-2x-*.png examples are in the Release render-fixtures folder. Horizontal normal/maximum-glow and left/right previews were visually inspected. These are synthetic readings, including RAM; no new hardware accuracy or live performance claim follows.

Manual check: launch the Release executable, select Top and 60 DIPs, inspect icon size and number alignment, then check Left/Right with either CPU shape setting. Temperatures appear only after valid provider readings; live RAM collection remains outside provider coverage. Exit via the tray or Settings → Exit Loadbar. No AppBar was launched during this work; interactive Windows, hardware accuracy, performance/soak and clean-machine packaging remain pending. No commit, push or release was performed.

Temperature design alignment (1.2.0, unpublished)

On 2026-10-09, the full strip was aligned to the original 1200 × 60-DIP component export. Debug, Release, RelWithDebInfo and ASan configure/builds succeeded; each passed all four CTest suites (13.21 s, 8.13 s, 8.55 s and 31.74 s respectively). Formatting, full clang-tidy and MSVC analysis passed with no actionable project diagnostics. Logs are out/temperature-spacing-tidy.log and out/temperature-spacing-analyze.log.

New deterministic tests assert the reference positions, margins, divider, icon/readout centers, equal device widths, readable temperature text, and non-overlap on all four edges at 40/60/120 DIPs and increased text scaling. DirectWrite tests check actual embedded-font ink bounds for 64°C, 100°C, 250°C and -99°C. Font-failure recovery now covers the additional cached temperature format. The shortened-Shell-edge fixture was updated from 600 to 640 DIPs: under the new spacing, shortening to 560 DIPs raises the readable minimum from 40 to 64 DIPs. The renegotiation and bounded-retry assertions remain intact.

The production renderer generated 24 normal/hot/critical comparison PNGs at 1×/2× density: temperature-reference-{top,left,right}-*-{1x,2x}-{normal,hot,critical}.png in each build's render-fixtures directory. Horizontal previews include 1200 × 60 and 1400 × 60 DIPs; vertical previews are 60 × 1400 DIPs. Release horizontal and both vertical previews were visually inspected, including comparison against the original 2× export. All example readings are synthetic. Existing temperature availability/retention/high-contrast previews were regenerated. Both local reference images match the supplied originals by SHA-256.

Commands run from the pinned developer environment:

. ./scripts/enter-dev-shell.ps1
foreach ($preset in @('windows-x64-debug', 'windows-x64-release', 'windows-x64-relwithdebinfo', 'windows-x64-asan')) {
    cmake --preset $preset
    cmake --build --preset $preset
    ctest --preset $preset --output-on-failure
}
cmake --build --preset windows-x64-debug --target format-check
cmake --build --preset windows-x64-debug --target tidy
cmake --build --preset windows-x64-debug --target analyze
dumpbin /dependents out/build/windows-x64-release/Loadbar.exe
git -c core.safecrlf=false diff --check

The Release EXE is 1,356,288 bytes, reports 1.2.0, and imports only Windows components. SHA-256: 4284D7FD9791D8E84ABB0B476155331D7E9F59DC4EE8C4943C99E61DB2752B8D. No live AppBar was launched or changed, no hardware provider was newly validated, and no performance/soak or clean-machine packaging gate was run. For manual comparison, launch the Release executable, select Top and 60 DIPs, then compare both CPU-shape settings and the Left/Right edges. Exit via the tray menu or Settings → Exit Loadbar. Real temperatures appear only when their provider has returned a valid reading; RAM collection remains unfinished. This change was not committed, published or released.

Temperatures and Geist Mono (1.2.0, unpublished)

Implemented and checked on 2026-10-09 with the unchanged pinned toolchain. Final complete CTest runs passed 4/4 in Debug (11.22 s), Release (5.88 s), RelWithDebInfo (5.12 s), and ASan (23.89 s). A subsequent test-only explicit-float correction passed the rendering suite again in all four presets. Formatting, full clang-tidy and MSVC analysis passed after fixing the optional-access, narrowing and cleanup-result findings. A final GPU error-reporting correction passed targeted clang-tidy, unit/resource tests in all four presets and MSVC analysis again. No diagnostics were disabled to obtain these results.

Coverage includes sensor selection/duplicate indices, malformed/truncated descriptors, signed values, thresholds, no initial fabricated zero, identity-scoped retention and age, one-Hz cadence, sleep/Hide/discovery failure, access errors, reconnect with unchanged identity, asynchronous completion, cancellation and retry backoff. Offscreen production-renderer tests cover all four edges, absent/normal/critical/retained/mixed readings, unchanged icons before first observation, high contrast, 100/150/200% DPI/text scaling, font weights/cache/recovery, and temperature-only repaint decisions. The repaint test uses a hidden HWND; it does not exercise live Shell behavior. Larger test fixtures were moved to owned heap storage to preserve the MSVC stack-usage check, without suppressing it.

