From 7e4079f845a2b2ad20ed567450b755cf7fb74bc5 Mon Sep 17 00:00:00 2001
From: CharlesKWON <charleskwonohjun@gmail.com>
Date: Sat, 18 Apr 2026 08:52:08 +0900
Subject: [PATCH] perf(gengo): reassociate left-leaning string-concat literal
 runs

`"a" + x + "b" + "c" + "d"` used to emit 4 Plus() calls because
the parser builds a left-leaning chain and no pair was
literal+literal. Add a reassociation step inside foldLiteralTree:
when the outer shape is `(Y + strlit1) + strlit2`, rewrite as
`Y + (strlit1+strlit2)` so the tail literals collapse. Also run
foldLiteralTree on the root BinaryExpr in emitExpr so the
outermost reassoc fires (was only running on children).

Verified: the 4-Plus case now emits 2 Plus calls (`"a" + x + "bcd"`).
FiveSql2 43/43, Harbour compat 56/56.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---
 compiler/gengo/gengo.go | 45 +++++++++++++++++++++++++++++++++--------
 1 file changed, 37 insertions(+), 8 deletions(-)

diff --git a/compiler/gengo/gengo.go b/compiler/gengo/gengo.go
index 640b812..86d7ceb 100644
--- a/compiler/gengo/gengo.go
+++ b/compiler/gengo/gengo.go
@@ -275,6 +275,13 @@ func negateLiteral(lit *ast.LiteralExpr) (*ast.LiteralExpr, bool) {
 // where both operands eventually collapse to literals. Non-foldable
 // subtrees come back unchanged. Used as a preorder pre-pass so the
 // caller can look at a flat LITERAL + LITERAL pair.
+//
+// For left-associative string-concat chains like "a" + x + "b" + "c",
+// the parser builds (((("a" + x) + "b") + "c")) and no pair is
+// literal+literal. We reassociate: if the LHS is `Y + strlit` and the
+// RHS is a string literal, rewrite as `Y + (strlit+rhslit)` so the
+// tail literals collapse. Only safe for STRING+STRING (numeric `+`
+// cares about types / overflow).
 func foldLiteralTree(e ast.Expr) ast.Expr {
 	be, ok := e.(*ast.BinaryExpr)
 	if !ok {
@@ -285,6 +292,26 @@ func foldLiteralTree(e ast.Expr) ast.Expr {
 	if folded, ok := tryFoldBinary(be); ok {
 		return folded
 	}
+	// String-concat reassociation for left-leaning chains.
+	if be.Op == token.PLUS {
+		if rLit, ok := be.Right.(*ast.LiteralExpr); ok && rLit.Kind == token.STRING {
+			if lBin, ok := be.Left.(*ast.BinaryExpr); ok && lBin.Op == token.PLUS {
+				if mLit, ok := lBin.Right.(*ast.LiteralExpr); ok && mLit.Kind == token.STRING {
+					fused := &ast.LiteralExpr{
+						ValuePos: mLit.ValuePos,
+						Kind:     token.STRING,
+						Value:    mLit.Value + rLit.Value,
+					}
+					return &ast.BinaryExpr{
+						OpPos: be.OpPos,
+						Op:   token.PLUS,
+						Left:  lBin.Left,
+						Right: fused,
+					}
+				}
+			}
+		}
+	}
 	return be
 }
 
@@ -1945,15 +1972,17 @@ func (g *Generator) emitExpr(expr ast.Expr) {
 	case *ast.IdentExpr:
 		g.emitIdent(e)
 	case *ast.BinaryExpr:
-		// Compile-time constant folding. Fold children first so
-		// `(2*3) + 5` collapses all the way to a single PushInt(11)
-		// instead of stopping at the inner `6`. Mutates the AST in
-		// place — safe because the generator owns the tree post-parse.
-		e.Left = foldLiteralTree(e.Left)
-		e.Right = foldLiteralTree(e.Right)
-		if folded, ok := tryFoldBinary(e); ok {
+		// Compile-time constant folding. Fold the whole subtree so
+		// `(2*3) + 5` collapses to PushInt(11) and so string-concat
+		// tails like `x + "b" + "c"` reassociate into `x + "bc"`.
+		// Mutates the AST in place — safe because the generator owns
+		// the tree post-parse.
+		switch folded := foldLiteralTree(e).(type) {
+		case *ast.LiteralExpr:
 			g.emitLiteral(folded)
-			break
+			return
+		case *ast.BinaryExpr:
+			e = folded
 		}
 		// Short-circuit AND/OR: Harbour evaluates right operand only if needed.
 		// With a literal LHS we can skip the PushBool/PopLogical roundtrip