Use HashSet instead of Set in the type suggestion algorithm (#3883)

* use HashSet instead of Set in the type suggestion algorithm

* fix tests

* optimize jaro

* reduce allocations

* fix

* use Stack

* back to ResizeArray, no rev

* refactor GetRecordLabelsForType

* use imperative loop instead of seq

Author: vasily-kirichenko

src/fsharp/ConstraintSolver.fs

@@ -2050,7 +2050,7 @@ and ReportNoCandidatesError (csenv:ConstraintSolverEnv) (nUnnamedCallerArgs, nNa
                 let predictFields() =
                     minfo.DeclaringEntityRef.AllInstanceFieldsAsList
                     |> List.map (fun p -> p.Name.Replace("@", ""))
-                    |> Set.ofList
+                    |> System.Collections.Generic.HashSet
 
                 ErrorWithSuggestions((msgNum, FSComp.SR.csCtorHasNoArgumentOrReturnProperty(methodName, id.idText, msgText)), id.idRange, id.idText, predictFields)
             else

src/fsharp/ErrorLogger.fs

@@ -54,9 +54,9 @@ let rec findOriginalException err =
     | WrappedError(err, _)  -> findOriginalException err
     | _ -> err
 
-type Suggestions = unit -> Set<string>
+type Suggestions = unit -> Collections.Generic.HashSet<string>
 
-let NoSuggestions : Suggestions = fun () -> Set.empty
+let NoSuggestions : Suggestions = fun () -> Collections.Generic.HashSet()
 
 /// Thrown when we stop processing the F# Interactive entry or #load.
 exception StopProcessingExn of exn option with

src/fsharp/NameResolution.fs

@@ -4458,16 +4458,16 @@ and TcTyparOrMeasurePar optKind cenv (env:TcEnv) newOk tpenv (Typar(id, _, _) as
                 let predictions1 =
                     env.eNameResEnv.eTypars
                     |> Seq.map (fun p -> "'" + p.Key)
-                    |> Set.ofSeq
 
                 let predictions2 =
                     match tpenv with
                     | UnscopedTyparEnv elements ->
                         elements
                         |> Seq.map (fun p -> "'" + p.Key)
-                        |> Set.ofSeq
 
-                Set.union predictions1 predictions2
+                [ yield! predictions1 
+                  yield! predictions2 ]
+                |> HashSet
 
             let reportedId = Ident("'" + id.idText, id.idRange)
             error (UndefinedName(0, FSComp.SR.undefinedNameTypeParameter, reportedId, predictTypeParameters))
@@ -6440,7 +6440,7 @@ and FreshenObjExprAbstractSlot cenv (env: TcEnv) (implty:TType) virtNameAndArity
             let suggestVirtualMembers() =
                 virtNameAndArityPairs
                 |> List.map (fst >> fst)
-                |> Set.ofList
+                |> HashSet
 
             if containsNonAbstractMemberWithSameName then
                 errorR(ErrorWithSuggestions(FSComp.SR.tcMemberFoundIsNotAbstractOrVirtual(tcref.DisplayName, bindName), mBinding, bindName, suggestVirtualMembers))

src/fsharp/TypeChecker.fs

@@ -3,73 +3,61 @@
 /// Functions to compute the edit distance between two strings
 module internal Internal.Utilities.EditDistance
 
