test(asm): pin the flag-list TEXT shapes against the toolchain

Assisted-by: GLM 5.3 Flash

asm/flaglist_integration_test.go

@@ -0,0 +1,127 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+package asm
+
+import (
+	"bytes"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
+)
+
+// flagListShapes are the TEXT shapes the flags operand decides: joined
+// names, the legacy numeric spellings, the arithmetic and immediate forms,
+// each with a frame size beside it.  A flag the operand carries suppresses
+// the stack-split guard; the toolchain reads one integer from the operand,
+// and after the front end folds it the bytes must not tell the difference.
+// The bodies are leaf-shaped on purpose: the frame engine's open divergences
+// (the NOFRAME prologue for a non-leaf, the big-frame guard's shape) sit
+// outside the flags operand and are somebody else's gap list.
+var flagListShapes = []struct {
+	name  string // the subtest's name
+	flags string // the flags operand as written
+	frame string // the frame operand as written
+	body  string // the function body
+}{
+	{"name", "NOSPLIT", "$4096-0", "\tMOVD R0, R1\n\tRET\n"},
+	{"numeric", "4", "$4096-0", "\tMOVD R0, R1\n\tRET\n"},
+	{"numericOR", "2|4", "$4096-0", "\tMOVD R0, R1\n\tRET\n"},
+	{"joinedFrame", "DUPOK|NOSPLIT", "$4096-0", "\tMOVD R0, R1\n\tRET\n"},
+	{"joinedFrameless", "NOSPLIT|TOPFRAME", "$0-0", "\tRET\n"},
+	{"parenthesised", "(NOSPLIT|NOFRAME)", "$0-0", "\tMOVD R0, R1\n\tRET\n"},
+}
+
+// TestFlagListAssemblyPathRejectsUnknown holds the front end's rejections
+// that mirror the toolchain's: an identifier outside the flag table
+// ("unexpected TYPO evaluating expression") and the immediate spelling of
+// the operand ("TEXT: expected integer constant"), each refused on the
+// assembly path before the encoder ever sees a tree.
+func TestFlagListAssemblyPathRejectsUnknown(t *testing.T) {
+	for _, c := range []struct{ src, want string }{
+		{"#include \"textflag.h\"\n\nTEXT f(SB), NOSPLIT|TYPO, $0-0\n\tRET\n",
+			"unexpected TYPO evaluating expression"},
+		{"#include \"textflag.h\"\n\nTEXT f(SB), $NOSPLIT, $0-0\n\tRET\n",
+			"TEXT: expected integer constant; found $NOSPLIT"},
+		{"#include \"textflag.h\"\n\nGLOBL g<>(SB), $8, $8\n",
+			"GLOBL: expected integer constant; found $8"},
+	} {
+		_, errs := parser.ParseWithOptions("f.s", c.src, parser.Options{Expand: true})
+		if len(errs) == 0 || !strings.Contains(errs[0].Error(), c.want) {
+			t.Errorf("errors = %v, want %q", errs, c.want)
+		}
+	}
+}
+
+// TestFlagListByteParity assembles every flag-list shape through the fixed
+// front end and holds the function's bytes equal to the installed
+// toolchain's, no guard words where the flag suppresses them and none
+// missing where it demands one.
+func TestFlagListByteParity(t *testing.T) {
+	if testing.Short() {
+		t.Skip("live go tool asm oracle: skipped in -short mode")
+	}
+	goBin, err := exec.LookPath("go")
+	if err != nil {
+		t.Skip("no Go toolchain available")
+	}
+	out, err := exec.Command(goBin, "env", "GOROOT").Output()
+	if err != nil {
+		t.Fatalf("go env GOROOT: %v", err)
+	}
+	include := filepath.Join(strings.TrimSpace(string(out)), "pkg", "include")
+
+	for _, c := range flagListShapes {
+		t.Run(c.name, func(t *testing.T) {
+			src := "#include \"textflag.h\"\n\nTEXT f(SB), " + c.flags + ", " + c.frame + "\n" + c.body
+			dir := t.TempDir()
+			file := filepath.Join(dir, "f.s")
+			if err := os.WriteFile(file, []byte(src), 0o644); err != nil {
+				t.Fatal(err)
+			}
+
+			af, errs := parser.ParseWithOptions(file, src, parser.Options{Expand: true})
+			if len(errs) > 0 {
+				t.Fatalf("parse: %v", errs[0])
+			}
+			img, err := AssembleFileARM64(af)
+			if err != nil {
+				t.Fatalf("AssembleFileARM64: %v", err)
+			}
+			if len(img.Funcs) != 1 {
+				t.Fatalf("%d functions, want 1", len(img.Funcs))
+			}
+
+			objPath := filepath.Join(t.TempDir(), "oracle.o")
+			cmd := exec.Command(goBin, "tool", "asm", "-std", "-I", include,
+				"-p", "flaglisttest", "-o", objPath, file)
+			cmd.Env = append(os.Environ(), "GOOS=linux", "GOARCH=arm64")
+			if oout, err := cmd.CombinedOutput(); err != nil {
+				t.Fatalf("go tool asm: %v\n%s", err, oout)
+			}
+			goCode, ok := oracleFuncText(t, mustRead(t, objPath))["f"]
+			if !ok {
+				t.Fatal("the oracle output carries no f")
+			}
+			fn := img.Funcs[0]
+			gasmCode := maskCode(append([]byte(nil), img.Code[fn.Offset:fn.Offset+fn.Size]...), fn.Relocs)
+			goCode = maskCode(append([]byte(nil), goCode...), fn.Relocs)
+			cmpLen := min(len(goCode), len(gasmCode))
+			if !bytes.Equal(gasmCode[:cmpLen], goCode[:cmpLen]) {
+				t.Errorf("bytes differ:\ngasm: % x\ngo:   % x", gasmCode[:cmpLen], goCode[:cmpLen])
+			}
+			if len(goCode) > len(gasmCode) {
+				for _, b := range goCode[len(gasmCode):] {
+					if b != 0 {
+						t.Errorf("non-zero trailing bytes in the oracle output")
+						break
+					}
+				}
+			}
+		})
+	}
+}

