motiejus/zig
1
Fork 0
Code Releases Activity

stage1: memoize strings in the AST

Browse Source 
Currently, stage1 runs astgen for every comptime function call,
resulting in identifier strings being allocated multiple times,
wasting memory. As a workaround until the code is adjusted to
make astgen run only once per source node, we memoize the
result into the AST.

 * Rename `ir_gen_*` to `astgen_*`
   - Oops, meant to do this in a separate commit. My bad.
 * tokenizer: avoid using designated initializer syntax.
   MSVC does not support it.
This commit is contained in:
Andrew Kelley
2021-05-28 15:22:03 -07:00
parent 52b3daa90e
commit f5d4fe3e17
5 changed files with 489 additions and 467 deletions
Download Patch File Download Diff File Expand all files Collapse all files

43
src/stage1/parser.cpp
View File

@@ -202,6 +202,19 @@ static void put_back_token(ParseContext *pc) {
pc->current_token -= 1;
}
static Buf *token_string_literal_buf(RootStruct *root_struct, TokenIndex token) {
Error err;
assert(root_struct->token_ids[token] == TokenIdStringLiteral);
const char *source = buf_ptr(root_struct->source_code);
size_t byte_offset = root_struct->token_locs[token].offset;
size_t bad_index;
Buf *str = buf_alloc();
if ((err = source_string_literal_buf(source + byte_offset, str, &bad_index))) {
zig_panic("TODO handle string literal parse error");
}
return str;
}
static Buf *token_buf(ParseContext *pc, TokenIndex token) {
if (token == 0)
return nullptr;
@@ -3465,19 +3478,6 @@ Error source_char_literal(const char *source, uint32_t *result, size_t *bad_inde
}
Buf *token_string_literal_buf(RootStruct *root_struct, TokenIndex token) {
Error err;
assert(root_struct->token_ids[token] == TokenIdStringLiteral);
const char *source = buf_ptr(root_struct->source_code);
size_t byte_offset = root_struct->token_locs[token].offset;
size_t bad_index;
Buf *str = buf_alloc();
if ((err = source_string_literal_buf(source + byte_offset, str, &bad_index))) {
zig_panic("TODO handle string literal parse error");
}
return str;
}
Buf *token_identifier_buf(RootStruct *root_struct, TokenIndex token) {
Error err;
const char *source = buf_ptr(root_struct->source_code);
@@ -3515,14 +3515,15 @@ Buf *token_identifier_buf(RootStruct *root_struct, TokenIndex token) {
Buf *node_identifier_buf(AstNode *node) {
assert(node->type == NodeTypeIdentifier);
RootStruct *root_struct = node->owner->data.structure.root_struct;
return token_identifier_buf(root_struct, node->main_token);
}
Buf *node_string_literal_buf(AstNode *node) {
assert(node->type == NodeTypeStringLiteral);
RootStruct *root_struct = node->owner->data.structure.root_struct;
return token_string_literal_buf(root_struct, node->main_token);
// Currently, stage1 runs astgen for every comptime function call,
// resulting the allocation here wasting memory. As a workaround until
// the code is adjusted to make astgen run only once per source node,
// we memoize the result into the AST here.
if (node->data.identifier.name == nullptr) {
RootStruct *root_struct = node->owner->data.structure.root_struct;
node->data.identifier.name = token_identifier_buf(root_struct, node->main_token);
}
return node->data.identifier.name;
}
void token_number_literal_bigint(RootStruct *root_struct, BigInt *result, TokenIndex token) {