Implement interpolated strings via String.Concat

docs/release-notes/.FSharp.Compiler.Service/11.0.100.md

@@ -100,5 +100,9 @@
 
 * Improvements in error and warning messages: new error FS3885 when `let!`/`use!` is the final expression in a computation expression; new warning FS3886 when a list literal contains a single tuple element (likely missing `;` separator); improved wording for FS0003, FS0025, FS0039, FS0072, FS0247, FS0597, FS0670, FS3082, and SRTP operator-not-in-scope hints. ([PR #19398](https://github.com/dotnet/fsharp/pull/19398))
 * Exception field serialization (`GetObjectData` and field-restoring constructor) is now gated behind `langversion:11` (`LanguageFeature.ExceptionFieldSerializationSupport`). With langversion ≤10, exception codegen is unchanged from pre-#19342 behavior. ([PR #19746](https://github.com/dotnet/fsharp/pull/19746))
+* Lower string-typed interpolated strings to `System.String.Concat` rather than the reflection-based `printf` engine, making them trim- and NativeAOT-compatible. This generalizes and ungates the previous all-string `String.Concat` optimization, so it now applies to every string-typed interpolation. ([Language suggestion #1108](https://github.com/fsharp/fslang-suggestions/issues/1108), [PR #19971](https://github.com/dotnet/fsharp/pull/19971))
+* Interpolated string holes (e.g. `$"{x}"`) are now formatted with invariant culture (via the `string` operator) instead of the current thread culture. ([PR #19971](https://github.com/dotnet/fsharp/pull/19971))
 
 ### Breaking Changes
+
+* `FSharp.Compiler.Syntax.SynInterpolatedStringPart.FillExpr` now carries a `SynInterpolationFormatting` value (separating .NET alignment/format from printf specifiers) instead of an `Ident option`. ([PR #19971](https://github.com/dotnet/fsharp/pull/19971))

src/Compiler/Checking/CheckFormatStrings.fsi

@@ -12,6 +12,12 @@ open FSharp.Compiler.TcGlobals
 open FSharp.Compiler.Text
 open FSharp.Compiler.TypedTree
 
+/// A flexible type variable constrained to the integer types accepted by the '%d'/'%i'/'%u' specifiers.
+val mkFlexibleIntFormatTypar: g: TcGlobals -> m: range -> TType
+
+/// A flexible type variable constrained to 'decimal', as accepted by the '%M' specifier.
+val mkFlexibleDecimalFormatTypar: g: TcGlobals -> m: range -> TType
+
 val ParseFormatString:
     m: range ->
     fragmentRanges: range list ->

src/Compiler/Service/SynExpr.fs

@@ -1090,7 +1090,7 @@ module SynExpr =
             | SynExpr.InterpolatedString(contents = contents), Dangling.Problematic _ ->
                 contents
                 |> List.exists (function
-                    | SynInterpolatedStringPart.FillExpr(qualifiers = Some _) -> true
+                    | SynInterpolatedStringPart.FillExpr(formatting = SynInterpolationFormatting.DotNet(format = Some _)) -> true
                     | _ -> false)
 
             // { (!x) with … }

src/Compiler/SyntaxTree/ParseHelpers.fs

@@ -69,6 +69,50 @@ let rhs2 (parseState: IParseState) i j =
 /// Get the range corresponding to one of the r.h.s. symbols of a grammar rule while it is being reduced
 let rhs parseState i = rhs2 parseState i i
 
