damus

nostr ios client
git clone git://jb55.com/damus
Log | Files | Refs | README | LICENSE

nostrdb.c (107753B)


      1 
      2 #include "nostrdb.h"
      3 #include "jsmn.h"
      4 #include "hex.h"
      5 #include "cursor.h"
      6 #include "random.h"
      7 #include "sha256.h"
      8 #include "lmdb.h"
      9 #include "util.h"
     10 #include "cpu.h"
     11 #include "threadpool.h"
     12 #include "protected_queue.h"
     13 #include "memchr.h"
     14 #include <stdlib.h>
     15 #include <limits.h>
     16 #include <assert.h>
     17 
     18 #include "bindings/c/profile_json_parser.h"
     19 #include "bindings/c/profile_builder.h"
     20 #include "bindings/c/meta_builder.h"
     21 #include "bindings/c/meta_reader.h"
     22 #include "bindings/c/profile_verifier.h"
     23 #include "secp256k1.h"
     24 #include "secp256k1_ecdh.h"
     25 #include "secp256k1_schnorrsig.h"
     26 
     27 #define max(a,b) ((a) > (b) ? (a) : (b))
     28 #define min(a,b) ((a) < (b) ? (a) : (b))
     29 
     30 // the maximum number of things threads pop and push in bulk
     31 static const int THREAD_QUEUE_BATCH = 4096;
     32 
     33 // the maximum size of inbox queues
     34 static const int DEFAULT_QUEUE_SIZE = 1000000;
     35 
     36 // increase if we need bigger filters
     37 #define NDB_FILTER_PAGES 64
     38 
     39 #define ndb_flag_set(flags, f) ((flags & f) == f)
     40 
     41 #define NDB_PARSED_ID           (1 << 0)
     42 #define NDB_PARSED_PUBKEY       (1 << 1)
     43 #define NDB_PARSED_SIG          (1 << 2)
     44 #define NDB_PARSED_CREATED_AT   (1 << 3)
     45 #define NDB_PARSED_KIND         (1 << 4)
     46 #define NDB_PARSED_CONTENT      (1 << 5)
     47 #define NDB_PARSED_TAGS         (1 << 6)
     48 #define NDB_PARSED_ALL          (NDB_PARSED_ID|NDB_PARSED_PUBKEY|NDB_PARSED_SIG|NDB_PARSED_CREATED_AT|NDB_PARSED_KIND|NDB_PARSED_CONTENT|NDB_PARSED_TAGS)
     49 
     50 typedef int (*ndb_migrate_fn)(struct ndb *);
     51 typedef int (*ndb_word_parser_fn)(void *, const char *word, int word_len,
     52 				  int word_index);
     53 
     54 struct ndb_migration {
     55 	ndb_migrate_fn fn;
     56 };
     57 
     58 struct ndb_profile_record_builder {
     59 	flatcc_builder_t *builder;
     60 	void *flatbuf;
     61 };
     62 
     63 // controls whether to continue or stop the json parser
     64 enum ndb_idres {
     65 	NDB_IDRES_CONT,
     66 	NDB_IDRES_STOP,
     67 };
     68 
     69 // closure data for the id-detecting ingest controller
     70 struct ndb_ingest_controller
     71 {
     72 	MDB_txn *read_txn;
     73 	struct ndb_lmdb *lmdb;
     74 };
     75 
     76 enum ndb_writer_msgtype {
     77 	NDB_WRITER_QUIT, // kill thread immediately
     78 	NDB_WRITER_NOTE, // write a note to the db
     79 	NDB_WRITER_PROFILE, // write a profile to the db
     80 	NDB_WRITER_DBMETA, // write ndb metadata
     81 	NDB_WRITER_PROFILE_LAST_FETCH, // when profiles were last fetched
     82 };
     83 
     84 // keys used for storing data in the NDB metadata database (NDB_DB_NDB_META)
     85 enum ndb_meta_key {
     86 	NDB_META_KEY_VERSION = 1
     87 };
     88 
     89 struct ndb_json_parser {
     90 	const char *json;
     91 	int json_len;
     92 	struct ndb_builder builder;
     93 	jsmn_parser json_parser;
     94 	jsmntok_t *toks, *toks_end;
     95 	int i;
     96 	int num_tokens;
     97 };
     98 
     99 // useful to pass to threads on its own
    100 struct ndb_lmdb {
    101 	MDB_env *env;
    102 	MDB_dbi dbs[NDB_DBS];
    103 };
    104 
    105 struct ndb_writer {
    106 	struct ndb_lmdb *lmdb;
    107 
    108 	void *queue_buf;
    109 	int queue_buflen;
    110 	pthread_t thread_id;
    111 
    112 	struct prot_queue inbox;
    113 };
    114 
    115 struct ndb_ingester {
    116 	uint32_t flags;
    117 	struct threadpool tp;
    118 	struct ndb_writer *writer;
    119 	void *filter_context;
    120 	ndb_ingest_filter_fn filter;
    121 };
    122 
    123 
    124 struct ndb {
    125 	struct ndb_lmdb lmdb;
    126 	struct ndb_ingester ingester;
    127 	struct ndb_writer writer;
    128 	int version;
    129 	uint32_t flags; // setting flags
    130 	// lmdb environ handles, etc
    131 };
    132 
    133 
    134 // A clustered key with an id and a timestamp
    135 struct ndb_tsid {
    136 	unsigned char id[32];
    137 	uint64_t timestamp;
    138 };
    139 
    140 // A u64 + timestamp id. Just using this for kinds at the moment.
