Files
Odin/src/parser.cpp
2024-07-22 16:11:33 +02:00

6626 lines
190 KiB
C++

#include "parser_pos.cpp"
gb_internal u64 ast_file_vet_flags(AstFile *f) {
if (f != nullptr && f->vet_flags_set) {
return f->vet_flags;
}
return build_context.vet_flags;
}
gb_internal bool ast_file_vet_style(AstFile *f) {
return (ast_file_vet_flags(f) & VetFlag_Style) != 0;
}
gb_internal bool ast_file_vet_deprecated(AstFile *f) {
return (ast_file_vet_flags(f) & VetFlag_Deprecated) != 0;
}
gb_internal bool file_allow_newline(AstFile *f) {
bool is_strict = build_context.strict_style || ast_file_vet_style(f);
return !is_strict;
}
gb_internal Token token_end_of_line(AstFile *f, Token tok) {
u8 const *start = f->tokenizer.start + tok.pos.offset;
u8 const *s = start;
while (*s && *s != '\n' && s < f->tokenizer.end) {
s += 1;
}
tok.pos.column += cast(i32)(s - start) - 1;
return tok;
}
gb_internal gbString get_file_line_as_string(TokenPos const &pos, i32 *offset_) {
AstFile *file = thread_safe_get_ast_file_from_id(pos.file_id);
if (file == nullptr) {
return nullptr;
}
u8 *start = file->tokenizer.start;
u8 *end = file->tokenizer.end;
if (start == end) {
return nullptr;
}
isize offset = pos.offset;
if (pos.line != 0 && offset == 0) {
for (i32 i = 1; i < pos.line; i++) {
while (start+offset < end) {
u8 c = start[offset++];
if (c == '\n') {
break;
}
}
}
for (i32 i = 1; i < pos.column; i++) {
u8 *ptr = start+offset;
u8 c = *ptr;
if (c & 0x80) {
offset += utf8_decode(ptr, end-ptr, nullptr);
} else {
offset++;
}
}
}
isize len = end-start;
if (len < offset) {
return nullptr;
}
u8 *pos_offset = start+offset;
u8 *line_start = pos_offset;
u8 *line_end = pos_offset;
if (offset > 0 && *line_start == '\n') {
// Prevent an error token that starts at the boundary of a line that
// leads to an empty line from advancing off its line.
line_start -= 1;
}
while (line_start >= start) {
if (*line_start == '\n') {
line_start += 1;
break;
}
line_start -= 1;
}
if (line_start == start - 1) {
// Prevent an error on the first line from stepping behind the boundary
// of the text.
line_start += 1;
}
while (line_end < end) {
if (*line_end == '\n') {
break;
}
line_end += 1;
}
String the_line = make_string(line_start, line_end-line_start);
the_line = string_trim_whitespace(the_line);
if (offset_) *offset_ = cast(i32)(pos_offset - the_line.text);
return gb_string_make_length(heap_allocator(), the_line.text, the_line.len);
}
gb_internal isize ast_node_size(AstKind kind) {
return align_formula_isize(gb_size_of(AstCommonStuff) + ast_variant_sizes[kind], gb_align_of(void *));
}
// gb_global std::atomic<isize> global_total_node_memory_allocated;
// NOTE(bill): And this below is why is I/we need a new language! Discriminated unions are a pain in C/C++
gb_internal Ast *alloc_ast_node(AstFile *f, AstKind kind) {
isize size = ast_node_size(kind);
Ast *node = cast(Ast *)arena_alloc(get_arena(ThreadArena_Permanent), size, 16);
node->kind = kind;
node->file_id = f ? f->id : 0;
// global_total_node_memory_allocated.fetch_add(size);
return node;
}
gb_internal Ast *clone_ast(Ast *node, AstFile *f = nullptr);
gb_internal Array<Ast *> clone_ast_array(Array<Ast *> const &array, AstFile *f) {
Array<Ast *> result = {};
if (array.count > 0) {
result = array_make<Ast *>(ast_allocator(nullptr), array.count);
for_array(i, array) {
result[i] = clone_ast(array[i], f);
}
}
return result;
}
gb_internal Slice<Ast *> clone_ast_array(Slice<Ast *> const &array, AstFile *f) {
Slice<Ast *> result = {};
if (array.count > 0) {
result = slice_clone(ast_allocator(nullptr), array);
for_array(i, array) {
result[i] = clone_ast(array[i], f);
}
}
return result;
}
gb_internal Ast *clone_ast(Ast *node, AstFile *f) {
if (node == nullptr) {
return nullptr;
}
if (f == nullptr) {
f = node->thread_safe_file();
}
Ast *n = alloc_ast_node(f, node->kind);
gb_memmove(n, node, ast_node_size(node->kind));
switch (n->kind) {
default: GB_PANIC("Unhandled Ast %.*s", LIT(ast_strings[n->kind])); break;
case Ast_Invalid: break;
case Ast_Ident:
n->Ident.entity = nullptr;
break;
case Ast_Implicit: break;
case Ast_Uninit: break;
case Ast_BasicLit: break;
case Ast_BasicDirective: break;
case Ast_PolyType:
n->PolyType.type = clone_ast(n->PolyType.type, f);
n->PolyType.specialization = clone_ast(n->PolyType.specialization, f);
break;
case Ast_Ellipsis:
n->Ellipsis.expr = clone_ast(n->Ellipsis.expr, f);
break;
case Ast_ProcGroup:
n->ProcGroup.args = clone_ast_array(n->ProcGroup.args, f);
break;
case Ast_ProcLit:
n->ProcLit.type = clone_ast(n->ProcLit.type, f);
n->ProcLit.body = clone_ast(n->ProcLit.body, f);
n->ProcLit.where_clauses = clone_ast_array(n->ProcLit.where_clauses, f);
break;
case Ast_CompoundLit:
n->CompoundLit.type = clone_ast(n->CompoundLit.type, f);
n->CompoundLit.elems = clone_ast_array(n->CompoundLit.elems, f);
break;
case Ast_BadExpr: break;
case Ast_TagExpr:
n->TagExpr.expr = clone_ast(n->TagExpr.expr, f);
break;
case Ast_UnaryExpr:
n->UnaryExpr.expr = clone_ast(n->UnaryExpr.expr, f);
break;
case Ast_BinaryExpr:
n->BinaryExpr.left = clone_ast(n->BinaryExpr.left, f);
n->BinaryExpr.right = clone_ast(n->BinaryExpr.right, f);
break;
case Ast_ParenExpr:
n->ParenExpr.expr = clone_ast(n->ParenExpr.expr, f);
break;
case Ast_SelectorExpr:
n->SelectorExpr.expr = clone_ast(n->SelectorExpr.expr, f);
n->SelectorExpr.selector = clone_ast(n->SelectorExpr.selector, f);
break;
case Ast_ImplicitSelectorExpr:
n->ImplicitSelectorExpr.selector = clone_ast(n->ImplicitSelectorExpr.selector, f);
break;
case Ast_SelectorCallExpr:
n->SelectorCallExpr.expr = clone_ast(n->SelectorCallExpr.expr, f);
n->SelectorCallExpr.call = clone_ast(n->SelectorCallExpr.call, f);
break;
case Ast_IndexExpr:
n->IndexExpr.expr = clone_ast(n->IndexExpr.expr, f);
n->IndexExpr.index = clone_ast(n->IndexExpr.index, f);
break;
case Ast_MatrixIndexExpr:
n->MatrixIndexExpr.expr = clone_ast(n->MatrixIndexExpr.expr, f);
n->MatrixIndexExpr.row_index = clone_ast(n->MatrixIndexExpr.row_index, f);
n->MatrixIndexExpr.column_index = clone_ast(n->MatrixIndexExpr.column_index, f);
break;
case Ast_DerefExpr:
n->DerefExpr.expr = clone_ast(n->DerefExpr.expr, f);
break;
case Ast_SliceExpr:
n->SliceExpr.expr = clone_ast(n->SliceExpr.expr, f);
n->SliceExpr.low = clone_ast(n->SliceExpr.low, f);
n->SliceExpr.high = clone_ast(n->SliceExpr.high, f);
break;
case Ast_CallExpr:
n->CallExpr.proc = clone_ast(n->CallExpr.proc, f);
n->CallExpr.args = clone_ast_array(n->CallExpr.args, f);
break;
case Ast_FieldValue:
n->FieldValue.field = clone_ast(n->FieldValue.field, f);
n->FieldValue.value = clone_ast(n->FieldValue.value, f);
break;
case Ast_EnumFieldValue:
n->EnumFieldValue.name = clone_ast(n->EnumFieldValue.name, f);
n->EnumFieldValue.value = clone_ast(n->EnumFieldValue.value, f);
break;
case Ast_TernaryIfExpr:
n->TernaryIfExpr.x = clone_ast(n->TernaryIfExpr.x, f);
n->TernaryIfExpr.cond = clone_ast(n->TernaryIfExpr.cond, f);
n->TernaryIfExpr.y = clone_ast(n->TernaryIfExpr.y, f);
break;
case Ast_TernaryWhenExpr:
n->TernaryWhenExpr.x = clone_ast(n->TernaryWhenExpr.x, f);
n->TernaryWhenExpr.cond = clone_ast(n->TernaryWhenExpr.cond, f);
n->TernaryWhenExpr.y = clone_ast(n->TernaryWhenExpr.y, f);
break;
case Ast_OrElseExpr:
n->OrElseExpr.x = clone_ast(n->OrElseExpr.x, f);
n->OrElseExpr.y = clone_ast(n->OrElseExpr.y, f);
break;
case Ast_OrReturnExpr:
n->OrReturnExpr.expr = clone_ast(n->OrReturnExpr.expr, f);
break;
case Ast_OrBranchExpr:
n->OrBranchExpr.label = clone_ast(n->OrBranchExpr.label, f);
n->OrBranchExpr.expr = clone_ast(n->OrBranchExpr.expr, f);
break;
case Ast_TypeAssertion:
n->TypeAssertion.expr = clone_ast(n->TypeAssertion.expr, f);
n->TypeAssertion.type = clone_ast(n->TypeAssertion.type, f);
break;
case Ast_TypeCast:
n->TypeCast.type = clone_ast(n->TypeCast.type, f);
n->TypeCast.expr = clone_ast(n->TypeCast.expr, f);
break;
case Ast_AutoCast:
n->AutoCast.expr = clone_ast(n->AutoCast.expr, f);
break;
case Ast_InlineAsmExpr:
n->InlineAsmExpr.param_types = clone_ast_array(n->InlineAsmExpr.param_types, f);
n->InlineAsmExpr.return_type = clone_ast(n->InlineAsmExpr.return_type, f);
n->InlineAsmExpr.asm_string = clone_ast(n->InlineAsmExpr.asm_string, f);
n->InlineAsmExpr.constraints_string = clone_ast(n->InlineAsmExpr.constraints_string, f);
break;
case Ast_BadStmt: break;
case Ast_EmptyStmt: break;
case Ast_ExprStmt:
n->ExprStmt.expr = clone_ast(n->ExprStmt.expr, f);
break;
case Ast_AssignStmt:
n->AssignStmt.lhs = clone_ast_array(n->AssignStmt.lhs, f);
n->AssignStmt.rhs = clone_ast_array(n->AssignStmt.rhs, f);
break;
case Ast_BlockStmt:
n->BlockStmt.label = clone_ast(n->BlockStmt.label, f);
n->BlockStmt.stmts = clone_ast_array(n->BlockStmt.stmts, f);
break;
case Ast_IfStmt:
n->IfStmt.label = clone_ast(n->IfStmt.label, f);
n->IfStmt.init = clone_ast(n->IfStmt.init, f);
n->IfStmt.cond = clone_ast(n->IfStmt.cond, f);
n->IfStmt.body = clone_ast(n->IfStmt.body, f);
n->IfStmt.else_stmt = clone_ast(n->IfStmt.else_stmt, f);
break;
case Ast_WhenStmt:
n->WhenStmt.cond = clone_ast(n->WhenStmt.cond, f);
n->WhenStmt.body = clone_ast(n->WhenStmt.body, f);
n->WhenStmt.else_stmt = clone_ast(n->WhenStmt.else_stmt, f);
break;
case Ast_ReturnStmt:
n->ReturnStmt.results = clone_ast_array(n->ReturnStmt.results, f);
break;
case Ast_ForStmt:
n->ForStmt.label = clone_ast(n->ForStmt.label, f);
n->ForStmt.init = clone_ast(n->ForStmt.init, f);
n->ForStmt.cond = clone_ast(n->ForStmt.cond, f);
n->ForStmt.post = clone_ast(n->ForStmt.post, f);
n->ForStmt.body = clone_ast(n->ForStmt.body, f);
break;
case Ast_RangeStmt:
n->RangeStmt.label = clone_ast(n->RangeStmt.label, f);
n->RangeStmt.vals = clone_ast_array(n->RangeStmt.vals, f);
n->RangeStmt.expr = clone_ast(n->RangeStmt.expr, f);
n->RangeStmt.body = clone_ast(n->RangeStmt.body, f);
break;
case Ast_UnrollRangeStmt:
n->UnrollRangeStmt.val0 = clone_ast(n->UnrollRangeStmt.val0, f);
n->UnrollRangeStmt.val1 = clone_ast(n->UnrollRangeStmt.val1, f);
n->UnrollRangeStmt.expr = clone_ast(n->UnrollRangeStmt.expr, f);
n->UnrollRangeStmt.body = clone_ast(n->UnrollRangeStmt.body, f);
break;
case Ast_CaseClause:
n->CaseClause.list = clone_ast_array(n->CaseClause.list, f);
n->CaseClause.stmts = clone_ast_array(n->CaseClause.stmts, f);
n->CaseClause.implicit_entity = nullptr;
break;
case Ast_SwitchStmt:
n->SwitchStmt.label = clone_ast(n->SwitchStmt.label, f);
n->SwitchStmt.init = clone_ast(n->SwitchStmt.init, f);
n->SwitchStmt.tag = clone_ast(n->SwitchStmt.tag, f);
n->SwitchStmt.body = clone_ast(n->SwitchStmt.body, f);
break;
case Ast_TypeSwitchStmt:
n->TypeSwitchStmt.label = clone_ast(n->TypeSwitchStmt.label, f);
n->TypeSwitchStmt.tag = clone_ast(n->TypeSwitchStmt.tag, f);
n->TypeSwitchStmt.body = clone_ast(n->TypeSwitchStmt.body, f);
break;
case Ast_DeferStmt:
n->DeferStmt.stmt = clone_ast(n->DeferStmt.stmt, f);
break;
case Ast_BranchStmt:
n->BranchStmt.label = clone_ast(n->BranchStmt.label, f);
break;
case Ast_UsingStmt:
n->UsingStmt.list = clone_ast_array(n->UsingStmt.list, f);
break;
case Ast_BadDecl: break;
case Ast_ForeignBlockDecl:
n->ForeignBlockDecl.foreign_library = clone_ast(n->ForeignBlockDecl.foreign_library, f);
n->ForeignBlockDecl.body = clone_ast(n->ForeignBlockDecl.body, f);
n->ForeignBlockDecl.attributes = clone_ast_array(n->ForeignBlockDecl.attributes, f);
break;
case Ast_Label:
n->Label.name = clone_ast(n->Label.name, f);
break;
case Ast_ValueDecl:
n->ValueDecl.names = clone_ast_array(n->ValueDecl.names, f);
n->ValueDecl.type = clone_ast(n->ValueDecl.type, f);
n->ValueDecl.values = clone_ast_array(n->ValueDecl.values, f);
n->ValueDecl.attributes = clone_ast_array(n->ValueDecl.attributes, f);
break;
case Ast_Attribute:
n->Attribute.elems = clone_ast_array(n->Attribute.elems, f);
break;
case Ast_Field:
n->Field.names = clone_ast_array(n->Field.names, f);
n->Field.type = clone_ast(n->Field.type, f);
break;
case Ast_BitFieldField:
n->BitFieldField.name = clone_ast(n->BitFieldField.name, f);
n->BitFieldField.type = clone_ast(n->BitFieldField.type, f);
n->BitFieldField.bit_size = clone_ast(n->BitFieldField.bit_size, f);
break;
case Ast_FieldList:
n->FieldList.list = clone_ast_array(n->FieldList.list, f);
break;
case Ast_TypeidType:
n->TypeidType.specialization = clone_ast(n->TypeidType.specialization, f);
break;
case Ast_HelperType:
n->HelperType.type = clone_ast(n->HelperType.type, f);
break;
case Ast_DistinctType:
n->DistinctType.type = clone_ast(n->DistinctType.type, f);
break;
case Ast_ProcType:
n->ProcType.params = clone_ast(n->ProcType.params, f);
n->ProcType.results = clone_ast(n->ProcType.results, f);
break;
case Ast_RelativeType:
n->RelativeType.tag = clone_ast(n->RelativeType.tag, f);
n->RelativeType.type = clone_ast(n->RelativeType.type, f);
break;
case Ast_PointerType:
n->PointerType.type = clone_ast(n->PointerType.type, f);
n->PointerType.tag = clone_ast(n->PointerType.tag, f);
break;
case Ast_MultiPointerType:
n->MultiPointerType.type = clone_ast(n->MultiPointerType.type, f);
break;
case Ast_ArrayType:
n->ArrayType.count = clone_ast(n->ArrayType.count, f);
n->ArrayType.elem = clone_ast(n->ArrayType.elem, f);
n->ArrayType.tag = clone_ast(n->ArrayType.tag, f);
break;
case Ast_DynamicArrayType:
n->DynamicArrayType.elem = clone_ast(n->DynamicArrayType.elem, f);
break;
case Ast_StructType:
n->StructType.fields = clone_ast_array(n->StructType.fields, f);
n->StructType.polymorphic_params = clone_ast(n->StructType.polymorphic_params, f);
n->StructType.align = clone_ast(n->StructType.align, f);
n->StructType.field_align = clone_ast(n->StructType.field_align, f);
n->StructType.where_clauses = clone_ast_array(n->StructType.where_clauses, f);
break;
case Ast_UnionType:
n->UnionType.variants = clone_ast_array(n->UnionType.variants, f);
n->UnionType.polymorphic_params = clone_ast(n->UnionType.polymorphic_params, f);
n->UnionType.where_clauses = clone_ast_array(n->UnionType.where_clauses, f);
break;
case Ast_EnumType:
n->EnumType.base_type = clone_ast(n->EnumType.base_type, f);
n->EnumType.fields = clone_ast_array(n->EnumType.fields, f);
break;
case Ast_BitSetType:
n->BitSetType.elem = clone_ast(n->BitSetType.elem, f);
n->BitSetType.underlying = clone_ast(n->BitSetType.underlying, f);
break;
case Ast_BitFieldType:
n->BitFieldType.backing_type = clone_ast(n->BitFieldType.backing_type, f);
n->BitFieldType.fields = clone_ast_array(n->BitFieldType.fields, f);
break;
case Ast_MapType:
n->MapType.count = clone_ast(n->MapType.count, f);
n->MapType.key = clone_ast(n->MapType.key, f);
n->MapType.value = clone_ast(n->MapType.value, f);
break;
case Ast_MatrixType:
n->MatrixType.row_count = clone_ast(n->MatrixType.row_count, f);
n->MatrixType.column_count = clone_ast(n->MatrixType.column_count, f);
n->MatrixType.elem = clone_ast(n->MatrixType.elem, f);
break;
}
return n;
}
gb_internal void error(Ast *node, char const *fmt, ...) {
Token token = {};
TokenPos end_pos = {};
if (node != nullptr) {
token = ast_token(node);
end_pos = ast_end_pos(node);
}
va_list va;
va_start(va, fmt);
error_va(token.pos, end_pos, fmt, va);
va_end(va);
if (node != nullptr && node->file_id != 0) {
AstFile *f = node->thread_safe_file();
f->error_count += 1;
}
}
gb_internal void syntax_error_with_verbose(Ast *node, char const *fmt, ...) {
Token token = {};
TokenPos end_pos = {};
if (node != nullptr) {
token = ast_token(node);
end_pos = ast_end_pos(node);
}
va_list va;
va_start(va, fmt);
syntax_error_with_verbose_va(token.pos, end_pos, fmt, va);
va_end(va);
if (node != nullptr && node->file_id != 0) {
AstFile *f = node->thread_safe_file();
f->error_count += 1;
}
}
gb_internal void error_no_newline(Ast *node, char const *fmt, ...) {
Token token = {};
if (node != nullptr) {
token = ast_token(node);
}
va_list va;
va_start(va, fmt);
error_no_newline_va(token.pos, fmt, va);
va_end(va);
if (node != nullptr && node->file_id != 0) {
AstFile *f = node->thread_safe_file();
f->error_count += 1;
}
}
gb_internal void warning(Ast *node, char const *fmt, ...) {
Token token = {};
TokenPos end_pos = {};
if (node != nullptr) {
token = ast_token(node);
end_pos = ast_end_pos(node);
}
va_list va;
va_start(va, fmt);
warning_va(token.pos, end_pos, fmt, va);
va_end(va);
}
gb_internal void syntax_error(Ast *node, char const *fmt, ...) {
Token token = {};
TokenPos end_pos = {};
if (node != nullptr) {
token = ast_token(node);
end_pos = ast_end_pos(node);
}
va_list va;
va_start(va, fmt);
syntax_error_va(token.pos, end_pos, fmt, va);
va_end(va);
if (node != nullptr && node->file_id != 0) {
AstFile *f = node->thread_safe_file();
f->error_count += 1;
}
}
gb_internal bool ast_node_expect(Ast *node, AstKind kind) {
if (node->kind != kind) {
syntax_error(node, "Expected %.*s, got %.*s", LIT(ast_strings[kind]), LIT(ast_strings[node->kind]));
return false;
}
return true;
}
gb_internal bool ast_node_expect2(Ast *node, AstKind kind0, AstKind kind1) {
if (node->kind != kind0 && node->kind != kind1) {
syntax_error(node, "Expected %.*s or %.*s, got %.*s", LIT(ast_strings[kind0]), LIT(ast_strings[kind1]), LIT(ast_strings[node->kind]));
return false;
}
return true;
}
gb_internal Ast *ast_bad_expr(AstFile *f, Token begin, Token end) {
Ast *result = alloc_ast_node(f, Ast_BadExpr);
result->BadExpr.begin = begin;
result->BadExpr.end = end;
return result;
}
gb_internal Ast *ast_tag_expr(AstFile *f, Token token, Token name, Ast *expr) {
Ast *result = alloc_ast_node(f, Ast_TagExpr);
result->TagExpr.token = token;
result->TagExpr.name = name;
result->TagExpr.expr = expr;
return result;
}
gb_internal Ast *ast_unary_expr(AstFile *f, Token op, Ast *expr) {
Ast *result = alloc_ast_node(f, Ast_UnaryExpr);
if (expr) switch (expr->kind) {
case Ast_OrReturnExpr:
syntax_error_with_verbose(expr, "'or_return' within an unary expression not wrapped in parentheses (...)");
break;
case Ast_OrBranchExpr:
syntax_error_with_verbose(expr, "'%.*s' within an unary expression not wrapped in parentheses (...)", LIT(expr->OrBranchExpr.token.string));
break;
}
result->UnaryExpr.op = op;
result->UnaryExpr.expr = expr;
return result;
}
gb_internal Ast *ast_binary_expr(AstFile *f, Token op, Ast *left, Ast *right) {
Ast *result = alloc_ast_node(f, Ast_BinaryExpr);
if (left == nullptr) {
syntax_error(op, "No lhs expression for binary expression '%.*s'", LIT(op.string));
left = ast_bad_expr(f, op, op);
}
if (right == nullptr) {
syntax_error(op, "No rhs expression for binary expression '%.*s'", LIT(op.string));
right = ast_bad_expr(f, op, op);
}
if (left) switch (left->kind) {
case Ast_OrReturnExpr:
syntax_error_with_verbose(left, "'or_return' within a binary expression not wrapped in parentheses (...)");
break;
case Ast_OrBranchExpr:
syntax_error_with_verbose(left, "'%.*s' within a binary expression not wrapped in parentheses (...)", LIT(left->OrBranchExpr.token.string));
break;
}
if (right) switch (right->kind) {
case Ast_OrReturnExpr:
syntax_error_with_verbose(right, "'or_return' within a binary expression not wrapped in parentheses (...)");
break;
case Ast_OrBranchExpr:
syntax_error_with_verbose(right, "'%.*s' within a binary expression not wrapped in parentheses (...)", LIT(right->OrBranchExpr.token.string));
break;
}
result->BinaryExpr.op = op;
result->BinaryExpr.left = left;
result->BinaryExpr.right = right;
return result;
}
gb_internal Ast *ast_paren_expr(AstFile *f, Ast *expr, Token open, Token close) {
Ast *result = alloc_ast_node(f, Ast_ParenExpr);
result->ParenExpr.expr = expr;
result->ParenExpr.open = open;
result->ParenExpr.close = close;
return result;
}
gb_internal Ast *ast_call_expr(AstFile *f, Ast *proc, Array<Ast *> const &args, Token open, Token close, Token ellipsis) {
Ast *result = alloc_ast_node(f, Ast_CallExpr);
result->CallExpr.proc = proc;
result->CallExpr.args = slice_from_array(args);
result->CallExpr.open = open;
result->CallExpr.close = close;
result->CallExpr.ellipsis = ellipsis;
return result;
}
gb_internal Ast *ast_selector_expr(AstFile *f, Token token, Ast *expr, Ast *selector) {
Ast *result = alloc_ast_node(f, Ast_SelectorExpr);
result->SelectorExpr.token = token;
result->SelectorExpr.expr = expr;
result->SelectorExpr.selector = selector;
return result;
}
gb_internal Ast *ast_implicit_selector_expr(AstFile *f, Token token, Ast *selector) {
Ast *result = alloc_ast_node(f, Ast_ImplicitSelectorExpr);
result->ImplicitSelectorExpr.token = token;
result->ImplicitSelectorExpr.selector = selector;
return result;
}
gb_internal Ast *ast_selector_call_expr(AstFile *f, Token token, Ast *expr, Ast *call) {
Ast *result = alloc_ast_node(f, Ast_SelectorCallExpr);
result->SelectorCallExpr.token = token;
result->SelectorCallExpr.expr = expr;
result->SelectorCallExpr.call = call;
return result;
}
gb_internal Ast *ast_index_expr(AstFile *f, Ast *expr, Ast *index, Token open, Token close) {
Ast *result = alloc_ast_node(f, Ast_IndexExpr);
result->IndexExpr.expr = expr;
result->IndexExpr.index = index;
result->IndexExpr.open = open;
result->IndexExpr.close = close;
return result;
}
gb_internal Ast *ast_slice_expr(AstFile *f, Ast *expr, Token open, Token close, Token interval, Ast *low, Ast *high) {
Ast *result = alloc_ast_node(f, Ast_SliceExpr);
result->SliceExpr.expr = expr;
result->SliceExpr.open = open;
result->SliceExpr.close = close;
result->SliceExpr.interval = interval;
result->SliceExpr.low = low;
result->SliceExpr.high = high;
return result;
}
gb_internal Ast *ast_deref_expr(AstFile *f, Ast *expr, Token op) {
Ast *result = alloc_ast_node(f, Ast_DerefExpr);
result->DerefExpr.expr = expr;
result->DerefExpr.op = op;
return result;
}
gb_internal Ast *ast_matrix_index_expr(AstFile *f, Ast *expr, Token open, Token close, Token interval, Ast *row, Ast *column) {
Ast *result = alloc_ast_node(f, Ast_MatrixIndexExpr);
result->MatrixIndexExpr.expr = expr;
result->MatrixIndexExpr.row_index = row;
result->MatrixIndexExpr.column_index = column;
result->MatrixIndexExpr.open = open;
result->MatrixIndexExpr.close = close;
return result;
}
gb_internal Ast *ast_ident(AstFile *f, Token token) {
Ast *result = alloc_ast_node(f, Ast_Ident);
result->Ident.token = token;
return result;
}
gb_internal Ast *ast_implicit(AstFile *f, Token token) {
Ast *result = alloc_ast_node(f, Ast_Implicit);
result->Implicit = token;
return result;
}
gb_internal Ast *ast_uninit(AstFile *f, Token token) {
Ast *result = alloc_ast_node(f, Ast_Uninit);
result->Uninit = token;
return result;
}
gb_internal ExactValue exact_value_from_token(AstFile *f, Token const &token) {
String s = token.string;
switch (token.kind) {
case Token_Rune:
if (!unquote_string(ast_allocator(f), &s, 0)) {
syntax_error(token, "Invalid rune literal");
}
break;
case Token_String:
if (!unquote_string(ast_allocator(f), &s, 0, s.text[0] == '`')) {
syntax_error(token, "Invalid string literal");
}
break;
}
ExactValue value = exact_value_from_basic_literal(token.kind, s);
if (value.kind == ExactValue_Invalid) {
switch (token.kind) {
case Token_Integer:
syntax_error(token, "Invalid integer literal");
break;
case Token_Float:
syntax_error(token, "Invalid float literal");
break;
default:
syntax_error(token, "Invalid token literal");
break;
}
}
return value;
}
gb_internal String string_value_from_token(AstFile *f, Token const &token) {
