gasm-sdk/format/fuzz_test.go

// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
// SPDX-License-Identifier: BSD-3-Clause

package format

import (
	"fmt"
	"os"
	"path/filepath"
	"reflect"
	"strings"
	"testing"

	"sourcedock.dev/petrbalvin/gasm-sdk/ast"
	"sourcedock.dev/petrbalvin/gasm-sdk/lexer"
	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
	"sourcedock.dev/petrbalvin/gasm-sdk/token"
)

// FuzzFormatIdempotency hammers the formatter with arbitrary input.  The
// contract: formatting twice equals formatting once, the output re-lexes to
// the same tokens as the input (so formatting changes layout, never meaning),
// input that parses cleanly still parses cleanly after formatting, and its
// tree, positions aside, is unchanged.  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")
	// Line-ending whitespace at the edge of a comment: a CR followed by more
	// trailing whitespace once survived the first pass and disappeared on
	// re-lexing, so formatting was not idempotent.
	f.Add("//\r ")
	f.Add("// loop \r\t\nMOVQ AX, BX\n")
	f.Add("TEXT ·f(SB), NOSPLIT, $0 // tail\r\n\tMOVQ AX, BX\r\n\tRET\r\n")
	// A block comment ahead of code on one line: the comment must not swallow
	// the statement that follows it.
	f.Add("/* head */ MOVQ AX, BX\n")
	f.Add("/* head */ DATA d<>+0(SB)/8, $1\n")
	// The exotic operand shapes the corpus taught the parser: register ranges,
	// PC-relative jumps with a negative displacement and the U+2215 package
	// path.
	f.Add("TEXT ·f(SB), $0\n\tV4FMADDPS [Z0-Z3], Z2, Z1\n\tRET\n")
	f.Add("TEXT ·f(SB), $0\n\tJMP -3(PC)\n\tRET\n")
	f.Add("TEXT ·f(SB), $0\n\tCALL internal∕runtime∕atomic·Xchg(SB)\n\tRET\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)
		}
		in, out := tokenView(src), tokenView(once)
		if !sameView(in, out) {
			t.Fatalf("formatting changed the token stream:\ninput:  %v\noutput: %v\n%s", viewString(in), viewString(out), once)
		}
		file, errs := parser.Parse("in.s", src)
		if len(errs) > 0 || hasStrayIllegal(src) {
			// A file the parser already reports on, or one whose stray
			// characters the formatter documents dropping, has no tree to
			// preserve; the token view above already pinned its tokens.
			return
		}
		reFile, errs := parser.Parse("out.s", once)
		if len(errs) > 0 {
			t.Fatalf("formatted output of clean input does not parse: %v\n%s", errs[0], once)
		}
		if got := scrub(reFile); !reflect.DeepEqual(got, scrub(file)) {
			t.Fatalf("formatting changed the tree:\ninput:  %s\noutput: %s\n%s", dump(scrub(file)), dump(got), once)
		}
	})
}

// hasStrayIllegal reports whether src lexes to an Illegal token other than
// the operand brackets, which the formatter drops by contract.
func hasStrayIllegal(src string) bool {
	for _, tok := range lexer.Tokenize(src) {
		if tok.Kind == token.Illegal && tok.Text != "[" && tok.Text != "]" {
			return true
		}
	}
	return false
}

// tokenView is the meaning of a source file as the assembler reads it: the
// token kinds and texts in order, ignoring line structure.  Stray Illegal
// runes are dropped because the formatter documents dropping them; comment
// texts are compared trailing-whitespace-trimmed because line comments lose
// exactly that on the way through the lexer, and an unterminated block
// comment runs to the end of the file, so the formatter's mandatory final
// newline (and any line ending it lands after) is layout, not content the
// formatter destroyed.
type viewToken struct {
	kind token.Kind
	text string
}