-/// Given an offset and a radius from that offset, 
-/// does mChar exist in that part of str?
-let inline existsInWin (mChar: char) (str: string) offset rad =
-    let startAt = max 0 (offset - rad)
-    let endAt = min (offset + rad) (String.length str - 1)  
+open System
+open System.Collections.Generic
+
+/// Given an offset and a radius from that offset, does mChar exist in that part of str?
+let inline existsInWin (mChar: char) (str: string) (offset: int) (rad: int) =
+    let startAt = Math.Max(0, offset - rad)
+    let endAt = Math.Min(offset + rad, str.Length - 1)  
     if endAt - startAt < 0 then false
     else
         let rec exists index =
             if str.[index] = mChar then true
             elif index = endAt then false
             else exists (index + 1)
         exists startAt
-
-/// The jaro distance between s1 and s2
-let jaro s1 s2 =    
-    // The radius is half of the lesser 
-    // of the two string lengths rounded up.
+    
+let jaro (s1: string) (s2: string) =    
+    // The radius is half of the lesser of the two string lengths rounded up.
     let matchRadius = 
-        let minLen = 
-            min (String.length s1) (String.length s2) in
-              minLen / 2 + minLen % 2
-
+        let minLen = Math.Min(s1.Length, s2.Length)
+        minLen / 2 + minLen % 2
+    
     // An inner function which recursively finds the number  
     // of matched characters within the radius.
     let commonChars (chars1: string) (chars2: string) =
-        let rec inner i result = 
-            match i with
-            | -1 -> result
-            | _ -> if existsInWin chars1.[i] chars2 i matchRadius
-                   then inner (i - 1) (chars1.[i] :: result)
-                   else inner (i - 1) result
-        inner (chars1.Length - 1) []
-
+        let result = ResizeArray(chars1.Length)
+        for i = 0 to chars1.Length - 1 do
+            let c = chars1.[i]
+            if existsInWin c chars2 i matchRadius then 
+                result.Add c
+        result
+    
     // The sets of common characters and their lengths as floats 
     let c1 = commonChars s1 s2
     let c2 = commonChars s2 s1
-    let c1length = float (List.length c1)
-    let c2length = float (List.length c2)
-    
-    // The number of transpositions within 
-    // the sets of common characters.
-    let transpositions = 
-        let rec inner cl1 cl2 result = 
-            match cl1, cl2 with
-            | [], _ | _, [] -> result
-            | c1h :: c1t, c2h :: c2t -> 
-                if c1h <> c2h
-                then inner c1t c2t (result + 1.0)
-                else inner c1t c2t result
-        let mismatches = inner c1 c2 0.0
+    let c1length = float c1.Count
+    let c2length = float c2.Count
+        
+    // The number of transpositions within the sets of common characters.
+    let transpositions =
+        let mutable mismatches = 0.0
+        for i = 0 to (Math.Min(c1.Count, c2.Count)) - 1 do
+            if c1.[i] <> c2.[i] then 
+                mismatches <- mismatches + 1.0
+                        
         // If one common string is longer than the other
         // each additional char counts as half a transposition
         (mismatches + abs (c1length - c2length)) / 2.0
-
-    let s1length = float (String.length s1)
-    let s2length = float (String.length s2)
-    let tLength = max c1length c2length
-
-    // The jaro distance as given by 
-    // 1/3 ( m2/|s1| + m1/|s2| + (mc-t)/mc )
-    let result = (c1length / s1length +
-                  c2length / s2length + 
-                  (tLength - transpositions) / tLength)
-                 / 3.0
-
+    
+    let tLength = Math.Max(c1length, c2length)
+    
+    // The jaro distance as given by 1/3 ( m2/|s1| + m1/|s2| + (mc-t)/mc )
+    let result = (c1length / float s1.Length + c2length / float s2.Length + (tLength - transpositions) / tLength) / 3.0
+    
     // This is for cases where |s1|, |s2| or m are zero 
-    if System.Double.IsNaN result then 0.0 else result
+    if Double.IsNaN result then 0.0 else result
 
 /// Calculates the Jaro-Winkler edit distance between two strings.
 /// The edit distance is a metric that allows to measure the amount of similarity between two strings.

src/utils/EditDistance.fs

@@ -5,8 +5,8 @@ neg01a.fs(22,8,23,7): typecheck error FS0913: Types cannot contain nested type d
 
 neg01b.fs(2,13,2,14): typecheck error FS0039: The value, constructor, namespace or type 'X' is not defined. Maybe you want one of the following:
 
+   z
+
    A
 
    B
-
-   z

tests/fsharp/typecheck/sigs/neg01.bsl

@@ -3,8 +3,8 @@ neg14a.fs(9,6,9,33): typecheck error FS0343: The type 'missingInterfaceInSignatu
 
 neg14b.fs(2,13,2,14): typecheck error FS0039: The value, constructor, namespace or type 'X' is not defined. Maybe you want one of the following:
 
+   z
+
    A
 
    B
-
-   z