ExactValue value = exact_value_from_token(f, token);
String str = {};
if (value.kind == ExactValue_String) {
str = value.value_string;
}
return str;
}
gb_internal Ast *ast_basic_lit(AstFile *f, Token basic_lit) {
Ast *result = alloc_ast_node(f, Ast_BasicLit);
result->BasicLit.token = basic_lit;
result->tav.mode = Addressing_Constant;
result->tav.value = exact_value_from_token(f, basic_lit);
return result;
}
gb_internal Ast *ast_basic_directive(AstFile *f, Token token, Token name) {
Ast *result = alloc_ast_node(f, Ast_BasicDirective);
result->BasicDirective.token = token;
result->BasicDirective.name = name;
if (string_starts_with(name.string, str_lit("load"))) {
f->seen_load_directive_count++;
}
return result;
}
gb_internal Ast *ast_ellipsis(AstFile *f, Token token, Ast *expr) {
Ast *result = alloc_ast_node(f, Ast_Ellipsis);
result->Ellipsis.token = token;
result->Ellipsis.expr = expr;
return result;
}
gb_internal Ast *ast_proc_group(AstFile *f, Token token, Token open, Token close, Array<Ast *> const &args) {
Ast *result = alloc_ast_node(f, Ast_ProcGroup);
result->ProcGroup.token = token;
result->ProcGroup.open = open;
result->ProcGroup.close = close;
result->ProcGroup.args = slice_from_array(args);
return result;
}
gb_internal Ast *ast_proc_lit(AstFile *f, Ast *type, Ast *body, u64 tags, Token where_token, Array<Ast *> const &where_clauses) {
Ast *result = alloc_ast_node(f, Ast_ProcLit);
result->ProcLit.type = type;
result->ProcLit.body = body;
result->ProcLit.tags = tags;
result->ProcLit.where_token = where_token;
result->ProcLit.where_clauses = slice_from_array(where_clauses);
return result;
}
gb_internal Ast *ast_field_value(AstFile *f, Ast *field, Ast *value, Token eq) {
Ast *result = alloc_ast_node(f, Ast_FieldValue);
result->FieldValue.field = field;
result->FieldValue.value = value;
result->FieldValue.eq = eq;
return result;
}
gb_internal Ast *ast_enum_field_value(AstFile *f, Ast *name, Ast *value, CommentGroup *docs, CommentGroup *comment) {
Ast *result = alloc_ast_node(f, Ast_EnumFieldValue);
result->EnumFieldValue.name = name;
result->EnumFieldValue.value = value;
result->EnumFieldValue.docs = docs;
result->EnumFieldValue.comment = comment;
return result;
}
gb_internal Ast *ast_compound_lit(AstFile *f, Ast *type, Array<Ast *> const &elems, Token open, Token close) {
Ast *result = alloc_ast_node(f, Ast_CompoundLit);
result->CompoundLit.type = type;
result->CompoundLit.elems = slice_from_array(elems);
result->CompoundLit.open = open;
result->CompoundLit.close = close;
return result;
}
gb_internal Ast *ast_ternary_if_expr(AstFile *f, Ast *x, Ast *cond, Ast *y) {
Ast *result = alloc_ast_node(f, Ast_TernaryIfExpr);
result->TernaryIfExpr.x = x;
result->TernaryIfExpr.cond = cond;
result->TernaryIfExpr.y = y;
return result;
}
gb_internal Ast *ast_ternary_when_expr(AstFile *f, Ast *x, Ast *cond, Ast *y) {
Ast *result = alloc_ast_node(f, Ast_TernaryWhenExpr);
result->TernaryWhenExpr.x = x;
result->TernaryWhenExpr.cond = cond;
result->TernaryWhenExpr.y = y;
return result;
}
gb_internal Ast *ast_or_else_expr(AstFile *f, Ast *x, Token const &token, Ast *y) {
Ast *result = alloc_ast_node(f, Ast_OrElseExpr);
result->OrElseExpr.x = x;
result->OrElseExpr.token = token;
result->OrElseExpr.y = y;
return result;
}
gb_internal Ast *ast_or_return_expr(AstFile *f, Ast *expr, Token const &token) {
Ast *result = alloc_ast_node(f, Ast_OrReturnExpr);
result->OrReturnExpr.expr = expr;
result->OrReturnExpr.token = token;
return result;
}
gb_internal Ast *ast_or_branch_expr(AstFile *f, Ast *expr, Token const &token, Ast *label) {
Ast *result = alloc_ast_node(f, Ast_OrBranchExpr);
result->OrBranchExpr.expr = expr;
result->OrBranchExpr.token = token;
result->OrBranchExpr.label = label;
return result;
}
gb_internal Ast *ast_type_assertion(AstFile *f, Ast *expr, Token dot, Ast *type) {
Ast *result = alloc_ast_node(f, Ast_TypeAssertion);
result->TypeAssertion.expr = expr;
result->TypeAssertion.dot = dot;
result->TypeAssertion.type = type;
return result;
}
gb_internal Ast *ast_type_cast(AstFile *f, Token token, Ast *type, Ast *expr) {
Ast *result = alloc_ast_node(f, Ast_TypeCast);
result->TypeCast.token = token;
result->TypeCast.type = type;
result->TypeCast.expr = expr;
return result;
}
gb_internal Ast *ast_auto_cast(AstFile *f, Token token, Ast *expr) {
Ast *result = alloc_ast_node(f, Ast_AutoCast);
result->AutoCast.token = token;
result->AutoCast.expr = expr;
return result;
}
gb_internal Ast *ast_inline_asm_expr(AstFile *f, Token token, Token open, Token close,
Array<Ast *> const &param_types,
Ast *return_type,
Ast *asm_string,
Ast *constraints_string,
bool has_side_effects,
bool is_align_stack,
InlineAsmDialectKind dialect) {
Ast *result = alloc_ast_node(f, Ast_InlineAsmExpr);
result->InlineAsmExpr.token = token;
result->InlineAsmExpr.open = open;
result->InlineAsmExpr.close = close;
result->InlineAsmExpr.param_types = slice_from_array(param_types);
result->InlineAsmExpr.return_type = return_type;
result->InlineAsmExpr.asm_string = asm_string;
result->InlineAsmExpr.constraints_string = constraints_string;
result->InlineAsmExpr.has_side_effects = has_side_effects;
result->InlineAsmExpr.is_align_stack = is_align_stack;
result->InlineAsmExpr.dialect = dialect;
return result;
}
gb_internal Ast *ast_bad_stmt(AstFile *f, Token begin, Token end) {
Ast *result = alloc_ast_node(f, Ast_BadStmt);
result->BadStmt.begin = begin;
result->BadStmt.end = end;
return result;
}
gb_internal Ast *ast_empty_stmt(AstFile *f, Token token) {
Ast *result = alloc_ast_node(f, Ast_EmptyStmt);
result->EmptyStmt.token = token;
return result;
}
gb_internal Ast *ast_expr_stmt(AstFile *f, Ast *expr) {
Ast *result = alloc_ast_node(f, Ast_ExprStmt);
result->ExprStmt.expr = expr;
return result;
}
gb_internal Ast *ast_assign_stmt(AstFile *f, Token op, Array<Ast *> const &lhs, Array<Ast *> const &rhs) {
Ast *result = alloc_ast_node(f, Ast_AssignStmt);
result->AssignStmt.op = op;
result->AssignStmt.lhs = slice_from_array(lhs);
result->AssignStmt.rhs = slice_from_array(rhs);
return result;
}
gb_internal Ast *ast_block_stmt(AstFile *f, Array<Ast *> const &stmts, Token open, Token close) {
Ast *result = alloc_ast_node(f, Ast_BlockStmt);
result->BlockStmt.stmts = slice_from_array(stmts);
result->BlockStmt.open = open;
result->BlockStmt.close = close;
return result;
}
gb_internal Ast *ast_if_stmt(AstFile *f, Token token, Ast *init, Ast *cond, Ast *body, Ast *else_stmt) {
Ast *result = alloc_ast_node(f, Ast_IfStmt);
result->IfStmt.token = token;
result->IfStmt.init = init;
result->IfStmt.cond = cond;
result->IfStmt.body = body;
result->IfStmt.else_stmt = else_stmt;
return result;
}
gb_internal Ast *ast_when_stmt(AstFile *f, Token token, Ast *cond, Ast *body, Ast *else_stmt) {
Ast *result = alloc_ast_node(f, Ast_WhenStmt);
result->WhenStmt.token = token;
result->WhenStmt.cond = cond;
result->WhenStmt.body = body;
result->WhenStmt.else_stmt = else_stmt;
return result;
}
gb_internal Ast *ast_return_stmt(AstFile *f, Token token, Array<Ast *> const &results) {
Ast *result = alloc_ast_node(f, Ast_ReturnStmt);
result->ReturnStmt.token = token;
result->ReturnStmt.results = slice_from_array(results);
return result;
}
gb_internal Ast *ast_for_stmt(AstFile *f, Token token, Ast *init, Ast *cond, Ast *post, Ast *body) {
Ast *result = alloc_ast_node(f, Ast_ForStmt);
result->ForStmt.token = token;
result->ForStmt.init = init;
result->ForStmt.cond = cond;
result->ForStmt.post = post;
result->ForStmt.body = body;
return result;
}
gb_internal Ast *ast_range_stmt(AstFile *f, Token token, Slice<Ast *> vals, Token in_token, Ast *expr, Ast *body) {
Ast *result = alloc_ast_node(f, Ast_RangeStmt);
result->RangeStmt.token = token;
result->RangeStmt.vals = vals;
result->RangeStmt.in_token = in_token;
result->RangeStmt.expr = expr;
result->RangeStmt.body = body;
return result;
}
gb_internal Ast *ast_unroll_range_stmt(AstFile *f, Token unroll_token, Token for_token, Ast *val0, Ast *val1, Token in_token, Ast *expr, Ast *body) {
Ast *result = alloc_ast_node(f, Ast_UnrollRangeStmt);
result->UnrollRangeStmt.unroll_token = unroll_token;
result->UnrollRangeStmt.for_token = for_token;
result->UnrollRangeStmt.val0 = val0;
result->UnrollRangeStmt.val1 = val1;
result->UnrollRangeStmt.in_token = in_token;
result->UnrollRangeStmt.expr = expr;
result->UnrollRangeStmt.body = body;
return result;
}
gb_internal Ast *ast_switch_stmt(AstFile *f, Token token, Ast *init, Ast *tag, Ast *body) {
Ast *result = alloc_ast_node(f, Ast_SwitchStmt);
result->SwitchStmt.token = token;
result->SwitchStmt.init = init;
result->SwitchStmt.tag = tag;
result->SwitchStmt.body = body;
result->SwitchStmt.partial = false;
return result;
}
gb_internal Ast *ast_type_switch_stmt(AstFile *f, Token token, Ast *tag, Ast *body) {
Ast *result = alloc_ast_node(f, Ast_TypeSwitchStmt);
result->TypeSwitchStmt.token = token;
result->TypeSwitchStmt.tag = tag;
result->TypeSwitchStmt.body = body;
result->TypeSwitchStmt.partial = false;
return result;
}
gb_internal Ast *ast_case_clause(AstFile *f, Token token, Array<Ast *> const &list, Array<Ast *> const &stmts) {
Ast *result = alloc_ast_node(f, Ast_CaseClause);
result->CaseClause.token = token;
result->CaseClause.list = slice_from_array(list);
result->CaseClause.stmts = slice_from_array(stmts);
return result;
}
gb_internal Ast *ast_defer_stmt(AstFile *f, Token token, Ast *stmt) {
Ast *result = alloc_ast_node(f, Ast_DeferStmt);
result->DeferStmt.token = token;
result->DeferStmt.stmt = stmt;
return result;
}
gb_internal Ast *ast_branch_stmt(AstFile *f, Token token, Ast *label) {
Ast *result = alloc_ast_node(f, Ast_BranchStmt);
result->BranchStmt.token = token;
result->BranchStmt.label = label;
return result;
}
gb_internal Ast *ast_using_stmt(AstFile *f, Token token, Array<Ast *> const &list) {
Ast *result = alloc_ast_node(f, Ast_UsingStmt);
result->UsingStmt.token = token;
result->UsingStmt.list = slice_from_array(list);
return result;
}
gb_internal Ast *ast_bad_decl(AstFile *f, Token begin, Token end) {
Ast *result = alloc_ast_node(f, Ast_BadDecl);
result->BadDecl.begin = begin;
result->BadDecl.end = end;
return result;
}
gb_internal Ast *ast_field(AstFile *f, Array<Ast *> const &names, Ast *type, Ast *default_value, u32 flags, Token tag,
CommentGroup *docs, CommentGroup *comment) {
Ast *result = alloc_ast_node(f, Ast_Field);
result->Field.names = slice_from_array(names);
result->Field.type = type;
result->Field.default_value = default_value;
result->Field.flags = flags;
result->Field.tag = tag;
result->Field.docs = docs;
result->Field.comment = comment;
return result;
}
gb_internal Ast *ast_bit_field_field(AstFile *f, Ast *name, Ast *type, Ast *bit_size, Token tag,
CommentGroup *docs, CommentGroup *comment) {
Ast *result = alloc_ast_node(f, Ast_BitFieldField);
result->BitFieldField.name = name;
result->BitFieldField.type = type;
result->BitFieldField.bit_size = bit_size;
result->BitFieldField.tag = tag;
result->BitFieldField.docs = docs;
result->BitFieldField.comment = comment;
return result;
}
gb_internal Ast *ast_field_list(AstFile *f, Token token, Array<Ast *> const &list) {
Ast *result = alloc_ast_node(f, Ast_FieldList);
result->FieldList.token = token;
result->FieldList.list = slice_from_array(list);
return result;
}
gb_internal Ast *ast_typeid_type(AstFile *f, Token token, Ast *specialization) {
Ast *result = alloc_ast_node(f, Ast_TypeidType);
result->TypeidType.token = token;
result->TypeidType.specialization = specialization;
return result;
}
gb_internal Ast *ast_helper_type(AstFile *f, Token token, Ast *type) {
Ast *result = alloc_ast_node(f, Ast_HelperType);
result->HelperType.token = token;
result->HelperType.type = type;
return result;
}
gb_internal Ast *ast_distinct_type(AstFile *f, Token token, Ast *type) {
Ast *result = alloc_ast_node(f, Ast_DistinctType);
result->DistinctType.token = token;
result->DistinctType.type = type;
return result;
}
gb_internal Ast *ast_poly_type(AstFile *f, Token token, Ast *type, Ast *specialization) {
Ast *result = alloc_ast_node(f, Ast_PolyType);
result->PolyType.token = token;
result->PolyType.type = type;
result->PolyType.specialization = specialization;
return result;
}
gb_internal Ast *ast_proc_type(AstFile *f, Token token, Ast *params, Ast *results, u64 tags, ProcCallingConvention calling_convention, bool generic, bool diverging) {
Ast *result = alloc_ast_node(f, Ast_ProcType);
result->ProcType.token = token;
result->ProcType.params = params;
result->ProcType.results = results;
result->ProcType.tags = tags;
result->ProcType.calling_convention = calling_convention;
result->ProcType.generic = generic;
result->ProcType.diverging = diverging;
return result;
}
gb_internal Ast *ast_relative_type(AstFile *f, Ast *tag, Ast *type) {
Ast *result = alloc_ast_node(f, Ast_RelativeType);
result->RelativeType.tag = tag;
result->RelativeType.type = type;
return result;
}
gb_internal Ast *ast_pointer_type(AstFile *f, Token token, Ast *type) {
Ast *result = alloc_ast_node(f, Ast_PointerType);
result->PointerType.token = token;
result->PointerType.type = type;
return result;
}
gb_internal Ast *ast_multi_pointer_type(AstFile *f, Token token, Ast *type) {
Ast *result = alloc_ast_node(f, Ast_MultiPointerType);
result->MultiPointerType.token = token;
result->MultiPointerType.type = type;
return result;
}
gb_internal Ast *ast_array_type(AstFile *f, Token token, Ast *count, Ast *elem) {
Ast *result = alloc_ast_node(f, Ast_ArrayType);
result->ArrayType.token = token;
result->ArrayType.count = count;
result->ArrayType.elem = elem;
return result;
}
gb_internal Ast *ast_dynamic_array_type(AstFile *f, Token token, Ast *elem) {
Ast *result = alloc_ast_node(f, Ast_DynamicArrayType);
result->DynamicArrayType.token = token;
result->DynamicArrayType.elem = elem;
return result;
}
gb_internal Ast *ast_struct_type(AstFile *f, Token token, Slice<Ast *> fields, isize field_count,
Ast *polymorphic_params, bool is_packed, bool is_raw_union, bool is_no_copy,
Ast *align, Ast *field_align,
Token where_token, Array<Ast *> const &where_clauses) {
Ast *result = alloc_ast_node(f, Ast_StructType);
result->StructType.token = token;
result->StructType.fields = fields;
result->StructType.field_count = field_count;
result->StructType.polymorphic_params = polymorphic_params;
result->StructType.is_packed = is_packed;
result->StructType.is_raw_union = is_raw_union;
result->StructType.is_no_copy = is_no_copy;
result->StructType.align = align;
result->StructType.field_align = field_align;
result->StructType.where_token = where_token;
result->StructType.where_clauses = slice_from_array(where_clauses);
return result;
}
gb_internal Ast *ast_union_type(AstFile *f, Token token, Array<Ast *> const &variants, Ast *polymorphic_params, Ast *align, UnionTypeKind kind,
Token where_token, Array<Ast *> const &where_clauses) {
Ast *result = alloc_ast_node(f, Ast_UnionType);
result->UnionType.token = token;
result->UnionType.variants = slice_from_array(variants);
result->UnionType.polymorphic_params = polymorphic_params;
result->UnionType.align = align;
result->UnionType.kind = kind;
result->UnionType.where_token = where_token;
result->UnionType.where_clauses = slice_from_array(where_clauses);
return result;
}
gb_internal Ast *ast_enum_type(AstFile *f, Token token, Ast *base_type, Array<Ast *> const &fields) {
Ast *result = alloc_ast_node(f, Ast_EnumType);
result->EnumType.token = token;
result->EnumType.base_type = base_type;
result->EnumType.fields = slice_from_array(fields);
return result;
}
gb_internal Ast *ast_bit_set_type(AstFile *f, Token token, Ast *elem, Ast *underlying) {
Ast *result = alloc_ast_node(f, Ast_BitSetType);
result->BitSetType.token = token;
result->BitSetType.elem = elem;
result->BitSetType.underlying = underlying;
return result;
}
gb_internal Ast *ast_bit_field_type(AstFile *f, Token token, Ast *backing_type, Token open, Array<Ast *> const &fields, Token close) {
Ast *result = alloc_ast_node(f, Ast_BitFieldType);
result->BitFieldType.token = token;
result->BitFieldType.backing_type = backing_type;
result->BitFieldType.open = open;
result->BitFieldType.fields = slice_from_array(fields);
result->BitFieldType.close = close;
return result;
}
gb_internal Ast *ast_map_type(AstFile *f, Token token, Ast *key, Ast *value) {
Ast *result = alloc_ast_node(f, Ast_MapType);
result->MapType.token = token;
result->MapType.key = key;
result->MapType.value = value;
return result;
}
gb_internal Ast *ast_matrix_type(AstFile *f, Token token, Ast *row_count, Ast *column_count, Ast *elem) {
Ast *result = alloc_ast_node(f, Ast_MatrixType);
result->MatrixType.token = token;
result->MatrixType.row_count = row_count;
result->MatrixType.column_count = column_count;
result->MatrixType.elem = elem;
return result;
}
gb_internal Ast *ast_foreign_block_decl(AstFile *f, Token token, Ast *foreign_library, Ast *body,
CommentGroup *docs) {
Ast *result = alloc_ast_node(f, Ast_ForeignBlockDecl);
result->ForeignBlockDecl.token = token;
result->ForeignBlockDecl.foreign_library = foreign_library;
result->ForeignBlockDecl.body = body;
result->ForeignBlockDecl.docs = docs;
result->ForeignBlockDecl.attributes.allocator = ast_allocator(f);
return result;
}
gb_internal Ast *ast_label_decl(AstFile *f, Token token, Ast *name) {
Ast *result = alloc_ast_node(f, Ast_Label);
result->Label.token = token;
result->Label.name = name;
return result;
}
gb_internal Ast *ast_value_decl(AstFile *f, Array<Ast *> const &names, Ast *type, Array<Ast *> const &values, bool is_mutable,
CommentGroup *docs, CommentGroup *comment) {
Ast *result = alloc_ast_node(f, Ast_ValueDecl);
result->ValueDecl.names = slice_from_array(names);
result->ValueDecl.type = type;
result->ValueDecl.values = slice_from_array(values);
result->ValueDecl.is_mutable = is_mutable;
result->ValueDecl.docs = docs;
result->ValueDecl.comment = comment;
result->ValueDecl.attributes.allocator = ast_allocator(f);
return result;
}
gb_internal Ast *ast_package_decl(AstFile *f, Token token, Token name, CommentGroup *docs, CommentGroup *comment) {
Ast *result = alloc_ast_node(f, Ast_PackageDecl);
result->PackageDecl.token = token;
result->PackageDecl.name = name;
result->PackageDecl.docs = docs;
result->PackageDecl.comment = comment;
return result;
}
gb_internal Ast *ast_import_decl(AstFile *f, Token token, Token relpath, Token import_name,
CommentGroup *docs, CommentGroup *comment) {
Ast *result = alloc_ast_node(f, Ast_ImportDecl);
result->ImportDecl.token = token;
result->ImportDecl.relpath = relpath;
result->ImportDecl.import_name = import_name;
result->ImportDecl.docs = docs;
result->ImportDecl.comment = comment;
result->ImportDecl.attributes.allocator = ast_allocator(f);
return result;
}
gb_internal Ast *ast_foreign_import_decl(AstFile *f, Token token, Array<Ast *> filepaths, Token library_name,
bool multiple_filepaths,
CommentGroup *docs, CommentGroup *comment) {
Ast *result = alloc_ast_node(f, Ast_ForeignImportDecl);
result->ForeignImportDecl.token = token;
result->ForeignImportDecl.filepaths = slice_from_array(filepaths);
result->ForeignImportDecl.library_name = library_name;
result->ForeignImportDecl.docs = docs;
result->ForeignImportDecl.comment = comment;
result->ForeignImportDecl.multiple_filepaths = multiple_filepaths;
result->ForeignImportDecl.attributes.allocator = ast_allocator(f);
return result;
}
gb_internal Ast *ast_attribute(AstFile *f, Token token, Token open, Token close, Array<Ast *> const &elems) {
Ast *result = alloc_ast_node(f, Ast_Attribute);
result->Attribute.token = token;
result->Attribute.open = open;
result->Attribute.elems = slice_from_array(elems);
result->Attribute.close = close;
return result;
}
gb_internal bool next_token0(AstFile *f) {
if (f->curr_token_index+1 < f->tokens.count) {
f->curr_token = f->tokens[++f->curr_token_index];
return true;
}
syntax_error(f->curr_token, "Token is EOF");
return false;
}
gb_internal Token consume_comment(AstFile *f, isize *end_line_) {
Token tok = f->curr_token;
GB_ASSERT(tok.kind == Token_Comment);
isize end_line = tok.pos.line;
if (tok.string[1] == '*') {
for (isize i = 2; i < tok.string.len; i++) {
if (tok.string[i] == '\n') {
end_line++;
}
}
}
if (end_line_) *end_line_ = end_line;
next_token0(f);
if (f->curr_token.pos.line > tok.pos.line || tok.kind == Token_EOF) {
end_line++;
}
return tok;
}
gb_internal CommentGroup *consume_comment_group(AstFile *f, isize n, isize *end_line_) {
Array<Token> list = {};
list.allocator = ast_allocator(f);
isize end_line = f->curr_token.pos.line;
if (f->curr_token_index == 1 &&
f->prev_token.kind == Token_Comment &&
f->prev_token.pos.line+1 == f->curr_token.pos.line) {
// NOTE(bill): Special logic for the first comment in the file
array_add(&list, f->prev_token);
}
while (f->curr_token.kind == Token_Comment &&
f->curr_token.pos.line <= end_line+n) {
array_add(&list, consume_comment(f, &end_line));
}
if (end_line_) *end_line_ = end_line;
CommentGroup *comments = nullptr;
if (list.count > 0) {
comments = gb_alloc_item(permanent_allocator(), CommentGroup);
comments->list = slice_from_array(list);
array_add(&f->comments, comments);
}
return comments;
}
gb_internal void consume_comment_groups(AstFile *f, Token prev) {
if (f->curr_token.kind == Token_Comment) {
CommentGroup *comment = nullptr;
isize end_line = 0;
if (f->curr_token.pos.line == prev.pos.line) {
comment = consume_comment_group(f, 0, &end_line);
if (f->curr_token.pos.line != end_line || f->curr_token.kind == Token_EOF) {
f->line_comment = comment;
}
}
end_line = -1;
while (f->curr_token.kind == Token_Comment) {
comment = consume_comment_group(f, 1, &end_line);
}
if (end_line+1 == f->curr_token.pos.line || end_line < 0) {
f->lead_comment = comment;
}
GB_ASSERT(f->curr_token.kind != Token_Comment);
}
}
gb_internal gb_inline bool ignore_newlines(AstFile *f) {
return f->expr_level > 0;
}
gb_internal Token advance_token(AstFile *f) {
f->lead_comment = nullptr;
f->line_comment = nullptr;
f->prev_token_index = f->curr_token_index;
Token prev = f->prev_token = f->curr_token;
bool ok = next_token0(f);
if (ok) {
switch (f->curr_token.kind) {
case Token_Comment:
consume_comment_groups(f, prev);
break;
case Token_Semicolon:
if (ignore_newlines(f) && f->curr_token.string == "\n") {
advance_token(f);
}
break;
}
}
return prev;
}
gb_internal Token peek_token(AstFile *f) {
for (isize i = f->curr_token_index+1; i < f->tokens.count; i++) {
Token tok = f->tokens[i];
if (tok.kind == Token_Comment) {
continue;
}
return tok;
}
return {};
}
gb_internal Token peek_token_n(AstFile *f, isize n) {
Token found = {};
for (isize i = f->curr_token_index+1; i < f->tokens.count; i++) {
Token tok = f->tokens[i];
if (tok.kind == Token_Comment) {
continue;
}
found = tok;
if (n-- == 0) {
return found;
}
}
return {};
}
gb_internal bool skip_possible_newline(AstFile *f) {
if (token_is_newline(f->curr_token)) {
advance_token(f);
return true;
}
return false;
}
gb_internal bool skip_possible_newline_for_literal(AstFile *f, bool ignore_strict_style=false) {
Token curr = f->curr_token;
if (token_is_newline(curr)) {
