Decide which target frameworks to support

[vaceslav]

The library targets net10.0 only. TriasDev.Templify targets net6–net10; many consumers are still on net8 (LTS).

• Try building for net8.0 and net9.0 and list what breaks (newer BCL APIs such as MemoryExtensions.Count, SearchValues, collection expressions over spans, …)
• Decide: net10 only, or net8 + net10 (LTS pair), or wider
• If multi-targeting: run the test suite on every target, and re-check the performance numbers on the oldest one

[vaceslav]

First check (2026-09-25, local, library changed only in Directory.Build.props):

• net9.0 and net8.0: the library builds with no errors — none of the newer BCL APIs it uses are net10-only.
• Test suite on net8.0: 348 of 350 pass. Two fail — SecondRoundFixesTests.StopsInsideARowFullOfElementsItWantsNothingFrom and StopsInsideAValueAssembledFromCountlessEmptyRuns — both expect an OperationCanceledException from inside a single read and get none. Needs investigating before net8 can be claimed: either the cancellation stride behaves differently there, or the tests depend on net10 timing/allocation behaviour.
• Test suite on net9.0: not completed in this check (tooling issue, not a code failure).

Support dates to weigh (per Microsoft's lifecycle): .NET 8 (LTS) and .NET 9 (STS) both reach end of support in November 2026; .NET 10 (LTS) runs to November 2028. Many consumers will still be on net8 for a while after that.

[vaceslav]

Decision (2026-09-25): target net8.0 and net10.0 — the two LTS releases. No netstandard: it would need System.Memory for spans (breaking the zero-dependency rule) and lose the BCL APIs the performance rests on. To do: multi-target in the root Directory.Build.props, fix the two cancellation tests that fail on net8, run the suite on both targets.

[vaceslav]

Also: .NET 11 is due within weeks and a release candidate is out. Install the RC locally (side by side, not in global.json), run the test suite and both benchmark projects on it, and compare with .NET 10 — does reading or analysis get faster, and does anything break? Decide then whether to add net11.0 as a target once it ships.

[vaceslav]

From the pre-release review: net8.0 + net10.0 is a one-line change in the root Directory.Build.props — library, tests, benchmarks and sample all build for both with 0 warnings; on net8 392/394 tests pass. The two failures are a test-fixture problem, not a reader problem (the runtime's deflate output decides whether the cancellation arms) — tracked in #16.

[vaceslav]

Done in 7acb0dc:

• src and tests target net8.0 + net10.0; benchmarks and samples stay on net10.
• The SonarQube workflow also installs the net8 runtime.
• Suite: 514/514 on both targets.
• The two net8 failures were fixed: their fixtures deflated to a few hundred bytes, so net8's zlib read everything before the token armed. Random attribute values keep the fixtures incompressible now. A mutation check (cancellation checks removed from the read loops) makes both tests fail on both runtimes.

Performance on the 572 MB csv:

	net8	net10
Import	6.05 s	5.25–5.32 s
Analysis	23.7–23.9 s	20.1–22.9 s

Allocations are the same on both runtimes. The numbers are recorded in the guide.

The .NET 11 RC trial stays as the optional item in #18.