+/// Split a trailing printf specifier (e.g. "%d") off an interpolated-string literal that precedes a
+/// hole. '%%' is a literal escape, not a specifier.
+let peelTrailingPrintfSpecifier (litText: string) : string * string option =
+    let n = litText.Length
+    let mutable i = 0
+    let mutable specStart = -1
+
+    while i < n && specStart < 0 do
+        if litText[i] = '%' then
+            if i + 1 < n && litText[i + 1] = '%' then
+                i <- i + 2 // '%%' escape, keep scanning
+            else
+                specStart <- i // start of a real specifier
+        else
+            i <- i + 1
+
+    // A real printf specifier ends, immediately before the hole, with a type character. Anything else
+    // (for example the explicit '%P(' placeholder syntax) is left in the literal untouched.
+    if specStart < 0 || "bscdiuxXoBeEfFgGMOAat".IndexOf litText[n - 1] < 0 then
+        litText, None
+    else
+        litText[.. specStart - 1], Some litText[specStart..]
+
+/// Build the [String literal; FillExpr hole] pair for one interpolation hole, splitting the '{x,n}'
+/// alignment out of its tuple encoding and peeling a trailing printf specifier onto the hole.
+let mkInterpolatedStringFillParts (litText: string, litRange: range, fill: SynExpr * Ident option) =
+    let fillExpr, qualifier = fill
+
+    let holeExpr, alignment =
+        match fillExpr with
+        | SynExpr.Tuple(false, [ e; (SynExpr.Const(SynConst.Int32 _, _) as n) ], _, _) -> e, Some n
+        | _ -> fillExpr, None
+
+    let litValue, formatting =
+        match qualifier, alignment with
+        | None, None ->
+            match peelTrailingPrintfSpecifier litText with
+            | lit, Some spec -> lit, SynInterpolationFormatting.Printf(spec, litRange)
+            | _, None -> litText, SynInterpolationFormatting.DotNet(None, None)
+        | _ -> litText, SynInterpolationFormatting.DotNet(alignment, qualifier)
+
+    [ SynInterpolatedStringPart.String(litValue, litRange)
+      SynInterpolatedStringPart.FillExpr(holeExpr, formatting) ]
+
 //------------------------------------------------------------------------
 // Parsing/lexing: status of #if/#endif processing in lexing, used for continuations
 // for whitespace tokens in parser specification.

src/Compiler/SyntaxTree/ParseHelpers.fsi

@@ -38,6 +38,16 @@ val rhs2: parseState: IParseState -> i: int -> j: int -> range
 
 val rhs: parseState: IParseState -> i: int -> range
 
+/// Peel a trailing printf specifier (e.g. "%d") off an interpolated-string literal that precedes a
+/// hole, returning the literal without it and the specifier text. '%%' is a literal escape.
+val peelTrailingPrintfSpecifier: litText: string -> string * string option
+
+/// Build the [String literal; FillExpr hole] pair for one interpolation hole, splitting the
+/// '{x,n}' alignment out of its tuple encoding and peeling a trailing printf specifier off the
+/// literal onto the hole.
+val mkInterpolatedStringFillParts:
+    litText: string * litRange: range * fill: (SynExpr * Ident option) -> SynInterpolatedStringPart list
+
 type LexerIfdefStackEntry =
     | IfDefIf
     | IfDefElse

src/Compiler/SyntaxTree/SyntaxTree.fs

@@ -875,7 +875,12 @@ type SynExprRecordField =
 [<NoEquality; NoComparison; RequireQualifiedAccess>]
 type SynInterpolatedStringPart =
     | String of value: string * range: range
-    | FillExpr of fillExpr: SynExpr * qualifiers: Ident option
+    | FillExpr of fillExpr: SynExpr * formatting: SynInterpolationFormatting
+
+[<NoEquality; NoComparison; RequireQualifiedAccess>]
+type SynInterpolationFormatting =
+    | DotNet of alignment: SynExpr option * format: Ident option
+    | Printf of specifier: string * range: range
 
 [<NoEquality; NoComparison; RequireQualifiedAccess>]
 type SynSimplePat =

src/Compiler/SyntaxTree/SyntaxTree.fsi

@@ -999,7 +999,16 @@ type SynExprRecordField =
 [<NoEquality; NoComparison; RequireQualifiedAccess>]
 type SynInterpolatedStringPart =
     | String of value: string * range: range
-    | FillExpr of fillExpr: SynExpr * qualifiers: Ident option
+    | FillExpr of fillExpr: SynExpr * formatting: SynInterpolationFormatting
+
+/// Represents how an interpolation hole in an interpolated string is formatted.
+[<NoEquality; NoComparison; RequireQualifiedAccess>]
+type SynInterpolationFormatting =
+    /// .NET-style formatting: optional alignment '{x,n}' and optional format '{x:fmt}'.
+    | DotNet of alignment: SynExpr option * format: Ident option
+
+    /// printf-style formatting: a single specifier, the '%d' in '%d{x}'.
+    | Printf of specifier: string * range: range
 