Token next = peek_token(f);
if (curr.pos.line+1 >= next.pos.line) {
switch (next.kind) {
case Token_OpenBrace:
case Token_else:
if (build_context.strict_style && !ignore_strict_style) {
syntax_error(next, "With '-strict-style' the attached brace style (1TBS) is enforced");
}
/*fallthrough*/
case Token_where:
advance_token(f);
return true;
}
}
}
return false;
}
gb_internal String token_to_string(Token const &tok) {
String p = token_strings[tok.kind];
if (token_is_newline(tok)) {
p = str_lit("newline");
}
return p;
}
gb_internal Token expect_token(AstFile *f, TokenKind kind) {
Token prev = f->curr_token;
if (prev.kind != kind) {
String c = token_strings[kind];
String p = token_to_string(prev);
begin_error_block();
syntax_error(f->curr_token, "Expected '%.*s', got '%.*s'", LIT(c), LIT(p));
if (kind == Token_Ident) switch (prev.kind) {
case Token_context:
error_line("\tSuggestion: '%.*s' is a keyword, would 'ctx' suffice?\n", LIT(prev.string));
break;
case Token_package:
error_line("\tSuggestion: '%.*s' is a keyword, would 'pkg' suffice?\n", LIT(prev.string));
break;
default:
if (token_is_keyword(prev.kind)) {
error_line("\tNote: '%.*s' is a keyword\n", LIT(prev.string));
}
break;
}
end_error_block();
if (prev.kind == Token_EOF) {
exit_with_errors();
}
}
advance_token(f);
return prev;
}
gb_internal Token expect_token_after(AstFile *f, TokenKind kind, char const *msg) {
Token prev = f->prev_token;
Token curr = f->curr_token;
if (curr.kind != kind) {
String p = token_to_string(curr);
Token token = f->curr_token;
if (token_is_newline(curr)) {
token = curr;
token.pos.column -= 1;
skip_possible_newline(f);
}
syntax_error(token, "Expected '%.*s' after %s, got '%.*s'",
LIT(token_strings[kind]),
msg,
LIT(p));
}
advance_token(f);
if (ast_file_vet_style(f) &&
prev.kind == Token_Comma &&
prev.pos.line == curr.pos.line) {
syntax_error(prev, "No need for a trailing comma followed by a %.*s on the same line", LIT(token_strings[kind]));
}
return curr;
}
gb_internal bool is_token_range(TokenKind kind) {
switch (kind) {
case Token_Ellipsis:
case Token_RangeFull:
case Token_RangeHalf:
return true;
}
return false;
}
gb_internal bool is_token_range(Token tok) {
return is_token_range(tok.kind);
}
gb_internal Token expect_operator(AstFile *f) {
Token prev = f->curr_token;
if ((prev.kind == Token_in || prev.kind == Token_not_in) && (f->expr_level >= 0 || f->allow_in_expr)) {
// okay
} else if (prev.kind == Token_if || prev.kind == Token_when) {
// okay
} else if (prev.kind == Token_or_else || prev.kind == Token_or_return ||
prev.kind == Token_or_break || prev.kind == Token_or_continue) {
// okay
} else if (!gb_is_between(prev.kind, Token__OperatorBegin+1, Token__OperatorEnd-1)) {
String p = token_to_string(prev);
syntax_error(prev, "Expected an operator, got '%.*s'",
LIT(p));
} else if (!f->allow_range && is_token_range(prev)) {
String p = token_to_string(prev);
syntax_error(prev, "Expected an non-range operator, got '%.*s'",
LIT(p));
}
if (prev.kind == Token_Ellipsis) {
syntax_error(prev, "'..' for ranges are not allowed, did you mean '..<' or '..='?");
f->tokens[f->curr_token_index].flags |= TokenFlag_Replace;
}
advance_token(f);
return prev;
}
gb_internal bool allow_token(AstFile *f, TokenKind kind) {
Token prev = f->curr_token;
if (prev.kind == kind) {
advance_token(f);
return true;
}
return false;
}
gb_internal Token expect_closing_brace_of_field_list(AstFile *f) {
Token token = f->curr_token;
if (allow_token(f, Token_CloseBrace)) {
return token;
}
bool ok = true;
if (f->allow_newline) {
ok = !skip_possible_newline(f);
}
if (ok && allow_token(f, Token_Semicolon)) {
String p = token_to_string(token);
syntax_error(token_end_of_line(f, f->prev_token), "Expected a comma, got a %.*s", LIT(p));
}
return expect_token(f, Token_CloseBrace);
}
gb_internal bool is_blank_ident(String str) {
if (str.len == 1) {
return str[0] == '_';
}
return false;
}
gb_internal bool is_blank_ident(Token token) {
if (token.kind == Token_Ident) {
return is_blank_ident(token.string);
}
return false;
}
gb_internal bool is_blank_ident(Ast *node) {
if (node->kind == Ast_Ident) {
ast_node(i, Ident, node);
return is_blank_ident(i->token.string);
}
return false;
}
// NOTE(bill): Go to next statement to prevent numerous error messages popping up
gb_internal void fix_advance_to_next_stmt(AstFile *f) {
for (;;) {
Token t = f->curr_token;
switch (t.kind) {
case Token_EOF:
case Token_Semicolon:
return;
case Token_package:
case Token_foreign:
case Token_import:
case Token_if:
case Token_for:
case Token_when:
case Token_return:
case Token_switch:
case Token_defer:
case Token_using:
case Token_break:
case Token_continue:
case Token_fallthrough:
case Token_Hash:
{
if (t.pos == f->fix_prev_pos &&
f->fix_count < PARSER_MAX_FIX_COUNT) {
f->fix_count++;
return;
}
if (f->fix_prev_pos < t.pos) {
f->fix_prev_pos = t.pos;
f->fix_count = 0; // NOTE(bill): Reset
return;
}
// NOTE(bill): Reaching here means there is a parsing bug
} break;
}
advance_token(f);
}
}
gb_internal Token expect_closing(AstFile *f, TokenKind kind, String const &context) {
if (f->curr_token.kind != kind &&
f->curr_token.kind == Token_Semicolon &&
(f->curr_token.string == "\n" || f->curr_token.kind == Token_EOF)) {
if (f->allow_newline) {
Token tok = f->prev_token;
tok.pos.column += cast(i32)tok.string.len;
syntax_error(tok, "Missing ',' before newline in %.*s", LIT(context));
}
advance_token(f);
}
return expect_token(f, kind);
}
gb_internal void assign_removal_flag_to_semicolon(AstFile *f) {
// NOTE(bill): this is used for rewriting files to strip unneeded semicolons
Token *prev_token = &f->tokens[f->prev_token_index];
Token *curr_token = &f->tokens[f->curr_token_index];
GB_ASSERT(prev_token->kind == Token_Semicolon);
if (prev_token->string != ";") {
return;
}
bool ok = false;
if (curr_token->pos.line > prev_token->pos.line) {
ok = true;
} else if (curr_token->pos.line == prev_token->pos.line) {
switch (curr_token->kind) {
case Token_CloseBrace:
case Token_CloseParen:
case Token_EOF:
ok = true;
break;
}
}
if (!ok) {
return;
}
if (build_context.strict_style || (ast_file_vet_flags(f) & VetFlag_Semicolon)) {
syntax_error(*prev_token, "Found unneeded semicolon");
}
prev_token->flags |= TokenFlag_Remove;
}
gb_internal void expect_semicolon(AstFile *f) {
Token prev_token = {};
if (allow_token(f, Token_Semicolon)) {
assign_removal_flag_to_semicolon(f);
return;
}
switch (f->curr_token.kind) {
case Token_CloseBrace:
case Token_CloseParen:
if (f->curr_token.pos.line == f->prev_token.pos.line) {
return;
}
break;
}
prev_token = f->prev_token;
if (prev_token.kind == Token_Semicolon) {
assign_removal_flag_to_semicolon(f);
return;
}
if (f->curr_token.kind == Token_EOF) {
return;
}
switch (f->curr_token.kind) {
case Token_EOF:
return;
}
if (f->curr_token.pos.line == f->prev_token.pos.line) {
String p = token_to_string(f->curr_token);
prev_token.pos = token_pos_end(prev_token);
syntax_error(prev_token, "Expected ';', got %.*s", LIT(p));
fix_advance_to_next_stmt(f);
}
}
gb_internal Ast * parse_expr(AstFile *f, bool lhs);
gb_internal Ast * parse_proc_type(AstFile *f, Token proc_token);
gb_internal Array<Ast *> parse_stmt_list(AstFile *f);
gb_internal Ast * parse_stmt(AstFile *f);
gb_internal Ast * parse_body(AstFile *f);
gb_internal Ast * parse_do_body(AstFile *f, Token const &token, char const *msg);
gb_internal Ast * parse_block_stmt(AstFile *f, b32 is_when);
gb_internal Ast *parse_ident(AstFile *f, bool allow_poly_names=false) {
Token token = f->curr_token;
if (token.kind == Token_Ident) {
advance_token(f);
} else if (allow_poly_names && token.kind == Token_Dollar) {
Token dollar = expect_token(f, Token_Dollar);
Ast *name = ast_ident(f, expect_token(f, Token_Ident));
if (is_blank_ident(name)) {
syntax_error(name, "Invalid polymorphic type definition with a blank identifier");
}
return ast_poly_type(f, dollar, name, nullptr);
} else {
token.string = str_lit("_");
expect_token(f, Token_Ident);
}
return ast_ident(f, token);
}
gb_internal Ast *parse_tag_expr(AstFile *f, Ast *expression) {
Token token = expect_token(f, Token_Hash);
Token name = expect_token(f, Token_Ident);
return ast_tag_expr(f, token, name, expression);
}
gb_internal Ast *unparen_expr(Ast *node) {
for (;;) {
if (node == nullptr) {
return nullptr;
}
if (node->kind != Ast_ParenExpr) {
return node;
}
node = node->ParenExpr.expr;
}
}
gb_internal Ast *unselector_expr(Ast *node) {
node = unparen_expr(node);
if (node == nullptr) {
return nullptr;
}
while (node->kind == Ast_SelectorExpr) {
node = node->SelectorExpr.selector;
}
return node;
}
gb_internal Ast *strip_or_return_expr(Ast *node) {
for (;;) {
if (node == nullptr) {
return node;
}
if (node->kind == Ast_OrReturnExpr) {
node = node->OrReturnExpr.expr;
} else if (node->kind == Ast_OrBranchExpr) {
node = node->OrBranchExpr.expr;
} else if (node->kind == Ast_ParenExpr) {
node = node->ParenExpr.expr;
} else {
return node;
}
}
}
gb_internal Ast *parse_value(AstFile *f);
gb_internal Array<Ast *> parse_element_list(AstFile *f) {
auto elems = array_make<Ast *>(ast_allocator(f));
while (f->curr_token.kind != Token_CloseBrace &&
f->curr_token.kind != Token_EOF) {
Ast *elem = parse_value(f);
if (f->curr_token.kind == Token_Eq) {
Token eq = expect_token(f, Token_Eq);
Ast *value = parse_value(f);
elem = ast_field_value(f, elem, value, eq);
}
array_add(&elems, elem);
if (!allow_field_separator(f)) {
break;
}
}
return elems;
}
gb_internal CommentGroup *consume_line_comment(AstFile *f) {
CommentGroup *comment = f->line_comment;
if (f->line_comment == f->lead_comment) {
f->lead_comment = nullptr;
}
f->line_comment = nullptr;
return comment;
}
gb_internal Array<Ast *> parse_enum_field_list(AstFile *f) {
auto elems = array_make<Ast *>(ast_allocator(f));
while (f->curr_token.kind != Token_CloseBrace &&
f->curr_token.kind != Token_EOF) {
CommentGroup *docs = f->lead_comment;
CommentGroup *comment = nullptr;
Ast *name = parse_value(f);
Ast *value = nullptr;
if (f->curr_token.kind == Token_Eq) {
Token eq = expect_token(f, Token_Eq);
value = parse_value(f);
}
comment = consume_line_comment(f);
Ast *elem = ast_enum_field_value(f, name, value, docs, comment);
array_add(&elems, elem);
if (!allow_field_separator(f)) {
break;
}
if (!elem->EnumFieldValue.comment) {
elem->EnumFieldValue.comment = consume_line_comment(f);
}
}
return elems;
}
gb_internal Ast *parse_literal_value(AstFile *f, Ast *type) {
Array<Ast *> elems = {};
Token open = expect_token(f, Token_OpenBrace);
isize expr_level = f->expr_level;
f->expr_level = 0;
if (f->curr_token.kind != Token_CloseBrace) {
elems = parse_element_list(f);
}
f->expr_level = expr_level;
Token close = expect_closing(f, Token_CloseBrace, str_lit("compound literal"));
return ast_compound_lit(f, type, elems, open, close);
}
gb_internal Ast *parse_value(AstFile *f) {
if (f->curr_token.kind == Token_OpenBrace) {
return parse_literal_value(f, nullptr);
}
Ast *value;
bool prev_allow_range = f->allow_range;
f->allow_range = true;
value = parse_expr(f, false);
f->allow_range = prev_allow_range;
return value;
}
gb_internal Ast *parse_type_or_ident(AstFile *f);
gb_internal void check_proc_add_tag(AstFile *f, Ast *tag_expr, u64 *tags, ProcTag tag, String const &tag_name) {
if (*tags & tag) {
syntax_error(tag_expr, "Procedure tag already used: %.*s", LIT(tag_name));
}
*tags |= tag;
}
gb_internal bool is_foreign_name_valid(String const &name) {
if (name.len == 0) {
return false;
}
isize offset = 0;
while (offset < name.len) {
Rune rune;
isize remaining = name.len - offset;
isize width = utf8_decode(name.text+offset, remaining, &rune);
if (rune == GB_RUNE_INVALID && width == 1) {
return false;
} else if (rune == GB_RUNE_BOM && remaining > 0) {
return false;
}
if (offset == 0) {
switch (rune) {
case '-':
case '$':
case '.':
case '_':
break;
default:
if (!gb_char_is_alpha(cast(char)rune))
return false;
break;
}
} else {
switch (rune) {
case '-':
case '$':
case '.':
case '_':
break;
default:
if (!gb_char_is_alphanumeric(cast(char)rune)) {
return false;
}
break;
}
}
offset += width;
}
return true;
}
gb_internal void parse_proc_tags(AstFile *f, u64 *tags) {
GB_ASSERT(tags != nullptr);
while (f->curr_token.kind == Token_Hash) {
Ast *tag_expr = parse_tag_expr(f, nullptr);
ast_node(te, TagExpr, tag_expr);
String tag_name = te->name.string;
#define ELSE_IF_ADD_TAG(name) \
else if (tag_name == #name) { \
check_proc_add_tag(f, tag_expr, tags, ProcTag_##name, tag_name); \
}
if (false) {}
ELSE_IF_ADD_TAG(optional_ok)
ELSE_IF_ADD_TAG(optional_allocator_error)
ELSE_IF_ADD_TAG(require_results)
ELSE_IF_ADD_TAG(bounds_check)
ELSE_IF_ADD_TAG(no_bounds_check)
ELSE_IF_ADD_TAG(type_assert)
ELSE_IF_ADD_TAG(no_type_assert)
else {
syntax_error(tag_expr, "Unknown procedure type tag #%.*s", LIT(tag_name));
}
#undef ELSE_IF_ADD_TAG
}
if ((*tags & ProcTag_bounds_check) && (*tags & ProcTag_no_bounds_check)) {
syntax_error(f->curr_token, "You cannot apply both #bounds_check and #no_bounds_check to a procedure");
}
if ((*tags & ProcTag_type_assert) && (*tags & ProcTag_no_type_assert)) {
syntax_error(f->curr_token, "You cannot apply both #type_assert and #no_type_assert to a procedure");
}
}
gb_internal Array<Ast *> parse_lhs_expr_list (AstFile *f);
gb_internal Array<Ast *> parse_rhs_expr_list (AstFile *f);
gb_internal Ast * parse_simple_stmt (AstFile *f, u32 flags);
gb_internal Ast * parse_type (AstFile *f);
gb_internal Ast * parse_call_expr (AstFile *f, Ast *operand);
gb_internal Ast * parse_struct_field_list(AstFile *f, isize *name_count_);
gb_internal Ast *parse_field_list(AstFile *f, isize *name_count_, u32 allowed_flags, TokenKind follow, bool allow_default_parameters, bool allow_typeid_token);
gb_internal Ast *parse_unary_expr(AstFile *f, bool lhs);
gb_internal Ast *convert_stmt_to_expr(AstFile *f, Ast *statement, String const &kind) {
if (statement == nullptr) {
return nullptr;
}
if (statement->kind == Ast_ExprStmt) {
return statement->ExprStmt.expr;
}
syntax_error(f->curr_token, "Expected '%.*s', found a simple statement.", LIT(kind));
Token end = f->curr_token;
if (f->tokens.count < f->curr_token_index) {
end = f->tokens[f->curr_token_index+1];
}
return ast_bad_expr(f, f->curr_token, end);
}
gb_internal Ast *convert_stmt_to_body(AstFile *f, Ast *stmt) {
if (stmt->kind == Ast_BlockStmt) {
syntax_error(stmt, "Expected a normal statement rather than a block statement");
return stmt;
}
if (stmt->kind == Ast_EmptyStmt) {
syntax_error(stmt, "Expected a non-empty statement");
}
GB_ASSERT(is_ast_stmt(stmt) || is_ast_decl(stmt));
Token open = ast_token(stmt);
Token close = ast_token(stmt);
auto stmts = array_make<Ast *>(ast_allocator(f), 0, 1);
array_add(&stmts, stmt);
return ast_block_stmt(f, stmts, open, close);
}
gb_internal void check_polymorphic_params_for_type(AstFile *f, Ast *polymorphic_params, Token token) {
if (polymorphic_params == nullptr) {
return;
}
if (polymorphic_params->kind != Ast_FieldList) {
return;
}
ast_node(fl, FieldList, polymorphic_params);
for (Ast *field : fl->list) {
if (field->kind != Ast_Field) {
continue;
}
for (Ast *name : field->Field.names) {
if (name->kind != field->Field.names[0]->kind) {
syntax_error(name, "Mixture of polymorphic names using both $ and not for %.*s parameters", LIT(token.string));
return;
}
}
}
}
gb_internal bool ast_on_same_line(Token const &x, Ast *yp) {
Token y = ast_token(yp);
return x.pos.line == y.pos.line;
}
gb_internal Ast *parse_force_inlining_operand(AstFile *f, Token token) {
Ast *expr = parse_unary_expr(f, false);
Ast *e = strip_or_return_expr(expr);
if (e == nullptr) {
return expr;
}
if (e->kind != Ast_ProcLit && e->kind != Ast_CallExpr) {
syntax_error(expr, "%.*s must be followed by a procedure literal or call, got %.*s", LIT(token.string), LIT(ast_strings[expr->kind]));
return ast_bad_expr(f, token, f->curr_token);
}
ProcInlining pi = ProcInlining_none;
if (token.kind == Token_Ident) {
if (token.string == "force_inline") {
pi = ProcInlining_inline;
} else if (token.string == "force_no_inline") {
pi = ProcInlining_no_inline;
}
}
if (pi != ProcInlining_none) {
if (e->kind == Ast_ProcLit) {
if (expr->ProcLit.inlining != ProcInlining_none &&
expr->ProcLit.inlining != pi) {
syntax_error(expr, "Cannot apply both '#force_inline' and '#force_no_inline' to a procedure literal");
}
expr->ProcLit.inlining = pi;
} else if (e->kind == Ast_CallExpr) {
if (expr->CallExpr.inlining != ProcInlining_none &&
expr->CallExpr.inlining != pi) {
syntax_error(expr, "Cannot apply both '#force_inline' and '#force_no_inline' to a procedure call");
}
expr->CallExpr.inlining = pi;
}
}
return expr;
}
gb_internal Ast *parse_check_directive_for_statement(Ast *s, Token const &tag_token, u16 state_flag) {
String name = tag_token.string;
if (s == nullptr) {
syntax_error(tag_token, "Invalid operand for #%.*s", LIT(name));
return nullptr;
}
if (s != nullptr && s->kind == Ast_EmptyStmt) {
if (s->EmptyStmt.token.string == "\n") {
syntax_error(tag_token, "#%.*s cannot be followed by a newline", LIT(name));
} else {
syntax_error(tag_token, "#%.*s cannot be applied to an empty statement ';'", LIT(name));
}
}
if (s->state_flags & state_flag) {
syntax_error(tag_token, "#%.*s has been applied multiple times", LIT(name));
}
s->state_flags |= state_flag;
switch (state_flag) {
case StateFlag_bounds_check:
if ((s->state_flags & StateFlag_no_bounds_check) != 0) {
syntax_error(tag_token, "#bounds_check and #no_bounds_check cannot be applied together");
}
break;
case StateFlag_no_bounds_check:
if ((s->state_flags & StateFlag_bounds_check) != 0) {
syntax_error(tag_token, "#bounds_check and #no_bounds_check cannot be applied together");
}
break;
case StateFlag_type_assert:
if ((s->state_flags & StateFlag_no_type_assert) != 0) {
syntax_error(tag_token, "#type_assert and #no_type_assert cannot be applied together");
}
break;
case StateFlag_no_type_assert:
if ((s->state_flags & StateFlag_type_assert) != 0) {
syntax_error(tag_token, "#type_assert and #no_type_assert cannot be applied together");
}
break;
}
switch (state_flag) {
case StateFlag_bounds_check:
case StateFlag_no_bounds_check:
case StateFlag_type_assert:
case StateFlag_no_type_assert:
switch (s->kind) {
case Ast_BlockStmt:
case Ast_IfStmt:
case Ast_WhenStmt:
case Ast_ForStmt:
case Ast_RangeStmt:
case Ast_UnrollRangeStmt:
case Ast_SwitchStmt:
case Ast_TypeSwitchStmt:
case Ast_ReturnStmt:
case Ast_DeferStmt:
case Ast_AssignStmt:
// Okay
break;
case Ast_ValueDecl:
if (!s->ValueDecl.is_mutable) {
syntax_error(tag_token, "#%.*s may only be applied to a variable declaration, and not a constant value declaration", LIT(name));
}
break;
default:
syntax_error(tag_token, "#%.*s may only be applied to the following statements: '{}', 'if', 'when', 'for', 'switch', 'return', 'defer', assignment, variable declaration", LIT(name));
break;
}
break;
}
return s;
}
gb_internal Array<Ast *> parse_union_variant_list(AstFile *f) {
auto variants = array_make<Ast *>(ast_allocator(f));
while (f->curr_token.kind != Token_CloseBrace &&
f->curr_token.kind != Token_EOF) {
Ast *type = parse_type(f);
if (type->kind != Ast_BadExpr) {
array_add(&variants, type);
}
if (!allow_field_separator(f)) {
break;
}
}
return variants;
}
gb_internal void parser_check_polymorphic_record_parameters(AstFile *f, Ast *polymorphic_params) {
if (polymorphic_params == nullptr) {
return;
}
if (polymorphic_params->kind != Ast_FieldList) {
return;
}
enum {Unknown, Dollar, Bare} prefix = Unknown;
gb_unused(prefix);
for (Ast *field : polymorphic_params->FieldList.list) {
if (field == nullptr || field->kind != Ast_Field) {
continue;
}
for (Ast *name : field->Field.names) {
if (name == nullptr) {
continue;
}
bool error = false;
if (name->kind == Ast_Ident) {
switch (prefix) {
case Unknown: prefix = Bare; break;
case Dollar: error = true; break;
case Bare: break;
}
} else if (name->kind == Ast_PolyType) {
switch (prefix) {
case Unknown: prefix = Dollar; break;
case Dollar: break;
case Bare: error = true; break;
}
}
if (error) {
syntax_error(name, "Mixture of polymorphic $ names and normal identifiers are not allowed within record parameters");
}
}
}
}
gb_internal Ast *parse_operand(AstFile *f, bool lhs) {
Ast *operand = nullptr; // Operand
switch (f->curr_token.kind) {
case Token_Ident:
return parse_ident(f);
case Token_Uninit:
return ast_uninit(f, expect_token(f, Token_Uninit));
case Token_context:
return ast_implicit(f, expect_token(f, Token_context));
case Token_Integer:
case Token_Float:
case Token_Imag:
case Token_Rune:
return ast_basic_lit(f, advance_token(f));
case Token_String:
return ast_basic_lit(f, advance_token(f));
case Token_OpenBrace:
if (!lhs) return parse_literal_value(f, nullptr);
break;
case Token_OpenParen: {
bool allow_newline;
isize prev_expr_level;
Token open, close;
// NOTE(bill): Skip the Paren Expression
open = expect_token(f, Token_OpenParen);
if (f->prev_token.kind == Token_CloseParen) {
close = expect_token(f, Token_CloseParen);
syntax_error(open, "Invalid parentheses expression with no inside expression");
return ast_bad_expr(f, open, close);
}
prev_expr_level = f->expr_level;
allow_newline = f->allow_newline;
if (f->expr_level < 0) {
f->allow_newline = false;
}
// NOTE(bill): enforce it to >0
f->expr_level = gb_max(f->expr_level, 0)+1;
operand = parse_expr(f, false);
f->allow_newline = allow_newline;
f->expr_level = prev_expr_level;
close = expect_token(f, Token_CloseParen);
return ast_paren_expr(f, operand, open, close);
}
case Token_distinct: {
Token token = expect_token(f, Token_distinct);
Ast *type = parse_type(f);
return ast_distinct_type(f, token, type);
}
case Token_Hash: {
Token token = expect_token(f, Token_Hash);
Token name = expect_token(f, Token_Ident);
if (name.string == "type") {
return ast_helper_type(f, token, parse_type(f));
} else if ( name.string == "simd") {
Ast *tag = ast_basic_directive(f, token, name);
Ast *original_type = parse_type(f);
Ast *type = unparen_expr(original_type);
switch (type->kind) {
case Ast_ArrayType: type->ArrayType.tag = tag; break;
default:
syntax_error(type, "Expected a fixed array type after #%.*s, got %.*s", LIT(name.string), LIT(ast_strings[type->kind]));
break;
}
return original_type;
} else if (name.string == "soa") {
Ast *tag = ast_basic_directive(f, token, name);
Ast *original_type = parse_type(f);
Ast *type = unparen_expr(original_type);
switch (type->kind) {
case Ast_ArrayType: type->ArrayType.tag = tag; break;
case Ast_DynamicArrayType: type->DynamicArrayType.tag = tag; break;
case Ast_PointerType: type->PointerType.tag = tag; break;
default:
syntax_error(type, "Expected an array or pointer type after #%.*s, got %.*s", LIT(name.string), LIT(ast_strings[type->kind]));
break;
}
return original_type;
} else if (name.string == "row_major" ||
name.string == "column_major") {
Ast *original_type = parse_type(f);
Ast *type = unparen_expr(original_type);
switch (type->kind) {
case Ast_MatrixType:
type->MatrixType.is_row_major = (name.string == "row_major");
break;
default:
syntax_error(type, "Expected a matrix type after #%.*s, got %.*s", LIT(name.string), LIT(ast_strings[type->kind]));
break;
}
return original_type;
