all: switch to LLVM 16

This commit adds support for LLVM 16 and switches to it by default. That
means three LLVM versions are supported at the same time: LLVM 14, 15,
and 16.

This commit includes work by QuLogic:

  * Part of this work was based on a PR by QuLogic:
    https://github.com/tinygo-org/tinygo/pull/3649
    But I also had parts of this already implemented in an old branch I
    already made for LLVM 16.
  * QuLogic also provided a CGo fix here, which is also incorporated in
    this commit:
    https://github.com/tinygo-org/tinygo/pull/3869

The difference with the original PR by QuLogic is that this commit is
more complete:
  * It switches to LLVM 16 by default.
  * It updates some things to also make it work with a self-built LLVM.
  * It fixes the CGo bug in a slightly different way, and also fixes
    another one not included in the original PR.
  * It does not keep compiler tests passing on older LLVM versions. I
    have found this to be quite burdensome and therefore don't generally
    do this - the smoke tests should hopefully catch most regressions.

6 files changed, 79 insertions, 42 deletions

diff --git a/cgo/libclang.go b/cgo/libclang.go
index 27ac1b576..1f2922c56 100644
--- a/cgo/libclang.go
+++ b/cgo/libclang.go
@@ -60,6 +60,7 @@ CXSourceRange tinygo_clang_getCursorExtent(GoCXCursor c);
 CXTranslationUnit tinygo_clang_Cursor_getTranslationUnit(GoCXCursor c);
 long long tinygo_clang_getEnumConstantDeclValue(GoCXCursor c);
 CXType tinygo_clang_getEnumDeclIntegerType(GoCXCursor c);
+unsigned tinygo_clang_Cursor_isAnonymous(GoCXCursor c);
 unsigned tinygo_clang_Cursor_isBitField(GoCXCursor c);
 
 int tinygo_clang_globals_visitor(GoCXCursor c, GoCXCursor parent, CXClientData client_data);
@@ -653,8 +654,19 @@ func (p *cgoPackage) addErrorAt(position token.Position, msg string) {
 func (f *cgoFile) makeDecayingASTType(typ C.CXType, pos token.Pos) ast.Expr {
 	// Strip typedefs, if any.
 	underlyingType := typ
+	if underlyingType.kind == C.CXType_Elaborated {
+		// Starting with LLVM 16, the elaborated type is used for more types.
+		// According to the Clang documentation, the elaborated type has no
+		// semantic meaning so can be stripped (it is used to better convey type
+		// name information).
+		// Source:
+		// https://clang.llvm.org/doxygen/classclang_1_1ElaboratedType.html#details
+		// > The type itself is always "sugar", used to express what was written
+		// > in the source code but containing no additional semantic information.
+		underlyingType = C.clang_Type_getNamedType(underlyingType)
+	}
 	if underlyingType.kind == C.CXType_Typedef {
-		c := C.tinygo_clang_getTypeDeclaration(typ)
+		c := C.tinygo_clang_getTypeDeclaration(underlyingType)
 		underlyingType = C.tinygo_clang_getTypedefDeclUnderlyingType(c)
 		// TODO: support a chain of typedefs. At the moment, it seems to get
 		// stuck in an endless loop when trying to get to the most underlying
@@ -788,6 +800,8 @@ func (f *cgoFile) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
 			return f.makeASTType(underlying, pos)
 		case C.CXType_Enum:
 			return f.makeASTType(underlying, pos)
+		case C.CXType_Typedef:
+			return f.makeASTType(underlying, pos)
 		default:
 			typeKindSpelling := getString(C.clang_getTypeKindSpelling(underlying.kind))
 			f.addError(pos, fmt.Sprintf("unknown elaborated type (libclang type kind %s)", typeKindSpelling))
@@ -806,7 +820,7 @@ func (f *cgoFile) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
 			// makeASTRecordType will create an appropriate error.
 			cgoRecordPrefix = "record_"
 		}
-		if name == "" {
+		if name == "" || C.tinygo_clang_Cursor_isAnonymous(cursor) != 0 {
 			// Anonymous record, probably inside a typedef.
 			location := f.getUniqueLocationID(pos, cursor)
 			name = f.getUnnamedDeclName("_Ctype_"+cgoRecordPrefix+"__", location)
diff --git a/cgo/libclang_config_llvm15.go b/cgo/libclang_config_llvm15.go
index 32daf0608..e105dfe72 100644
--- a/cgo/libclang_config_llvm15.go
+++ b/cgo/libclang_config_llvm15.go
@@ -1,4 +1,4 @@
-//go:build !byollvm && !llvm14
+//go:build !byollvm && llvm15
 
