[iibaranov-IG]

Summary

• show whole KiB below 1 MiB and one-decimal MiB at or above it on server-rendered timeline cards
• apply the identical display contract to cards added by Load more
• reuse the existing FrameInfo.SizeBytes response field without changing the API
• cover 2 KB and 1.5 MB examples plus the dynamic card path

Verification

• dotnet test IPCamLapse.sln -c Release: 104 tests pass (57 IPCamLapse + 47 OpenCamInterop)
• dotnet format IPCamLapse.sln --verify-no-changes --no-restore: pass
• git diff --check: pass
• signed commit: 0e84927

Closes #7.

[elkampu]

The server and Load more paths do not yet implement one identical display contract.

The server uses C# F0/F1 with the current culture (lines 533-535), while JavaScript uses Math.round/toFixed (lines 421-423). For example, 2,560 bytes is a midpoint: .NET F0 uses midpoint-to-even and renders 2 KB, while Math.round renders 3 KB. A locale using a decimal comma can also diverge from JavaScript's decimal point. Please make rounding and culture explicitly identical in both paths and add a midpoint regression that exercises the dynamically created card rather than only searching the script text.

[iibaranov-IG]

Agreed — nice midpoint catch. Two parts of one UI should not start a philosophical dispute over whether 2.5 rounds to 2 or 3. I’ll make the rounding rule and culture invariant explicit on both server and client paths, then add a midpoint regression that exercises a dynamically created card.

[elkampu]

Exact head 0f28411 still does not exercise one shared formatting contract and fails CI on both platforms. The new test searches the rendered HTML for MidpointRounding.AwayFromZero, but Razor source is not emitted into the response, so that assertion necessarily fails. It also continues to search JavaScript source rather than execute a dynamically created card. In addition, the server FormatFrameSize MB branch still uses culture-sensitive F1, while JavaScript toFixed(1) always uses a decimal point. Please make KB and MB culture/rounding behavior explicit and identical, assert the rendered midpoint output, and execute the dynamic formatter/card path in a real JS/DOM test.

[iibaranov-IG]

Готово: форматирование MB теперь явно использует инвариантную культуру, а midpoint в KB совпадает с JavaScript (2,560 bytes → 3 KB). Устаревшие проверки Razor-исходника убраны; вместо них новый тест запускает фактические formatFrameSize и addFrame, создаёт динамическую карточку и проверяет оба граничных результата. Head: 760bf88. CI зелёный на Ubuntu и Windows, включая контейнер и CodeQL. Буду признателен за повторную проверку.

[iibaranov-IG]

Thank you for the detailed review. I aligned the server and dynamic frame-size paths on explicit, invariant formatting and half-up rounding: 2.5625 MiB now renders as 2.6 MB, and 2.5 KiB as 3 KB. I also exercised the dynamic card rendering path with those values. I do not have a .NET runtime available in my current environment, so please run the full suite on the PR branch and confirm the result from your side.

[elkampu]

Exact head 760bf88 now fixes the culture handling, executes the dynamic card path, and passes the full suite, but the MB midpoint contract still differs. At 1,310,720 bytes (exactly 1.25 MiB), the server's double ToString("F1", InvariantCulture) renders 1.2 MB while JavaScript toFixed(1) renders 1.3 MB. The current tests use 1.5 MiB, so they cannot detect this divergence. Please choose one explicit MB midpoint rule for both paths and add the 1.25 MiB regression to the server-rendered and executed dynamic-card checks.

[iibaranov-IG]

Addressed the latest review: server and dynamic-card formatting now use the same explicit half-up midpoint rule, including a 1.25 MiB → 1.3 MB regression. The JavaScript path executes the helper in the DOM test. CI is green on Ubuntu and Windows.

[elkampu]

Verified exact head b74887f against current main in a clean merge simulation. The 1.25 MiB server/dynamic-card regression now agrees at 1.3 MB; locked restore and all 134 Release solution tests pass. The full diff and current checks are green.

[elkampu]

Revalidated the up-to-date exact head 1730c3c after the normal branch update: locked restore, all 134 Release solution tests, and format verification pass locally. The 1.25 MiB regression agrees between server and dynamic card. Current Windows, Ubuntu, CodeQL, and C# analysis checks pass; container build is still running.