} else if (name.string == "partial") {
Ast *tag = ast_basic_directive(f, token, name);
Ast *original_expr = parse_expr(f, lhs);
Ast *expr = unparen_expr(original_expr);
if (expr == nullptr) {
syntax_error(name, "Expected a compound literal after #%.*s", LIT(name.string));
return ast_bad_expr(f, token, name);
}
switch (expr->kind) {
case Ast_CompoundLit:
expr->CompoundLit.tag = tag;
break;
default:
syntax_error(expr, "Expected a compound literal after #%.*s, got %.*s", LIT(name.string), LIT(ast_strings[expr->kind]));
break;
}
return original_expr;
} else if (name.string == "sparse") {
Ast *tag = ast_basic_directive(f, token, name);
Ast *original_type = parse_type(f);
Ast *type = unparen_expr(original_type);
switch (type->kind) {
case Ast_ArrayType: type->ArrayType.tag = tag; break;
default:
syntax_error(type, "Expected an enumerated array type after #%.*s, got %.*s", LIT(name.string), LIT(ast_strings[type->kind]));
break;
}
return original_type;
} else if (name.string == "bounds_check") {
Ast *operand = parse_expr(f, lhs);
return parse_check_directive_for_statement(operand, name, StateFlag_bounds_check);
} else if (name.string == "no_bounds_check") {
Ast *operand = parse_expr(f, lhs);
return parse_check_directive_for_statement(operand, name, StateFlag_no_bounds_check);
} else if (name.string == "type_assert") {
Ast *operand = parse_expr(f, lhs);
return parse_check_directive_for_statement(operand, name, StateFlag_type_assert);
} else if (name.string == "no_type_assert") {
Ast *operand = parse_expr(f, lhs);
return parse_check_directive_for_statement(operand, name, StateFlag_no_type_assert);
} else if (name.string == "relative") {
Ast *tag = ast_basic_directive(f, token, name);
if (f->curr_token.kind != Token_OpenParen) {
syntax_error(tag, "expected #relative(<integer type>) <type>");
} else {
tag = parse_call_expr(f, tag);
}
Ast *type = parse_type(f);
return ast_relative_type(f, tag, type);
} else if (name.string == "force_inline" ||
name.string == "force_no_inline") {
return parse_force_inlining_operand(f, name);
}
return ast_basic_directive(f, token, name);
}
// Parse Procedure Type or Literal or Group
case Token_proc: {
Token token = expect_token(f, Token_proc);
if (f->curr_token.kind == Token_OpenBrace) { // ProcGroup
Token open = expect_token(f, Token_OpenBrace);
auto args = array_make<Ast *>(ast_allocator(f));
while (f->curr_token.kind != Token_CloseBrace &&
f->curr_token.kind != Token_EOF) {
Ast *elem = parse_expr(f, false);
array_add(&args, elem);
if (!allow_field_separator(f)) {
break;
}
}
Token close = expect_token(f, Token_CloseBrace);
if (args.count == 0) {
syntax_error(token, "Expected a least 1 argument in a procedure group");
}
return ast_proc_group(f, token, open, close, args);
}
Ast *type = parse_proc_type(f, token);
Token where_token = {};
Array<Ast *> where_clauses = {};
u64 tags = 0;
skip_possible_newline_for_literal(f);
if (f->curr_token.kind == Token_where) {
where_token = expect_token(f, Token_where);
isize prev_level = f->expr_level;
f->expr_level = -1;
where_clauses = parse_rhs_expr_list(f);
f->expr_level = prev_level;
}
parse_proc_tags(f, &tags);
if ((tags & ProcTag_require_results) != 0) {
syntax_error(f->curr_token, "#require_results has now been replaced as an attribute @(require_results) on the declaration");
tags &= ~ProcTag_require_results;
}
GB_ASSERT(type->kind == Ast_ProcType);
type->ProcType.tags = tags;
if (f->allow_type && f->expr_level < 0) {
if (tags != 0) {
syntax_error(token, "A procedure type cannot have suffix tags");
}
if (where_token.kind != Token_Invalid) {
syntax_error(where_token, "'where' clauses are not allowed on procedure types");
}
return type;
}
skip_possible_newline_for_literal(f, where_token.kind == Token_where);
if (allow_token(f, Token_Uninit)) {
if (where_token.kind != Token_Invalid) {
syntax_error(where_token, "'where' clauses are not allowed on procedure literals without a defined body (replaced with ---)");
}
return ast_proc_lit(f, type, nullptr, tags, where_token, where_clauses);
} else if (f->curr_token.kind == Token_OpenBrace) {
Ast *curr_proc = f->curr_proc;
Ast *body = nullptr;
f->curr_proc = type;
body = parse_body(f);
f->curr_proc = curr_proc;
// Apply the tags directly to the body rather than the type
if (tags & ProcTag_no_bounds_check) {
body->state_flags |= StateFlag_no_bounds_check;
}
if (tags & ProcTag_bounds_check) {
body->state_flags |= StateFlag_bounds_check;
}
if (tags & ProcTag_no_type_assert) {
body->state_flags |= StateFlag_no_type_assert;
}
if (tags & ProcTag_type_assert) {
body->state_flags |= StateFlag_type_assert;
}
return ast_proc_lit(f, type, body, tags, where_token, where_clauses);
} else if (allow_token(f, Token_do)) {
Ast *curr_proc = f->curr_proc;
Ast *body = nullptr;
f->curr_proc = type;
body = convert_stmt_to_body(f, parse_stmt(f));
f->curr_proc = curr_proc;
syntax_error(body, "'do' for procedure bodies is not allowed, prefer {}");
return ast_proc_lit(f, type, body, tags, where_token, where_clauses);
}
if (tags != 0) {
syntax_error(token, "A procedure type cannot have suffix tags");
}
if (where_token.kind != Token_Invalid) {
syntax_error(where_token, "'where' clauses are not allowed on procedure types");
}
return type;
}
// Check for Types
case Token_Dollar: {
Token token = expect_token(f, Token_Dollar);
Ast *type = parse_ident(f);
if (is_blank_ident(type)) {
syntax_error(type, "Invalid polymorphic type definition with a blank identifier");
}
Ast *specialization = nullptr;
if (allow_token(f, Token_Quo)) {
specialization = parse_type(f);
}
return ast_poly_type(f, token, type, specialization);
} break;
case Token_typeid: {
Token token = expect_token(f, Token_typeid);
return ast_typeid_type(f, token, nullptr);
} break;
case Token_Pointer: {
Token token = expect_token(f, Token_Pointer);
Ast *elem = parse_type(f);
return ast_pointer_type(f, token, elem);
} break;
case Token_Mul:
return parse_unary_expr(f, true);
case Token_OpenBracket: {
Token token = expect_token(f, Token_OpenBracket);
Ast *count_expr = nullptr;
if (f->curr_token.kind == Token_Pointer) {
expect_token(f, Token_Pointer);
expect_token(f, Token_CloseBracket);
return ast_multi_pointer_type(f, token, parse_type(f));
} else if (f->curr_token.kind == Token_Question) {
count_expr = ast_unary_expr(f, expect_token(f, Token_Question), nullptr);
} else if (allow_token(f, Token_dynamic)) {
expect_token(f, Token_CloseBracket);
return ast_dynamic_array_type(f, token, parse_type(f));
} else if (f->curr_token.kind != Token_CloseBracket) {
f->expr_level++;
count_expr = parse_expr(f, false);
f->expr_level--;
}
expect_token(f, Token_CloseBracket);
return ast_array_type(f, token, count_expr, parse_type(f));
} break;
case Token_map: {
Token token = expect_token(f, Token_map);
Ast *key = nullptr;
Ast *value = nullptr;
Token open, close;
open = expect_token_after(f, Token_OpenBracket, "map");
key = parse_expr(f, true);
close = expect_token(f, Token_CloseBracket);
value = parse_type(f);
return ast_map_type(f, token, key, value);
} break;
case Token_matrix: {
Token token = expect_token(f, Token_matrix);
Ast *row_count = nullptr;
Ast *column_count = nullptr;
Ast *type = nullptr;
Token open, close;
open = expect_token_after(f, Token_OpenBracket, "matrix");
row_count = parse_expr(f, true);
expect_token(f, Token_Comma);
column_count = parse_expr(f, true);
close = expect_token(f, Token_CloseBracket);
type = parse_type(f);
return ast_matrix_type(f, token, row_count, column_count, type);
} break;
case Token_bit_field: {
Token token = expect_token(f, Token_bit_field);
isize prev_level;
prev_level = f->expr_level;
f->expr_level = -1;
Ast *backing_type = parse_type_or_ident(f);
if (backing_type == nullptr) {
Token token = advance_token(f);
syntax_error(token, "Expected a backing type for a 'bit_field'");
backing_type = ast_bad_expr(f, token, f->curr_token);
}
skip_possible_newline_for_literal(f);
Token open = expect_token_after(f, Token_OpenBrace, "bit_field");
auto fields = array_make<Ast *>(ast_allocator(f), 0, 0);
while (f->curr_token.kind != Token_CloseBrace &&
f->curr_token.kind != Token_EOF) {
CommentGroup *docs = nullptr;
CommentGroup *comment = nullptr;
Ast *name = parse_ident(f);
bool err_once = false;
while (allow_token(f, Token_Comma)) {
Ast *dummy_name = parse_ident(f);
if (!err_once) {
error(dummy_name, "'bit_field' fields do not support multiple names per field");
err_once = true;
}
}
expect_token(f, Token_Colon);
Ast *type = parse_type(f);
expect_token(f, Token_Or);
Ast *bit_size = parse_expr(f, true);
Token tag = {};
if (f->curr_token.kind == Token_String) {
tag = expect_token(f, Token_String);
}
Ast *bf_field = ast_bit_field_field(f, name, type, bit_size, tag, docs, comment);
array_add(&fields, bf_field);
if (!allow_field_separator(f)) {
break;
}
}
Token close = expect_closing_brace_of_field_list(f);
f->expr_level = prev_level;
return ast_bit_field_type(f, token, backing_type, open, fields, close);
}
case Token_struct: {
Token token = expect_token(f, Token_struct);
Ast *polymorphic_params = nullptr;
bool is_packed = false;
bool is_raw_union = false;
bool no_copy = false;
Ast *align = nullptr;
Ast *field_align = nullptr;
if (allow_token(f, Token_OpenParen)) {
isize param_count = 0;
polymorphic_params = parse_field_list(f, &param_count, 0, Token_CloseParen, true, true);
if (param_count == 0) {
syntax_error(polymorphic_params, "Expected at least 1 polymorphic parameter");
polymorphic_params = nullptr;
}
expect_token_after(f, Token_CloseParen, "parameter list");
check_polymorphic_params_for_type(f, polymorphic_params, token);
}
isize prev_level;
prev_level = f->expr_level;
f->expr_level = -1;
while (allow_token(f, Token_Hash)) {
Token tag = expect_token_after(f, Token_Ident, "#");
if (tag.string == "packed") {
if (is_packed) {
syntax_error(tag, "Duplicate struct tag '#%.*s'", LIT(tag.string));
}
is_packed = true;
} else if (tag.string == "align") {
if (align) {
syntax_error(tag, "Duplicate struct tag '#%.*s'", LIT(tag.string));
}
align = parse_expr(f, true);
if (align && align->kind != Ast_ParenExpr) {
ERROR_BLOCK();
gbString s = expr_to_string(align);
syntax_warning(tag, "#align requires parentheses around the expression");
error_line("\tSuggestion: #align(%s)", s);
gb_string_free(s);
}
} else if (tag.string == "field_align") {
if (field_align) {
syntax_error(tag, "Duplicate struct tag '#%.*s'", LIT(tag.string));
}
field_align = parse_expr(f, true);
if (field_align && field_align->kind != Ast_ParenExpr) {
ERROR_BLOCK();
gbString s = expr_to_string(field_align);
syntax_warning(tag, "#field_align requires parentheses around the expression");
error_line("\tSuggestion: #field_align(%s)", s);
gb_string_free(s);
}
} else if (tag.string == "raw_union") {
if (is_raw_union) {
syntax_error(tag, "Duplicate struct tag '#%.*s'", LIT(tag.string));
}
is_raw_union = true;
} else if (tag.string == "no_copy") {
if (no_copy) {
syntax_error(tag, "Duplicate struct tag '#%.*s'", LIT(tag.string));
}
no_copy = true;
} else {
syntax_error(tag, "Invalid struct tag '#%.*s'", LIT(tag.string));
}
}
f->expr_level = prev_level;
if (is_raw_union && is_packed) {
is_packed = false;
syntax_error(token, "'#raw_union' cannot also be '#packed'");
}
Token where_token = {};
Array<Ast *> where_clauses = {};
skip_possible_newline_for_literal(f);
if (f->curr_token.kind == Token_where) {
where_token = expect_token(f, Token_where);
prev_level = f->expr_level;
f->expr_level = -1;
where_clauses = parse_rhs_expr_list(f);
f->expr_level = prev_level;
}
skip_possible_newline_for_literal(f);
Token open = expect_token_after(f, Token_OpenBrace, "struct");
isize name_count = 0;
Ast *fields = parse_struct_field_list(f, &name_count);
Token close = expect_closing_brace_of_field_list(f);
Slice<Ast *> decls = {};
if (fields != nullptr) {
GB_ASSERT(fields->kind == Ast_FieldList);
decls = fields->FieldList.list;
}
parser_check_polymorphic_record_parameters(f, polymorphic_params);
return ast_struct_type(f, token, decls, name_count, polymorphic_params, is_packed, is_raw_union, no_copy, align, field_align, where_token, where_clauses);
} break;
case Token_union: {
Token token = expect_token(f, Token_union);
Ast *polymorphic_params = nullptr;
Ast *align = nullptr;
bool no_nil = false;
bool maybe = false;
bool shared_nil = false;
UnionTypeKind union_kind = UnionType_Normal;
Token start_token = f->curr_token;
if (allow_token(f, Token_OpenParen)) {
isize param_count = 0;
polymorphic_params = parse_field_list(f, &param_count, 0, Token_CloseParen, true, true);
if (param_count == 0) {
syntax_error(polymorphic_params, "Expected at least 1 polymorphic parametric");
polymorphic_params = nullptr;
}
expect_token_after(f, Token_CloseParen, "parameter list");
check_polymorphic_params_for_type(f, polymorphic_params, token);
}
while (allow_token(f, Token_Hash)) {
Token tag = expect_token_after(f, Token_Ident, "#");
if (tag.string == "align") {
if (align) {
syntax_error(tag, "Duplicate union tag '#%.*s'", LIT(tag.string));
}
align = parse_expr(f, true);
if (align && align->kind != Ast_ParenExpr) {
ERROR_BLOCK();
gbString s = expr_to_string(align);
syntax_warning(tag, "#align requires parentheses around the expression");
error_line("\tSuggestion: #align(%s)", s);
gb_string_free(s);
}
} else if (tag.string == "no_nil") {
if (no_nil) {
syntax_error(tag, "Duplicate union tag '#%.*s'", LIT(tag.string));
}
no_nil = true;
} else if (tag.string == "shared_nil") {
if (shared_nil) {
syntax_error(tag, "Duplicate union tag '#%.*s'", LIT(tag.string));
}
shared_nil = true;
} else if (tag.string == "maybe") {
if (maybe) {
syntax_error(tag, "Duplicate union tag '#%.*s'", LIT(tag.string));
}
maybe = true;
}else {
syntax_error(tag, "Invalid union tag '#%.*s'", LIT(tag.string));
}
}
if (no_nil && shared_nil) {
syntax_error(f->curr_token, "#shared_nil and #no_nil cannot be applied together");
}
if (maybe) {
syntax_error(f->curr_token, "#maybe functionality has now been merged with standard 'union' functionality");
}
if (no_nil) {
union_kind = UnionType_no_nil;
} else if (shared_nil) {
union_kind = UnionType_shared_nil;
}
skip_possible_newline_for_literal(f);
Token where_token = {};
Array<Ast *> where_clauses = {};
if (f->curr_token.kind == Token_where) {
where_token = expect_token(f, Token_where);
isize prev_level = f->expr_level;
f->expr_level = -1;
where_clauses = parse_rhs_expr_list(f);
f->expr_level = prev_level;
}
skip_possible_newline_for_literal(f);
Token open = expect_token_after(f, Token_OpenBrace, "union");
auto variants = parse_union_variant_list(f);
Token close = expect_closing_brace_of_field_list(f);
parser_check_polymorphic_record_parameters(f, polymorphic_params);
return ast_union_type(f, token, variants, polymorphic_params, align, union_kind, where_token, where_clauses);
} break;
case Token_enum: {
Token token = expect_token(f, Token_enum);
Ast *base_type = nullptr;
if (f->curr_token.kind != Token_OpenBrace) {
base_type = parse_type(f);
}
skip_possible_newline_for_literal(f);
Token open = expect_token(f, Token_OpenBrace);
Array<Ast *> values = parse_enum_field_list(f);
Token close = expect_closing_brace_of_field_list(f);
return ast_enum_type(f, token, base_type, values);
} break;
case Token_bit_set: {
Token token = expect_token(f, Token_bit_set);
expect_token(f, Token_OpenBracket);
Ast *elem = nullptr;
Ast *underlying = nullptr;
bool prev_allow_range = f->allow_range;
f->allow_range = true;
elem = parse_expr(f, true);
f->allow_range = prev_allow_range;
if (elem == nullptr) {
syntax_error(token, "Expected a type or range, got nothing");
}
if (allow_token(f, Token_Semicolon)) {
underlying = parse_type(f);
} else if (allow_token(f, Token_Comma)) {
String p = token_to_string(f->prev_token);
syntax_error(token_end_of_line(f, f->prev_token), "Expected a semicolon, got a %.*s", LIT(p));
underlying = parse_type(f);
}
expect_token(f, Token_CloseBracket);
return ast_bit_set_type(f, token, elem, underlying);
}
case Token_asm: {
Token token = expect_token(f, Token_asm);
Array<Ast *> param_types = {};
Ast *return_type = nullptr;
if (allow_token(f, Token_OpenParen)) {
param_types = array_make<Ast *>(ast_allocator(f));
while (f->curr_token.kind != Token_CloseParen && f->curr_token.kind != Token_EOF) {
Ast *t = parse_type(f);
array_add(&param_types, t);
if (f->curr_token.kind != Token_Comma ||
f->curr_token.kind == Token_EOF) {
break;
}
advance_token(f);
}
expect_token(f, Token_CloseParen);
if (allow_token(f, Token_ArrowRight)) {
return_type = parse_type(f);
}
}
bool has_side_effects = false;
bool is_align_stack = false;
InlineAsmDialectKind dialect = InlineAsmDialect_Default;
while (f->curr_token.kind == Token_Hash) {
advance_token(f);
if (f->curr_token.kind == Token_Ident) {
Token token = advance_token(f);
String name = token.string;
if (name == "side_effects") {
if (has_side_effects) {
syntax_error(token, "Duplicate directive on inline asm expression: '#side_effects'");
}
has_side_effects = true;
} else if (name == "align_stack") {
if (is_align_stack) {
syntax_error(token, "Duplicate directive on inline asm expression: '#align_stack'");
}
is_align_stack = true;
} else if (name == "att") {
if (dialect == InlineAsmDialect_ATT) {
syntax_error(token, "Duplicate directive on inline asm expression: '#att'");
} else if (dialect != InlineAsmDialect_Default) {
syntax_error(token, "Conflicting asm dialects");
} else {
dialect = InlineAsmDialect_ATT;
}
} else if (name == "intel") {
if (dialect == InlineAsmDialect_Intel) {
syntax_error(token, "Duplicate directive on inline asm expression: '#intel'");
} else if (dialect != InlineAsmDialect_Default) {
syntax_error(token, "Conflicting asm dialects");
} else {
dialect = InlineAsmDialect_Intel;
}
}
} else {
syntax_error(f->curr_token, "Expected an identifier after hash");
}
}
skip_possible_newline_for_literal(f);
Token open = expect_token(f, Token_OpenBrace);
Ast *asm_string = parse_expr(f, false);
expect_token(f, Token_Comma);
Ast *constraints_string = parse_expr(f, false);
allow_token(f, Token_Comma);
Token close = expect_closing_brace_of_field_list(f);
return ast_inline_asm_expr(f, token, open, close, param_types, return_type, asm_string, constraints_string, has_side_effects, is_align_stack, dialect);
}
}
return nullptr;
}
gb_internal bool is_literal_type(Ast *node) {
node = unparen_expr(node);
switch (node->kind) {
case Ast_BadExpr:
case Ast_Ident:
case Ast_SelectorExpr:
case Ast_ArrayType:
case Ast_StructType:
case Ast_UnionType:
case Ast_EnumType:
case Ast_DynamicArrayType:
case Ast_MapType:
case Ast_BitSetType:
case Ast_MatrixType:
case Ast_CallExpr:
return true;
case Ast_MultiPointerType:
// For better error messages
return true;
}
return false;
}
gb_internal Ast *parse_call_expr(AstFile *f, Ast *operand) {
auto args = array_make<Ast *>(ast_allocator(f));
Token open_paren, close_paren;
Token ellipsis = {};
isize prev_expr_level = f->expr_level;
bool prev_allow_newline = f->allow_newline;
f->expr_level = 0;
f->allow_newline = file_allow_newline(f);
open_paren = expect_token(f, Token_OpenParen);
bool seen_ellipsis = false;
while (f->curr_token.kind != Token_CloseParen &&
f->curr_token.kind != Token_EOF) {
if (f->curr_token.kind == Token_Comma) {
syntax_error(f->curr_token, "Expected an expression not ,");
} else if (f->curr_token.kind == Token_Eq) {
syntax_error(f->curr_token, "Expected an expression not =");
}
bool prefix_ellipsis = false;
if (f->curr_token.kind == Token_Ellipsis) {
prefix_ellipsis = true;
ellipsis = expect_token(f, Token_Ellipsis);
}
Ast *arg = parse_expr(f, false);
if (f->curr_token.kind == Token_Eq) {
Token eq = expect_token(f, Token_Eq);
if (prefix_ellipsis) {
syntax_error(ellipsis, "'..' must be applied to value rather than the field name");
}
Ast *value = parse_value(f);
arg = ast_field_value(f, arg, value, eq);
} else if (seen_ellipsis) {
syntax_error(arg, "Positional arguments are not allowed after '..'");
}
array_add(&args, arg);
if (ellipsis.pos.line != 0) {
seen_ellipsis = true;
}
if (!allow_field_separator(f)) {
break;
}
}
f->allow_newline = prev_allow_newline;
f->expr_level = prev_expr_level;
close_paren = expect_closing(f, Token_CloseParen, str_lit("argument list"));
Ast *call = ast_call_expr(f, operand, args, open_paren, close_paren, ellipsis);
Ast *o = unparen_expr(operand);
if (o && o->kind == Ast_SelectorExpr && o->SelectorExpr.token.kind == Token_ArrowRight) {
return ast_selector_call_expr(f, o->SelectorExpr.token, o, call);
}
return call;
}
gb_internal void parse_check_or_return(Ast *operand, char const *msg) {
if (operand == nullptr) {
return;
}
switch (operand->kind) {
case Ast_OrReturnExpr:
syntax_error_with_verbose(operand, "'or_return' use within %s is not wrapped in parentheses (...)", msg);
break;
case Ast_OrBranchExpr:
syntax_error_with_verbose(operand, "'%.*s' use within %s is not wrapped in parentheses (...)", msg, LIT(operand->OrBranchExpr.token.string));
break;
}
}
gb_internal Ast *parse_atom_expr(AstFile *f, Ast *operand, bool lhs) {
if (operand == nullptr) {
if (f->allow_type) return nullptr;
Token begin = f->curr_token;
syntax_error(begin, "Expected an operand");
fix_advance_to_next_stmt(f);
operand = ast_bad_expr(f, begin, f->curr_token);
}
bool loop = true;
while (loop) {
switch (f->curr_token.kind) {
case Token_OpenParen:
parse_check_or_return(operand, "call expression");
operand = parse_call_expr(f, operand);
break;
case Token_Period: {
Token token = advance_token(f);
switch (f->curr_token.kind) {
case Token_Ident:
parse_check_or_return(operand, "selector expression");
operand = ast_selector_expr(f, token, operand, parse_ident(f));
break;
case Token_OpenParen: {
parse_check_or_return(operand, "type assertion");
Token open = expect_token(f, Token_OpenParen);
Ast *type = parse_type(f);
Token close = expect_token(f, Token_CloseParen);
operand = ast_type_assertion(f, operand, token, type);
} break;
case Token_Question: {
parse_check_or_return(operand, ".? based type assertion");
Token question = expect_token(f, Token_Question);
Ast *type = ast_unary_expr(f, question, nullptr);
operand = ast_type_assertion(f, operand, token, type);
} break;
default:
syntax_error(f->curr_token, "Expected a selector");
advance_token(f);
operand = ast_bad_expr(f, ast_token(operand), f->curr_token);
// operand = ast_selector_expr(f, f->curr_token, operand, nullptr);
break;
}
} break;
case Token_ArrowRight: {
parse_check_or_return(operand, "-> based call expression");
Token token = advance_token(f);
operand = ast_selector_expr(f, token, operand, parse_ident(f));
// Ast *call = parse_call_expr(f, sel);
// operand = ast_selector_call_expr(f, token, sel, call);
break;
}
case Token_OpenBracket: {
bool prev_allow_range = f->allow_range;
f->allow_range = false;
Token open = {}, close = {}, interval = {};
Ast *indices[2] = {};
bool is_interval = false;
f->expr_level++;
open = expect_token(f, Token_OpenBracket);
switch (f->curr_token.kind) {
case Token_Ellipsis:
case Token_RangeFull:
case Token_RangeHalf:
// NOTE(bill): Do not err yet
case Token_Colon:
break;
default:
indices[0] = parse_expr(f, false);
break;
}
switch (f->curr_token.kind) {
case Token_Ellipsis:
case Token_RangeFull:
