Implement RequireNamedArgumentsAttribute (FS-1095) - #20340

Round-1 review council (NoBloat, CodeCompaction, Expert-review, interop
test-coverage) unanimously flagged two blocking issues in the initial
RequireNamedArgumentAttribute implementation:

[16] Recognition was broadened generically via resolveAttribPath, which
made ~30 unrelated BCL well-known attributes (CallerLineNumber, IsReadOnly,
Extension, ...) newly recognised for same-compilation-unit definitions under
released langversions - a real behavioural regression. Revert the generic
resolveAttribPath broadening and the well-known-attribute bit machinery;
recognise the attribute with a scoped, name-based MethInfoHasAttributeByName
(mirrors the FS-1053 TyconRefHasAttributeByName precedent). One uniform
by-name check covers all provenance: runtime assembly, same compilation unit,
and any other assembly - with zero blast radius.

[17] ParamArray hole: TotalNumUnnamedCallerArgs is 0 when every positional
caller argument is captured by a ParamArray, so 'C.Sum(1,2,3)' skipped the
check. Enforce on NumUnnamedCallerArgs > 0 OR non-empty ParamArrayCallerArgs.

Tests: add negative-recognition (same simple name, different namespace; F#
and C#/IL), mixed named/positional, and ParamArray positional/named cases.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 0ba0e7d9-1886-4d3b-9dd8-dee813ee144b

…ion, first-class)

Round-1 council tie-breaks (ducks A/B/C across gpt-5.6-sol, gemini-3.1-pro,
claude-opus-4.8):
- [7] overload resolution -> ADDRESS (A+B): enforcement runs after overload
  selection; positional call to the unannotated overload succeeds, to the
  annotated overload fails.
- [8] extension methods -> proven green and kept: F#-style extension member and
  C#-style extension method (ILMeth); the receiver 'this' is not counted as a
  positional caller argument.
- [18] first-class use -> ADDRESS (A+C, codify rejection): 'let f = C.Add'
  errors FS3910; explicit named forwarding 'let f x y = C.Add(x=x, y=y)' works.
- [11],[14],[15] -> REJECT (majority): by-name check is provenance-agnostic;
  named-before-positional is invalid F# syntax; module comment is accurate.

24 RequireNamedArgument tests pass; well-known-attribute regression suites
(byref/readonly/callerinfo/ORP 27, Span/Attribute/Extension 643) green.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 0ba0e7d9-1886-4d3b-9dd8-dee813ee144b

…diagnostic

Round-2 fixes:
- Guard RequireNamedArgument enforcement with AssociatedPropertyInfo.IsNone and
  NumArgSets <= 1 so property/indexer accessors and curried members (which have no
  named-argument form) are not made uncallable.
- Constructor diagnostics now name the type (via ApparentEnclosingTyconRef.DisplayName)
  instead of leaking the internal '.ctor' name.
- Fix stale test fixture comments referencing the reverted cached-flags path.
- Add regression tests: indexer getter/setter, property setter, curried member,
  constructor positional (names type) and named-call success.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 0ba0e7d9-1886-4d3b-9dd8-dee813ee144b

- Trim enforcement comment (council [102], 2-1) keeping the guard rationale.
- Add C# extension-method + ParamArray coverage (council [106]).
- Add C# optional-parameter coverage (council [108]).
- Add generic annotated-method coverage (council [109]).
All 36 feature tests pass.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 0ba0e7d9-1886-4d3b-9dd8-dee813ee144b

Adds three regression tests approved by the Round-3 review council, all
locking distinct MethInfo resolution paths against a positional-call bypass:

- C# interface slot + class override (virtual/interface-slot path)
- C# imported value-type (struct) constructor path
- method-group-to-delegate coercion (Func<int,int>(C.Ping)) still enforces

Also documents the computation-expression custom-operation limitation
(no named form, intentionally not carved out) noted by the expert reviewer.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 0ba0e7d9-1886-4d3b-9dd8-dee813ee144b

Red-green verification showed the F# primary-constructor test never guarded
the .ctor-name fix (an F# ctor's LogicalName is already the type name). The
imported C# (ILMeth) constructor is the path whose LogicalName is '.ctor', so
assert the struct-constructor diagnostic names the enclosing type 'S' - this
is the test that genuinely fails if the constructor-name branch regresses.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 0ba0e7d9-1886-4d3b-9dd8-dee813ee144b

…+ imported attribute)

Every prior test co-located the attribute with the annotated method. This adds
the missing off-diagonal: a local F# method (same compilation unit) annotated
with the attribute *type* imported from a referenced assembly - the canonical
scenario once the BCL ships the attribute and you annotate your own method.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 0ba0e7d9-1886-4d3b-9dd8-dee813ee144b

Recognise RequireNamedArgumentAttribute via the cached well-known-attribute
flags (WellKnownILAttributes bit 27 + WellKnownValAttributes bit 43 with the
matching classifyILAttrib/classifyValAttrib name entries) instead of a
per-call by-name attribute scan.