    141 struct ndb_u64_tsid {
    142 	uint64_t u64; // kind, etc
    143 	uint64_t timestamp;
    144 };
    145 
    146 struct ndb_word
    147 {
    148 	const char *word;
    149 	int word_len;
    150 };
    151 
    152 struct ndb_search_words
    153 {
    154 	struct ndb_word words[MAX_TEXT_SEARCH_WORDS];
    155 	int num_words;
    156 };
    157 
    158 // ndb_text_search_key
    159 //
    160 // This is compressed when in lmdb:
    161 //
    162 //   note_id:    varint
    163 //   strlen:     varint
    164 //   str:        cstr
    165 //   timestamp:  varint
    166 //   word_index: varint
    167 //   
    168 static int ndb_make_text_search_key(unsigned char *buf, int bufsize,
    169 				    int word_index, int word_len, const char *str,
    170 				    uint64_t timestamp, uint64_t note_id,
    171 				    int *keysize)
    172 {
    173 	struct cursor cur;
    174 	int size, pad;
    175 	make_cursor(buf, buf + bufsize, &cur);
    176 
    177 	// TODO: need update this to uint64_t
    178 	// we push this first because our query function can pull this off
    179 	// quicky to check matches
    180 	if (!push_varint(&cur, (int32_t)note_id))
    181 		return 0;
    182 
    183 	// string length
    184 	if (!push_varint(&cur, word_len))
    185 		return 0;
    186 
    187 	// non-null terminated, lowercase string
    188 	if (!cursor_push_lowercase(&cur, str, word_len))
    189 		return 0;
    190 
    191 	// TODO: need update this to uint64_t
    192 	if (!push_varint(&cur, (int)timestamp))
    193 		return 0;
    194 
    195 	// the index of the word in the content so that we can do more accurate
    196 	// phrase searches
    197 	if (!push_varint(&cur, word_index))
    198 		return 0;
    199 
    200 	size = cur.p - cur.start;
    201 
    202 	// pad to 8-byte alignment
    203 	pad = ((size + 7) & ~7) - size;
    204 	if (pad > 0) {
    205 		if (!cursor_memset(&cur, 0, pad)) {
    206 			return 0;
    207 		}
    208 	}
    209 
    210 	*keysize = cur.p - cur.start;
    211 	assert((*keysize % 8) == 0);
    212 
    213 	return 1;
    214 }
    215 
    216 static int ndb_make_noted_text_search_key(unsigned char *buf, int bufsize,
    217 					  int wordlen, const char *word,
    218 					  uint64_t timestamp, uint64_t note_id,
    219 					  int *keysize)
    220 {
    221 	return ndb_make_text_search_key(buf, bufsize, 0, wordlen, word,
    222 					timestamp, note_id, keysize);
    223 }
    224 
    225 static int ndb_make_text_search_key_low(unsigned char *buf, int bufsize,
    226 					int wordlen, const char *word,
    227 					int *keysize)
    228 {
    229 	uint64_t timestamp, note_id;
    230 	timestamp = 0;
    231 	note_id = 0;
    232 	return ndb_make_text_search_key(buf, bufsize, 0, wordlen, word,
    233 					timestamp, note_id, keysize);
    234 }
    235 
    236 static int ndb_make_text_search_key_high(unsigned char *buf, int bufsize,
    237 					 int wordlen, const char *word,
    238 					 int *keysize)
    239 {
    240 	uint64_t timestamp, note_id;
    241 	timestamp = INT32_MAX;
    242 	note_id = INT32_MAX;
    243 	return ndb_make_text_search_key(buf, bufsize, 0, wordlen, word,