case Token_RangeHalf:
syntax_error(f->curr_token, "Expected a colon, not a range");
/* fallthrough */
case Token_Comma: // matrix index
case Token_Colon:
interval = advance_token(f);
is_interval = true;
if (f->curr_token.kind != Token_CloseBracket &&
f->curr_token.kind != Token_EOF) {
indices[1] = parse_expr(f, false);
}
break;
}
f->expr_level--;
close = expect_token(f, Token_CloseBracket);
if (is_interval) {
if (interval.kind == Token_Comma) {
if (indices[0] == nullptr || indices[1] == nullptr) {
syntax_error(open, "Matrix index expressions require both row and column indices");
}
parse_check_or_return(operand, "matrix index expression");
operand = ast_matrix_index_expr(f, operand, open, close, interval, indices[0], indices[1]);
} else {
parse_check_or_return(operand, "slice expression");
operand = ast_slice_expr(f, operand, open, close, interval, indices[0], indices[1]);
}
} else {
parse_check_or_return(operand, "index expression");
operand = ast_index_expr(f, operand, indices[0], open, close);
}
f->allow_range = prev_allow_range;
} break;
case Token_Pointer: // Deference
parse_check_or_return(operand, "dereference");
operand = ast_deref_expr(f, operand, expect_token(f, Token_Pointer));
break;
case Token_or_return:
operand = ast_or_return_expr(f, operand, expect_token(f, Token_or_return));
break;
case Token_or_break:
case Token_or_continue:
{
Token token = advance_token(f);
Ast *label = nullptr;
if (f->curr_token.kind == Token_Ident) {
label = parse_ident(f);
}
operand = ast_or_branch_expr(f, operand, token, label);
}
break;
case Token_OpenBrace:
if (!lhs && is_literal_type(operand) && f->expr_level >= 0) {
operand = parse_literal_value(f, operand);
} else {
loop = false;
}
break;
case Token_Increment:
case Token_Decrement:
if (!lhs) {
Token token = advance_token(f);
syntax_error(token, "Postfix '%.*s' operator is not supported", LIT(token.string));
} else {
loop = false;
}
break;
default:
loop = false;
break;
}
lhs = false; // NOTE(bill): 'tis not lhs anymore
}
return operand;
}
gb_internal Ast *parse_unary_expr(AstFile *f, bool lhs) {
switch (f->curr_token.kind) {
case Token_transmute:
case Token_cast: {
Token token = advance_token(f);
expect_token(f, Token_OpenParen);
Ast *type = parse_type(f);
expect_token(f, Token_CloseParen);
Ast *expr = parse_unary_expr(f, lhs);
return ast_type_cast(f, token, type, expr);
}
case Token_auto_cast: {
Token token = advance_token(f);
Ast *expr = parse_unary_expr(f, lhs);
return ast_auto_cast(f, token, expr);
}
case Token_Add:
case Token_Sub:
case Token_Xor:
case Token_And:
case Token_Not:
case Token_Mul: // Used for error handling when people do C-like things
{
Token token = advance_token(f);
Ast *expr = parse_unary_expr(f, lhs);
return ast_unary_expr(f, token, expr);
}
case Token_Increment:
case Token_Decrement: {
Token token = advance_token(f);
syntax_error(token, "Unary '%.*s' operator is not supported", LIT(token.string));
Ast *expr = parse_unary_expr(f, lhs);
return ast_unary_expr(f, token, expr);
}
case Token_Period: {
Token token = expect_token(f, Token_Period);
Ast *ident = parse_ident(f);
return ast_implicit_selector_expr(f, token, ident);
}
}
return parse_atom_expr(f, parse_operand(f, lhs), lhs);
}
gb_internal bool is_ast_range(Ast *expr) {
if (expr == nullptr) {
return false;
}
if (expr->kind != Ast_BinaryExpr) {
return false;
}
return is_token_range(expr->BinaryExpr.op.kind);
}
// NOTE(bill): result == priority
gb_internal i32 token_precedence(AstFile *f, TokenKind t) {
switch (t) {
case Token_Question:
case Token_if:
case Token_when:
case Token_or_else:
return 1;
case Token_Ellipsis:
case Token_RangeFull:
case Token_RangeHalf:
if (!f->allow_range) {
return 0;
}
return 2;
case Token_CmpOr:
return 3;
case Token_CmpAnd:
return 4;
case Token_CmpEq:
case Token_NotEq:
case Token_Lt:
case Token_Gt:
case Token_LtEq:
case Token_GtEq:
return 5;
case Token_in:
case Token_not_in:
if (f->expr_level < 0 && !f->allow_in_expr) {
return 0;
}
/*fallthrough*/
case Token_Add:
case Token_Sub:
case Token_Or:
case Token_Xor:
return 6;
case Token_Mul:
case Token_Quo:
case Token_Mod:
case Token_ModMod:
case Token_And:
case Token_AndNot:
case Token_Shl:
case Token_Shr:
return 7;
}
return 0;
}
gb_internal Ast *parse_binary_expr(AstFile *f, bool lhs, i32 prec_in) {
Ast *expr = parse_unary_expr(f, lhs);
for (;;) {
Token op = f->curr_token;
i32 op_prec = token_precedence(f, op.kind);
if (op_prec < prec_in) {
// NOTE(bill): This will also catch operators that are not valid "binary" operators
break;
}
Token prev = f->prev_token;
switch (op.kind) {
case Token_if:
case Token_when:
if (prev.pos.line < op.pos.line) {
// NOTE(bill): Check to see if the `if` or `when` is on the same line of the `lhs` condition
goto loop_end;
}
break;
}
expect_operator(f); // NOTE(bill): error checks too
if (op.kind == Token_Question) {
Ast *cond = expr;
// Token_Question
Ast *x = parse_expr(f, lhs);
Token token_c = expect_token(f, Token_Colon);
Ast *y = parse_expr(f, lhs);
expr = ast_ternary_if_expr(f, x, cond, y);
} else if (op.kind == Token_if || op.kind == Token_when) {
Ast *x = expr;
Ast *cond = parse_expr(f, lhs);
Token tok_else = expect_token(f, Token_else);
Ast *y = parse_expr(f, lhs);
switch (op.kind) {
case Token_if:
expr = ast_ternary_if_expr(f, x, cond, y);
break;
case Token_when:
expr = ast_ternary_when_expr(f, x, cond, y);
break;
}
} else {
Ast *right = parse_binary_expr(f, false, op_prec+1);
if (right == nullptr) {
syntax_error(op, "Expected expression on the right-hand side of the binary operator '%.*s'", LIT(op.string));
}
if (op.kind == Token_or_else) {
// NOTE(bill): easier to handle its logic different with its own AST kind
expr = ast_or_else_expr(f, expr, op, right);
} else {
expr = ast_binary_expr(f, op, expr, right);
}
}
lhs = false;
}
loop_end:;
return expr;
}
gb_internal Ast *parse_expr(AstFile *f, bool lhs) {
return parse_binary_expr(f, lhs, 0+1);
}
gb_internal Array<Ast *> parse_expr_list(AstFile *f, bool lhs) {
bool allow_newline = f->allow_newline;
f->allow_newline = file_allow_newline(f);
auto list = array_make<Ast *>(ast_allocator(f));
for (;;) {
Ast *e = parse_expr(f, lhs);
array_add(&list, e);
if (f->curr_token.kind != Token_Comma ||
f->curr_token.kind == Token_EOF) {
break;
}
advance_token(f);
}
f->allow_newline = allow_newline;
return list;
}
gb_internal Array<Ast *> parse_lhs_expr_list(AstFile *f) {
return parse_expr_list(f, true);
}
gb_internal Array<Ast *> parse_rhs_expr_list(AstFile *f) {
return parse_expr_list(f, false);
}
gb_internal Array<Ast *> parse_ident_list(AstFile *f, bool allow_poly_names) {
auto list = array_make<Ast *>(ast_allocator(f));
for (;;) {
array_add(&list, parse_ident(f, allow_poly_names));
if (f->curr_token.kind != Token_Comma ||
f->curr_token.kind == Token_EOF) {
break;
}
advance_token(f);
}
return list;
}
gb_internal Ast *parse_type(AstFile *f) {
Ast *type = parse_type_or_ident(f);
if (type == nullptr) {
Token prev_token = f->curr_token;
Token token = {};
if (f->curr_token.kind == Token_OpenBrace) {
token = f->curr_token;
} else {
token = advance_token(f);
}
syntax_error(token, "Expected a type, got '%.*s'", LIT(prev_token.string));
return ast_bad_expr(f, token, f->curr_token);
} else if (type->kind == Ast_ParenExpr &&
unparen_expr(type) == nullptr) {
syntax_error(type, "Expected a type within the parentheses");
return ast_bad_expr(f, type->ParenExpr.open, type->ParenExpr.close);
}
return type;
}
gb_internal void parse_foreign_block_decl(AstFile *f, Array<Ast *> *decls) {
Ast *decl = parse_stmt(f);
switch (decl->kind) {
case Ast_EmptyStmt:
case Ast_BadStmt:
case Ast_BadDecl:
return;
case Ast_WhenStmt:
case Ast_ValueDecl:
array_add(decls, decl);
return;
default:
syntax_error(decl, "Foreign blocks only allow procedure and variable declarations");
return;
}
}
gb_internal Ast *parse_foreign_block(AstFile *f, Token token) {
CommentGroup *docs = f->lead_comment;
Ast *foreign_library = nullptr;
if (f->curr_token.kind == Token_OpenBrace) {
foreign_library = ast_ident(f, blank_token);
} else {
foreign_library = parse_ident(f);
}
Token open = {};
Token close = {};
auto decls = array_make<Ast *>(ast_allocator(f));
bool prev_in_foreign_block = f->in_foreign_block;
defer (f->in_foreign_block = prev_in_foreign_block);
f->in_foreign_block = true;
skip_possible_newline_for_literal(f);
open = expect_token(f, Token_OpenBrace);
while (f->curr_token.kind != Token_CloseBrace &&
f->curr_token.kind != Token_EOF) {
parse_foreign_block_decl(f, &decls);
}
close = expect_token(f, Token_CloseBrace);
Ast *body = ast_block_stmt(f, decls, open, close);
Ast *decl = ast_foreign_block_decl(f, token, foreign_library, body, docs);
expect_semicolon(f);
return decl;
}
gb_internal void print_comment_group(CommentGroup *group) {
if (group) {
for (Token const &token : group->list) {
gb_printf_err("%.*s\n", LIT(token.string));
}
gb_printf_err("\n");
}
}
gb_internal Ast *parse_value_decl(AstFile *f, Array<Ast *> names, CommentGroup *docs) {
bool is_mutable = true;
Array<Ast *> values = {};
Ast *type = parse_type_or_ident(f);
if (f->curr_token.kind == Token_Eq ||
f->curr_token.kind == Token_Colon) {
Token sep = {};
if (!is_mutable) {
sep = expect_token_after(f, Token_Colon, "type");
} else {
sep = advance_token(f);
is_mutable = sep.kind != Token_Colon;
}
values = parse_rhs_expr_list(f);
if (values.count > names.count) {
syntax_error(f->curr_token, "Too many values on the right hand side of the declaration");
} else if (values.count < names.count && !is_mutable) {
syntax_error(f->curr_token, "All constant declarations must be defined");
} else if (values.count == 0) {
syntax_error(f->curr_token, "Expected an expression for this declaration");
}
}
if (is_mutable) {
if (type == nullptr && values.count == 0) {
syntax_error(f->curr_token, "Missing variable type or initialization");
return ast_bad_decl(f, f->curr_token, f->curr_token);
}
} else {
if (type == nullptr && values.count == 0 && names.count > 0) {
syntax_error(f->curr_token, "Missing constant value");
return ast_bad_decl(f, f->curr_token, f->curr_token);
}
}
if (values.data == nullptr) {
values.allocator = ast_allocator(f);
}
CommentGroup *end_comment = f->lead_comment;
if (f->expr_level >= 0) {
if (f->curr_token.kind == Token_CloseBrace &&
f->curr_token.pos.line == f->prev_token.pos.line) {
} else {
expect_semicolon(f);
}
}
if (f->curr_proc == nullptr) {
if (values.count > 0 && names.count != values.count) {
syntax_error(
values[0],
"Expected %td expressions on the right hand side, got %td\n"
"\tNote: Global declarations do not allow for multi-valued expressions",
names.count, values.count
);
}
}
return ast_value_decl(f, names, type, values, is_mutable, docs, end_comment);
}
gb_internal Ast *parse_simple_stmt(AstFile *f, u32 flags) {
Token token = f->curr_token;
CommentGroup *docs = f->lead_comment;
Array<Ast *> lhs = parse_lhs_expr_list(f);
token = f->curr_token;
switch (token.kind) {
case Token_Eq:
case Token_AddEq:
case Token_SubEq:
case Token_MulEq:
case Token_QuoEq:
case Token_ModEq:
case Token_ModModEq:
case Token_AndEq:
case Token_OrEq:
case Token_XorEq:
case Token_ShlEq:
case Token_ShrEq:
case Token_AndNotEq:
case Token_CmpAndEq:
case Token_CmpOrEq:
{
if (f->curr_proc == nullptr) {
syntax_error(f->curr_token, "You cannot use a simple statement in the file scope");
return ast_bad_stmt(f, f->curr_token, f->curr_token);
}
advance_token(f);
Array<Ast *> rhs = parse_rhs_expr_list(f);
if (rhs.count == 0) {
syntax_error(token, "No right-hand side in assignment statement.");
return ast_bad_stmt(f, token, f->curr_token);
}
return ast_assign_stmt(f, token, lhs, rhs);
} break;
case Token_in:
if (flags&StmtAllowFlag_In) {
allow_token(f, Token_in);
bool prev_allow_range = f->allow_range;
f->allow_range = true;
Ast *expr = parse_expr(f, true);
f->allow_range = prev_allow_range;
auto rhs = array_make<Ast *>(ast_allocator(f), 0, 1);
array_add(&rhs, expr);
return ast_assign_stmt(f, token, lhs, rhs);
}
break;
case Token_Colon:
expect_token_after(f, Token_Colon, "identifier list");
if ((flags&StmtAllowFlag_Label) && lhs.count == 1) {
bool is_partial = false;
bool is_reverse = false;
Token partial_token = {};
if (f->curr_token.kind == Token_Hash) {
// NOTE(bill): This is purely for error messages
Token name = peek_token_n(f, 0);
if (name.kind == Token_Ident && name.string == "partial" &&
peek_token_n(f, 1).kind == Token_switch) {
partial_token = expect_token(f, Token_Hash);
expect_token(f, Token_Ident);
is_partial = true;
} else if (name.kind == Token_Ident && name.string == "reverse" &&
peek_token_n(f, 1).kind == Token_for) {
partial_token = expect_token(f, Token_Hash);
expect_token(f, Token_Ident);
is_reverse = true;
}
}
switch (f->curr_token.kind) {
case Token_OpenBrace: // block statement
case Token_if:
case Token_for:
case Token_switch: {
Ast *name = lhs[0];
Ast *label = ast_label_decl(f, ast_token(name), name);
Ast *stmt = parse_stmt(f);
#define _SET_LABEL(Kind_, label_) case GB_JOIN2(Ast_, Kind_): (stmt->Kind_).label = label_; break
switch (stmt->kind) {
_SET_LABEL(BlockStmt, label);
_SET_LABEL(IfStmt, label);
_SET_LABEL(ForStmt, label);
_SET_LABEL(RangeStmt, label);
_SET_LABEL(SwitchStmt, label);
_SET_LABEL(TypeSwitchStmt, label);
default:
syntax_error(token, "Labels can only be applied to a loop or switch statement");
break;
}
#undef _SET_LABEL
if (is_partial) {
switch (stmt->kind) {
case Ast_SwitchStmt:
stmt->SwitchStmt.partial = true;
break;
case Ast_TypeSwitchStmt:
stmt->TypeSwitchStmt.partial = true;
break;
default:
syntax_error(partial_token, "Incorrect use of directive, use '%.*s: #partial switch'", LIT(ast_token(name).string));
break;
}
} else if (is_reverse) {
switch (stmt->kind) {
case Ast_RangeStmt:
if (stmt->RangeStmt.reverse) {
syntax_error(token, "#reverse already applied to a 'for in' statement");
}
stmt->RangeStmt.reverse = true;
break;
default:
syntax_error(token, "#reverse can only be applied to a 'for in' statement");
break;
}
}
return stmt;
} break;
}
}
return parse_value_decl(f, lhs, docs);
}
if (lhs.count > 1) {
syntax_error(token, "Expected 1 expression");
return ast_bad_stmt(f, token, f->curr_token);
}
switch (token.kind) {
case Token_Increment:
case Token_Decrement:
advance_token(f);
syntax_error(token, "Postfix '%.*s' statement is not supported", LIT(token.string));
break;
}
#if 0
switch (token.kind) {
case Token_Inc:
case Token_Dec:
advance_token(f);
return ast_inc_dec_stmt(f, token, lhs[0]);
}
#endif
return ast_expr_stmt(f, lhs[0]);
}
gb_internal Ast *parse_block_stmt(AstFile *f, b32 is_when) {
skip_possible_newline_for_literal(f);
if (!is_when && f->curr_proc == nullptr) {
syntax_error(f->curr_token, "You cannot use a block statement in the file scope");
return ast_bad_stmt(f, f->curr_token, f->curr_token);
}
return parse_body(f);
}
gb_internal Ast *parse_results(AstFile *f, bool *diverging) {
if (!allow_token(f, Token_ArrowRight)) {
return nullptr;
}
if (allow_token(f, Token_Not)) {
if (diverging) *diverging = true;
return nullptr;
}
isize prev_level = f->expr_level;
defer (f->expr_level = prev_level);
if (f->curr_token.kind != Token_OpenParen) {
Token begin_token = f->curr_token;
Array<Ast *> empty_names = {};
auto list = array_make<Ast *>(ast_allocator(f), 0, 1);
Ast *type = parse_type(f);
Token tag = {};
array_add(&list, ast_field(f, empty_names, type, nullptr, 0, tag, nullptr, nullptr));
return ast_field_list(f, begin_token, list);
}
Ast *list = nullptr;
expect_token(f, Token_OpenParen);
list = parse_field_list(f, nullptr, FieldFlag_Results, Token_CloseParen, true, false);
if (file_allow_newline(f)) {
skip_possible_newline(f);
}
expect_token_after(f, Token_CloseParen, "parameter list");
return list;
}
gb_internal ProcCallingConvention string_to_calling_convention(String const &s) {
if (s == "odin") return ProcCC_Odin;
if (s == "contextless") return ProcCC_Contextless;
if (s == "cdecl") return ProcCC_CDecl;
if (s == "c") return ProcCC_CDecl;
if (s == "stdcall") return ProcCC_StdCall;
if (s == "std") return ProcCC_StdCall;
if (s == "fastcall") return ProcCC_FastCall;
if (s == "fast") return ProcCC_FastCall;
if (s == "none") return ProcCC_None;
if (s == "naked") return ProcCC_Naked;
if (s == "win64") return ProcCC_Win64;
if (s == "sysv") return ProcCC_SysV;
if (s == "system") {
if (build_context.metrics.os == TargetOs_windows) {
return ProcCC_StdCall;
}
return ProcCC_CDecl;
}
return ProcCC_Invalid;
}
gb_internal Ast *parse_proc_type(AstFile *f, Token proc_token) {
Ast *params = nullptr;
Ast *results = nullptr;
bool diverging = false;
ProcCallingConvention cc = ProcCC_Invalid;
if (f->curr_token.kind == Token_String) {
Token token = expect_token(f, Token_String);
auto c = string_to_calling_convention(string_value_from_token(f, token));
if (c == ProcCC_Invalid) {
syntax_error(token, "Unknown procedure calling convention: '%.*s'", LIT(token.string));
} else {
cc = c;
}
}
if (cc == ProcCC_Invalid) {
if (f->in_foreign_block) {
cc = ProcCC_ForeignBlockDefault;
} else {
cc = default_calling_convention();
}
}
expect_token(f, Token_OpenParen);
f->expr_level += 1;
params = parse_field_list(f, nullptr, FieldFlag_Signature, Token_CloseParen, true, true);
if (file_allow_newline(f)) {
skip_possible_newline(f);
}
f->expr_level -= 1;
expect_token_after(f, Token_CloseParen, "parameter list");
results = parse_results(f, &diverging);
u64 tags = 0;
bool is_generic = false;
for (Ast *param : params->FieldList.list) {
ast_node(field, Field, param);
if (field->type != nullptr) {
if (field->type->kind == Ast_PolyType) {
is_generic = true;
goto end;
}
for (Ast *name : field->names) {
if (name->kind == Ast_PolyType) {
is_generic = true;
goto end;
}
}
}
}
end:
return ast_proc_type(f, proc_token, params, results, tags, cc, is_generic, diverging);
}
gb_internal Ast *parse_var_type(AstFile *f, bool allow_ellipsis, bool allow_typeid_token) {
if (allow_ellipsis && f->curr_token.kind == Token_Ellipsis) {
Token tok = advance_token(f);
Ast *type = parse_type_or_ident(f);
if (type == nullptr) {
syntax_error(tok, "variadic field missing type after '..'");
type = ast_bad_expr(f, tok, f->curr_token);
}
return ast_ellipsis(f, tok, type);
}
Ast *type = nullptr;
if (allow_typeid_token &&
f->curr_token.kind == Token_typeid) {
Token token = expect_token(f, Token_typeid);
Ast *specialization = nullptr;
if (allow_token(f, Token_Quo)) {
specialization = parse_type(f);
}
type = ast_typeid_type(f, token, specialization);
} else {
type = parse_type(f);
}
return type;
}
struct ParseFieldPrefixMapping {
String name;
TokenKind token_kind;
FieldFlag flag;
};
gb_global ParseFieldPrefixMapping const parse_field_prefix_mappings[] = {
{str_lit("using"), Token_using, FieldFlag_using},
{str_lit("no_alias"), Token_Hash, FieldFlag_no_alias},
{str_lit("no_capture"), Token_Hash, FieldFlag_no_capture},
{str_lit("c_vararg"), Token_Hash, FieldFlag_c_vararg},
{str_lit("const"), Token_Hash, FieldFlag_const},
{str_lit("any_int"), Token_Hash, FieldFlag_any_int},
{str_lit("subtype"), Token_Hash, FieldFlag_subtype},
{str_lit("by_ptr"), Token_Hash, FieldFlag_by_ptr},
{str_lit("no_broadcast"), Token_Hash, FieldFlag_no_broadcast},
};
gb_internal FieldFlag is_token_field_prefix(AstFile *f) {
switch (f->curr_token.kind) {
case Token_EOF:
return FieldFlag_Invalid;
case Token_using:
return FieldFlag_using;
case Token_Hash:
advance_token(f);
switch (f->curr_token.kind) {
case Token_Ident:
for (i32 i = 0; i < gb_count_of(parse_field_prefix_mappings); i++) {
auto const &mapping = parse_field_prefix_mappings[i];
if (mapping.token_kind == Token_Hash) {
if (f->curr_token.string == mapping.name) {
return mapping.flag;
}
}
}
break;
}
return FieldFlag_Unknown;
}
return FieldFlag_Invalid;
}
gb_internal u32 parse_field_prefixes(AstFile *f) {
i32 counts[gb_count_of(parse_field_prefix_mappings)] = {};
for (;;) {
FieldFlag flag = is_token_field_prefix(f);
if (flag & FieldFlag_Invalid) {
break;
}
if (flag & FieldFlag_Unknown) {
syntax_error(f->curr_token, "Unknown prefix kind '#%.*s'", LIT(f->curr_token.string));
advance_token(f);
continue;
}
for (i32 i = 0; i < gb_count_of(parse_field_prefix_mappings); i++) {
if (parse_field_prefix_mappings[i].flag == flag) {
counts[i] += 1;
advance_token(f);
break;
}
}
}
u32 field_flags = 0;
for (i32 i = 0; i < gb_count_of(parse_field_prefix_mappings); i++) {
if (counts[i] > 0) {
field_flags |= parse_field_prefix_mappings[i].flag;
if (counts[i] != 1) {
auto const &mapping = parse_field_prefix_mappings[i];
String name = mapping.name;
char const *prefix = "";
if (mapping.token_kind == Token_Hash) {
prefix = "#";
}
syntax_error(f->curr_token, "Multiple '%s%.*s' in this field list", prefix, LIT(name));
}
}
}
return field_flags;
}
gb_internal u32 check_field_prefixes(AstFile *f, isize name_count, u32 allowed_flags, u32 set_flags) {
for (i32 i = 0; i < gb_count_of(parse_field_prefix_mappings); i++) {
bool err = false;
auto const &m = parse_field_prefix_mappings[i];
if ((set_flags & m.flag) != 0) {
if (m.flag == FieldFlag_using && name_count > 1) {
err = true;
syntax_error(f->curr_token, "Cannot apply 'using' to more than one of the same type");
}
if ((allowed_flags & m.flag) == 0) {
err = true;
char const *prefix = "";
if (m.token_kind == Token_Hash) {
prefix = "#";
}
syntax_error(f->curr_token, "'%s%.*s' in not allowed within this field list", prefix, LIT(m.name));
}
}
if (err) {
set_flags &= ~m.flag;
}
}
return set_flags;
}
struct AstAndFlags {
Ast *node;
u32 flags;
};
gb_internal Array<Ast *> convert_to_ident_list(AstFile *f, Array<AstAndFlags> list, bool ignore_flags, bool allow_poly_names) {
auto idents = array_make<Ast *>(ast_allocator(f), 0, list.count);
// Convert to ident list
isize i = 0;
for (AstAndFlags const &item : list) {
Ast *ident = item.node;
if (!ignore_flags) {
if (i != 0) {
syntax_error(ident, "Illegal use of prefixes in parameter list");
}
}
switch (ident->kind) {
case Ast_Ident:
case Ast_BadExpr:
break;
case Ast_Implicit:
begin_error_block();
syntax_error(ident, "Expected an identifier, '%.*s' which is a keyword", LIT(ident->Implicit.string));
if (ident->Implicit.kind == Token_context) {
error_line("\tSuggestion: Would 'ctx' suffice as an alternative name?\n");
}
end_error_block();
ident = ast_ident(f, blank_token);
break;
case Ast_PolyType:
if (allow_poly_names) {
if (ident->PolyType.specialization == nullptr) {
break;
} else {
syntax_error(ident, "Expected a polymorphic identifier without any specialization");
}
} else {
syntax_error(ident, "Expected a non-polymorphic identifier");
}
/*fallthrough*/
default:
syntax_error(ident, "Expected an identifier");
ident = ast_ident(f, blank_token);
break;
}
array_add(&idents, ident);
i += 1;
}
return idents;
}
gb_internal bool allow_field_separator(AstFile *f) {
Token token = f->curr_token;
if (allow_token(f, Token_Comma)) {
return true;
}
if (token.kind == Token_Semicolon) {
bool ok = false;
if (file_allow_newline(f) && token_is_newline(token)) {
TokenKind next = peek_token(f).kind;
switch (next) {
case Token_CloseBrace:
case Token_CloseParen:
ok = true;
break;
}
}
if (!ok) {
String p = token_to_string(token);
