Fix #18841: let _ = &expr now compiles like let x = &expr

[T-Gro]

Summary

Fixes #18841

let _ = &expr was incorrectly triggering FS0421 ("The address of the variable cannot be used at this point") while the equivalent let x = &expr compiled fine.

Root Cause

The wildcard short-circuit in TcLetBinding rewrote let _ = e as a Sequential expression, which caused PostInferenceChecks to evaluate � with PermitByRefExpr.No and emit FS0421 for address-of expressions.

Fix

Skip the wildcard short-circuit when the RHS has a byref-like type so the binding keeps its real Expr.Let shape and the byref-permitting context is preserved.

Tests

Added component tests in AddressOf.fs covering:

• let _ = &expr (the reported bug)
• Various forms: struct fields, array elements, mutable locals
• Regression guards confirming named bindings and plain discards still work

[github-actions]

❗ Release notes required

---

✅ Found changes and release notes in following paths:

Change path	Release notes path	Description
src/Compiler	docs/release-notes/.FSharp.Compiler.Service/11.0.100.md

[T-Gro]

Review Summary

Verdict: LGTM ✅

Root Cause Analysis

The fix correctly identifies the problem: TcLetBinding has a shortcut that rewrites let _ = expr into Expr.Sequential(expr, continuation). In PostInferenceChecks.CheckExprLinear, the first expression of a NormalSeq sequential is checked via CheckExprNoByrefs (line 1038), which uses PermitByRefExpr.No. This rejects &expr with FS0421.

In contrast, Expr.Let bindings (line 1042-1055) set bindingContext = PermitByRefExpr.YesReturnable when the variable has a byref type, correctly allowing the RHS to be a byref expression.

Fix Correctness

The guard not (isByrefLikeTy g m tauTy) is the right check:

• isByrefLikeTy (via stripTyEqns + isByrefLikeTyconRef) covers both actual byrefs (byref<T>, inref<T>, outref<T>) and [<IsByRefLike>] structs like Span<T>.
• When the shortcut is skipped, the code falls through to the | _ -> case which creates a patternInput temporary with the byref-like type, producing a proper Expr.Let that PostInferenceChecks handles correctly.

Comment Quality

The inline comment explaining the rationale is excellent — it references the issue, explains why the exception is needed, and describes the expected invariant.

Test Coverage

Good coverage of the key scenarios (struct values, mutable locals, struct fields, parenthesized forms, function parameters) plus regression guards for named bindings and non-byref discards.

Minor style note: the new tests use Fsx + typecheck while existing tests in the file use getCompilation + compile. This is actually fine per the ComponentTests guidelines which recommend typecheck as the cheapest action for type-error-only scenarios.

No Concerns

• No performance concern — isByrefLikeTy is cached via TryIsByRefLike and called once per let binding.
• No soundness gap — the TPat_const(Const.Unit, _) arm is harmlessly guarded too (a unit-typed expr can't be byref-like).
• No binary compatibility impact.