nostrdb

commit 47b016f7ff7e655b5a2acabf61a3fdb3605cad52
parent 82fcacd19fbe4fd58234b75f1fe13fdc6df6dfd7
Author: William Casarin <jb55@jb55.com>
Date:   Sun, 23 Jul 2023 12:07:52 -0700

huge perf boost when switching from strncmp to memcmp

not sure why this is so slow.

Diffstat:
M
nostrdb.c
 | 
13
++++++++-----
M
test.c
 | 
2
+-

2 files changed, 9 insertions(+), 6 deletions(-)
diff --git a/nostrdb.c b/nostrdb.c
@@ -136,7 +136,7 @@ static inline int ndb_builder_find_str(struct ndb_builder *builder,
 		uint32_t index = ((uint32_t*)builder->str_indices.start)[i];
 		const char *some_str = (const char*)builder->strings.start + index;
 
-		if (!strncmp(some_str, str, len)) {
+		if (!memcmp(some_str, str, len)) {
 			// found an existing matching str, use that index
 			*pstr = ndb_offset_str(index);
 			return 1;
@@ -171,10 +171,13 @@ static int ndb_builder_push_packed_id(struct ndb_builder *builder,
 				      unsigned char *id,
 				      union ndb_packed_str *pstr)
 {
-	if (ndb_builder_find_str(builder, (const char*)id, 32, pstr)) {
-		pstr->packed.flag = NDB_PACKED_ID;
-		return 1;
-	}
+	// Don't both find id duplicates. very rarely are they duplicated
+	// and it slows things down quite a bit. If we really care about this
+	// We can switch to a hash table.
+	//if (ndb_builder_find_str(builder, (const char*)id, 32, pstr)) {
+	//	pstr->packed.flag = NDB_PACKED_ID;
+	//	return 1;
+	//}
 
 	if (ndb_builder_push_str(builder, (const char*)id, 32, pstr)) {
 		pstr->packed.flag = NDB_PACKED_ID;
diff --git a/test.c b/test.c
@@ -126,7 +126,7 @@ static void test_parse_contact_list()
 	size = ndb_note_from_json((const char*)json, written, &note, buf, alloc_size);
 	printf("ndb_note_from_json size %d\n", size);
 	assert(size > 0);
-	assert(size == 34058);
+	assert(size == 34322);
 
 	const char* expected_content = 
 	"{\"wss://nos.lol\":{\"write\":true,\"read\":true},"