- Generalise the existing MethInfoHasWellKnownAttribute to take a
  BuiltinAttribInfo voption, so a polyfill with no BuiltinAttribInfo handle
  passes ValueNone; drop the hand-rolled MethInfoHasRequireNamedArgumentAttribute
  that duplicated its MethInfo dispatch.
- Extend resolveAttribPath so a same-compilation-unit polyfill's local ref
  resolves its PublicPath (symmetric with the compilingFSharpCore branch).
- Extract the call-site guard into CalledMeth.TryGetRequireNamedArgumentViolationName,
  reusing the existing TotalNumUnnamedCallerArgs/ParamArrayCallerArgs accessors
  rather than re-folding ArgSets; the O(1) attribute test runs first.
- Remove the dead tname_RequireNamedArgumentAttribute literal and trim
  restating comments across the feature and its tests.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 0ba0e7d9-1886-4d3b-9dd8-dee813ee144b

… C# polyfill

Rubber-duck panel (3 cross-model) consensus, zero-locality-cost wins only:
- Extract rejectsPositional/acceptsNamed/rejectsCompiled/acceptsCompiled to
  collapse the ~40 repeated 5-line assertion tails (kills magic 3910 and the
  trailing |> ignore noise); the 3 bespoke tails (v9.0 gate, two message
  assertions) stay explicit.
- Extract csPolyfill/csharpWithPolyfill for the Method-target C# fixtures.
- Fix a factually wrong test name (interface implementation, not override).

43 tests unchanged and green; -140 LOC.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 0ba0e7d9-1886-4d3b-9dd8-dee813ee144b

The F# (fsPolyfill / withPolyfillCtor) and C# (csPolyfill / csStructCtorLib)
attribute declarations were near-identical, differing only in AttributeTargets.
Introduce parameterized templates (fsPolyfillTargeting / csPolyfillTargeting +
csharpWithPolyfillTargeting) so each attribute source has a single definition.
Only the two intentional wrong-namespace (MyApp) negatives remain as separate
declarations, by design.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 0ba0e7d9-1886-4d3b-9dd8-dee813ee144b

The FS-1095 RequireNamedArgumentAttribute does not exist in the BCL
(dotnet/runtime#51451 is only a proposal); the feature recognises a
user-defined attribute by full type name. Two comments wrongly implied a
"real Method-only BCL attribute" exists. Reword both to describe the
mechanism (recognition by full type name / an AttributeUsage guard) without
asserting a BCL original.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 0ba0e7d9-1886-4d3b-9dd8-dee813ee144b

The compile-path tests each pay C#-lib-compile + full-compile overhead, and
many referenced the same C# library. Merge, per shared library, the
positional-error call sites into one compile and the named-success call sites
into one compile:
- csAnnotatedLib: 4 tests -> 2 (plain IL method + optional-arg method)
- csExtensionLib: 4 tests -> 2 (plain extension + ParamArray)
- csInterfaceLib: 3 tests -> 2 (slot + concrete receiver)

Rigor is preserved via a new count-exact helper `rejectsAllCompiled`, which
uses `withErrorMessages` (asserts exact error count AND each FS3910 message,
whose method name identifies each merged call site). 43 -> 38 tests.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 0ba0e7d9-1886-4d3b-9dd8-dee813ee144b

Merge the ~22 same-compilation-unit typecheck cases (each repeating the
polyfill + a near-identical annotated type) into a single shared `annotatedApi`
zoo plus one positive and one negative grouped test. Distinct member names keep
the merged negative fully diagnostic via count-exact `withErrorMessages`.

38 -> 18 tests, all coverage lines preserved.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 0ba0e7d9-1886-4d3b-9dd8-dee813ee144b

…rgument-research

# Conflicts:
#	src/Compiler/FSComp.txt
#	src/Compiler/xlf/FSComp.txt.cs.xlf
#	src/Compiler/xlf/FSComp.txt.de.xlf
#	src/Compiler/xlf/FSComp.txt.es.xlf
#	src/Compiler/xlf/FSComp.txt.fr.xlf
#	src/Compiler/xlf/FSComp.txt.it.xlf
#	src/Compiler/xlf/FSComp.txt.ja.xlf
#	src/Compiler/xlf/FSComp.txt.ko.xlf
#	src/Compiler/xlf/FSComp.txt.pl.xlf
#	src/Compiler/xlf/FSComp.txt.pt-BR.xlf
#	src/Compiler/xlf/FSComp.txt.ru.xlf
#	src/Compiler/xlf/FSComp.txt.tr.xlf
#	src/Compiler/xlf/FSComp.txt.zh-Hans.xlf
#	src/Compiler/xlf/FSComp.txt.zh-Hant.xlf

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

Keep the single-diagnostic assertion and include collection contents in its failure output.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

…rgument-research

Keep forwarding explicitly non-inline after List.forall2 became inline. Check only Test.test, retain a call-site local-closure positive control, and reject unrelated helper allocations and missing methods.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

The Unix ILDASM driver accepts only a hyphen-prefixed switch; a slash is parsed as a second input filename. Reproduced the failure with the native Linux ILDASM and verified method isolation, allocation controls, and missing-method output with -item=.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>