Commands used from the developer shell, repeated after relevant fixes:

. ./scripts/enter-dev-shell.ps1
foreach ($preset in @('windows-x64-debug', 'windows-x64-release', 'windows-x64-relwithdebinfo', 'windows-x64-asan')) {
    cmake --preset $preset
    cmake --build --preset $preset
    ctest --preset $preset --output-on-failure
}
cmake --build --preset windows-x64-debug --target format-check tidy analyze
cmake --build --preset windows-x64-debug --target analyze
clang-tidy -p out/build/windows-x64-debug/tidy-commands src/telemetry/temperature.cpp tests/settings_window.cpp tests/discovery.cpp tests/continuity.cpp
foreach ($preset in @('windows-x64-debug', 'windows-x64-release', 'windows-x64-relwithdebinfo', 'windows-x64-asan')) {
    cmake --build --preset $preset
    ctest --preset $preset -R loadbar-rendering --output-on-failure
}
cmake --build --preset windows-x64-debug --target format-check analyze
cmake --build --preset windows-x64-debug --target tidy
clang-tidy -p out/build/windows-x64-debug/tidy-commands src/telemetry/temperature.cpp
foreach ($preset in @('windows-x64-debug', 'windows-x64-release', 'windows-x64-relwithdebinfo', 'windows-x64-asan')) {
    cmake --build --preset $preset
    ctest --preset $preset -R 'loadbar-unit|loadbar-resources' --output-on-failure
}
cmake --build --preset windows-x64-debug --target format-check analyze
cmake --build --preset windows-x64-release --target loadbar_temperature_probe
& .\out\build\windows-x64-release\loadbar_temperature_probe.exe --read-only
foreach ($asset in @('GeistMono.ttf', 'OFL.txt')) {
    cmake -E touch "resources/fonts/$asset"
    cmake --build --preset windows-x64-release --target loadbar
}
dumpbin /dependents out/build/windows-x64-release/Loadbar.exe
dumpbin /headers out/build/windows-x64-release/Loadbar.exe
rg -n 'LoadLibrary|LoadPackagedLibrary|GetProcAddress|DELAYLOAD' src CMakeLists.txt cmake
git -c core.safecrlf=false diff --check

The explicitly invoked read-only probe ran without elevation and returned GPU 57.3°C (main sensor, physical adapter 0), drive 1 59°C, and drive 2 41°C (device sensor 0). It opened no AppBar/window and generated no workload. These observations establish provider availability on this host only, not sensor accuracy or coverage of other hardware. Ordinary tests use fake providers. This initial check preceded the ASUS CPU/vendor GPU extension recorded below. RAM temperature remains outside provider coverage. No release was published.

Release synthetic previews are in out/build/windows-x64-release/render-fixtures/: temperature-top-60-mixed.png shows available GPU/one-drive temperatures; temperature-*-60-normal.png and temperature-*-60-critical.png demonstrate every component, including synthetic CPU/RAM. Both horizontal and vertical Release previews were visually inspected.

For the first manual test, exit the existing Loadbar through its tray/bar menu, then launch out/build/windows-x64-release/Loadbar.exe. Check the first-reading icon transition, CPU, GPU and both drive readouts, Settings sensor/status details, independent hover/tooltip toggles, Hide, all edges and thicknesses, and the new Licenses dialog. Use Exit Loadbar in Settings or Exit in the tray/bar menu for normal cleanup. Compare temperatures with the same sensor scope in a trusted hardware tool; do not equate GPU hotspot and main temperatures.

Live AppBar/DPI/text-scale/accessibility checks, actual sensor-failure retention, sleep/resume, removable-drive cancellation, integrated/linked GPUs, storage bridges, HDD sleep behavior, clean-machine/minimum-Windows execution and controlled performance/soak remain pending.

ASUS CPU and vendor GPU extension

On 2026-10-09, the final incremental builds passed all four CTest suites in Debug (12.83 s), Release (4.63 s), RelWithDebInfo (5.60 s) and ASan (25.08 s), with the same pinned toolchain. Format-check, full clang-tidy and MSVC analysis passed. The vendor regression tests inject API tables: LUID/physical-member matching, sensor filtering, missing exports, failed initialization, malformed enumeration, non-finite readings, cached bindings, exception cleanup, fallback, backoff, Hide and retained source/age. ASUS tests cover the exact read-only request, parser bounds, cadence, pending I/O, timeout, reset and shutdown. These fakes do not establish real driver ABI compatibility or sensor accuracy.

Additional commands (the four-preset build/CTest loop above was repeated):

cmake --build --preset windows-x64-debug --target format-check
cmake --build --preset windows-x64-debug --target analyze
cmake --build --preset windows-x64-debug --target tidy
cmake --build --preset windows-x64-release --target loadbar_temperature_probe
& ./out/build/windows-x64-release/loadbar_temperature_probe.exe --read-only --vendors
dumpbin /dependents out/build/windows-x64-release/Loadbar.exe
dumpbin /headers out/build/windows-x64-release/Loadbar.exe
ninja -C out/build/windows-x64-release -t query CMakeFiles/loadbar.dir/resources/fonts.rc.res
rg -n 'LoadLibrary|LoadPackagedLibrary|GetProcAddress|DELAYLOAD' src CMakeLists.txt cmake
git -c core.safecrlf=false diff --check

The Release probe ran with a non-elevated token and returned ASUS firmware CPU 97°C, Windows GPU 59.6°C, NVIDIA NVAPI GPU core 59°C, and drives 61/45°C. These are transient observations during development checks, not an idle baseline or cross-sensor accuracy comparison. IGCL initialized but reported no temperature sensors on this host; AMD hardware/runtime was unavailable for a real ADLX check. NVIDIA first bind/read took 13.48 ms; Intel initialization/enumeration took 259.79 ms. Neither is steady-state overhead.

An earlier Debug vendor probe exposed a CFG failure when invoking a cached GetProcAddress pointer after loading the NVIDIA shim. Keeping the import call inside an application lambda fixed the reproduction; subsequent Debug and Release probes exited successfully with CFG still enabled. Static-analysis findings were fixed without disabling checks. The Release import table contains Windows components only; PE headers retain x64, ASLR, DEP, high-entropy VA and CFG. Optional vendor DLLs are loaded from System32; their installed-driver dependencies still require clean-machine validation. All 15 vendor material hashes matched third_party/SHA256SUMS.