 package cgo
 
diff --git a/cgo/libclang_config_llvm16.go b/cgo/libclang_config_llvm16.go
new file mode 100644
index 000000000..79aacd2f2
--- /dev/null
+++ b/cgo/libclang_config_llvm16.go
@@ -0,0 +1,21 @@
+//go:build !byollvm && !llvm14 && !llvm15
+
+package cgo
+
+// As of 2023-05-05, there is a packaging issue on Debian:
+// https://github.com/llvm/llvm-project/issues/62199
+// A workaround is to fix this locally, using something like this:
+//
+//   ln -sf ../../x86_64-linux-gnu/libclang-16.so.1 /usr/lib/llvm-16/lib/libclang.so
+
+/*
+#cgo linux        CFLAGS:  -I/usr/lib/llvm-16/include
+#cgo darwin,amd64 CFLAGS:  -I/usr/local/opt/llvm@16/include
+#cgo darwin,arm64 CFLAGS:  -I/opt/homebrew/opt/llvm@16/include
+#cgo freebsd      CFLAGS:  -I/usr/local/llvm16/include
+#cgo linux        LDFLAGS: -L/usr/lib/llvm-16/lib -lclang
+#cgo darwin,amd64 LDFLAGS: -L/usr/local/opt/llvm@16/lib -lclang -lffi
+#cgo darwin,arm64 LDFLAGS: -L/opt/homebrew/opt/llvm@16/lib -lclang -lffi
+#cgo freebsd      LDFLAGS: -L/usr/local/llvm16/lib -lclang
+*/
+import "C"
diff --git a/cgo/libclang_stubs.c b/cgo/libclang_stubs.c
index b0668ed71..1b157d0aa 100644
--- a/cgo/libclang_stubs.c
+++ b/cgo/libclang_stubs.c
@@ -77,6 +77,10 @@ CXType tinygo_clang_getEnumDeclIntegerType(CXCursor c) {
 	return clang_getEnumDeclIntegerType(c);
 }
 
+unsigned tinygo_clang_Cursor_isAnonymous(CXCursor c) {
+	return clang_Cursor_isAnonymous(c);
+}
+
 unsigned tinygo_clang_Cursor_isBitField(CXCursor c) {
 	return clang_Cursor_isBitField(c);
-}
\ No newline at end of file
+}
diff --git a/cgo/testdata/errors.out.go b/cgo/testdata/errors.out.go
index 716fd771b..b1646a2e0 100644
--- a/cgo/testdata/errors.out.go
+++ b/cgo/testdata/errors.out.go
@@ -51,10 +51,10 @@ type (
 	C.longlong  int64
 	C.ulonglong uint64
 )
-type C._Ctype_struct___0 struct {
+type C.struct_point_t struct {
 	x C.int
 	y C.int
 }
-type C.point_t = C._Ctype_struct___0
+type C.point_t = C.struct_point_t
 
 const C.SOME_CONST_3 = 1234
diff --git a/cgo/testdata/types.out.go b/cgo/testdata/types.out.go
index bd7b3fc42..a35d3733c 100644
--- a/cgo/testdata/types.out.go
+++ b/cgo/testdata/types.out.go
@@ -38,11 +38,11 @@ type (
 	C.ulonglong uint64
 )
 type C.myint = C.int
-type C._Ctype_struct___0 struct {
+type C.struct_point2d_t struct {
 	x C.int
 	y C.int
 }
-type C.point2d_t = C._Ctype_struct___0
+type C.point2d_t = C.struct_point2d_t
 type C.struct_point3d struct {
 	x C.int
 	y C.int
@@ -55,21 +55,19 @@ type C.struct_type1 struct {
 	___type C.int
 }
 type C.struct_type2 struct{ _type C.int }
-type C._Ctype_union___1 struct{ i C.int }
-type C.union1_t = C._Ctype_union___1
-type C._Ctype_union___2 struct{ $union uint64 }
+type C.union_union1_t struct{ i C.int }
+type C.union1_t = C.union_union1_t
+type C.union_union3_t struct{ $union uint64 }
 
-func (union *C._Ctype_union___2) unionfield_i() *C.int {
-	return (*C.int)(unsafe.Pointer(&union.$union))
-}
-func (union *C._Ctype_union___2) unionfield_d() *float64 {
+func (union *C.union_union3_t) unionfield_i() *C.int { return (*C.int)(unsafe.Pointer(&union.$union)) }
+func (union *C.union_union3_t) unionfield_d() *float64 {
 	return (*float64)(unsafe.Pointer(&union.$union))
 }
-func (union *C._Ctype_union___2) unionfield_s() *C.short {
+func (union *C.union_union3_t) unionfield_s() *C.short {
 	return (*C.short)(unsafe.Pointer(&union.$union))
 }
 