syntax_error(token_end_of_line(f, f->prev_token), "Expected a comma, got a %.*s", LIT(p));
}
advance_token(f);
return true;
}
return false;
}
gb_internal Ast *parse_struct_field_list(AstFile *f, isize *name_count_) {
Token start_token = f->curr_token;
auto decls = array_make<Ast *>(ast_allocator(f));
isize total_name_count = 0;
Ast *params = parse_field_list(f, &total_name_count, FieldFlag_Struct, Token_CloseBrace, false, false);
if (name_count_) *name_count_ = total_name_count;
return params;
}
// Returns true if any are polymorphic names
gb_internal bool check_procedure_name_list(Array<Ast *> const &names) {
if (names.count == 0) {
return false;
}
bool first_is_polymorphic = names[0]->kind == Ast_PolyType;
bool any_polymorphic_names = first_is_polymorphic;
for (isize i = 1; i < names.count; i++) {
Ast *name = names[i];
if (first_is_polymorphic) {
if (name->kind == Ast_PolyType) {
any_polymorphic_names = true;
} else {
syntax_error(name, "Mixture of polymorphic and non-polymorphic identifiers");
return any_polymorphic_names;
}
} else {
if (name->kind == Ast_PolyType) {
any_polymorphic_names = true;
syntax_error(name, "Mixture of polymorphic and non-polymorphic identifiers");
return any_polymorphic_names;
} else {
// Okay
}
}
}
return any_polymorphic_names;
}
gb_internal Ast *parse_field_list(AstFile *f, isize *name_count_, u32 allowed_flags, TokenKind follow, bool allow_default_parameters, bool allow_typeid_token) {
bool prev_allow_newline = f->allow_newline;
defer (f->allow_newline = prev_allow_newline);
f->allow_newline = file_allow_newline(f);
Token start_token = f->curr_token;
CommentGroup *docs = f->lead_comment;
auto params = array_make<Ast *>(ast_allocator(f));
auto list = array_make<AstAndFlags>(temporary_allocator());
bool allow_poly_names = allow_typeid_token;
isize total_name_count = 0;
bool allow_ellipsis = allowed_flags&FieldFlag_ellipsis;
bool seen_ellipsis = false;
bool is_signature = (allowed_flags & FieldFlag_Signature) == FieldFlag_Signature;
while (f->curr_token.kind != follow &&
f->curr_token.kind != Token_Colon &&
f->curr_token.kind != Token_EOF) {
u32 flags = parse_field_prefixes(f);
Ast *param = parse_var_type(f, allow_ellipsis, allow_typeid_token);
if (param->kind == Ast_Ellipsis) {
if (seen_ellipsis) syntax_error(param, "Extra variadic parameter after ellipsis");
seen_ellipsis = true;
} else if (seen_ellipsis) {
syntax_error(param, "Extra parameter after ellipsis");
}
AstAndFlags naf = {param, flags};
array_add(&list, naf);
if (!allow_field_separator(f)) {
break;
}
}
if (f->curr_token.kind == Token_Colon) {
Array<Ast *> names = convert_to_ident_list(f, list, true, allow_poly_names); // Copy for semantic reasons
if (names.count == 0) {
syntax_error(f->curr_token, "Empty field declaration");
}
bool any_polymorphic_names = check_procedure_name_list(names);
u32 set_flags = 0;
if (list.count > 0) {
set_flags = list[0].flags;
}
set_flags = check_field_prefixes(f, names.count, allowed_flags, set_flags);
total_name_count += names.count;
Ast *type = nullptr;
Ast *default_value = nullptr;
Token tag = {};
expect_token_after(f, Token_Colon, "field list");
if (f->curr_token.kind != Token_Eq) {
type = parse_var_type(f, allow_ellipsis, allow_typeid_token);
Ast *tt = unparen_expr(type);
if (tt == nullptr) {
syntax_error(f->prev_token, "Invalid type expression in field list");
} else if (is_signature && !any_polymorphic_names && tt->kind == Ast_TypeidType && tt->TypeidType.specialization != nullptr) {
syntax_error(type, "Specialization of typeid is not allowed without polymorphic names");
}
}
if (allow_token(f, Token_Eq)) {
default_value = parse_expr(f, false);
if (!allow_default_parameters) {
syntax_error(f->curr_token, "Default parameters are only allowed for procedures");
default_value = nullptr;
}
}
if (default_value != nullptr && names.count > 1) {
syntax_error(f->curr_token, "Default parameters can only be applied to single values");
}
if (allowed_flags == FieldFlag_Struct && default_value != nullptr) {
syntax_error(default_value, "Default parameters are not allowed for structs");
default_value = nullptr;
}
if (type != nullptr && type->kind == Ast_Ellipsis) {
if (seen_ellipsis) syntax_error(type, "Extra variadic parameter after ellipsis");
seen_ellipsis = true;
if (names.count != 1) {
syntax_error(type, "Variadic parameters can only have one field name");
}
} else if (seen_ellipsis && default_value == nullptr) {
syntax_error(f->curr_token, "Extra parameter after ellipsis without a default value");
}
if (type != nullptr && default_value == nullptr) {
if (f->curr_token.kind == Token_String) {
tag = expect_token(f, Token_String);
if ((allowed_flags & FieldFlag_Tags) == 0) {
syntax_error(tag, "Field tags are only allowed within structures");
}
}
}
allow_field_separator(f);
Ast *param = ast_field(f, names, type, default_value, set_flags, tag, docs, f->line_comment);
array_add(&params, param);
while (f->curr_token.kind != follow &&
f->curr_token.kind != Token_EOF &&
f->curr_token.kind != Token_Semicolon) {
CommentGroup *docs = f->lead_comment;
u32 set_flags = parse_field_prefixes(f);
Token tag = {};
Array<Ast *> names = parse_ident_list(f, allow_poly_names);
if (names.count == 0) {
syntax_error(f->curr_token, "Empty field declaration");
break;
}
bool any_polymorphic_names = check_procedure_name_list(names);
set_flags = check_field_prefixes(f, names.count, allowed_flags, set_flags);
total_name_count += names.count;
Ast *type = nullptr;
Ast *default_value = nullptr;
expect_token_after(f, Token_Colon, "field list");
if (f->curr_token.kind != Token_Eq) {
type = parse_var_type(f, allow_ellipsis, allow_typeid_token);
Ast *tt = unparen_expr(type);
if (is_signature && !any_polymorphic_names &&
tt != nullptr &&
tt->kind == Ast_TypeidType && tt->TypeidType.specialization != nullptr) {
syntax_error(type, "Specialization of typeid is not allowed without polymorphic names");
}
}
if (allow_token(f, Token_Eq)) {
default_value = parse_expr(f, false);
if (!allow_default_parameters) {
syntax_error(f->curr_token, "Default parameters are only allowed for procedures");
default_value = nullptr;
}
}
if (default_value != nullptr && names.count > 1) {
syntax_error(f->curr_token, "Default parameters can only be applied to single values");
}
if (type != nullptr && type->kind == Ast_Ellipsis) {
if (seen_ellipsis) syntax_error(type, "Extra variadic parameter after ellipsis");
seen_ellipsis = true;
if (names.count != 1) {
syntax_error(type, "Variadic parameters can only have one field name");
}
} else if (seen_ellipsis && default_value == nullptr) {
syntax_error(f->curr_token, "Extra parameter after ellipsis without a default value");
}
if (type != nullptr && default_value == nullptr) {
if (f->curr_token.kind == Token_String) {
tag = expect_token(f, Token_String);
if ((allowed_flags & FieldFlag_Tags) == 0) {
syntax_error(tag, "Field tags are only allowed within structures");
}
}
}
bool ok = allow_field_separator(f);
Ast *param = ast_field(f, names, type, default_value, set_flags, tag, docs, f->line_comment);
array_add(&params, param);
if (!ok) {
break;
}
}
if (name_count_) *name_count_ = total_name_count;
return ast_field_list(f, start_token, params);
}
for (AstAndFlags const &item : list) {
Ast *type = item.node;
Token token = blank_token;
if (allowed_flags&FieldFlag_Results) {
// NOTE(bill): Make this nothing and not `_`
token.string = str_lit("");
}
auto names = array_make<Ast *>(ast_allocator(f), 1);
token.pos = ast_token(type).pos;
names[0] = ast_ident(f, token);
u32 flags = check_field_prefixes(f, list.count, allowed_flags, item.flags);
Token tag = {};
Ast *param = ast_field(f, names, item.node, nullptr, flags, tag, docs, f->line_comment);
array_add(&params, param);
}
if (name_count_) *name_count_ = total_name_count;
return ast_field_list(f, start_token, params);
}
gb_internal Ast *parse_type_or_ident(AstFile *f) {
bool prev_allow_type = f->allow_type;
isize prev_expr_level = f->expr_level;
defer ({
f->allow_type = prev_allow_type;
f->expr_level = prev_expr_level;
});
f->allow_type = true;
f->expr_level = -1;
bool lhs = true;
Ast *operand = parse_operand(f, lhs);
Ast *type = parse_atom_expr(f, operand, lhs);
return type;
}
gb_internal Ast *parse_body(AstFile *f) {
Array<Ast *> stmts = {};
Token open, close;
isize prev_expr_level = f->expr_level;
bool prev_allow_newline = f->allow_newline;
// NOTE(bill): The body may be within an expression so reset to zero
f->expr_level = 0;
// f->allow_newline = false;
open = expect_token(f, Token_OpenBrace);
stmts = parse_stmt_list(f);
close = expect_token(f, Token_CloseBrace);
f->expr_level = prev_expr_level;
f->allow_newline = prev_allow_newline;
return ast_block_stmt(f, stmts, open, close);
}
gb_internal Ast *parse_do_body(AstFile *f, Token const &token, char const *msg) {
Token open, close;
isize prev_expr_level = f->expr_level;
bool prev_allow_newline = f->allow_newline;
// NOTE(bill): The body may be within an expression so reset to zero
f->expr_level = 0;
f->allow_newline = false;
Ast *body = convert_stmt_to_body(f, parse_stmt(f));
if (build_context.disallow_do) {
syntax_error(body, "'do' has been disallowed");
} else if (token.pos.file_id != 0 && !ast_on_same_line(token, body)) {
syntax_error(body, "The body of a 'do' must be on the same line as %s", msg);
}
f->expr_level = prev_expr_level;
f->allow_newline = prev_allow_newline;
return body;
}
gb_internal bool parse_control_statement_semicolon_separator(AstFile *f) {
Token tok = peek_token(f);
if (tok.kind != Token_OpenBrace) {
return allow_token(f, Token_Semicolon);
}
if (f->curr_token.string == ";") {
return allow_token(f, Token_Semicolon);
}
return false;
}
gb_internal Ast *parse_if_stmt(AstFile *f) {
if (f->curr_proc == nullptr) {
syntax_error(f->curr_token, "You cannot use an if statement in the file scope");
return ast_bad_stmt(f, f->curr_token, f->curr_token);
}
Token token = expect_token(f, Token_if);
Ast *init = nullptr;
Ast *cond = nullptr;
Ast *body = nullptr;
Ast *else_stmt = nullptr;
isize prev_level = f->expr_level;
f->expr_level = -1;
bool prev_allow_in_expr = f->allow_in_expr;
f->allow_in_expr = true;
if (allow_token(f, Token_Semicolon)) {
cond = parse_expr(f, false);
} else {
init = parse_simple_stmt(f, StmtAllowFlag_None);
if (parse_control_statement_semicolon_separator(f)) {
cond = parse_expr(f, false);
} else {
cond = convert_stmt_to_expr(f, init, str_lit("boolean expression"));
init = nullptr;
}
}
f->expr_level = prev_level;
f->allow_in_expr = prev_allow_in_expr;
if (cond == nullptr) {
syntax_error(f->curr_token, "Expected condition for if statement");
}
if (allow_token(f, Token_do)) {
body = parse_do_body(f, cond ? ast_token(cond) : token, "the if statement");
} else {
body = parse_block_stmt(f, false);
}
bool ignore_strict_style = false;
if (token.pos.line == ast_end_token(body).pos.line) {
ignore_strict_style = true;
}
skip_possible_newline_for_literal(f, ignore_strict_style);
if (f->curr_token.kind == Token_else) {
Token else_token = expect_token(f, Token_else);
switch (f->curr_token.kind) {
case Token_if:
else_stmt = parse_if_stmt(f);
break;
case Token_OpenBrace:
else_stmt = parse_block_stmt(f, false);
break;
case Token_do:
expect_token(f, Token_do);
else_stmt = parse_do_body(f, else_token, "'else'");
break;
default:
syntax_error(f->curr_token, "Expected if statement block statement");
else_stmt = ast_bad_stmt(f, f->curr_token, f->tokens[f->curr_token_index+1]);
break;
}
}
return ast_if_stmt(f, token, init, cond, body, else_stmt);
}
gb_internal Ast *parse_when_stmt(AstFile *f) {
Token token = expect_token(f, Token_when);
Ast *cond = nullptr;
Ast *body = nullptr;
Ast *else_stmt = nullptr;
isize prev_level = f->expr_level;
f->expr_level = -1;
bool prev_allow_in_expr = f->allow_in_expr;
f->allow_in_expr = true;
cond = parse_expr(f, false);
f->allow_in_expr = prev_allow_in_expr;
f->expr_level = prev_level;
if (cond == nullptr) {
syntax_error(f->curr_token, "Expected condition for when statement");
}
bool was_in_when_statement = f->in_when_statement;
f->in_when_statement = true;
if (allow_token(f, Token_do)) {
body = parse_do_body(f, cond ? ast_token(cond) : token, "then when statement");
} else {
body = parse_block_stmt(f, true);
}
bool ignore_strict_style = false;
if (token.pos.line == ast_end_token(body).pos.line) {
ignore_strict_style = true;
}
skip_possible_newline_for_literal(f, ignore_strict_style);
if (f->curr_token.kind == Token_else) {
Token else_token = expect_token(f, Token_else);
switch (f->curr_token.kind) {
case Token_when:
else_stmt = parse_when_stmt(f);
break;
case Token_OpenBrace:
else_stmt = parse_block_stmt(f, true);
break;
case Token_do: {
expect_token(f, Token_do);
else_stmt = parse_do_body(f, else_token, "'else'");
} break;
default:
syntax_error(f->curr_token, "Expected when statement block statement");
else_stmt = ast_bad_stmt(f, f->curr_token, f->tokens[f->curr_token_index+1]);
break;
}
}
f->in_when_statement = was_in_when_statement;
return ast_when_stmt(f, token, cond, body, else_stmt);
}
gb_internal Ast *parse_return_stmt(AstFile *f) {
Token token = expect_token(f, Token_return);
if (f->curr_proc == nullptr) {
syntax_error(f->curr_token, "You cannot use a return statement in the file scope");
return ast_bad_stmt(f, token, f->curr_token);
}
if (f->expr_level > 0) {
syntax_error(f->curr_token, "You cannot use a return statement within an expression");
return ast_bad_stmt(f, token, f->curr_token);
}
auto results = array_make<Ast *>(ast_allocator(f));
while (f->curr_token.kind != Token_Semicolon && f->curr_token.kind != Token_CloseBrace) {
Ast *arg = parse_expr(f, false);
array_add(&results, arg);
if (f->curr_token.kind != Token_Comma ||
f->curr_token.kind == Token_EOF) {
break;
}
advance_token(f);
}
expect_semicolon(f);
return ast_return_stmt(f, token, results);
}
gb_internal Ast *parse_for_stmt(AstFile *f) {
if (f->curr_proc == nullptr) {
syntax_error(f->curr_token, "You cannot use a for statement in the file scope");
return ast_bad_stmt(f, f->curr_token, f->curr_token);
}
Token token = expect_token(f, Token_for);
Ast *init = nullptr;
Ast *cond = nullptr;
Ast *post = nullptr;
Ast *body = nullptr;
bool is_range = false;
if (f->curr_token.kind != Token_OpenBrace &&
f->curr_token.kind != Token_do) {
isize prev_level = f->expr_level;
defer (f->expr_level = prev_level);
f->expr_level = -1;
if (f->curr_token.kind == Token_in) {
Token in_token = expect_token(f, Token_in);
syntax_error(in_token, "Prefer 'for _ in' over 'for in'");
Ast *rhs = nullptr;
bool prev_allow_range = f->allow_range;
f->allow_range = true;
rhs = parse_expr(f, false);
f->allow_range = prev_allow_range;
if (allow_token(f, Token_do)) {
body = parse_do_body(f, token, "the for statement");
} else {
body = parse_block_stmt(f, false);
}
return ast_range_stmt(f, token, {}, in_token, rhs, body);
}
if (f->curr_token.kind != Token_Semicolon) {
cond = parse_simple_stmt(f, StmtAllowFlag_In);
if (cond->kind == Ast_AssignStmt && cond->AssignStmt.op.kind == Token_in) {
is_range = true;
}
}
if (!is_range && parse_control_statement_semicolon_separator(f)) {
init = cond;
cond = nullptr;
if (f->curr_token.kind == Token_OpenBrace || f->curr_token.kind == Token_do) {
syntax_error(f->curr_token, "Expected ';', followed by a condition expression and post statement, got %.*s", LIT(token_strings[f->curr_token.kind]));
} else {
if (f->curr_token.kind != Token_Semicolon) {
cond = parse_simple_stmt(f, StmtAllowFlag_None);
}
if (f->curr_token.string != ";") {
syntax_error(f->curr_token, "Expected ';', got %.*s", LIT(token_to_string(f->curr_token)));
} else {
expect_token(f, Token_Semicolon);
}
if (f->curr_token.kind != Token_OpenBrace &&
f->curr_token.kind != Token_do) {
post = parse_simple_stmt(f, StmtAllowFlag_None);
}
}
}
}
if (allow_token(f, Token_do)) {
body = parse_do_body(f, token, "the for statement");
} else {
body = parse_block_stmt(f, false);
}
if (is_range) {
GB_ASSERT(cond->kind == Ast_AssignStmt);
Token in_token = cond->AssignStmt.op;
Slice<Ast *> vals = cond->AssignStmt.lhs;
Ast *rhs = nullptr;
if (cond->AssignStmt.rhs.count > 0) {
rhs = cond->AssignStmt.rhs[0];
}
return ast_range_stmt(f, token, vals, in_token, rhs, body);
}
cond = convert_stmt_to_expr(f, cond, str_lit("boolean expression"));
if (init != nullptr &&
cond == nullptr &&
post == nullptr) {
syntax_error(init, "'for init; ; {' without an explicit condition nor post statement is not allowed, please prefer something like 'for init; true; /**/{'");
}
return ast_for_stmt(f, token, init, cond, post, body);
}
gb_internal Ast *parse_case_clause(AstFile *f, bool is_type) {
Token token = f->curr_token;
Array<Ast *> list = {};
expect_token(f, Token_case);
bool prev_allow_range = f->allow_range;
bool prev_allow_in_expr = f->allow_in_expr;
f->allow_range = !is_type;
f->allow_in_expr = !is_type;
if (f->curr_token.kind != Token_Colon) {
list = parse_rhs_expr_list(f);
}
f->allow_range = prev_allow_range;
f->allow_in_expr = prev_allow_in_expr;
expect_token(f, Token_Colon);
Array<Ast *> stmts = parse_stmt_list(f);
return ast_case_clause(f, token, list, stmts);
}
gb_internal Ast *parse_switch_stmt(AstFile *f) {
if (f->curr_proc == nullptr) {
syntax_error(f->curr_token, "You cannot use a switch statement in the file scope");
return ast_bad_stmt(f, f->curr_token, f->curr_token);
}
Token token = expect_token(f, Token_switch);
Ast *init = nullptr;
Ast *tag = nullptr;
Ast *body = nullptr;
Token open, close;
bool is_type_switch = false;
auto list = array_make<Ast *>(ast_allocator(f));
if (f->curr_token.kind != Token_OpenBrace) {
isize prev_level = f->expr_level;
f->expr_level = -1;
defer (f->expr_level = prev_level);
if (f->curr_token.kind == Token_in) {
Token in_token = expect_token(f, Token_in);
syntax_error(in_token, "Prefer 'switch _ in' over 'switch in'");
auto lhs = array_make<Ast *>(ast_allocator(f), 0, 1);
auto rhs = array_make<Ast *>(ast_allocator(f), 0, 1);
Token blank_ident = token;
blank_ident.kind = Token_Ident;
blank_ident.string = str_lit("_");
Ast *blank = ast_ident(f, blank_ident);
array_add(&lhs, blank);
array_add(&rhs, parse_expr(f, true));
tag = ast_assign_stmt(f, token, lhs, rhs);
is_type_switch = true;
} else {
tag = parse_simple_stmt(f, StmtAllowFlag_In);
if (tag->kind == Ast_AssignStmt && tag->AssignStmt.op.kind == Token_in) {
is_type_switch = true;
} else if (parse_control_statement_semicolon_separator(f)) {
init = tag;
tag = nullptr;
if (f->curr_token.kind != Token_OpenBrace) {
tag = parse_simple_stmt(f, StmtAllowFlag_None);
}
}
}
}
skip_possible_newline(f);
open = expect_token(f, Token_OpenBrace);
while (f->curr_token.kind == Token_case) {
array_add(&list, parse_case_clause(f, is_type_switch));
}
close = expect_token(f, Token_CloseBrace);
body = ast_block_stmt(f, list, open, close);
if (is_type_switch) {
return ast_type_switch_stmt(f, token, tag, body);
}
tag = convert_stmt_to_expr(f, tag, str_lit("switch expression"));
return ast_switch_stmt(f, token, init, tag, body);
}
gb_internal Ast *parse_defer_stmt(AstFile *f) {
if (f->curr_proc == nullptr) {
syntax_error(f->curr_token, "You cannot use a defer statement in the file scope");
return ast_bad_stmt(f, f->curr_token, f->curr_token);
}
Token token = expect_token(f, Token_defer);
Ast *stmt = parse_stmt(f);
switch (stmt->kind) {
case Ast_EmptyStmt:
syntax_error(token, "Empty statement after defer (e.g. ';')");
break;
case Ast_DeferStmt:
syntax_error(token, "You cannot defer a defer statement");
stmt = stmt->DeferStmt.stmt;
break;
case Ast_ReturnStmt:
syntax_error(token, "You cannot defer a return statement");
break;
}
return ast_defer_stmt(f, token, stmt);
}
enum ImportDeclKind {
ImportDecl_Standard,
ImportDecl_Using,
};
gb_internal Ast *parse_import_decl(AstFile *f, ImportDeclKind kind) {
CommentGroup *docs = f->lead_comment;
Token token = expect_token(f, Token_import);
Token import_name = {};
switch (f->curr_token.kind) {
case Token_Ident:
import_name = advance_token(f);
break;
default:
import_name.pos = f->curr_token.pos;
break;
}
Token file_path = expect_token_after(f, Token_String, "import");
Ast *s = nullptr;
if (f->curr_proc != nullptr) {
syntax_error(import_name, "Cannot use 'import' within a procedure. This must be done at the file scope");
s = ast_bad_decl(f, import_name, file_path);
} else {
s = ast_import_decl(f, token, file_path, import_name, docs, f->line_comment);
array_add(&f->imports, s);
}
if (f->in_when_statement) {
syntax_error(import_name, "Cannot use 'import' within a 'when' statement. Prefer using the file suffixes (e.g. foo_windows.odin) or '//+build' tags");
}
if (kind != ImportDecl_Standard) {
syntax_error(import_name, "'using import' is not allowed, please use the import name explicitly");
}
expect_semicolon(f);
return s;
}
gb_internal Ast *parse_foreign_decl(AstFile *f) {
CommentGroup *docs = f->lead_comment;
Token token = expect_token(f, Token_foreign);
switch (f->curr_token.kind) {
case Token_Ident:
case Token_OpenBrace:
return parse_foreign_block(f, token);
case Token_import: {
Token import_token = expect_token(f, Token_import);
Token lib_name = {};
switch (f->curr_token.kind) {
case Token_Ident:
lib_name = advance_token(f);
break;
default:
lib_name.pos = token.pos;
break;
}
if (is_blank_ident(lib_name)) {
syntax_error(lib_name, "Illegal foreign import name: '_'");
}
bool multiple_filepaths = false;
Array<Ast *> filepaths = {};
if (allow_token(f, Token_OpenBrace)) {
multiple_filepaths = true;
array_init(&filepaths, ast_allocator(f));
while (f->curr_token.kind != Token_CloseBrace &&
f->curr_token.kind != Token_EOF) {
Ast *path = parse_expr(f, false);
array_add(&filepaths, path);
if (!allow_field_separator(f)) {
break;
}
}
expect_closing_brace_of_field_list(f);
} else {
filepaths = array_make<Ast *>(ast_allocator(f), 0, 1);
Token path = expect_token(f, Token_String);
Ast *lit = ast_basic_lit(f, path);
array_add(&filepaths, lit);
}
Ast *s = nullptr;
if (filepaths.count == 0) {
syntax_error(lib_name, "foreign import without any paths");
s = ast_bad_decl(f, lib_name, f->curr_token);
} else if (f->curr_proc != nullptr) {
syntax_error(lib_name, "You cannot use foreign import within a procedure. This must be done at the file scope");
s = ast_bad_decl(f, lib_name, ast_token(filepaths[0]));
} else {
s = ast_foreign_import_decl(f, token, filepaths, lib_name, multiple_filepaths, docs, f->line_comment);
}
expect_semicolon(f);
return s;
}
}
syntax_error(token, "Invalid foreign declaration");
return ast_bad_decl(f, token, f->curr_token);
}
gb_internal Ast *parse_attribute(AstFile *f, Token token, TokenKind open_kind, TokenKind close_kind, CommentGroup *docs) {
Array<Ast *> elems = {};
Token open = {};
Token close = {};
if (f->curr_token.kind == Token_Ident) {
elems = array_make<Ast *>(ast_allocator(f), 0, 1);
Ast *elem = parse_ident(f);
array_add(&elems, elem);
} else {
open = expect_token(f, open_kind);
f->expr_level++;
if (f->curr_token.kind != close_kind) {
elems = array_make<Ast *>(ast_allocator(f));
while (f->curr_token.kind != close_kind &&
f->curr_token.kind != Token_EOF) {
Ast *elem = nullptr;
elem = parse_ident(f);
if (f->curr_token.kind == Token_Eq) {
Token eq = expect_token(f, Token_Eq);