 /// Represents a syntax tree for simple F# patterns
 [<NoEquality; NoComparison; RequireQualifiedAccess>]

src/Compiler/TypedTree/TcGlobals.fs

@@ -792,6 +792,7 @@ type TcGlobals(
 
   let v_byte_operator_info         = makeIntrinsicValRef(fslib_MFOperators_nleref,                             "byte"                                 , None                 , Some "ToByte",    [vara],   ([[varaTy]], v_byte_ty))
   let v_sbyte_operator_info        = makeIntrinsicValRef(fslib_MFOperators_nleref,                             "sbyte"                                , None                 , Some "ToSByte",   [vara],   ([[varaTy]], v_sbyte_ty))
+  let v_string_operator_info       = makeIntrinsicValRef(fslib_MFOperators_nleref,                             "string"                               , None                 , Some "ToString",  [vara],   ([[varaTy]], v_string_ty))
   let v_int16_operator_info        = makeIntrinsicValRef(fslib_MFOperators_nleref,                             "int16"                                , None                 , Some "ToInt16",   [vara],   ([[varaTy]], v_int16_ty))
   let v_uint16_operator_info       = makeIntrinsicValRef(fslib_MFOperators_nleref,                             "uint16"                               , None                 , Some "ToUInt16",  [vara],   ([[varaTy]], v_uint16_ty))
   let v_int32_operator_info        = makeIntrinsicValRef(fslib_MFOperators_nleref,                             "int32"                                , None                 , Some "ToInt32",   [vara],   ([[varaTy]], v_int32_ty))
@@ -1594,6 +1595,7 @@ type TcGlobals(
 
   member _.byte_operator_info       = v_byte_operator_info
   member _.sbyte_operator_info      = v_sbyte_operator_info
+  member _.string_operator_info     = v_string_operator_info
   member _.int16_operator_info      = v_int16_operator_info
   member _.uint16_operator_info     = v_uint16_operator_info
   member _.int32_operator_info      = v_int32_operator_info
@@ -1707,7 +1709,6 @@ type TcGlobals(
   member _.seq_map_info               = v_seq_map_info
   member _.seq_singleton_info         = v_seq_singleton_info
   member _.seq_empty_info             = v_seq_empty_info
-  member _.sprintf_info               = v_sprintf_info
   member _.new_format_info            = v_new_format_info
   member _.unbox_info                 = v_unbox_info
   member _.get_generic_comparer_info  = v_get_generic_comparer_info

src/Compiler/TypedTree/TcGlobals.fsi

@@ -939,6 +939,8 @@ type internal TcGlobals =
 
     member sbyte_operator_info: IntrinsicValRef
 
+    member string_operator_info: IntrinsicValRef
+
     member sbyte_tcr: TypedTree.EntityRef
 
     member sbyte_ty: TypedTree.TType
@@ -1003,8 +1005,6 @@ type internal TcGlobals =
 
     member splice_raw_expr_vref: TypedTree.ValRef
 
-    member sprintf_info: IntrinsicValRef
-
     member sprintf_vref: TypedTree.ValRef
 
     member string_ty: TypedTree.TType

src/Compiler/TypedTree/TypedTreeOps.ExprOps.fs

@@ -1366,6 +1366,9 @@ module internal Makers =
     let mkCallNewFormat (g: TcGlobals) m aty bty cty dty ety formatStringExpr =
         mkApps g (typedExprForIntrinsic g m g.new_format_info, [ [ aty; bty; cty; dty; ety ] ], [ formatStringExpr ], m)
 
+    let mkCallStringOperator (g: TcGlobals) m argTy e =
+        mkApps g (typedExprForIntrinsic g m g.string_operator_info, [ [ argTy ] ], [ e ], m)
+
     let tryMkCallBuiltInWitness (g: TcGlobals) traitInfo argExprs m =
         let info, tinst = g.MakeBuiltInWitnessInfo traitInfo
         let vref = ValRefForIntrinsic info
@@ -1449,9 +1452,6 @@ module internal Makers =
     let mkCallSeqEmpty g m ty1 =
         mkApps g (typedExprForIntrinsic g m g.seq_empty_info, [ [ ty1 ] ], [], m)
 
-    let mkCall_sprintf (g: TcGlobals) m funcTy fmtExpr fillExprs =
-        mkApps g (typedExprForIntrinsic g m g.sprintf_info, [ [ funcTy ] ], fmtExpr :: fillExprs, m)
-
     let mkCallDeserializeQuotationFSharp20Plus g m e1 e2 e3 e4 =
         let args = [ e1; e2; e3; e4 ]
         mkApps g (typedExprForIntrinsic g m g.deserialize_quoted_FSharp_20_plus_info, [], [ mkRefTupledNoTypes g m args ], m)

src/Compiler/TypedTree/TypedTreeOps.ExprOps.fsi

@@ -208,6 +208,9 @@ module internal Makers =
     val mkCallNewFormat:
         TcGlobals -> range -> TType -> TType -> TType -> TType -> TType -> formatStringExpr: Expr -> Expr
 