func tokenView(src string) []viewToken {
	var out []viewToken
	for _, tok := range lexer.Tokenize(src) {
		switch tok.Kind {
		case token.EOF, token.Newline:
			continue
		case token.Illegal:
			if tok.Text != "[" && tok.Text != "]" {
				continue
			}
		case token.Comment:
			text := strings.TrimRight(tok.Text, " \t\r")
			if strings.HasPrefix(text, "/*") && !strings.HasSuffix(text, "*/") {
				text = strings.TrimRight(text, " \t\r\n")
			}
			out = append(out, viewToken{tok.Kind, text})
			continue
		}
		out = append(out, viewToken{tok.Kind, tok.Text})
	}
	return out
}

func sameView(a, b []viewToken) bool {
	if len(a) != len(b) {
		return false
	}
	for i := range a {
		if a[i] != b[i] {
			return false
		}
	}
	return true
}

func viewString(v []viewToken) string {
	var b strings.Builder
	for i, t := range v {
		if i > 0 {
			b.WriteByte(' ')
		}
		b.WriteString(t.kind.String() + "(" + t.text + ")")
	}
	return b.String()
}

// scrub reduces a parsed file to what formatting must preserve: every field
// except source positions, the file path, and a preprocessor line's Raw text.
// Positions move with layout, the path is an input of the call, and Raw is a
// single-space join of the directive's tokens, whose input spelling the
// canonical operand spacing may legitimately re-glue ("#define A (x)" formats
// to "#define A(x)").  The token view already pins those tokens.
func scrub(f *ast.File) *ast.File {
	v := reflect.ValueOf(f).Elem()
	scrubValue(v)
	v.FieldByName("Path").SetString("")
	return f
}

// scrubValue walks v and zeroes every token.Position it reaches, so two trees
// taken from the same text at different layouts compare equal.
func scrubValue(v reflect.Value) {
	switch v.Kind() {
	case reflect.Pointer, reflect.Interface:
		if !v.IsNil() {
			scrubValue(v.Elem())
		}
	case reflect.Struct:
		switch v.Type() {
		case reflect.TypeFor[token.Position]():
			v.Set(reflect.Zero(v.Type()))
			return
		case reflect.TypeFor[ast.Preproc]():
			v.FieldByName("Raw").SetString("")
		}
		for _, field := range v.Fields() {
			scrubValue(field)
		}
	case reflect.Slice, reflect.Array:
		for i := 0; i < v.Len(); i++ {
			scrubValue(v.Index(i))
		}
	case reflect.Map:
		for _, k := range v.MapKeys() {
			scrubValue(v.MapIndex(k))
		}
	}
}

// dump renders a scrubbed tree as text, following pointers, because the fmt
// verbs stop at the first address inside a slice of interfaces.
func dump(v any) string {
	var b strings.Builder
	dumpValue(reflect.ValueOf(v), &b)
	return b.String()
}

func dumpValue(v reflect.Value, b *strings.Builder) {
	switch v.Kind() {
	case reflect.Pointer, reflect.Interface:
		if v.IsNil() {
			b.WriteString("nil")
			return
		}
		dumpValue(v.Elem(), b)
	case reflect.Struct:
		b.WriteString(v.Type().Name() + "{")
		for i := 0; i < v.NumField(); i++ {
			if i > 0 {
				b.WriteString(", ")
			}
			b.WriteString(v.Type().Field(i).Name + ":")
			dumpValue(v.Field(i), b)
		}
		b.WriteString("}")
	case reflect.Slice:
		if v.IsNil() {
			b.WriteString("nil")
			return
		}
		b.WriteString("[")
		for i := 0; i < v.Len(); i++ {
			if i > 0 {
				b.WriteString(", ")
			}
			dumpValue(v.Index(i), b)
		}
		b.WriteString("]")
	case reflect.Map:
		b.WriteString("map{")
		for _, k := range v.MapKeys() {
			dumpValue(k, b)
			b.WriteString(":")
			dumpValue(v.MapIndex(k), b)
		}
		b.WriteString("}")
	default:
		b.WriteString(fmt.Sprintf("%v", v))
	}
}