parser/flags.go

@@ -77,7 +77,8 @@ func flagsRun(g []token.Token) []token.Token {
 // evalFlags evaluates one flags operand.  Every identifier must name a
 // flag in textFlags; the toolchain rejects any other with "unexpected NAME
 // evaluating expression", and so does this.  The whole operand folds to
-// one integer, which is returned as the value.
+// one integer, which is returned as the value.  pseudo names the directive
+// in the toolchain's own diagnostics ("TEXT", "GLOBL").
 //
 // The names returned follow the operand's own spelling: an operand written
 // purely as names keeps them in written order, duplicates included, the
@@ -89,17 +90,21 @@ func flagsRun(g []token.Token) []token.Token {
 // After a failure the names collected so far are returned with value 0:
 // the tree stays usable beside the diagnostic, and value 0 is exactly the
 // flags the assembler can trust the operand for.
-func (p *state) evalFlags(g []token.Token, keepWritten bool) ([]string, int64) {
+func (p *state) evalFlags(g []token.Token, keepWritten bool, pseudo string) ([]string, int64) {
 	if len(g) == 0 {
 		return nil, 0
 	}
-	// An explicit '$' makes the operand an immediate ('$NOSPLIT' over the
-	// preprocessor's '$4'); the toolchain reads it as the same constant.
+	// The flags operand is a bare constant expression; the immediate
+	// spelling is the toolchain's "TEXT: expected integer constant".
 	if g[0].Kind == token.Dollar {
-		g = g[1:]
-		if len(g) == 0 {
-			return nil, 0
+		if p.expand {
+			found := "$"
+			if len(g) > 1 {
+				found += g[1].Text
+			}
+			p.errorf(g[0].Pos, "%s: expected integer constant; found %s", pseudo, found)
 		}
+		return nil, 0
 	}
 
 	sub := make([]token.Token, 0, len(g))

parser/flags_test.go

@@ -53,7 +53,10 @@ func TestTextFlagsExpression(t *testing.T) {
 		{"(NOSPLIT|NOFRAME)", []string{"NOSPLIT", "NOFRAME"}, 516},
 		{"NOSPLIT|4", []string{"NOSPLIT"}, 4},
 		{"NOSPLIT&DUPOK", nil, 0},
-		{"$NOSPLIT", []string{"NOSPLIT"}, 4},
+		// The immediate spelling is the toolchain's "expected integer
+		// constant" on the assembly path; the raw view reads nothing from
+		// it and stays quiet.
+		{"$NOSPLIT", nil, 0},
 	} {
 		txt := parseTextHeader(t, "TEXT \u00b7f(SB), "+c.operand+", $0")
 		if !slices.Equal(txt.Flags, c.wantFlag) {

parser/parser.go

@@ -261,7 +261,7 @@ func (p *state) parseText(line []token.Token) {
 	var frame []token.Token
 	switch {
 	case len(ops) >= 2:
-		text.Flags, text.FlagVal = p.evalFlags(flagsRun(ops[0]), false)
+		text.Flags, text.FlagVal = p.evalFlags(flagsRun(ops[0]), false, "TEXT")
 		frame = ops[1]
 		if len(ops) > 2 {
 			p.errorf(line[0].Pos, "expect two or three operands for TEXT")
@@ -355,7 +355,7 @@ func (p *state) parseGlobl(line []token.Token) *ast.Globl {
 	ops := splitOperands(rest)
 	switch {
 	case len(ops) >= 2:
-		g.Flags, g.FlagVal = p.evalFlags(flagsRun(ops[0]), true)
+		g.Flags, g.FlagVal = p.evalFlags(flagsRun(ops[0]), true, "GLOBL")
 		g.Size = parseOperand(ops[1], false)
 	case len(ops) == 1:
 		g.Size = parseOperand(ops[0], false)