Pending: AMD hardware ABI/edge readings, Intel hardware that exposes GPU-domain sensors, other ASUS firmware variants, linked GPUs, actual failure/recovery and cancellation latency, sensor-scope accuracy comparisons, live UI, controlled performance/soak and clean-machine packaging. No AppBar was launched and no existing Loadbar process was stopped for these checks.

1.1.5 publication build

On 2026-10-09, pwsh -NoProfile -File scripts/package-release.ps1 built version 1.1.5, passed all four Release CTest suites (3.87 s), and verified the single-entry ZIP against the tested executable. Debug, RelWithDebInfo and ASan were reconfigured/rebuilt with the release version and passed 4/4 suites each (10.41 s, 4.35 s and 14.54 s respectively). Formatting and MSVC analysis passed. Clang-tidy evidence for the unchanged implementation is recorded below; release preparation changed version metadata and documentation only.

pwsh -NoProfile -File scripts/package-release.ps1
. ./scripts/enter-dev-shell.ps1
foreach ($preset in @('windows-x64-debug', 'windows-x64-relwithdebinfo', 'windows-x64-asan')) {
    cmake --preset $preset
    cmake --build --preset $preset
    ctest --preset $preset --output-on-failure
}
cmake --build --preset windows-x64-debug --target format-check analyze
dumpbin /dependents out/release/1.1.5/Loadbar.exe
dumpbin /headers out/release/1.1.5/Loadbar.exe
rg -n 'LoadLibrary|LoadPackagedLibrary|GetProcAddress|DELAYLOAD' src CMakeLists.txt cmake
git -c core.safecrlf=false diff --check

Package hashes and import/header inspection are in packaging.md. The owner approved the tooltip changes after manual use. The detailed interactive matrix, hardware accuracy, controlled performance/soak and clean-machine execution remain pending.

Independent tooltips and compact quantities (1.1.5)

Implemented on 2026-10-09. Model tests cover one-decimal MB/GB formatting for capacities, rates and peaks, zero versus tiny positive readings, unit boundaries, invalid values, warm-up, failures and retained observations. RAM/GPU popup tests keep matching capacities and scope. Ordinary inline/Settings formats remain unchanged.

Settings tests exercise all 16 combinations of four preference flags, schema 1–10 migration, schema 11 validation, checkbox draft/cancellation, background pixels, keyboard order and responsive layout. A hidden native tooltip records pop/activation requests for all four inline/popup combinations and verifies cache invalidation. Offscreen rendering tests verify that tooltip enablement does not alter pixels and that inline-only and tooltip-only modes work. Existing cache tests cover sample/layout changes and retention-deadline expiry.

Commands run from the pinned developer environment:

. ./scripts/enter-dev-shell.ps1
cmake --build --preset windows-x64-debug
ctest --preset windows-x64-debug --output-on-failure
foreach ($preset in @('windows-x64-release', 'windows-x64-relwithdebinfo', 'windows-x64-asan')) {
    cmake --preset $preset
    cmake --build --preset $preset
    ctest --preset $preset --output-on-failure
}
cmake --build --preset windows-x64-debug --target format-check tidy analyze
cmake --build out/build/windows-x64-debug/analyze
cmake --build --preset windows-x64-debug --target format-check analyze
clang-tidy -p out/build/windows-x64-debug/tidy-commands src/model/model.cpp tests/metrics_upgrade.cpp tests/settings_window.cpp
clang-tidy -p out/build/windows-x64-debug/tidy-commands tests/settings_window.cpp
foreach ($preset in @('windows-x64-debug', 'windows-x64-release', 'windows-x64-relwithdebinfo', 'windows-x64-asan')) {
    cmake --build --preset $preset
    ctest --preset $preset -R loadbar-unit --output-on-failure
}
foreach ($preset in @('windows-x64-debug', 'windows-x64-release', 'windows-x64-relwithdebinfo', 'windows-x64-asan')) {
    cmake --build --preset $preset
    ctest --preset $preset -R loadbar-settings --output-on-failure
}
dumpbin /dependents out/build/windows-x64-release/Loadbar.exe
dumpbin /headers out/build/windows-x64-release/Loadbar.exe
rg -n 'LoadLibrary|LoadPackagedLibrary|GetProcAddress|DELAYLOAD' src CMakeLists.txt cmake
git -c core.safecrlf=false diff --check

All four configurations passed all four CTest suites. Formatting and MSVC analysis passed. A full clang-tidy scan found one callback conversion in the new test; using the existing Win32 pointer-conversion helper resolved it. The targeted follow-up passed. A null-handle diagnostic in that test was fixed with an explicit failure branch. Final affected unit/Settings reruns passed in all four configurations, including signed-zero formatting and checkbox keyboard order. Release dependency/header inspection found only Windows imports, x64, CFG, ASLR, DEP, high-entropy VA and an empty delay-import directory. No explicit dynamic loads were found. The owner subsequently reported that the changes look good and requested release as 1.1.5.

Pending manual checks: open Settings and try each inline/popup combination with Apply; Cancel/Close should discard unapplied changes. Confirm Show tooltips is below Show info on hover, blends into the window background and remains reachable at narrow sizes and higher DPI/text scale. Hover RAM, GPU, each disk and network: expect compact units, zero/tiny-positive distinction, current device/scope/status and no raw capacity/peak byte counts. With inline numbers off and tooltips on, check the 800 ms popup delay, sample changes and retained-age refresh while stationary. With tooltips off, confirm the tray tooltip is still available. Restart to verify persistence. The tests did not launch an AppBar, write real settings or alter the desktop; live behavior and fresh production performance remain unverified.