+    /// Build a call to the 'string' operator (Operators.ToString) at the given argument type.
+    val mkCallStringOperator: TcGlobals -> range -> argTy: TType -> Expr -> Expr
+
     val mkCallGetGenericComparer: TcGlobals -> range -> Expr
 
     val mkCallGetGenericEREqualityComparer: TcGlobals -> range -> Expr
@@ -401,9 +404,6 @@ module internal Makers =
 
     val mkCallSeqEmpty: TcGlobals -> range -> TType -> Expr
 
-    /// Make a call to the 'isprintf' function for string interpolation
-    val mkCall_sprintf: g: TcGlobals -> m: range -> funcTy: TType -> fmtExpr: Expr -> fillExprs: Expr list -> Expr
-
     val mkCallDeserializeQuotationFSharp20Plus: TcGlobals -> range -> Expr -> Expr -> Expr -> Expr -> Expr
 
     val mkCallDeserializeQuotationFSharp40Plus: TcGlobals -> range -> Expr -> Expr -> Expr -> Expr -> Expr -> Expr

src/Compiler/pars.fsy

@@ -7162,7 +7162,7 @@ interpolatedStringParts:
      { [ SynInterpolatedStringPart.String(fst $1, rhs parseState 1) ] }
 
   | INTERP_STRING_PART interpolatedStringFill interpolatedStringParts
-     { SynInterpolatedStringPart.String(fst $1, rhs parseState 1) :: SynInterpolatedStringPart.FillExpr $2 :: $3 }
+     { mkInterpolatedStringFillParts (fst $1, rhs parseState 1, $2) @ $3 }
 
   | INTERP_STRING_PART interpolatedStringParts
      { let rbrace = parseState.InputEndPosition 1
@@ -7176,7 +7176,7 @@ interpolatedStringParts:
 interpolatedString:
   | INTERP_STRING_BEGIN_PART interpolatedStringFill interpolatedStringParts
     { let s, synStringKind, _ = $1
-      SynInterpolatedStringPart.String(s, rhs parseState 1) :: SynInterpolatedStringPart.FillExpr $2 :: $3, synStringKind }
+      mkInterpolatedStringFillParts (s, rhs parseState 1, $2) @ $3, synStringKind }
 
   | INTERP_STRING_BEGIN_END
     { let s, synStringKind, _ = $1

tests/AheadOfTime/NativeAOT/NativeAOT_Test.fsproj

@@ -0,0 +1,36 @@
+<Project Sdk="Microsoft.NET.Sdk">
+
+  <PropertyGroup>
+    <OutputType>Exe</OutputType>
+    <TargetFrameworks>net9.0</TargetFrameworks>
+    <LangVersion>preview</LangVersion>
+    <TreatWarningsAsErrors>true</TreatWarningsAsErrors>
+  </PropertyGroup>
+
+  <PropertyGroup>
+    <DisableImplicitFSharpCoreReference>true</DisableImplicitFSharpCoreReference>
+    <DisableImplicitNuGetFallbackFolder>true</DisableImplicitNuGetFallbackFolder>
+    <DisableImplicitLibraryPacksFolder>true</DisableImplicitLibraryPacksFolder>
+    <PublishAot>true</PublishAot>
+    <RuntimeIdentifier>win-x64</RuntimeIdentifier>
+  </PropertyGroup>
+
+  <PropertyGroup>
+    <FSharpBuildAssemblyFile>$(LocalFSharpBuildBinPath)/FSharp.Build.dll</FSharpBuildAssemblyFile>
+    <DotnetFscCompilerPath>$(LocalFSharpBuildBinPath)/fsc.dll</DotnetFscCompilerPath>
+    <Fsc_DotNET_DotnetFscCompilerPath>$(LocalFSharpBuildBinPath)/fsc.dll</Fsc_DotNET_DotnetFscCompilerPath>
+    <FSharpPreferNetFrameworkTools>False</FSharpPreferNetFrameworkTools>
+    <FSharpPrefer64BitTools>True</FSharpPrefer64BitTools>
+  </PropertyGroup>
+
+  <ItemGroup>
+    <Compile Include="Program.fs" />
+  </ItemGroup>
+
+  <Import Project="$(MSBuildThisFileDirectory)../../../eng/Versions.props" />
+
+  <ItemGroup>
+    <PackageReference Include="FSharp.Core" Version="$(FSharpCorePreviewPackageVersionValue)" />
+  </ItemGroup>
+
+</Project>