-type C.union3_t = C._Ctype_union___2
+type C.union3_t = C.union_union3_t
 type C.union_union2d struct{ $union [2]uint64 }
 
 func (union *C.union_union2d) unionfield_i() *C.int { return (*C.int)(unsafe.Pointer(&union.$union)) }
@@ -78,50 +76,50 @@ func (union *C.union_union2d) unionfield_d() *[2]float64 {
 }
 
 type C.union2d_t = C.union_union2d
-type C._Ctype_union___3 struct{ arr [10]C.uchar }
-type C.unionarray_t = C._Ctype_union___3
-type C._Ctype_union___5 struct{ $union [3]uint32 }
+type C.union_unionarray_t struct{ arr [10]C.uchar }
+type C.unionarray_t = C.union_unionarray_t
+type C._Ctype_union___0 struct{ $union [3]uint32 }
 
-func (union *C._Ctype_union___5) unionfield_area() *C.point2d_t {
+func (union *C._Ctype_union___0) unionfield_area() *C.point2d_t {
 	return (*C.point2d_t)(unsafe.Pointer(&union.$union))
 }
-func (union *C._Ctype_union___5) unionfield_solid() *C.point3d_t {
+func (union *C._Ctype_union___0) unionfield_solid() *C.point3d_t {
 	return (*C.point3d_t)(unsafe.Pointer(&union.$union))
 }
 
-type C._Ctype_struct___4 struct {
+type C.struct_struct_nested_t struct {
 	begin C.point2d_t
 	end   C.point2d_t
 	tag   C.int
 
-	coord C._Ctype_union___5
+	coord C._Ctype_union___0
 }
-type C.struct_nested_t = C._Ctype_struct___4
-type C._Ctype_union___6 struct{ $union [2]uint64 }
+type C.struct_nested_t = C.struct_struct_nested_t
+type C.union_union_nested_t struct{ $union [2]uint64 }
 
-func (union *C._Ctype_union___6) unionfield_point() *C.point3d_t {
+func (union *C.union_union_nested_t) unionfield_point() *C.point3d_t {
 	return (*C.point3d_t)(unsafe.Pointer(&union.$union))
 }
-func (union *C._Ctype_union___6) unionfield_array() *C.unionarray_t {
+func (union *C.union_union_nested_t) unionfield_array() *C.unionarray_t {
 	return (*C.unionarray_t)(unsafe.Pointer(&union.$union))
 }
-func (union *C._Ctype_union___6) unionfield_thing() *C.union3_t {
+func (union *C.union_union_nested_t) unionfield_thing() *C.union3_t {
 	return (*C.union3_t)(unsafe.Pointer(&union.$union))
 }
 
-type C.union_nested_t = C._Ctype_union___6
+type C.union_nested_t = C.union_union_nested_t
 type C.enum_option = C.int
 type C.option_t = C.enum_option
-type C._Ctype_enum___7 = C.uint
-type C.option2_t = C._Ctype_enum___7
-type C._Ctype_struct___8 struct {
+type C.enum_option2_t = C.uint
+type C.option2_t = C.enum_option2_t
+type C.struct_types_t struct {
 	f   float32
 	d   float64
 	ptr *C.int
 }
-type C.types_t = C._Ctype_struct___8
+type C.types_t = C.struct_types_t
 type C.myIntArray = [10]C.int
-type C._Ctype_struct___9 struct {
+type C.struct_bitfield_t struct {
 	start        C.uchar
 	__bitfield_1 C.uchar
 
@@ -129,21 +127,21 @@ type C._Ctype_struct___9 struct {
 	e C.uchar
 }
 
-func (s *C._Ctype_struct___9) bitfield_a() C.uchar { return s.__bitfield_1 & 0x1f }
-func (s *C._Ctype_struct___9) set_bitfield_a(value C.uchar) {
+func (s *C.struct_bitfield_t) bitfield_a() C.uchar { return s.__bitfield_1 & 0x1f }
+func (s *C.struct_bitfield_t) set_bitfield_a(value C.uchar) {
 	s.__bitfield_1 = s.__bitfield_1&^0x1f | value&0x1f<<0
 }
-func (s *C._Ctype_struct___9) bitfield_b() C.uchar {
+func (s *C.struct_bitfield_t) bitfield_b() C.uchar {
 	return s.__bitfield_1 >> 5 & 0x1
 }
-func (s *C._Ctype_struct___9) set_bitfield_b(value C.uchar) {
+func (s *C.struct_bitfield_t) set_bitfield_b(value C.uchar) {
 	s.__bitfield_1 = s.__bitfield_1&^0x20 | value&0x1<<5
 }
-func (s *C._Ctype_struct___9) bitfield_c() C.uchar {
+func (s *C.struct_bitfield_t) bitfield_c() C.uchar {
 	return s.__bitfield_1 >> 6
 }
-func (s *C._Ctype_struct___9) set_bitfield_c(value C.uchar,
+func (s *C.struct_bitfield_t) set_bitfield_c(value C.uchar,
 
 ) { s.__bitfield_1 = s.__bitfield_1&0x3f | value<<6 }
 
-type C.bitfield_t = C._Ctype_struct___9
+type C.bitfield_t = C.struct_bitfield_t