Ast *value = parse_value(f);
elem = ast_field_value(f, elem, value, eq);
}
array_add(&elems, elem);
if (!allow_field_separator(f)) {
break;
}
}
}
f->expr_level--;
close = expect_closing(f, close_kind, str_lit("attribute"));
}
Ast *attribute = ast_attribute(f, token, open, close, elems);
skip_possible_newline(f);
Ast *decl = parse_stmt(f);
if (decl->kind == Ast_ValueDecl) {
if (decl->ValueDecl.docs == nullptr && docs != nullptr) {
decl->ValueDecl.docs = docs;
}
array_add(&decl->ValueDecl.attributes, attribute);
} else if (decl->kind == Ast_ForeignBlockDecl) {
array_add(&decl->ForeignBlockDecl.attributes, attribute);
} else if (decl->kind == Ast_ForeignImportDecl) {
array_add(&decl->ForeignImportDecl.attributes, attribute);
} else if (decl->kind == Ast_ImportDecl) {
array_add(&decl->ImportDecl.attributes, attribute);
} else {
syntax_error(decl, "Expected a value or foreign declaration after an attribute, got %.*s", LIT(ast_strings[decl->kind]));
return ast_bad_stmt(f, token, f->curr_token);
}
return decl;
}
gb_internal Ast *parse_unrolled_for_loop(AstFile *f, Token unroll_token) {
Token for_token = expect_token(f, Token_for);
Ast *val0 = nullptr;
Ast *val1 = nullptr;
Token in_token = {};
Ast *expr = nullptr;
Ast *body = nullptr;
bool bad_stmt = false;
if (f->curr_token.kind != Token_in) {
Array<Ast *> idents = parse_ident_list(f, false);
switch (idents.count) {
case 1:
val0 = idents[0];
break;
case 2:
val0 = idents[0];
val1 = idents[1];
break;
default:
syntax_error(for_token, "Expected either 1 or 2 identifiers");
bad_stmt = true;
break;
}
}
in_token = expect_token(f, Token_in);
bool prev_allow_range = f->allow_range;
isize prev_level = f->expr_level;
f->allow_range = true;
f->expr_level = -1;
expr = parse_expr(f, false);
f->expr_level = prev_level;
f->allow_range = prev_allow_range;
if (allow_token(f, Token_do)) {
body = parse_do_body(f, for_token, "the for statement");
} else {
body = parse_block_stmt(f, false);
}
if (bad_stmt) {
return ast_bad_stmt(f, unroll_token, f->curr_token);
}
return ast_unroll_range_stmt(f, unroll_token, for_token, val0, val1, in_token, expr, body);
}
gb_internal Ast *parse_stmt(AstFile *f) {
Ast *s = nullptr;
Token token = f->curr_token;
switch (token.kind) {
// Operands
case Token_context: // Also allows for `context =`
case Token_proc:
case Token_Ident:
case Token_Integer:
case Token_Float:
case Token_Imag:
case Token_Rune:
case Token_String:
case Token_OpenParen:
case Token_Pointer:
case Token_asm: // Inline assembly
// Unary Operators
case Token_Add:
case Token_Sub:
case Token_Xor:
case Token_Not:
case Token_And:
case Token_Mul: // Used for error handling when people do C-like things
s = parse_simple_stmt(f, StmtAllowFlag_Label);
expect_semicolon(f);
return s;
case Token_foreign:
return parse_foreign_decl(f);
case Token_import:
return parse_import_decl(f, ImportDecl_Standard);
case Token_if: return parse_if_stmt(f);
case Token_when: return parse_when_stmt(f);
case Token_for: return parse_for_stmt(f);
case Token_switch: return parse_switch_stmt(f);
case Token_defer: return parse_defer_stmt(f);
case Token_return: return parse_return_stmt(f);
case Token_break:
case Token_continue:
case Token_fallthrough: {
Token token = advance_token(f);
Ast *label = nullptr;
if (token.kind != Token_fallthrough &&
f->curr_token.kind == Token_Ident) {
label = parse_ident(f);
}
s = ast_branch_stmt(f, token, label);
expect_semicolon(f);
return s;
}
case Token_using: {
CommentGroup *docs = f->lead_comment;
Token token = expect_token(f, Token_using);
if (f->curr_token.kind == Token_import) {
return parse_import_decl(f, ImportDecl_Using);
}
Ast *decl = nullptr;
Array<Ast *> list = parse_lhs_expr_list(f);
if (list.count == 0) {
syntax_error(token, "Illegal use of 'using' statement");
expect_semicolon(f);
return ast_bad_stmt(f, token, f->curr_token);
}
if (f->curr_token.kind != Token_Colon) {
expect_semicolon(f);
return ast_using_stmt(f, token, list);
}
expect_token_after(f, Token_Colon, "identifier list");
decl = parse_value_decl(f, list, docs);
if (decl != nullptr && decl->kind == Ast_ValueDecl) {
decl->ValueDecl.is_using = true;
return decl;
}
syntax_error(token, "Illegal use of 'using' statement");
return ast_bad_stmt(f, token, f->curr_token);
} break;
case Token_At: {
CommentGroup *docs = f->lead_comment;
Token token = expect_token(f, Token_At);
return parse_attribute(f, token, Token_OpenParen, Token_CloseParen, docs);
}
case Token_Hash: {
Ast *s = nullptr;
Token hash_token = expect_token(f, Token_Hash);
Token name = expect_token(f, Token_Ident);
String tag = name.string;
if (tag == "bounds_check") {
s = parse_stmt(f);
return parse_check_directive_for_statement(s, name, StateFlag_bounds_check);
} else if (tag == "no_bounds_check") {
s = parse_stmt(f);
return parse_check_directive_for_statement(s, name, StateFlag_no_bounds_check);
} else if (tag == "type_assert") {
s = parse_stmt(f);
return parse_check_directive_for_statement(s, name, StateFlag_type_assert);
} else if (tag == "no_type_assert") {
s = parse_stmt(f);
return parse_check_directive_for_statement(s, name, StateFlag_no_type_assert);
} else if (tag == "partial") {
s = parse_stmt(f);
switch (s->kind) {
case Ast_SwitchStmt:
s->SwitchStmt.partial = true;
break;
case Ast_TypeSwitchStmt:
s->TypeSwitchStmt.partial = true;
break;
case Ast_EmptyStmt:
return parse_check_directive_for_statement(s, name, 0);
default:
syntax_error(token, "#partial can only be applied to a switch statement");
break;
}
return s;
} else if (tag == "assert" || tag == "panic") {
Ast *t = ast_basic_directive(f, hash_token, name);
Ast *stmt = ast_expr_stmt(f, parse_call_expr(f, t));
expect_semicolon(f);
return stmt;
} else if (name.string == "force_inline" ||
name.string == "force_no_inline") {
Ast *expr = parse_force_inlining_operand(f, name);
Ast *stmt = ast_expr_stmt(f, expr);
expect_semicolon(f);
return stmt;
} else if (tag == "unroll") {
return parse_unrolled_for_loop(f, name);
} else if (tag == "reverse") {
Ast *for_stmt = parse_stmt(f);
if (for_stmt->kind == Ast_RangeStmt) {
if (for_stmt->RangeStmt.reverse) {
syntax_error(token, "#reverse already applied to a 'for in' statement");
}
for_stmt->RangeStmt.reverse = true;
} else {
syntax_error(token, "#reverse can only be applied to a 'for in' statement");
}
return for_stmt;
} else if (tag == "include") {
syntax_error(token, "#include is not a valid import declaration kind. Did you mean 'import'?");
s = ast_bad_stmt(f, token, f->curr_token);
} else if (tag == "define") {
s = ast_bad_stmt(f, token, f->curr_token);
if (name.pos.line == f->curr_token.pos.line) {
bool call_like = false;
Ast *macro_expr = nullptr;
Token ident = f->curr_token;
if (allow_token(f, Token_Ident) &&
name.pos.line == f->curr_token.pos.line) {
if (f->curr_token.kind == Token_OpenParen && f->curr_token.pos.column == ident.pos.column+ident.string.len) {
call_like = true;
(void)parse_call_expr(f, nullptr);
}
if (name.pos.line == f->curr_token.pos.line && f->curr_token.kind != Token_Semicolon) {
macro_expr = parse_expr(f, false);
}
}
ERROR_BLOCK();
syntax_error(ident, "#define is not a valid declaration, Odin does not have a C-like preprocessor.");
if (macro_expr == nullptr || call_like) {
error_line("\tNote: Odin does not support macros\n");
} else {
gbString s = expr_to_string(macro_expr);
error_line("\tSuggestion: Did you mean '%.*s :: %s'?\n", LIT(ident.string), s);
gb_string_free(s);
}
} else {
syntax_error(token, "#define is not a valid declaration, Odin does not have a C-like preprocessor.");
}
} else {
syntax_error(token, "Unknown tag directive used: '%.*s'", LIT(tag));
s = ast_bad_stmt(f, token, f->curr_token);
}
fix_advance_to_next_stmt(f);
return s;
} break;
case Token_OpenBrace:
return parse_block_stmt(f, false);
case Token_Semicolon:
s = ast_empty_stmt(f, token);
expect_semicolon(f);
return s;
}
// Error correction statements
switch (token.kind) {
case Token_else:
expect_token(f, Token_else);
syntax_error(token, "'else' unattached to an 'if' statement");
switch (f->curr_token.kind) {
case Token_if:
return parse_if_stmt(f);
case Token_when:
return parse_when_stmt(f);
case Token_OpenBrace:
return parse_block_stmt(f, true);
case Token_do: {
expect_token(f, Token_do);
Ast *stmt = parse_do_body(f, {}, "the for statement");
if (build_context.disallow_do) {
syntax_error(stmt, "'do' has been disallowed");
}
return stmt;
} break;
default:
fix_advance_to_next_stmt(f);
return ast_bad_stmt(f, token, f->curr_token);
}
}
syntax_error(token, "Expected a statement, got '%.*s'", LIT(token_strings[token.kind]));
fix_advance_to_next_stmt(f);
return ast_bad_stmt(f, token, f->curr_token);
}
gb_internal u64 check_vet_flags(AstFile *file) {
if (file && file->vet_flags_set) {
return file->vet_flags;
}
return build_context.vet_flags;
}
gb_internal void parse_enforce_tabs(AstFile *f) {
Token prev = f->prev_token;
Token curr = f->curr_token;
if (prev.pos.line < curr.pos.line) {
u8 *start = f->tokenizer.start+prev.pos.offset;
u8 *end = f->tokenizer.start+curr.pos.offset;
u8 *it = end;
while (it > start) {
if (*it == '\n') {
it++;
break;
}
it--;
}
isize len = end-it;
for (isize i = 0; i < len; i++) {
if (it[i] == ' ') {
syntax_error(curr, "With '-vet-tabs', tabs must be used for indentation");
break;
}
}
}
}
gb_internal Array<Ast *> parse_stmt_list(AstFile *f) {
auto list = array_make<Ast *>(ast_allocator(f));
while (f->curr_token.kind != Token_case &&
f->curr_token.kind != Token_CloseBrace &&
f->curr_token.kind != Token_EOF) {
// Checks to see if tabs have been used for indentation
if (check_vet_flags(f) & VetFlag_Tabs) {
parse_enforce_tabs(f);
}
Ast *stmt = parse_stmt(f);
if (stmt && stmt->kind != Ast_EmptyStmt) {
array_add(&list, stmt);
if (stmt->kind == Ast_ExprStmt &&
stmt->ExprStmt.expr != nullptr &&
stmt->ExprStmt.expr->kind == Ast_ProcLit) {
syntax_error(stmt, "Procedure literal evaluated but not used");
}
}
}
return list;
}
gb_internal ParseFileError init_ast_file(AstFile *f, String const &fullpath, TokenPos *err_pos) {
GB_ASSERT(f != nullptr);
f->fullpath = string_trim_whitespace(fullpath); // Just in case
f->filename = remove_directory_from_path(f->fullpath);
f->directory = directory_from_path(f->fullpath);
set_file_path_string(f->id, f->fullpath);
thread_safe_set_ast_file_from_id(f->id, f);
if (!string_ends_with(f->fullpath, str_lit(".odin"))) {
return ParseFile_WrongExtension;
}
gb_zero_item(&f->tokenizer);
f->tokenizer.curr_file_id = f->id;
TokenizerInitError err = init_tokenizer_from_fullpath(&f->tokenizer, f->fullpath, build_context.copy_file_contents);
if (err != TokenizerInit_None) {
switch (err) {
case TokenizerInit_Empty:
break;
case TokenizerInit_NotExists:
return ParseFile_NotFound;
case TokenizerInit_Permission:
return ParseFile_Permission;
case TokenizerInit_FileTooLarge:
return ParseFile_FileTooLarge;
default:
return ParseFile_InvalidFile;
}
}
isize file_size = f->tokenizer.end - f->tokenizer.start;
// NOTE(bill): Determine allocation size required for tokens
isize token_cap = file_size/3ll;
isize pow2_cap = gb_max(cast(isize)prev_pow2(cast(i64)token_cap)/2, 16);
token_cap = ((token_cap + pow2_cap-1)/pow2_cap) * pow2_cap;
isize init_token_cap = gb_max(token_cap, 16);
array_init(&f->tokens, ast_allocator(f), 0, gb_max(init_token_cap, 16));
if (err == TokenizerInit_Empty) {
Token token = {Token_EOF};
token.pos.file_id = f->id;
token.pos.line = 1;
token.pos.column = 1;
array_add(&f->tokens, token);
return ParseFile_None;
}
u64 start = time_stamp_time_now();
for (;;) {
Token *token = array_add_and_get(&f->tokens);
tokenizer_get_token(&f->tokenizer, token);
if (token->kind == Token_Invalid) {
err_pos->line = token->pos.line;
err_pos->column = token->pos.column;
return ParseFile_InvalidToken;
}
if (token->kind == Token_EOF) {
break;
}
}
u64 end = time_stamp_time_now();
f->time_to_tokenize = cast(f64)(end-start)/cast(f64)time_stamp__freq();
f->prev_token_index = 0;
f->curr_token_index = 0;
f->prev_token = f->tokens[f->prev_token_index];
f->curr_token = f->tokens[f->curr_token_index];
array_init(&f->comments, ast_allocator(f), 0, 0);
array_init(&f->imports, ast_allocator(f), 0, 0);
f->curr_proc = nullptr;
return ParseFile_None;
}
gb_internal void destroy_ast_file(AstFile *f) {
GB_ASSERT(f != nullptr);
array_free(&f->tokens);
array_free(&f->comments);
array_free(&f->imports);
}
gb_internal bool init_parser(Parser *p) {
GB_ASSERT(p != nullptr);
string_set_init(&p->imported_files);
array_init(&p->packages, permanent_allocator());
return true;
}
gb_internal void destroy_parser(Parser *p) {
GB_ASSERT(p != nullptr);
for (AstPackage *pkg : p->packages) {
for (AstFile *file : pkg->files) {
destroy_ast_file(file);
}
array_free(&pkg->files);
array_free(&pkg->foreign_files);
}
array_free(&p->packages);
string_set_destroy(&p->imported_files);
}
gb_internal void parser_add_package(Parser *p, AstPackage *pkg) {
MUTEX_GUARD_BLOCK(&p->packages_mutex) {
pkg->id = p->packages.count+1;
array_add(&p->packages, pkg);
}
}
gb_internal ParseFileError process_imported_file(Parser *p, ImportedFile imported_file);
gb_internal WORKER_TASK_PROC(parser_worker_proc) {
ParserWorkerData *wd = cast(ParserWorkerData *)data;
ParseFileError err = process_imported_file(wd->parser, wd->imported_file);
if (err != ParseFile_None) {
auto *node = gb_alloc_item(permanent_allocator(), ParseFileErrorNode);
node->err = err;
MUTEX_GUARD_BLOCK(&wd->parser->file_error_mutex) {
if (wd->parser->file_error_tail != nullptr) {
wd->parser->file_error_tail->next = node;
}
wd->parser->file_error_tail = node;
if (wd->parser->file_error_head == nullptr) {
wd->parser->file_error_head = node;
}
}
}
return cast(isize)err;
}
gb_internal void parser_add_file_to_process(Parser *p, AstPackage *pkg, FileInfo fi, TokenPos pos) {
ImportedFile f = {pkg, fi, pos, p->file_to_process_count++};
f.pos.file_id = cast(i32)(f.index+1);
auto wd = gb_alloc_item(permanent_allocator(), ParserWorkerData);
wd->parser = p;
wd->imported_file = f;
thread_pool_add_task(parser_worker_proc, wd);
}
gb_internal WORKER_TASK_PROC(foreign_file_worker_proc) {
ForeignFileWorkerData *wd = cast(ForeignFileWorkerData *)data;
ImportedFile *imp = &wd->imported_file;
AstPackage *pkg = imp->pkg;
AstForeignFile foreign_file = {wd->foreign_kind};
String fullpath = string_trim_whitespace(imp->fi.fullpath); // Just in case
char *c_str = alloc_cstring(temporary_allocator(), fullpath);
gbFileContents fc = gb_file_read_contents(permanent_allocator(), true, c_str);
foreign_file.source.text = (u8 *)fc.data;
foreign_file.source.len = fc.size;
switch (wd->foreign_kind) {
case AstForeignFile_S:
// TODO(bill): Actually do something with it
break;
}
MUTEX_GUARD_BLOCK(&pkg->foreign_files_mutex) {
array_add(&pkg->foreign_files, foreign_file);
}
return 0;
}
gb_internal void parser_add_foreign_file_to_process(Parser *p, AstPackage *pkg, AstForeignFileKind kind, FileInfo fi, TokenPos pos) {
// TODO(bill): Use a better allocator
ImportedFile f = {pkg, fi, pos, p->file_to_process_count++};
f.pos.file_id = cast(i32)(f.index+1);
auto wd = gb_alloc_item(permanent_allocator(), ForeignFileWorkerData);
wd->parser = p;
wd->imported_file = f;
wd->foreign_kind = kind;
thread_pool_add_task(foreign_file_worker_proc, wd);
}
// NOTE(bill): Returns true if it's added
gb_internal AstPackage *try_add_import_path(Parser *p, String path, String const &rel_path, TokenPos pos, PackageKind kind = Package_Normal) {
String const FILE_EXT = str_lit(".odin");
MUTEX_GUARD_BLOCK(&p->imported_files_mutex) {
if (string_set_update(&p->imported_files, path)) {
return nullptr;
}
}
path = copy_string(permanent_allocator(), path);
AstPackage *pkg = gb_alloc_item(permanent_allocator(), AstPackage);
pkg->kind = kind;
pkg->fullpath = path;
array_init(&pkg->files, permanent_allocator());
pkg->foreign_files.allocator = permanent_allocator();
// NOTE(bill): Single file initial package
if (kind == Package_Init && !path_is_directory(path) && string_ends_with(path, FILE_EXT)) {
FileInfo fi = {};
fi.name = filename_from_path(path);
fi.fullpath = path;
fi.size = get_file_size(path);
fi.is_dir = false;
array_reserve(&pkg->files, 1);
pkg->is_single_file = true;
parser_add_package(p, pkg);
parser_add_file_to_process(p, pkg, fi, pos);
return pkg;
}
Array<FileInfo> list = {};
ReadDirectoryError rd_err = read_directory(path, &list);
defer (array_free(&list));
if (list.count == 1) {
GB_ASSERT(path != list[0].fullpath);
}
switch (rd_err) {
case ReadDirectory_InvalidPath:
syntax_error(pos, "Invalid path: %.*s", LIT(rel_path));
return nullptr;
case ReadDirectory_NotExists:
syntax_error(pos, "Path does not exist: %.*s", LIT(rel_path));
return nullptr;
case ReadDirectory_Permission:
syntax_error(pos, "Unknown error whilst reading path %.*s", LIT(rel_path));
return nullptr;
case ReadDirectory_NotDir:
syntax_error(pos, "Expected a directory for a package, got a file: %.*s", LIT(rel_path));
return nullptr;
case ReadDirectory_Empty:
syntax_error(pos, "Empty directory: %.*s", LIT(rel_path));
return nullptr;
case ReadDirectory_Unknown:
syntax_error(pos, "Unknown error whilst reading path %.*s", LIT(rel_path));
return nullptr;
}
isize files_with_ext = 0;
isize files_to_reserve = 1; // always reserve 1
for (FileInfo fi : list) {
String name = fi.name;
String ext = path_extension(name);
if (ext == FILE_EXT) {
files_with_ext += 1;
}
if (ext == FILE_EXT && !is_excluded_target_filename(name)) {
files_to_reserve += 1;
}
}
if (files_with_ext == 0 || files_to_reserve == 1) {
ERROR_BLOCK();
if (files_with_ext != 0) {
syntax_error(pos, "Directory contains no .odin files for the specified platform: %.*s", LIT(rel_path));
} else {
syntax_error(pos, "Empty directory that contains no .odin files: %.*s", LIT(rel_path));
}
if (build_context.command_kind == Command_test) {
error_line("\tSuggestion: Make an .odin file that imports packages to test and use the `-all-packages` flag.");
}
return nullptr;
}
array_reserve(&pkg->files, files_to_reserve);
for (FileInfo fi : list) {
String name = fi.name;
String ext = path_extension(name);
if (ext == FILE_EXT) {
if (is_excluded_target_filename(name)) {
continue;
}
parser_add_file_to_process(p, pkg, fi, pos);
} else if (ext == ".S" || ext ==".s") {
if (is_excluded_target_filename(name)) {
continue;
}
parser_add_foreign_file_to_process(p, pkg, AstForeignFile_S, fi, pos);
}
}
parser_add_package(p, pkg);
return pkg;
}
gb_global Rune illegal_import_runes[] = {
'"', '\'', '`',
'\t', '\r', '\n', '\v', '\f',
'\\', // NOTE(bill): Disallow windows style filepaths
'!', '$', '%', '^', '&', '*', '(', ')', '=',
'[', ']', '{', '}',
';',
':', // NOTE(bill): Disallow windows style absolute filepaths
'#',
'|', ',', '<', '>', '?',
};
gb_internal bool is_import_path_valid(String const &path) {
if (path.len > 0) {
u8 *start = path.text;
u8 *end = path.text + path.len;
u8 *curr = start;
while (curr < end) {
isize width = 1;
Rune r = *curr;
if (r >= 0x80) {
width = utf8_decode(curr, end-curr, &r);
if (r == GB_RUNE_INVALID && width == 1) {
return false;
}
else if (r == GB_RUNE_BOM && curr-start > 0) {
return false;
}
}
for (isize i = 0; i < gb_count_of(illegal_import_runes); i++) {
if (r == illegal_import_runes[i]) {
return false;
}
}
curr += width;
}
return true;
}
return false;
}
gb_internal bool is_build_flag_path_valid(String const &path) {
if (path.len > 0) {
u8 *start = path.text;
u8 *end = path.text + path.len;
u8 *curr = start;
isize index = 0;
while (curr < end) {
isize width = 1;
Rune r = *curr;
if (r >= 0x80) {
width = utf8_decode(curr, end-curr, &r);
if (r == GB_RUNE_INVALID && width == 1) {
return false;
}
else if (r == GB_RUNE_BOM && curr-start > 0) {
return false;
}
}
for (isize i = 0; i < gb_count_of(illegal_import_runes); i++) {
#if defined(GB_SYSTEM_WINDOWS)
if (r == '\\') {
break;
} else if (r == ':') {
break;
}
#endif
if (r == illegal_import_runes[i]) {
return false;
}
}
curr += width;
index += 1;
}
return true;
}
return false;
}
gb_internal bool is_package_name_reserved(String const &name) {
if (name == "builtin") {
return true;
} else if (name == "intrinsics") {
return true;
}
return false;
}
gb_internal bool determine_path_from_string(BlockingMutex *file_mutex, Ast *node, String base_dir, String const &original_string, String *path, bool use_check_errors=false) {
GB_ASSERT(path != nullptr);
void (*do_error)(Ast *, char const *, ...);
void (*do_warning)(Token const &, char const *, ...);
do_error = &syntax_error;
do_warning = &syntax_warning;
if (use_check_errors) {
do_error = &error;
do_error = &warning;
}
// NOTE(bill): if file_mutex == nullptr, this means that the code is used within the semantics stage
String collection_name = {};
isize colon_pos = -1;
for (isize j = 0; j < original_string.len; j++) {
if (original_string[j] == ':') {
colon_pos = j;
break;
}
}
bool has_windows_drive = false;
#if defined(GB_SYSTEM_WINDOWS)
if (file_mutex == nullptr) {
if (colon_pos == 1 && original_string.len > 2) {
if (original_string[2] == '/' || original_string[2] == '\\') {
colon_pos = -1;
has_windows_drive = true;
}
}
}
#endif
String file_str = {};
if (colon_pos == 0) {
do_error(node, "Expected a collection name");
return false;
}
if (original_string.len > 0 && colon_pos > 0) {
collection_name = substring(original_string, 0, colon_pos);
file_str = substring(original_string, colon_pos+1, original_string.len);
} else {
file_str = original_string;
}
if (has_windows_drive) {
String sub_file_path = substring(file_str, 3, file_str.len);
if (!is_import_path_valid(sub_file_path)) {
do_error(node, "Invalid import path: '%.*s'", LIT(file_str));
return false;
}
} else if (!is_import_path_valid(file_str)) {
do_error(node, "Invalid import path: '%.*s'", LIT(file_str));
return false;
}
if (collection_name.len > 0) {
// NOTE(bill): `base:runtime` == `core:runtime`
if (collection_name == "core") {
bool replace_with_base = false;
if (string_starts_with(file_str, str_lit("runtime"))) {
replace_with_base = true;
} else if (string_starts_with(file_str, str_lit("intrinsics"))) {
replace_with_base = true;
} if (string_starts_with(file_str, str_lit("builtin"))) {
replace_with_base = true;
}
if (replace_with_base) {
collection_name = str_lit("base");
}
if (replace_with_base) {
if (ast_file_vet_deprecated(node->file())) {
do_error(node, "import \"core:%.*s\" has been deprecated in favour of \"base:%.*s\"", LIT(file_str), LIT(file_str));
} else {
do_warning(ast_token(node), "import \"core:%.*s\" has been deprecated in favour of \"base:%.*s\"", LIT(file_str), LIT(file_str));
}
}
}
if (collection_name == "system") {
if (node->kind != Ast_ForeignImportDecl) {
do_error(node, "The library collection 'system' is restrict for 'foreign import'");
return false;
} else {
*path = file_str;
return true;
}
} else if (!find_library_collection_path(collection_name, &base_dir)) {
// NOTE(bill): It's a naughty name
do_error(node, "Unknown library collection: '%.*s'", LIT(collection_name));
return false;
}
} else {
#if !defined(GB_SYSTEM_WINDOWS)
// @NOTE(vassvik): foreign imports of shared libraries that are not in the system collection on
// linux/mac have to be local to the executable for consistency with shared libraries.