Settings checkbox backgrounds and Close controls (1.1.4)

On 2026-10-09, the final 1.1.4 builds passed all four CTest suites in Debug (8.76 s), Release (5.25 s), RelWithDebInfo (3.28 s) and ASan (15.08 s). The hidden Settings tests paint each preference checkbox into a DIB using WM_PRINTCLIENT, verify that empty pixels match COLOR_WINDOW, and repeat for checked, unchecked, disabled and theme-reset states. No screenshot, AppBar, collector or registry write is needed for these tests.

Footer tests cover separate left/right action groups, non-overlap, narrow/scrolling layouts, DPI/text scaling, native forward/reverse Tab order, and conditional Retry. Close discards the draft without setting application shutdown or destroying the bar, both with and without a tray icon and from a reparented scrolling footer. Exit Loadbar retains its full label.

Commands run from the pinned developer environment:

. ./scripts/enter-dev-shell.ps1
cmake --build --preset windows-x64-debug
ctest --preset windows-x64-debug --output-on-failure
cmake --preset windows-x64-release
cmake --build --preset windows-x64-release
ctest --preset windows-x64-release --output-on-failure
cmake --build --preset windows-x64-debug --target format-check tidy analyze
cmake --build out/build/windows-x64-debug/analyze
cmake --build --preset windows-x64-debug --target format-check analyze
git -c core.safecrlf=false diff --check

MSVC analysis initially could not prove the test helper's bitmap handle was non-null after its generic test assertion. An explicit failure branch resolved the diagnostic; the analysis rebuild passed without suppression. Full clang-tidy completed without actionable diagnostics. The final format/analysis rerun and diff check passed. The production change adds no drawing resources or timers. Final publication checks also ran:

. ./scripts/enter-dev-shell.ps1
cmake --preset windows-x64-debug
cmake --build --preset windows-x64-debug
ctest --preset windows-x64-debug --output-on-failure
pwsh -NoProfile -File scripts/package-release.ps1
foreach ($preset in @('windows-x64-relwithdebinfo', 'windows-x64-asan')) {
    cmake --preset $preset
    cmake --build --preset $preset
    ctest --preset $preset --output-on-failure
}
cmake --build --preset windows-x64-debug --target format-check analyze
clang-tidy -p out/build/windows-x64-debug/tidy-commands src/app/settings_ui.cpp src/app/application.cpp src/model/settings_form.cpp tests/settings_window.cpp
dumpbin /dependents out/release/1.1.4/Loadbar.exe
dumpbin /headers out/release/1.1.4/Loadbar.exe
rg -n 'LoadLibrary|LoadPackagedLibrary|GetProcAddress|DELAYLOAD' src CMakeLists.txt cmake
git -c core.safecrlf=false diff --check

All builds/tests, final format/MSVC analysis and targeted clang-tidy passed. The source search found no explicit dynamic/delayed loads. Packaging results are in packaging.md. Independent adversarial and code-level performance reviews found no actionable issues. Live high-contrast/DPI interaction and production performance measurements were not rerun.

Pending manual check: open Settings, confirm that all three preference checkboxes blend into the surrounding background, and verify the footer at normal and narrow sizes. Change a setting, choose Close, reopen Settings and confirm the unapplied change was discarded while monitoring continued. Exit Loadbar should stop monitoring normally. Repeat with available high-contrast, DPI and text-scale settings. No live desktop interaction was performed here.

1.1.3 publication build

On 2026-10-08, pwsh -NoProfile -File scripts/package-release.ps1 built version 1.1.3, passed all four Release CTest suites (2.84 s), and verified the single-entry ZIP against the tested executable. Debug, RelWithDebInfo and ASan were reconfigured and rebuilt with the release version and passed 4/4 suites each (8.05 s, 2.92 s and 14.10 s respectively).

Additional commands run:

. ./scripts/enter-dev-shell.ps1
foreach ($preset in @('windows-x64-debug', 'windows-x64-relwithdebinfo', 'windows-x64-asan')) {
    cmake --preset $preset
    cmake --build --preset $preset
    ctest --preset $preset --output-on-failure
}
cmake --build --preset windows-x64-debug --target format-check analyze
dumpbin /dependents out/release/1.1.3/Loadbar.exe
dumpbin /headers out/release/1.1.3/Loadbar.exe
rg -n 'LoadLibrary|LoadPackagedLibrary|GetProcAddress|DELAYLOAD' src CMakeLists.txt cmake
git -c core.safecrlf=false diff --check

Formatting and MSVC analysis passed. The implementation's full/targeted clang-tidy evidence is recorded below; publication changed version metadata and documentation only. Import/PE inspection and package hashes are in packaging.md. A fresh independent adversarial review and final package review were clean. The owner reports successful manual use; the detailed interactive matrix, hardware accuracy, performance/soak and clean-machine packaging checks remain pending and are disclosed in the release notes.

Edge-aware vertical layout (1.1.3)

On 2026-10-08, Debug, Release, RelWithDebInfo and ASan builds and all four CTest suites passed with the pinned toolchain. Geometry tests cover all four edges, 128 logical processors across groups, both shape settings, 40–640-DIP requests and text scales 1/1.5/2.25. They assert complete readable tiles, containment, squares, P/E sides, strip-thickness fitting and equal remaining device allocations. A review-found 128-thread uniform/unknown-class case reproduced unused cross-strip space; the fix fits occupied rows and its regression passes on all four edges. A strict equality assertion was corrected after observing that square mode can enlarge tiles further than rectangle mode at 120 DIPs; the regression requires that square mode never shrinks their cross-strip dimension.