    244 					timestamp, note_id, keysize);
    245 }
    246 
    247 typedef int (*ndb_text_search_key_order_fn)(unsigned char *buf, int bufsize, int wordlen, const char *word, int *keysize);
    248 
    249 /** From LMDB: Compare two items lexically */
    250 static int mdb_cmp_memn(const MDB_val *a, const MDB_val *b) {
    251 	int diff;
    252 	ssize_t len_diff;
    253 	unsigned int len;
    254 
    255 	len = a->mv_size;
    256 	len_diff = (ssize_t) a->mv_size - (ssize_t) b->mv_size;
    257 	if (len_diff > 0) {
    258 		len = b->mv_size;
    259 		len_diff = 1;
    260 	}
    261 
    262 	diff = memcmp(a->mv_data, b->mv_data, len);
    263 	return diff ? diff : len_diff<0 ? -1 : len_diff;
    264 }
    265 
    266 static int ndb_text_search_key_compare(const MDB_val *a, const MDB_val *b)
    267 {
    268 	struct cursor ca, cb;
    269 	int sa, sb;
    270 	int nid_a, nid_b;
    271 	MDB_val a2, b2;
    272 
    273 	make_cursor(a->mv_data, a->mv_data + a->mv_size, &ca);
    274 	make_cursor(b->mv_data, b->mv_data + b->mv_size, &cb);
    275 
    276 	// note_id
    277 	if (unlikely(!pull_varint(&ca, &nid_a) || !pull_varint(&cb, &nid_b)))
    278 		return 0;
    279 
    280 	// string size
    281 	if (unlikely(!pull_varint(&ca, &sa) || !pull_varint(&cb, &sb)))
    282 		return 0;
    283 
    284 	a2.mv_data = ca.p;
    285 	a2.mv_size = sa;
    286 
    287 	b2.mv_data = cb.p;
    288 	b2.mv_size = sb;
    289 
    290 	int cmp = mdb_cmp_memn(&a2, &b2);
    291 	if (cmp) return cmp;
    292 
    293 	// skip over string
    294 	ca.p += sa;
    295 	cb.p += sb;
    296 
    297 	// timestamp
    298 	if (unlikely(!pull_varint(&ca, &sa) || !pull_varint(&cb, &sb)))
    299 		return 0;
    300 
    301 	if      (sa < sb) return -1;
    302 	else if (sa > sb) return 1;
    303 
    304 	// note_id
    305 	if      (nid_a < nid_b) return -1;
    306 	else if (nid_a > nid_b) return 1;
    307 
    308 	// word index
    309 	if (unlikely(!pull_varint(&ca, &sa) || !pull_varint(&cb, &sb)))
    310 		return 0;
    311 
    312 	if      (sa < sb) return -1;
    313 	else if (sa > sb) return 1;
    314 
    315 	return 0;
    316 }
    317 
    318 static inline int ndb_unpack_text_search_key_noteid(
    319 		struct cursor *cur, uint64_t *note_id)
    320 {
    321 	int inote_id;
    322 	if (!pull_varint(cur, &inote_id))
    323 		return 0;
    324 
    325 	*note_id = inote_id;
    326 	return 1;
    327 }
    328 
    329 // faster peek of just the string instead of unpacking everything
    330 // this is used to quickly discard range query matches if there is no
    331 // common prefix
    332 static inline int ndb_unpack_text_search_key_string(struct cursor *cur,
    333 						    const char **str,
    334 						    int *str_len)
    335 {
    336 	if (!pull_varint(cur, str_len))
    337 		return 0;
    338 
    339 	*str = (const char *)cur->p;
    340 
    341 	if (!cursor_skip(cur, *str_len))
    342 		return 0;
    343 	
    344 	return 1;
    345 }
    346 
    347 // should be called after ndb_unpack_text_search_key_string. It continues
    348 // the unpacking of a text search key if we've already started it.
