follow-up: extract Ca2022Reflection.LoadHandlerType helper across CA2022 fixtures

[ottobolyos]

Context

Surfaced by the dime cycle-4 review on PR #219 (finding F-SIMP-002 LOW).

Three CA2022 test fixtures each reinvent the same reflection dance to reach the internal sealed class MTConnect.Servers.MTConnectPostResponseHandler:

• tests/MTConnect.NET-Common-Tests/Http/CA2022ShortReadTests.cs — LoadHandlerType() private method.
• tests/MTConnect.NET-Common-Tests/Http/CA2022ShortReadEdgeCaseTests.cs — inline anchor+GetType inside Invoke().
• tests/MTConnect.NET-Common-Tests/Http/CA2022NoTrimEndOnFixedBufferTests.cs — inline anchor+GetType in the IL-scan test.

All three do the same thing: anchor on a public type from MTConnect.NET-HTTP.dll (MTConnectHttpServer or MTConnectHttpResponse) to force assembly load, then asm.GetType(\"MTConnect.Servers.MTConnectPostResponseHandler\", ...). Every refactor of that internal name (which the anchor comment already flags as "the class may have been renamed") requires three fix sites.

Proposal

Extract a shared internal helper — e.g. tests/MTConnect.NET-Common-Tests/Http/Ca2022Reflection.cs:

internal static class Ca2022Reflection
{
    public static Type LoadHandlerType()
    {
        var anchor = typeof(MTConnect.Servers.Http.MTConnectHttpServer).Assembly;
        return anchor.GetType(\"MTConnect.Servers.MTConnectPostResponseHandler\", throwOnError: false)
            ?? throw new InvalidOperationException(\"MTConnectPostResponseHandler not visible via reflection.\");
    }
}

Update the three fixtures to call Ca2022Reflection.LoadHandlerType(). One rename site if the internal ever moves.

Cost / benefit

• Removes ~15 lines per site (3 sites → 1 site + 3 one-liners).
• Single blast radius for the "internal was renamed" scenario the fixtures already anticipate.
• Behaviour-equivalent — no test assertions change.

Provenance

Dime cycle-4 review on PR #219, subagent report 2026-08-21. Filed per CONVENTIONS §1.0d-trigies-septies (MEDIUM+ findings must be CLOSED, SKIP-rationale, or TRACKED before Ready flip).

[ottobolyos]

Cycle-5 scope broadening (dime cycle 5 on PR #219, findings F-SIMP-003 + F-SIMP-004 + F-IMP-009):

The scope of this issue should absorb two adjacent test-hygiene items surfaced in the cycle-5 review:

1. Signature-preference Invoke helper. After PR fix: clean every Release-pack warning and add multi-TFM CI gate #219 merges, both CA2022ShortReadTests and CA2022ShortReadEdgeCaseTests will still carry the "prefer (Stream, CancellationToken), fall back to (Stream)" reflection dance. The fallback is only load-bearing during a bisect across the F-IMP-001 fix commit; once merged, the two-branch dance can collapse to one call each — with the TargetInvocationException unwrap via ExceptionDispatchInfo folded into the shared helper. Extend Ca2022Reflection to expose:

   public static Task<byte[]?> InvokeReadRequestBytes(
       Stream? inputStream, CancellationToken cancellationToken = default);

   so both fixtures collapse to await Ca2022Reflection.InvokeReadRequestBytes(stream, ct).

2. Legacy-fallback dead code. The legacyMethod branch in both Invoke helpers becomes permanent dead code once fix: clean every Release-pack warning and add multi-TFM CI gate #219 merges. Drop the fallback in this issue's PR so the fixtures do not accumulate transient bisect-scaffolding.

3. Asymmetric unwrap. Right now CA2022ShortReadEdgeCaseTests.Invoke unwraps TargetInvocationException via ExceptionDispatchInfo.Capture(...).Throw() but CA2022ShortReadTests.Invoke does not — a latent asymmetry the next contributor will trip on. The shared helper resolves this atomically.

Net effect: Ca2022Reflection grows from a single LoadHandlerType() helper (the original F-SIMP-002 scope) to a two-method surface that covers both loader and invoker. Both fixtures shrink; one place to fix when the internal is renamed OR when a further signature revision lands.

[ottobolyos]

Cycle-6 scope extension (dime cycle 6 on PR #219, finding F-SIMP-005):

libraries/MTConnect.NET-HTTP/Ceen/Mvc/RestApiHelper.cs — cycle 5's F-IMP-005 fix landed identical 16-line blocks in both Post (lines 139-158) and PutDetail (lines 224-244): the same 10-line rationale comment + TimeoutCancellationToken.ThrowIfCancellationRequested() precheck + StreamReader construction + #if NET7_0_OR_GREATER guarded read + JsonSerializer.Deserialize<TData> call. Extract to a private helper:

private static async Task<TData> ReadJsonBodyAsync(IHttpContext context)
{
    context.Request.TimeoutCancellationToken.ThrowIfCancellationRequested();
    using var sr = new StreamReader(context.Request.Body, System.Text.Encoding.UTF8, false);
#if NET7_0_OR_GREATER
    var str = await sr.ReadToEndAsync(context.Request.TimeoutCancellationToken);
#else
    var str = await sr.ReadToEndAsync();
#endif
    return JsonSerializer.Deserialize<TData>(str);
}

Removes ~16 lines of drift risk; keeps the CA2022-precheck comment single-sourced so a future TFM prune (dropping net4x / net6.0) touches one site, not two. Behaviour-equivalent — both call sites collapse to item = await ReadJsonBodyAsync(context);.

Fold this alongside the existing Ca2022Reflection helper scope so the issue's PR closes both fixture duplication and controller duplication in one atomic simplification pass.