Offscreen tests check Left/Right cache switches against a fresh renderer, text rotation, hover suppression, DPI/text-scale transitions and restoration after injected text failure. A high-contrast CPU overlay comparison verifies opposite quarter-turns at pixel level. New 60 × 1400-pixel fixtures were visually inspected on both edges: render-fixtures/{left,right}-60-{squares,rectangles}-{normal,hover}.png under each build. These use synthetic readings through the production renderer, not hardware observations.

Commands for this update (from the x64 developer environment):

. ./scripts/enter-dev-shell.ps1
cmake --preset windows-x64-debug
cmake --build --preset windows-x64-debug
ctest --preset windows-x64-debug --output-on-failure
foreach ($preset in @('windows-x64-relwithdebinfo', 'windows-x64-asan')) {
    cmake --preset $preset
    cmake --build --preset $preset
    ctest --preset $preset --output-on-failure
}
cmake --build --preset windows-x64-debug --target loadbar_readme_frames format-check tidy analyze
# After the review-found uniform-grid fix, rebuild/test the three configurations above, then:
cmake --preset windows-x64-release
cmake --build --preset windows-x64-release
ctest --preset windows-x64-release --output-on-failure
cmake --build --preset windows-x64-debug --target loadbar_readme_frames format-check analyze
clang-tidy -p out/build/windows-x64-debug/tidy-commands src/model/layout.cpp tests/design.cpp
dumpbin /dependents out/build/windows-x64-release/Loadbar.exe
git -c core.safecrlf=false diff --check

Final CTest results were 4/4 in Debug (8.59 s), Release (3.00 s), RelWithDebInfo (3.06 s) and ASan (14.15 s). Format-check, MSVC analysis, the optional README-frame helper build and the final targeted clang-tidy rerun passed. Release dependency inspection found only Windows components, with no VC++ runtime DLL import. Diff checking passed.

The complete clang-tidy pass succeeded before the uniform-grid follow-up. Its targeted rerun then flagged an intentional integer row calculation embedded in a floating-point expression; storing the row as an integer before conversion resolved it without suppression or behavior change. Two independent consecutive reviews were clean after the fitting fix and both confirmed that final expression adjustment.

Manual acceptance: use Right at 60 DIPs, toggle squares, hover each widget, then switch to Left. Confirm top-down/right and bottom-up/left inline text, opposite P/E sides, squares using the strip thickness, equal device allocations, upward meters and upright icons below. Repeat at 120 DIPs and available mixed-DPI/text-scale settings. Check disabled hover, menu, Settings and clean Exit/work-area restoration. Live AppBar, hardware accuracy, production performance/soak and standalone packaging checks remain pending; no live bar was launched.

GPU fallback, CPU squares and interaction preferences (1.1.3)

All four suites passed in Debug, Release, RelWithDebInfo and ASan on 2026-10-08 using the pinned toolchain. New regression coverage includes source priority, engine-query failure isolation, memory retention with original scope/capacity/age, adapter changes, schema-10 migration and malformed values, checkbox drafts/Cancel, a stubbed Task Manager launcher, square P/E tiles, vertical 0/50/100% pixels, hover suppression and cached-layout transitions. Tests use fake readings and offscreen/hidden windows; they do not validate hardware accuracy.

Commands run from the repository root:

. ./scripts/enter-dev-shell.ps1
cmake --preset windows-x64-debug
cmake --preset windows-x64-release
cmake --preset windows-x64-relwithdebinfo
cmake --preset windows-x64-asan
foreach ($preset in @('windows-x64-debug', 'windows-x64-release', 'windows-x64-relwithdebinfo', 'windows-x64-asan')) {
    cmake --build --preset $preset
    ctest --preset $preset --output-on-failure
}
cmake --build --preset windows-x64-debug --target format-check tidy analyze
clang-tidy -p out/build/windows-x64-debug/tidy-commands tests/continuity.cpp
cmake --build --preset windows-x64-debug --target format-check analyze
dumpbin /dependents out/build/windows-x64-release/Loadbar.exe
git diff --check

The first tidy pass found two test-code issues (unchecked optional access and vector capacity); both were corrected. A Debug configure overlapping the analysis target encountered a Ninja metadata lock; a subsequent isolated configure succeeded. Format-check and MSVC analysis passed. All project files were checked by clang-tidy; the last remaining optional-access warning needed a named reference for the analyzer to prove its guard, and the final targeted rerun above passed. No diagnostics were suppressed to obtain a pass. Release imports contain only Windows components, with no VC++ runtime DLL dependency; this is not a clean-machine packaging test. Two independent consecutive final adversarial reviews were clean; both confirmed the final test-only guard adjustment.

CTest writes previews to out/build/windows-x64-release/render-fixtures/. New files use horizontal-60- or vertical-60-, then squares-/rectangles- and vram.png/shared.png/engines-only.png. They are 1400 × 60 or 60 × 1400 pixels at 96 DPI, rendered through production code with synthetic readings and no hover text. The square-mode horizontal/vertical previews were visually inspected. No live bar was launched or altered; manual AppBar/DPI interaction, GPU accuracy, performance/soak and standalone packaging remain pending. The existing running copy was left alone.

1.1.1 publication build

pwsh -NoProfile -File scripts/package-release.ps1 configured and built the normal Release preset with version 1.1.1; all four CTest suites passed in 3.20 s. The script verified the single-entry ZIP against the tested executable. Artifact hashes and import/PE inspection are recorded in packaging.md. The production source is unchanged from the reviewed and tested CPU-sizing update below; this pass updates version resources, release documentation and media. No live AppBar was launched.

