feat(format): fuzz targets for the parser and formatter

Assisted-by: GLM 5.3 Flash
2026-09-19 20:41:43 +02:00
parent f37f183577
commit bc3f448738
3 changed files with 184 additions and 9 deletions
@@ -50,12 +50,31 @@ func Source(src string) string {
 				}
 			case len(line) >= 2 && line[1].Kind == token.Colon:
 				inf.kind = kLabel
+				// Peel stacked labels exactly as the render pass does; the
+				// instruction after the last one is rendered at the
+				// function's alignment width, so its mnemonic counts here.
+				rest := line[2:]
+				for len(rest) >= 2 && rest[0].Kind == token.Ident && rest[1].Kind == token.Colon &&
+					!isDirective(rest[0].Text) {
+					rest = rest[2:]
+				}
+				if len(rest) > 0 && rest[0].Kind == token.Ident && !isDirective(rest[0].Text) {
+					inf.mnemLen = len(rest[0].Text)
+					if funcID >= 0 && inf.mnemLen > maxWidth[funcID] {
+						maxWidth[funcID] = inf.mnemLen
+					}
+				}
 			default:
 				inf.kind = kInstr
 				inf.funcID = funcID
-				inf.mnemLen = len(line[0].Text)
-				if funcID >= 0 && inf.mnemLen > maxWidth[funcID] {
-					maxWidth[funcID] = inf.mnemLen
+				// Only an identifier mnemonic takes the alignment width; a
+				// line starting with anything else renders unpadded, so its
+				// length must not enter the width either.
+				if line[0].Kind == token.Ident {
+					inf.mnemLen = len(line[0].Text)
+					if funcID >= 0 && inf.mnemLen > maxWidth[funcID] {
+						maxWidth[funcID] = inf.mnemLen
+					}
 				}
 			}
 		}
@@ -83,12 +102,35 @@ func Source(src string) string {
 			out = line[0].Text + " " + renderOps(line[1:])
 			inBody = line[0].Text == "TEXT"
 		case kLabel:
-			out = line[0].Text + ":"
-			// A label may share its line with an instruction; emit the
-			// instruction on the following line.
-			if rest := line[2:]; len(rest) > 0 {
-				out += "\n" + renderInstr(rest, maxWidth[inf.funcID])
+			// Every label, and a trailing instruction, becomes its own
+			// output line: separate outLines keep the blank-line pass
+			// honest about what it is looking at.
+			outs = append(outs, outLine{kind: kLabel, text: line[0].Text + ":"})
+			rest := line[2:]
+			for len(rest) >= 2 && rest[0].Kind == token.Ident && rest[1].Kind == token.Colon &&
+				!isDirective(rest[0].Text) {
+				outs = append(outs, outLine{kind: kLabel, text: rest[0].Text + ":"})
+				rest = rest[2:]
 			}
+			// A label may share its line with an instruction; the canonical
+			// form puts the instruction on the following line.  Trailing
+			// content that does not start an instruction (a stray operand
+			// token) stays on the label line: splitting it off would produce
+			// a line the parser rejects.
+			if len(rest) > 0 && rest[0].Kind == token.Ident && isDirective(rest[0].Text) {
+				// A bare directive cannot start a line of its own (the
+				// parser wants a symbol per line), so a directive sharing
+				// the label's line stays there.
+				outs[len(outs)-1].text += " " + strings.TrimRight(renderOps(rest), " \t")
+			} else if len(rest) > 0 && rest[0].Kind == token.Ident {
+				outs = append(outs, outLine{kind: kInstr, text: strings.TrimRight(renderInstr(rest, maxWidth[inf.funcID]), " \t")})
+				if strings.EqualFold(rest[0].Text, "RET") {
+					inBody = false
+				}
+			} else if len(rest) > 0 {
+				outs[len(outs)-1].text += " " + renderOps(rest)
+			}
+			continue
 		case kInstr:
 			out = renderInstr(line, maxWidth[inf.funcID])
 			// A RET ends the body for indentation purposes: comments that
@@ -192,6 +234,13 @@ func renderInstr(line []token.Token, width int) string {
 	if ops == "" {
 		return "\t" + mnem
 	}
+	// Alignment is a mnemonic convention: a line that does not start with
+	// an identifier (a stray operand token the parser tolerates) renders
+	// unpadded, so that no alignment width can depend on it and the output
+	// stays stable across passes.
