[holonsoftFounder]

What

Both conversion directions opened with holonsoft.FluentConditions:

IsPrepared.Requires("Prepare()").IsTrue();
data.Requires(nameof(data)).IsNotNull();

Replaced with a plain compare and the throw in a separate non inlined method. Prepare() keeps the fluent version, it runs once per converter.

Why

holonsoft.FluentConditions 3.0.1 is published to NuGet as a non optimized build. Its assembly carries DebuggableAttribute with IsJITOptimizerDisabled, so the JIT neither optimizes nor inlines anything inside it, in every process that consumes it. BenchmarkDotNet refuses to measure against such a reference at all, which is how this was found.

Separately from that: a fluent validator chain is the wrong tool for a per message guard clause regardless of how it was built.

Numbers

A/B of the same tree with and without the change, back to back on the same machine:

Method	fluent guards	plain guards	delta
Read LE	133.07 ns	68.17 ns	−48.8%
Read BE	125.90 ns	67.37 ns	−46.5%
Read reused	125.75 ns	59.59 ns	−52.6%
Read with string	66.76 ns	37.70 ns	−43.5%
Write LE	129.14 ns	110.74 ns	−14.2%
Write BE	131.19 ns	112.02 ns	−14.6%
Write with string	138.26 ns	119.97 ns	−13.2%

Allocation is unchanged in every row, the validator was a struct.

What the A/B shows is that every row improved and that reading gained far more than writing. What it does not show is why. Both directions removed exactly the same two guards, one on a reference type and one on bool, yet reading saved about 65 ns and writing about 18 ns, so the saving is not simply the cost of the two calls and nothing measured here isolates the rest of it. The likely remainder is an inlining cascade, where an opaque call at the top of a method blocks the JIT from inlining what follows. That is a hypothesis, not a result.

Compatibility

• Exception types unchanged: ArgumentNullException for a null argument, ArgumentOutOfRangeException for calling before Prepare(). The existing test that pins the second one still passes.
• The guard order on the array overloads is the original one again, so passing null to an unprepared converter still reports the unprepared state first.
• The original message text ('data' is null!) is kept, now with the parameter name attached as well, which FluentConditions dropped.
• 96/96 tests green, golden vectors unmoved.

Heads up beyond this repo

holonsoft.FluentConditions 3.0.1, holonsoft.FluentDateTime 2.1.1 and holonsoft.Utils 1.10.1 are all published as non optimized builds. All three set GeneratePackageOnBuild, so the .nupkg contains whatever configuration was built last, and an IDE build defaults to Debug. This library packs with an explicit -c Release in publish.yml and is not affected itself.

🤖 Generated with Claude Code