Logical-processor/performance and CPU sizing update

The 2026-10-08 update adds one tile per logical processor, reusable GPU aggregation scratch, explicit provider deactivation, bounded glow masks and NIC retry backoff. The subsequent CPU sizing correction allocates width to the tile grid and shares the remainder among devices. Its regression tests check four/eight logical tiles, spacing, thickness growth, hidden widgets and alignment. The RAM ellipsis fixture uses an 800-DIP edge so its ordinary capacity fits while oversized capacities still exercise truncation with full tooltip values. Regressions cover independent SMT colors/status/retention/high contrast, reordered identities, full processor-group visibility, scratch release versus ordinary reset, backoff deadlines/recovery, glow bounds, cache reuse and native-resource failure recovery. The four-core/eight-thread preview is render-fixtures/uniform4-8threads-60-normal.png; the four-thread version is render-fixtures/uniform4-60-normal.png, both under the candidate build: synthetic readings, 1400 × 60 pixels at 96 DPI, produced by the normal renderer without an AppBar or hover text.

Final warning-clean incremental builds and CTest results after the CPU sizing correction:

Configuration Result Duration
Debug 4/4 passed 9.09 s
Release candidate 4/4 passed 4.52 s
RelWithDebInfo 4/4 passed 4.52 s
ASan 4/4 passed; no reported sanitizer errors 14.67 s

Commands ran from the pinned developer environment. Configure was followed by build/test; small subsequent fixes used incremental builds and repeated all four suites:

. ./scripts/enter-dev-shell.ps1
cmake --preset windows-x64-debug
cmake --build --preset windows-x64-debug
ctest --preset windows-x64-debug --output-on-failure
cmake --preset windows-x64-relwithdebinfo
cmake --build --preset windows-x64-relwithdebinfo
ctest --preset windows-x64-relwithdebinfo --output-on-failure
cmake --preset windows-x64-asan
cmake --build --preset windows-x64-asan
ctest --preset windows-x64-asan --output-on-failure
cmake --preset windows-x64-release -B out/build/windows-x64-release-candidate
cmake --build out/build/windows-x64-release-candidate
ctest --test-dir out/build/windows-x64-release-candidate --output-on-failure
cmake --build --preset windows-x64-debug --target format-check
cmake --build --preset windows-x64-debug --target tidy
cmake --build --preset windows-x64-debug --target analyze
dumpbin /dependents out/build/windows-x64-release-candidate/Loadbar.exe
git diff --check

Format-check, MSVC analysis and git diff --check passed. The full clang-tidy run found one implicit integer-to-float conversion in the new test assertion. After changing its literals to 6.0F/48.0F, the affected file passed the same configured checks:

clang-tidy -p out/build/windows-x64-debug/tidy-commands tests/design.cpp

All four configurations were rebuilt and all suites passed again after that test-only fix; the unchanged files had passed the full tidy run. Final repetitions used cmake --build out/build/<directory> and ctest --test-dir out/build/<directory> --output-on-failure, for windows-x64-debug, windows-x64-release-candidate, windows-x64-relwithdebinfo and windows-x64-asan.

The isolated Release folder leaves the running published executable untouched. Dependency inspection lists Windows components only, with no VC++ runtime DLL import; source inspection found no explicit LoadLibrary, GetProcAddress or /DELAYLOAD path. This does not replace a clean-machine packaging check. The candidate executable SHA-256 is 133FD4466BDBEC0759D6F37EC703E020B73565CBEB9E6675E2DF7D85DBD3EDAA (1,068,032 bytes).

Two consecutive independent reviews of the thread/performance changes were clean. Two further independent CPU-sizing reviews and final documentation/fixture rechecks found no actionable defects. Both reviewers ran the Debug model tests; no assertions were removed or relaxed. No warnings or tests were disabled. No live AppBar was launched/stopped/reconfigured, and no user registry settings were changed. Live SMT hardware rendering, enlarged-glow appearance, CPU-to-RAM spacing on the live desktop, controlled process/GPU performance, one-hour soak and clean-machine packaging remain pending. The performance evidence separates synthetic probe results from production measurements; review.md records independent review coverage.

Drive visibility update

The drive-checkbox implementation adds schema-9 exclusion persistence (including escaped IDs, maximum bounds and migration), hidden-provider continuity and query lifecycle tests, hidden freshness deadlines, dynamic disk-index/render-cache checks, and native checklist draft, keyboard, focus, scroll, empty-inventory and device-refresh coverage. The final builds passed all four suites: Debug 15.78 s, Release 3.95 s, RelWithDebInfo 3.95 s, ASan 30.93 s, with no reported sanitizer errors. All presets were configured using the pinned toolchain; subsequent fixes used incremental builds/tests. Format-check, full clang-tidy, MSVC analysis and git diff --check passed. After fixing the first review finding, two independent consecutive adversarial reviews were clean.

Analysis findings were fixed without weakening checks: the second large discovery fixture moved to the heap; exclusion storage uses a sorted unique vector with allocation-free empty state and nonthrowing moves, avoiding the MSVC tree container's allocating move constructor. The first review found a meaningless checkbox could appear on the empty informational row; the list is now disabled while empty and re-enabled on discovery, with native regression tests. Live hot-plug, restart/persistence, Shell placement and accessibility acceptance remain pending. No live AppBar was launched and no user registry settings or Startup executable were changed.

Earlier 1.1.0 GPU/network and CPU verification

On 2026-10-08, the unchanged pinned toolchain built Debug, Release, RelWithDebInfo and ASan. All four CTest suites passed in each configuration; ASan reported no memory errors. Final suite times were 15.76 s (Debug), 4.72 s (Release), 4.68 s (RelWithDebInfo), and 29.88 s (ASan). format-check, full tidy, MSVC analyze, and git diff --check passed. Tidy findings in new test code (unsigned literals, reserved capacity, widened multiplication) were corrected before the final passing run. Two independent consecutive adversarial reviews were clean; scope and additional probes are recorded in review.md. The configure/build/test commands are the four-preset loop below. No Loadbar instance was launched, replaced in Startup, or stopped; tests used hidden windows and offscreen rendering.