tests/AheadOfTime/NativeAOT/Program.fs

@@ -0,0 +1,34 @@
+module Program
+
+open System
+
+// Check a rendering against an expected string literal; a mismatch prints a "FAILED" line.
+let check (actual: string, expected: string) =
+    if actual <> expected then
+        Console.WriteLine $"FAILED: expected '{expected}' but got '{actual}'"
+
+let runChecks () =
+    let x = 42
+    let name = "world"
+    let pi = 3.14159
+    let initial = 'F'
+    check ($"answer = {x}", "answer = 42")
+    check ($"hello {name}", "hello world")
+    check ($"pi ~ {pi:F2}", "pi ~ 3.14")
+    check ($"padded:{x,6}", "padded:    42")
+    check ($"greeting %s{name}", "greeting world")
+    // Bare '%d'/'%i'/'%c'/'%M' specifiers lower to the same reflection-free path as a plain hole.
+    check ($"answer = %d{x}", "answer = 42")
+    check ($"initial = %c{initial}", "initial = F")
+
+    // The following use printf specifiers that still route through 'sprintf', so they would make the
+    // NativeAOT publish fail with IL2026/IL2070/IL3050.
+    // check ($"pi ~ %.2f{pi}", "pi ~ 3.14")
+    // check ($"value = %A{x}", "value = 42")
+
+[<EntryPoint>]
+let main _ =
+    runChecks ()
+    // Success sentinel; a failed check above printed a "FAILED" line first, so the output won't be just this.
+    Console.WriteLine "Finished"
+    0

tests/AheadOfTime/NativeAOT/check.cmd

@@ -0,0 +1,2 @@
+@echo off
+powershell -ExecutionPolicy ByPass -NoProfile -command "& """%~dp0check.ps1""""

tests/AheadOfTime/NativeAOT/check.ps1

@@ -0,0 +1,37 @@
+# Publish the test project with NativeAOT and check that it runs.
+#
+# The point of this check is that the publish succeeds: a string-typed interpolated string
+# must lower to a reflection-free form (System.String.Concat), not the reflection-based
+# printf engine. If it regresses to printf, FSharp.Reflection becomes statically reachable,
+# NativeAOT analysis emits IL2026/IL2070/IL3050, TreatWarningsAsErrors turns them into errors,
+# and this publish fails.
+
+$ErrorActionPreference = "Stop"
+
+$root = "NativeAOT_Test"
+$tfm = "net9.0"
+
+$cwd = Get-Location
+Set-Location $PSScriptRoot
+
+dotnet publish -restore -c release -f:$tfm "$root.fsproj" -bl:"$PSScriptRoot/../../../artifacts/log/Release/AheadOfTime/NativeAOT/$root.binlog"
+if (-not ($LASTEXITCODE -eq 0)) {
+    Set-Location $cwd
+    Write-Error "NativeAOT publish failed with exit code $LASTEXITCODE" -ErrorAction Stop
+}
+
+$exe = Join-Path $PSScriptRoot "bin/release/$tfm/win-x64/publish/$root.exe"
+$output = (& $exe) -join "`n"
+$exitCode = $LASTEXITCODE
+Set-Location $cwd
+
+# The app prints a "FAILED" line per mismatch and "Finished" last, so its output is exactly "Finished" only if all checks passed.
+if (-not ($exitCode -eq 0)) {
+    Write-Error "NativeAOT app crashed with exit code $exitCode.`nOutput:`n$output" -ErrorAction Stop
+}
+
+if ($output.Trim() -ne "Finished") {
+    Write-Error "NativeAOT interpolation checks failed.`nOutput:`n$output" -ErrorAction Stop
+}
+
+Write-Host "NativeAOT interpolated-string test passed."

tests/AheadOfTime/check.ps1

@@ -2,3 +2,4 @@ Write-Host "AheadOfTime: check1.ps1"
 
 Equality\check.ps1
 Trimming\check.ps1
+NativeAOT\check.ps1