+	if line[0].Kind != token.Ident {
+		return "\t" + mnem + " " + ops
+	}
 	if width < len(mnem) {
 		width = len(mnem)
 	}
@@ -217,7 +266,19 @@ func renderPreproc(line []token.Token) string {
 func renderOps(toks []token.Token) string {
 	var b strings.Builder
 	for i, t := range toks {
-		if i > 0 && spaceBetween(toks[i-1], t) {
+		sp := i > 0 && spaceBetween(toks[i-1], t)
+		// The accumulated text ending in '/' must never meet a '/' or '*':
+		// the pair would re-lex as a comment and the next pass would see a
+		// different line, whatever the token boundaries were.
+		if !sp && i > 0 && (t.Kind == token.Slash || t.Kind == token.Star) && strings.HasSuffix(b.String(), "/") {
+			sp = true
+		}
+		if sp {
+			b.WriteByte(' ')
+		} else if i > 0 && wouldMerge(toks[i-1], t) {
+			// The tight spelling would re-lex as something else ('/'
+			// before '*' opens a comment), which would make the next
+			// formatting pass see a different line.
 			b.WriteByte(' ')
 		}
 		b.WriteString(t.Text)
@@ -225,8 +286,27 @@ func renderOps(toks []token.Token) string {
 	return b.String()
 }

+// wouldMerge reports whether writing prev immediately before cur would
+// re-lex as something other than those two tokens: a '/' before a '*' opens
+// a comment, '>' before '>' shifts, and adjacent operators regroup.
+func wouldMerge(prev, cur token.Token) bool {
+	var kinds []token.Kind
+	for _, t := range lexer.Tokenize(prev.Text + cur.Text) {
+		if t.Kind == token.EOF {
+			break
+		}
+		kinds = append(kinds, t.Kind)
+	}
+	return len(kinds) != 2 || kinds[0] != prev.Kind || kinds[1] != cur.Kind
+}
+
 // spaceBetween decides whether a single space separates prev and cur.
 func spaceBetween(prev, cur token.Token) bool {
+	// '/' beside '/' or '*' would form a comment opener in the output and
+	// make the next pass see a different line; keep them separated.
+	if prev.Kind == token.Slash && (cur.Kind == token.Slash || cur.Kind == token.Star) {
+		return true
+	}
 	switch cur.Kind {
 	case token.RParen:
 		return false
@@ -274,6 +354,13 @@ func splitLines(toks []token.Token) [][]token.Token {
 		if t.Kind == token.EOF {
 			break
 		}
+		if t.Kind == token.Illegal {
+			// Illegal tokens carry no canonical spelling: the parser
+			// reports them as errors where they matter, and the formatter
+			// drops them so that a stray character cannot survive into the
+			// output and make the next pass render a different file.
+			continue
+		}
 		if t.Kind == token.Newline {
 			lines = append(lines, cur)
 			cur = nil
@@ -0,0 +1,47 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+package format
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+
+	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+)
+
+// FuzzFormatIdempotency hammers the formatter with arbitrary input.  The
+// contract: formatting twice equals formatting once, and input that parses
+// cleanly still parses cleanly after formatting.  The seed corpus carries the
+// repository's kernels, so a plain `go test` run replays every seed as a
+// regression case and CI exercises them without any fuzzing budget.
+func FuzzFormatIdempotency(f *testing.F) {
+	for _, pattern := range []string{
+		"../testdata/*.s",
+		"../testdata/verify/*.s",
+	} {
+		files, _ := filepath.Glob(pattern)
+		for _, path := range files {
+			if b, err := os.ReadFile(path); err == nil {
+				f.Add(string(b))
+			}
+		}
+	}
+	f.Add("TEXT ·f(SB), NOSPLIT, $0\n\tMOVQ AX, BX\n\tRET\n")
+	f.Add("TEXT ·f(SB),NOSPLIT,$0\n\tMOVQ   AX,BX\n\n\n\tRET\n")
+	f.Add("garbage ### ???\n")
+
+	f.Fuzz(func(t *testing.T, src string) {
+		once := Source(src)
+		twice := Source(once)
+		if once != twice {
+			t.Fatalf("formatting is not idempotent:\nfirst:  %q\nsecond: %q", once, twice)
+		}
+		if _, errs := parser.Parse("in.s", src); len(errs) == 0 {
+			if _, errs := parser.Parse("out.s", once); len(errs) > 0 {
+				t.Fatalf("formatted output of clean input does not parse: %v\n%s", errs[0], once)
+			}
+		}
+	})
+}