Added regressions cover all GPU/network visibility combinations, schema-8 persistence and older-settings migration, hidden/disconnected draft choices, selective provider reset and last-value/peak retention, hidden freshness deadlines, uniform SMT/grouped CPU squares, DPI/text scaling, and cached-versus-fresh rendering after visibility changes.

The requested four-core previews are render-fixtures/uniform4-60-normal.png and render-fixtures/uniform4-60-hidden.png under each build preset. They use the production renderer with synthetic readings at 1400 × 60 pixels, 96 DPI, no hover text.

Recorded 1.0.0 validation

The 1.0.0 preparation on 2026-10-08 used the pinned toolchain. All four configurations built successfully and passed all four CTest suites:

Preset CTest Test duration
windows-x64-debug 4/4 passed 14.21 s
windows-x64-release 4/4 passed 3.41 s
windows-x64-relwithdebinfo 4/4 passed 3.62 s
windows-x64-asan 4/4 passed; no reported ASan errors 26.93 s

Format-check, full clang-tidy and MSVC analysis passed. The optional README renderer was also built and analyzed. These are build/test results, not application-overhead measurements. Release import, resource and package inspection is recorded in packaging.md.

Documentation cleanup was rechecked on the same date with cmake --build --preset <preset> and ctest --preset <preset> --output-on-failure for Debug and Release: both incremental builds passed, with 4/4 tests in 14.95 s and 3.53 s respectively. Debug format-check passed. README, contract, source, tests, build configuration and artwork matched the pre-cleanup hashes; only documentation and an obsolete icon generator changed. All 43 local Markdown links resolved. ASan, full analysis and live checks were not repeated for this documentation-only cleanup.

Earlier opt-in, read-only provider smoke runs exercised CPU/RAM, both discovered disks, GPU and the selected network interface after priming. A disk idle counter occasionally exceeded 100%, correctly producing an active-time error while read/write remained valid. These checks demonstrate provider execution on the build host, not hardware accuracy or support across other systems. No load was generated or AppBar registered by those runs.

Pending: live Shell/fullscreen behavior, hardware accuracy, accessibility acceptance, performance/soak and clean-machine packaging. Native hidden-window tests and offscreen images do not establish these outcomes.

Reproduce automated checks

Run in PowerShell 7 from the repository root:

. .\scripts\enter-dev-shell.ps1
foreach ($preset in @('windows-x64-debug','windows-x64-release','windows-x64-relwithdebinfo','windows-x64-asan')) {
    cmake --preset $preset
    cmake --build --preset $preset
    ctest --preset $preset --output-on-failure
}
cmake --build --preset windows-x64-debug --target format-check
cmake --build --preset windows-x64-debug --target tidy
cmake --build --preset windows-x64-debug --target analyze

CTest uses a small native runner with checks active in Release. Ordinary suites need no specific hardware and never register a live AppBar:

Suite Coverage
loadbar-unit Rates/statuses, settings migration, topology/identity parsing, GPU aggregation, bounded delivery, discovery failures, continuity and session peaks
loadbar-resources Embedded icon, manifest, exact license bytes and version/author metadata
loadbar-rendering Production Direct2D rendering into WIC bitmaps, layout/scale, 0/50/100% colors/fills, high contrast, hover and resource failure/recovery
loadbar-settings Hidden native controls, draft/apply/cancel, keyboard/scrolling layout, monitor refresh, tray flags and resize invalidation

Deterministic cases include irregular/zero intervals, reset/reconnect, non-finite data, 64-bit counters, multiple processor groups, heterogeneous/SMT cores, invalid/duplicate/reordered PDH instances, separate GPU engines/adapters, and incomplete discovery versus confirmed removal. Geometry covers all edges, negative origins, mixed DPI, Shell-adjusted rectangles, bounded renegotiation, readable minimums and complete core visibility. Lifecycle tests cover balanced registration, reentrancy, partial startup, worker replacement and late callbacks.

Synthetic render fixtures are written to out/build/<preset>/render-fixtures/. Hidden resize tests observe the production invalidation request because invisible HWNDs do not retain an observable paint region. They do not prove compositor behavior.

Incremental resource regressions

. .\scripts\enter-dev-shell.ps1
& .\scripts\check-manifest-incremental.ps1
& .\scripts\check-manifest-incremental.ps1 -Resource License

Both modes passed during implementation with the pinned Ninja generator. They copy sources under out/build/, build, alter only the copied input and verify the resulting embedded resource after an incremental rebuild. License mode first confirms the stale artifact fails the exact-byte check. Manifest mode tests both the app and Settings test executable. Neither launches Loadbar. Icon/version/license dependency checks also remain in CMake/tests.

Optional real-provider smoke check

.\out\build\windows-x64-release\loadbar_tests.exe --live-providers

This explicitly reads real device counters and is not part of CTest. Run only when hardware access is intended; it does not create a bar or generate workload. Compare scope/interval separately before treating any reading as accurate.

First manual test

  1. Use the Release executable and default 1 Hz sampling. Record Windows build, monitor/DPI, selected devices and executable hash.
  2. Confirm one bar, every core/disk, device identities, warm-up zero and truthful unavailable states. Hover for exact values; check raw status and retained age in Settings.
  3. Try Settings and tray with keyboard only. Apply/Cancel should be disabled for an unchanged form. Resize Settings and confirm the readings area grows and the footer stays reachable.
  4. Exit through the tray/bar menu or Settings Exit Loadbar. Confirm work-area restoration and no lingering tray icon before proceeding to fault tests.

