fix(asm): reject duplicate symbol declarations like the toolchain

Assisted-by: GLM 5.3 Flash

asm/link_test.go

@@ -12,9 +12,107 @@ import (
 	"strings"
 	"testing"
 
+	"sourcedock.dev/petrbalvin/gasm-sdk/ast"
 	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
+// TestDuplicateDecls pins the duplicate-declaration rule against the
+// toolchain's, measured with go tool asm (Go 1.27.1): a symbol declared
+// twice in one file is rejected at the second declaration whether the pair
+// is two TEXTs, two GLOBLs or one of each, and DUPOK never legalises it
+// within a file (the toolchain's duperror.s testdata asserts the plain
+// pair).  DUPOK legalises duplicates across files, so each file of a pair
+// assembles cleanly on its own, and symbol identity keeps the package
+// prefix and the <> static marker: foo(SB), foo<>(SB) and other·foo(SB)
+// are three symbols, not one.  Every case runs through all four
+// per-architecture entry points.
+func TestDuplicateDecls(t *testing.T) {
+	targets := []struct {
+		goarch string
+		file   string
+		asm    func(*ast.File) (*Image, error)
+	}{
+		{"amd64", "dup_amd64.s", func(f *ast.File) (*Image, error) { return AssembleFile(f) }},
+		{"arm64", "dup_arm64.s", AssembleFileARM64},
+		{"riscv64", "dup_riscv64.s", AssembleFileRISCV},
+		{"loong64", "dup_loong64.s", AssembleFileLOONG64},
+	}
+	cases := []struct {
+		name string
+		src  string
+		want string // substring of the error, "" for no error
+	}{
+		{
+			"plain TEXT duplicate rejected",
+			"TEXT foo(SB), NOSPLIT, $0\n\tRET\nTEXT foo(SB), NOSPLIT, $0\n\tRET\n",
+			`symbol "foo" redeclared (other declaration on line 1)`,
+		},
+		{
+			"DUPOK pair in one file rejected",
+			"TEXT foo(SB), DUPOK, $0\n\tRET\nTEXT foo(SB), DUPOK, $0\n\tRET\n",
+			`symbol "foo" redeclared (other declaration on line 1)`,
+		},
+		{
+			"TEXT then GLOBL rejected",
+			"TEXT foo(SB), NOSPLIT, $0\n\tRET\nGLOBL foo(SB), NOPTR, $8\n",
+			`symbol "foo" redeclared`,
+		},
+		{
+			"GLOBL then TEXT rejected",
+			"GLOBL foo(SB), NOPTR, $8\nTEXT foo(SB), NOSPLIT, $0\n\tRET\n",
+			`symbol "foo" redeclared`,
+		},
+		{
+			"plain GLOBL duplicate rejected",
+			"GLOBL bar(SB), NOPTR, $8\nGLOBL bar(SB), NOPTR, $8\n",
+			`symbol "bar" redeclared`,
+		},
+		{
+			"static and package markers separate symbols",
+			"TEXT foo(SB), NOSPLIT, $0\n\tRET\nTEXT foo<>(SB), NOSPLIT, $0\n\tRET\nTEXT other\u00b7foo(SB), NOSPLIT, $0\n\tRET\n",
+			"",
+		},
+		{
+			"single DUPOK TEXT assembles",
+			"TEXT foo(SB), DUPOK, $0\n\tRET\n",
+			"",
+		},
+	}
+	for _, tg := range targets {
+		for _, c := range cases {
+			f, errs := parser.Parse(tg.file, c.src)
+			if len(errs) > 0 {
+				t.Fatalf("%s/%s: parse: %v", tg.goarch, c.name, errs)
+			}
+			_, err := tg.asm(f)
+			if c.want == "" {
+				if err != nil {
+					t.Errorf("%s/%s: error %v, want none", tg.goarch, c.name, err)
+				}
+				continue
+			}
+			if err == nil || !strings.Contains(err.Error(), c.want) {
+				t.Errorf("%s/%s: error %v, want substring %q", tg.goarch, c.name, err, c.want)
+			}
+		}
+	}
+
+	// A DUPOK pair across files is the case DUPOK exists for: the linker
+	// resolves it, the assembler never sees both sides, so each file of the
+	// pair assembles cleanly on its own.
+	for _, tg := range targets {
+		for _, name := range []string{"first", "second"} {
+			f, errs := parser.Parse(tg.file, "TEXT foo(SB), DUPOK, $0\n\tRET\n")
+			if len(errs) > 0 {
+				t.Fatalf("%s/dupok cross-file %s: parse: %v", tg.goarch, name, errs)
+			}
+			if _, err := tg.asm(f); err != nil {
+				t.Errorf("%s/dupok cross-file %s: error %v, want none", tg.goarch, name, err)
+			}
+		}
+	}
+}
+
 // TestAssembleFileStaticData checks the whole-image layout; code, padding
 // and the data section; and that the RIP-relative displacements of static
 // symbol loads resolve to the right bytes.