// Unix does not have a concept of "import library" for shared/dynamic libraries,
// so we need to pass the relative path to the linker, and add the current
// working directory of the exe to the library search paths.
// Static libraries can be linked directly with the full pathname
//
if (node->kind == Ast_ForeignImportDecl && (string_ends_with(file_str, str_lit(".so")) || string_contains_string(file_str, str_lit(".so.")))) {
*path = file_str;
return true;
}
#endif
}
if (is_package_name_reserved(file_str)) {
*path = file_str;
if (collection_name == "core" || collection_name == "base") {
return true;
} else {
do_error(node, "The package '%.*s' must be imported with the 'base' library collection: 'base:%.*s'", LIT(file_str), LIT(file_str));
return false;
}
}
if (file_mutex) mutex_lock(file_mutex);
defer (if (file_mutex) mutex_unlock(file_mutex));
if (node->kind == Ast_ForeignImportDecl) {
node->ForeignImportDecl.collection_name = collection_name;
}
if (has_windows_drive) {
*path = file_str;
} else {
bool ok = false;
String fullpath = string_trim_whitespace(get_fullpath_relative(permanent_allocator(), base_dir, file_str, &ok));
*path = fullpath;
}
return true;
}
gb_internal void parse_setup_file_decls(Parser *p, AstFile *f, String const &base_dir, Slice<Ast *> &decls);
gb_internal void parse_setup_file_when_stmt(Parser *p, AstFile *f, String const &base_dir, AstWhenStmt *ws) {
if (ws->body != nullptr) {
auto stmts = ws->body->BlockStmt.stmts;
parse_setup_file_decls(p, f, base_dir, stmts);
}
if (ws->else_stmt != nullptr) {
switch (ws->else_stmt->kind) {
case Ast_BlockStmt: {
auto stmts = ws->else_stmt->BlockStmt.stmts;
parse_setup_file_decls(p, f, base_dir, stmts);
} break;
case Ast_WhenStmt:
parse_setup_file_when_stmt(p, f, base_dir, &ws->else_stmt->WhenStmt);
break;
}
}
}
gb_internal void parse_setup_file_decls(Parser *p, AstFile *f, String const &base_dir, Slice<Ast *> &decls) {
for_array(i, decls) {
Ast *node = decls[i];
if (!is_ast_decl(node) &&
node->kind != Ast_WhenStmt &&
node->kind != Ast_BadStmt &&
node->kind != Ast_EmptyStmt) {
// NOTE(bill): Sanity check
if (node->kind == Ast_ExprStmt) {
Ast *expr = node->ExprStmt.expr;
if (expr->kind == Ast_CallExpr &&
expr->CallExpr.proc->kind == Ast_BasicDirective) {
f->directive_count += 1;
continue;
}
}
syntax_error(node, "Only declarations are allowed at file scope, got %.*s", LIT(ast_strings[node->kind]));
} else if (node->kind == Ast_ImportDecl) {
ast_node(id, ImportDecl, node);
String original_string = string_trim_whitespace(string_value_from_token(f, id->relpath));
String import_path = {};
bool ok = determine_path_from_string(&p->file_decl_mutex, node, base_dir, original_string, &import_path);
if (!ok) {
decls[i] = ast_bad_decl(f, id->relpath, id->relpath);
continue;
}
import_path = string_trim_whitespace(import_path);
id->fullpath = import_path;
if (is_package_name_reserved(import_path)) {
continue;
}
try_add_import_path(p, import_path, original_string, ast_token(node).pos);
} else if (node->kind == Ast_ForeignImportDecl) {
ast_node(fl, ForeignImportDecl, node);
if (fl->filepaths.count == 0) {
syntax_error(decls[i], "No foreign paths found");
decls[i] = ast_bad_decl(f, ast_token(fl->filepaths[0]), ast_end_token(fl->filepaths[fl->filepaths.count-1]));
goto end;
} else if (!fl->multiple_filepaths &&
fl->filepaths.count == 1) {
Ast *fp = fl->filepaths[0];
GB_ASSERT(fp->kind == Ast_BasicLit);
Token fp_token = fp->BasicLit.token;
String file_str = string_trim_whitespace(string_value_from_token(f, fp_token));
String fullpath = file_str;
if (!is_arch_wasm() || string_ends_with(fullpath, str_lit(".o"))) {
String foreign_path = {};
bool ok = determine_path_from_string(&p->file_decl_mutex, node, base_dir, file_str, &foreign_path);
if (!ok) {
decls[i] = ast_bad_decl(f, fp_token, fp_token);
goto end;
}
fullpath = foreign_path;
}
fl->fullpaths = slice_make<String>(permanent_allocator(), 1);
fl->fullpaths[0] = fullpath;
}
} else if (node->kind == Ast_WhenStmt) {
ast_node(ws, WhenStmt, node);
parse_setup_file_when_stmt(p, f, base_dir, ws);
}
end:;
}
}
gb_internal String build_tag_get_token(String s, String *out) {
s = string_trim_whitespace(s);
isize n = 0;
while (n < s.len) {
Rune rune = 0;
isize width = utf8_decode(&s[n], s.len-n, &rune);
if (n == 0 && rune == '!') {
} else if (!rune_is_letter(rune) && !rune_is_digit(rune)) {
isize k = gb_max(gb_max(n, width), 1);
*out = substring(s, k, s.len);
return substring(s, 0, k);
}
n += width;
}
out->len = 0;
return s;
}
gb_internal bool parse_build_tag(Token token_for_pos, String s) {
String const prefix = str_lit("+build");
GB_ASSERT(string_starts_with(s, prefix));
s = string_trim_whitespace(substring(s, prefix.len, s.len));
if (s.len == 0) {
return true;
}
bool any_correct = false;
while (s.len > 0) {
bool this_kind_correct = true;
do {
String p = string_trim_whitespace(build_tag_get_token(s, &s));
if (p.len == 0) break;
if (p == ",") break;
bool is_notted = false;
if (p[0] == '!') {
is_notted = true;
p = substring(p, 1, p.len);
if (p.len == 0) {
syntax_error(token_for_pos, "Expected a build platform after '!'");
break;
}
}
if (p.len == 0) {
continue;
}
if (p == "ignore") {
this_kind_correct = false;
continue;
}
TargetOsKind os = get_target_os_from_string(p);
TargetArchKind arch = get_target_arch_from_string(p);
if (os != TargetOs_Invalid) {
GB_ASSERT(arch == TargetArch_Invalid);
if (is_notted) {
this_kind_correct = this_kind_correct && (os != build_context.metrics.os);
} else {
this_kind_correct = this_kind_correct && (os == build_context.metrics.os);
}
} else if (arch != TargetArch_Invalid) {
if (is_notted) {
this_kind_correct = this_kind_correct && (arch != build_context.metrics.arch);
} else {
this_kind_correct = this_kind_correct && (arch == build_context.metrics.arch);
}
}
if (os == TargetOs_Invalid && arch == TargetArch_Invalid) {
syntax_error(token_for_pos, "Invalid build tag platform: %.*s", LIT(p));
break;
}
} while (s.len > 0);
any_correct = any_correct || this_kind_correct;
}
return any_correct;
}
gb_internal String vet_tag_get_token(String s, String *out) {
s = string_trim_whitespace(s);
isize n = 0;
while (n < s.len) {
Rune rune = 0;
isize width = utf8_decode(&s[n], s.len-n, &rune);
if (n == 0 && rune == '!') {
} else if (!rune_is_letter(rune) && !rune_is_digit(rune) && rune != '-') {
isize k = gb_max(gb_max(n, width), 1);
*out = substring(s, k, s.len);
return substring(s, 0, k);
}
n += width;
}
out->len = 0;
return s;
}
gb_internal u64 parse_vet_tag(Token token_for_pos, String s) {
String const prefix = str_lit("+vet");
GB_ASSERT(string_starts_with(s, prefix));
s = string_trim_whitespace(substring(s, prefix.len, s.len));
if (s.len == 0) {
return VetFlag_All;
}
u64 vet_flags = 0;
u64 vet_not_flags = 0;
while (s.len > 0) {
String p = string_trim_whitespace(vet_tag_get_token(s, &s));
if (p.len == 0) {
break;
}
bool is_notted = false;
if (p[0] == '!') {
is_notted = true;
p = substring(p, 1, p.len);
if (p.len == 0) {
syntax_error(token_for_pos, "Expected a vet flag name after '!'");
return build_context.vet_flags;
}
}
u64 flag = get_vet_flag_from_name(p);
if (flag != VetFlag_NONE) {
if (is_notted) {
vet_not_flags |= flag;
} else {
vet_flags |= flag;
}
} else {
ERROR_BLOCK();
syntax_error(token_for_pos, "Invalid vet flag name: %.*s", LIT(p));
error_line("\tExpected one of the following\n");
error_line("\tunused\n");
error_line("\tshadowing\n");
error_line("\tusing-stmt\n");
error_line("\tusing-param\n");
error_line("\textra\n");
return build_context.vet_flags;
}
}
if (vet_flags == 0 && vet_not_flags == 0) {
return build_context.vet_flags;
}
if (vet_flags == 0 && vet_not_flags != 0) {
return build_context.vet_flags &~ vet_not_flags;
}
if (vet_flags != 0 && vet_not_flags == 0) {
return vet_flags;
}
GB_ASSERT(vet_flags != 0 && vet_not_flags != 0);
return vet_flags &~ vet_not_flags;
}
gb_internal String dir_from_path(String path) {
String base_dir = path;
for (isize i = path.len-1; i >= 0; i--) {
if (base_dir[i] == '\\' ||
base_dir[i] == '/') {
break;
}
base_dir.len--;
}
return base_dir;
}
gb_internal isize calc_decl_count(Ast *decl) {
isize count = 0;
switch (decl->kind) {
case Ast_BlockStmt:
for (Ast *stmt : decl->BlockStmt.stmts) {
count += calc_decl_count(stmt);
}
break;
case Ast_WhenStmt:
{
isize inner_count = calc_decl_count(decl->WhenStmt.body);
if (decl->WhenStmt.else_stmt) {
inner_count = gb_max(inner_count, calc_decl_count(decl->WhenStmt.else_stmt));
}
count += inner_count;
}
break;
case Ast_ValueDecl:
count = decl->ValueDecl.names.count;
break;
case Ast_ForeignBlockDecl:
count = calc_decl_count(decl->ForeignBlockDecl.body);
break;
case Ast_ImportDecl:
case Ast_ForeignImportDecl:
count = 1;
break;
}
return count;
}
gb_internal bool parse_build_project_directory_tag(Token token_for_pos, String s) {
String const prefix = str_lit("+build-project-name");
GB_ASSERT(string_starts_with(s, prefix));
s = string_trim_whitespace(substring(s, prefix.len, s.len));
if (s.len == 0) {
return true;
}
bool any_correct = false;
while (s.len > 0) {
bool this_kind_correct = true;
do {
String p = string_trim_whitespace(build_tag_get_token(s, &s));
if (p.len == 0) break;
if (p == ",") break;
bool is_notted = false;
if (p[0] == '!') {
is_notted = true;
p = substring(p, 1, p.len);
if (p.len == 0) {
syntax_error(token_for_pos, "Expected a build-project-name after '!'");
break;
}
}
if (p.len == 0) {
continue;
}
if (is_notted) {
this_kind_correct = this_kind_correct && (p != build_context.ODIN_BUILD_PROJECT_NAME);
} else {
this_kind_correct = this_kind_correct && (p == build_context.ODIN_BUILD_PROJECT_NAME);
}
} while (s.len > 0);
any_correct = any_correct || this_kind_correct;
}
return any_correct;
}
gb_internal bool parse_file(Parser *p, AstFile *f) {
if (f->tokens.count == 0) {
return true;
}
if (f->tokens.count > 0 && f->tokens[0].kind == Token_EOF) {
return true;
}
u64 start = time_stamp_time_now();
String filepath = f->tokenizer.fullpath;
String base_dir = dir_from_path(filepath);
if (f->curr_token.kind == Token_Comment) {
consume_comment_groups(f, f->prev_token);
}
CommentGroup *docs = f->lead_comment;
if (f->curr_token.kind != Token_package) {
ERROR_BLOCK();
syntax_error(f->curr_token, "Expected a package declaration at the beginning of the file");
// IMPORTANT NOTE(bill): this is technically a race condition with the suggestion, but it's ony a suggession
// so in practice is should be "fine"
if (f->pkg && f->pkg->name != "") {
error_line("\tSuggestion: Add 'package %.*s' to the top of the file\n", LIT(f->pkg->name));
}
return false;
}
f->package_token = expect_token(f, Token_package);
if (f->package_token.kind != Token_package) {
return false;
}
if (docs != nullptr) {
TokenPos end = token_pos_end(docs->list[docs->list.count-1]);
if (end.line == f->package_token.pos.line || end.line+1 == f->package_token.pos.line) {
// Okay
} else {
docs = nullptr;
}
}
Token package_name = expect_token_after(f, Token_Ident, "package");
if (package_name.kind == Token_Ident) {
if (package_name.string == "_") {
syntax_error(package_name, "Invalid package name '_'");
} else if (f->pkg->kind != Package_Runtime && package_name.string == "runtime") {
syntax_error(package_name, "Use of reserved package name '%.*s'", LIT(package_name.string));
} else if (is_package_name_reserved(package_name.string)) {
syntax_error(package_name, "Use of reserved package name '%.*s'", LIT(package_name.string));
}
}
f->package_name = package_name.string;
if (!f->pkg->is_single_file && docs != nullptr && docs->list.count > 0) {
for (Token const &tok : docs->list) {
GB_ASSERT(tok.kind == Token_Comment);
String str = tok.string;
if (string_starts_with(str, str_lit("//"))) {
String lc = string_trim_whitespace(substring(str, 2, str.len));
if (lc.len > 0 && lc[0] == '+') {
if (string_starts_with(lc, str_lit("+build-project-name"))) {
if (!parse_build_project_directory_tag(tok, lc)) {
return false;
}
} else if (string_starts_with(lc, str_lit("+build"))) {
if (!parse_build_tag(tok, lc)) {
return false;
}
} else if (string_starts_with(lc, str_lit("+vet"))) {
f->vet_flags = parse_vet_tag(tok, lc);
f->vet_flags_set = true;
} else if (string_starts_with(lc, str_lit("+ignore"))) {
return false;
} else if (string_starts_with(lc, str_lit("+private"))) {
f->flags |= AstFile_IsPrivatePkg;
String command = string_trim_starts_with(lc, str_lit("+private "));
command = string_trim_whitespace(command);
if (lc == "+private") {
f->flags |= AstFile_IsPrivatePkg;
} else if (command == "package") {
f->flags |= AstFile_IsPrivatePkg;
} else if (command == "file") {
f->flags |= AstFile_IsPrivateFile;
}
} else if (lc == "+lazy") {
if (build_context.ignore_lazy) {
// Ignore
} else if (f->pkg->kind == Package_Init && build_context.command_kind == Command_doc) {
// Ignore
} else {
f->flags |= AstFile_IsLazy;
}
} else if (lc == "+no-instrumentation") {
f->flags |= AstFile_NoInstrumentation;
} else {
warning(tok, "Ignoring unknown tag '%.*s'", LIT(lc));
}
}
}
}
}
Ast *pd = ast_package_decl(f, f->package_token, package_name, docs, f->line_comment);
expect_semicolon(f);
f->pkg_decl = pd;
if (f->error_count == 0) {
auto decls = array_make<Ast *>(ast_allocator(f));
while (f->curr_token.kind != Token_EOF) {
Ast *stmt = parse_stmt(f);
if (stmt && stmt->kind != Ast_EmptyStmt) {
array_add(&decls, stmt);
if (stmt->kind == Ast_ExprStmt &&
stmt->ExprStmt.expr != nullptr &&
stmt->ExprStmt.expr->kind == Ast_ProcLit) {
syntax_error(stmt, "Procedure literal evaluated but not used");
}
f->total_file_decl_count += calc_decl_count(stmt);
if (stmt->kind == Ast_WhenStmt || stmt->kind == Ast_ExprStmt || stmt->kind == Ast_ImportDecl) {
f->delayed_decl_count += 1;
}
}
}
f->decls = slice_from_array(decls);
parse_setup_file_decls(p, f, base_dir, f->decls);
}
u64 end = time_stamp_time_now();
f->time_to_parse = cast(f64)(end-start)/cast(f64)time_stamp__freq();
for (int i = 0; i < AstDelayQueue_COUNT; i++) {
array_init(f->delayed_decls_queues+i, ast_allocator(f), 0, f->delayed_decl_count);
}
return f->error_count == 0;
}
gb_internal ParseFileError process_imported_file(Parser *p, ImportedFile imported_file) {
AstPackage *pkg = imported_file.pkg;
FileInfo fi = imported_file.fi;
TokenPos pos = imported_file.pos;
AstFile *file = gb_alloc_item(permanent_allocator(), AstFile);
file->pkg = pkg;
file->id = cast(i32)(imported_file.index+1);
TokenPos err_pos = {0};
ParseFileError err = init_ast_file(file, fi.fullpath, &err_pos);
err_pos.file_id = file->id;
file->last_error = err;
if (err != ParseFile_None) {
if (err == ParseFile_EmptyFile) {
if (fi.fullpath == p->init_fullpath) {
syntax_error(pos, "Initial file is empty - %.*s\n", LIT(p->init_fullpath));
exit_with_errors();
}
} else {
switch (err) {
case ParseFile_WrongExtension:
syntax_error(pos, "Failed to parse file: %.*s; invalid file extension: File must have the extension '.odin'", LIT(fi.name));
break;
case ParseFile_InvalidFile:
syntax_error(pos, "Failed to parse file: %.*s; invalid file or cannot be found", LIT(fi.name));
break;
case ParseFile_Permission:
syntax_error(pos, "Failed to parse file: %.*s; file permissions problem", LIT(fi.name));
break;
case ParseFile_NotFound:
syntax_error(pos, "Failed to parse file: %.*s; file cannot be found ('%.*s')", LIT(fi.name), LIT(fi.fullpath));
break;
case ParseFile_InvalidToken:
syntax_error(err_pos, "Failed to parse file: %.*s; invalid token found in file", LIT(fi.name));
break;
case ParseFile_EmptyFile:
syntax_error(pos, "Failed to parse file: %.*s; file contains no tokens", LIT(fi.name));
break;
case ParseFile_FileTooLarge:
syntax_error(pos, "Failed to parse file: %.*s; file is too large, exceeds maximum file size of 2 GiB", LIT(fi.name));
break;
}
return err;
}
}
{
String name = file->fullpath;
name = remove_directory_from_path(name);
name = remove_extension_from_path(name);
if (string_starts_with(name, str_lit("_"))) {
syntax_error(pos, "Files cannot start with '_', got '%.*s'", LIT(file->fullpath));
}
}
if (build_context.command_kind == Command_test) {
String name = file->fullpath;
name = remove_extension_from_path(name);
}
if (parse_file(p, file)) {
MUTEX_GUARD_BLOCK(&pkg->files_mutex) {
array_add(&pkg->files, file);
}
mutex_lock(&pkg->name_mutex);
if (pkg->name.len == 0) {
pkg->name = file->package_name;
} else if (pkg->name != file->package_name) {
if (file->tokens.count > 0 && file->tokens[0].kind != Token_EOF) {
Token tok = file->package_token;
tok.pos.file_id = file->id;
tok.pos.line = gb_max(tok.pos.line, 1);
tok.pos.column = gb_max(tok.pos.column, 1);
syntax_error(tok, "Different package name, expected '%.*s', got '%.*s'", LIT(pkg->name), LIT(file->package_name));
}
}
mutex_unlock(&pkg->name_mutex);
p->total_line_count.fetch_add(file->tokenizer.line_count);
p->total_token_count.fetch_add(file->tokens.count);
}
return ParseFile_None;
}
gb_internal ParseFileError parse_packages(Parser *p, String init_filename) {
GB_ASSERT(init_filename.text[init_filename.len] == 0);
String init_fullpath = path_to_full_path(permanent_allocator(), init_filename);
if (!path_is_directory(init_fullpath)) {
String const ext = str_lit(".odin");
if (!string_ends_with(init_fullpath, ext)) {
error({}, "Expected either a directory or a .odin file, got '%.*s'\n", LIT(init_filename));
return ParseFile_WrongExtension;
}
} else if (init_fullpath.len != 0) {
String path = init_fullpath;
if (path[path.len-1] == '/') {
path.len -= 1;
}
if ((build_context.command_kind & Command__does_build) &&
build_context.build_mode == BuildMode_Executable) {
String output_path = path_to_string(temporary_allocator(), build_context.build_paths[8]);
char *cpath = alloc_cstring(temporary_allocator(), output_path);
if (path_is_directory(output_path) && gb_file_exists(cpath)) {
error({}, "Please specify the executable name with -out:<string> as a directory exists with the same name in the current working directory");
return ParseFile_DirectoryAlreadyExists;
}
}
}
{ // Add these packages serially and then process them parallel
TokenPos init_pos = {};
{
bool ok = false;
String s = get_fullpath_base_collection(permanent_allocator(), str_lit("runtime"), &ok);
if (!ok) {
compiler_error("Unable to find The 'base:runtime' package. Is the ODIN_ROOT set up correctly?");
}
try_add_import_path(p, s, s, init_pos, Package_Runtime);
}
try_add_import_path(p, init_fullpath, init_fullpath, init_pos, Package_Init);
p->init_fullpath = init_fullpath;
if (build_context.command_kind == Command_test) {
bool ok = false;
String s = get_fullpath_core_collection(permanent_allocator(), str_lit("testing"), &ok);
if (!ok) {
compiler_error("Unable to find The 'core:testing' package. Is the ODIN_ROOT set up correctly?");
}
try_add_import_path(p, s, s, init_pos, Package_Normal);
}
for (String const &path : build_context.extra_packages) {
String fullpath = path_to_full_path(permanent_allocator(), path); // LEAK?
if (!path_is_directory(fullpath)) {
String const ext = str_lit(".odin");
if (!string_ends_with(fullpath, ext)) {
error({}, "Expected either a directory or a .odin file, got '%.*s'\n", LIT(fullpath));
return ParseFile_WrongExtension;
}
}
AstPackage *pkg = try_add_import_path(p, fullpath, fullpath, init_pos, Package_Normal);
if (pkg) {
pkg->is_extra = true;
}
}
}
thread_pool_wait();
for (ParseFileErrorNode *node = p->file_error_head; node != nullptr; node = node->next) {
if (node->err != ParseFile_None) {
return node->err;
}
}
for (isize i = p->packages.count-1; i >= 0; i--) {
AstPackage *pkg = p->packages[i];
for (isize j = pkg->files.count-1; j >= 0; j--) {
AstFile *file = pkg->files[j];
if (file->error_count != 0) {
if (file->last_error != ParseFile_None) {
return file->last_error;
}
return ParseFile_GeneralError;
}
}
}
for (AstPackage *pkg : p->packages) {
for (AstFile *file : pkg->files) {
p->total_seen_load_directive_count += file->seen_load_directive_count;
}
}
return ParseFile_None;
}