Interactive matrix — pending

Procedure Expected observation
Toggle CPU squares, hover info and click action; Cancel, Apply, restart Defaults off/on/on; draft and persistence work; squares retain P/E sizes; disabled hover has no overlays/bar tooltip; disabled click does nothing; tray/Settings/Exit remain usable
Left/right edges at 40/60/120 DIPs Every thread remains visible; icons below graphics; all meters fill upward; non-CPU widgets share remaining height
Dedicated memory absent, shared available; both fail after a valid observation Shared label/capacity used; dual failure retains last observation with original age; never-observed memory uses two-engine layout
Maximize/Snap on three displays with taskbar/another AppBar Only the selected monitor reserves the negotiated strip; no drift/overlap
In Drives, uncheck one, Cancel, uncheck again and Apply; hide all, then show one Checkbox draft/actions work; remaining widgets expand in both orientations; shared counters continue until all are hidden, then prime on showing
With Settings open, connect a new physical disk; disconnect/reconnect an excluded disk; restart New identity is checked; excluded identity stays unchecked; catalog refresh preserves draft/focus/scroll; saved exclusions survive restart
With no discovered disks, use Tab/Space and reconnect a disk Informational row has no checkbox and is skipped by Tab; arrival re-enables checked rows
Hide GPU, Network, then both; show each again and restart Removed widgets free space; choices persist; hidden collection pauses; only shown family re-primes, preserving session peaks
Uniform 4/8/many-core CPU, both orientations and text scaling Equal square cells, single row when possible, wrapping with every core present
Every edge, portrait rotation and thickness 40 → 60 → 80 → 40 Full edge and all widgets remain visible; graphics scale until capped; explicit edge survives rotation
Another AppBar shortens/lengthens the edge with unchanged readings Immediate reflow and recomputed minimum independent of new telemetry
100/125/150/200% DPI, text scale, light/dark high contrast Consistent reservation/content; distinguishable CPU load; controls and tooltips readable
Preferred monitor disconnect, primary change, reconnect Single primary fallback, retained preference, automatic return; open Settings choices refresh without losing edits
Explorer restart Tray and AppBar restored once; repeated failure leaves Settings/Exit
Second launch Existing-instance report; no additional reservation
Lock/unlock and suspend/resume, including resume while locked No catch-up bursts; re-primed counters; sampling paused until unlocked
Exclusive/borderless fullscreen on selected/other monitors No focus/topmost fight; retained reservation and restoration without activation
Normal Exit; separately forced termination in a disposable session Normal cleanup releases work area; Shell/next-launch recovery after forced termination
NIC/disk disconnect, counter failure, rediscovery failure Last-valid values retain original age; no identity swap; removal only after completed discovery
GPU reset/render-target loss Resource recovery with other providers responsive
Malformed settings or denied registry writes in a disposable profile Defaults for invalid data; save failure visible with active session retained
Narrow/tall Settings, large text, Tab/Shift+Tab and Narrator Readings resize; footer/scrolling controls reachable; stable accessible details
Tray icon hover and keyboard focus Standard Loadbar tooltip, usable menu and Exit
English/non-English Windows; integrated/discrete/multiple GPUs Language-neutral counters and correctly scoped device/memory selection

AppBar launch, display changes, Explorer restart, fault injection and load generation are opt-in manual operations, not automatic tests.

Hardware accuracy — pending

Compare CPU with matching Processor Information logical instances and intervals. Compare each disk's read/write and 100 - % Idle Time against that exact PhysicalDisk instance. Compare NIC octet deltas over the same elapsed interval and selected identity. Match GPU LUID, physical node, engine type, interval and memory scope; exercise multiple processes and reconnects.

Use Task Manager only as a sanity reference. Valid idle requires actual valid observations; record anomalous raw values before display clipping. Performance/soak procedures are in performance.md; clean-machine checks are in packaging.md.

Documentation media

The optional fixture and encoding commands are in animation.md. The 1.3.0 fixture renders paired readout-off/on sequences at 1920 × 60 with temperatures enabled. Assertions require eight P tiles and sixteen E tiles, plus equal content, CPU-tile, icon and meter-height scales. Each core in this synthetic example has one logical processor; loads vary independently using a fixed pseudorandom seed. Both GIFs were decoded as 24 one-second frames, infinitely looping; every frame matched its quantized source. Media is synthetic and is not a desktop recording. GitHub rendering remains unverified.

The final square-mode fixture passed its Release build and offscreen generation, repository format-check, targeted clang-tidy and MSVC analysis. Commands, after entering the developer shell:

cmake --build --preset windows-x64-release --target loadbar_readme_frames
./out/build/windows-x64-release/loadbar_readme_frames.exe ./out/readme-frames
python scripts/make-readme-animation.py --frames out/readme-frames --output docs/media
cmake --build --preset windows-x64-debug --target format-check
clang-tidy -p out/build/windows-x64-debug/tidy-commands tests/readme_frames.cpp
cmake --build out/build/windows-x64-debug/analyze --target loadbar_readme_frames
ctest --preset windows-x64-release --output-on-failure
git diff --check

All completed successfully; Release CTest passed 4/4 in 6.25 seconds. Clang-tidy reported only suppressed non-project/filtered diagnostics. This task changed documentation and its optional fixture/encoder; application code was unchanged. The independent performance review is recorded in performance.md.

The follow-up 8-P/16-E fixture correction was rebuilt and regenerated; format-check, targeted clang-tidy, MSVC analysis and diff checks passed again. Both corrected animations were visually inspected. Application tests were not repeated for this fixture-only correction.