    349 static inline int
    350 ndb_unpack_remaining_text_search_key(struct cursor *cur,
    351 				     struct ndb_text_search_key *key)
    352 {
    353 	int timestamp;
    354 
    355 	if (!pull_varint(cur, &timestamp))
    356 		return 0;
    357 
    358 	if (!pull_varint(cur, &key->word_index))
    359 		return 0;
    360 
    361 	key->timestamp = timestamp;
    362 
    363 	return 1;
    364 }
    365 
    366 // unpack a fulltext search key
    367 //
    368 // full version of string + unpack remaining. This is split up because text
    369 // searching only requires to pull the string for prefix searching, and the
    370 // remaining is optional
    371 static inline int ndb_unpack_text_search_key(unsigned char *p, int len,
    372 				      struct ndb_text_search_key *key)
    373 {
    374 	struct cursor c;
    375 	make_cursor(p, p + len, &c);
    376 
    377 	if (!ndb_unpack_text_search_key_noteid(&c, &key->note_id))
    378 		return 0;
    379 
    380 	if (!ndb_unpack_text_search_key_string(&c, &key->str, &key->str_len))
    381 		return 0;
    382 
    383 	return ndb_unpack_remaining_text_search_key(&c, key);
    384 }
    385 
    386 // Copies only lowercase characters to the destination string and fills the rest with null bytes.
    387 // `dst` and `src` are pointers to the destination and source strings, respectively.
    388 // `n` is the maximum number of characters to copy.
    389 static void lowercase_strncpy(char *dst, const char *src, int n) {
    390 	int j = 0, i = 0;
    391 
    392 	if (!dst || !src || n == 0) {
    393 		return;
    394 	}
    395 
    396 	while (src[i] != '\0' && j < n) {
    397 		dst[j++] = tolower(src[i++]);
    398 	}
    399 
    400 	// Null-terminate and fill the destination string
    401 	while (j < n) {
    402 		dst[j++] = '\0';
    403 	}
    404 }
    405 
    406 int ndb_filter_init(struct ndb_filter *filter)
    407 {
    408 	struct cursor cur;
    409 	int page_size, elem_pages, data_pages, buf_size;
    410 
    411 	page_size = 4096; // assuming this, not a big deal if we're wrong
    412 	elem_pages = NDB_FILTER_PAGES / 4;
    413 	data_pages = NDB_FILTER_PAGES - elem_pages;
    414 	buf_size = page_size * NDB_FILTER_PAGES;
    415 
    416 	unsigned char *buf = malloc(buf_size);
    417 	if (!buf)
    418 		return 0;
    419 
    420 	// init memory arena for the cursor
    421 	make_cursor(buf, buf + buf_size, &cur);
    422 
    423 	cursor_slice(&cur, &filter->elem_buf, page_size * elem_pages);
    424 	cursor_slice(&cur, &filter->data_buf, page_size * data_pages);
    425 
    426 	// make sure we are fully allocated
    427 	assert(cur.p == cur.end);
    428 
    429 	// make sure elem_buf is the start of the buffer
    430 	assert(filter->elem_buf.start == cur.start);
    431 
    432 	filter->num_elements = 0;
    433 	filter->elements[0] = (struct ndb_filter_elements*) buf;
    434 	filter->current = NULL;
    435 
    436 	return 1;
    437 }
    438 
    439 void ndb_filter_reset(struct ndb_filter *filter)
    440 {
    441 	filter->num_elements = 0;
    442 	filter->elem_buf.p = filter->elem_buf.start;
    443 	filter->data_buf.p = filter->data_buf.start;
    444 	filter->current = NULL;
    445 }
    446 
    447 void ndb_filter_free(struct ndb_filter *filter)
    448 {
    449 	if (filter->elem_buf.start)
    450 		free(filter->elem_buf.start);
    451 
    452 	memset(filter, 0, sizeof(*filter));
    453 }
    454 
    455 static const char *ndb_filter_field_name(enum ndb_filter_fieldtype field)
    456 {
    457 	switch (field) {
    458 	case NDB_FILTER_IDS: return "ids";
    459 	case NDB_FILTER_AUTHORS: return "authors";
    460 	case NDB_FILTER_KINDS: return "kinds";
    461 	case NDB_FILTER_GENERIC: return "generic";
    462 	case NDB_FILTER_SINCE: return "since";
    463 	case NDB_FILTER_UNTIL: return "until";
    464 	case NDB_FILTER_LIMIT: return "limit";
    465 	}
    466 
    467 	return "unknown";
    468 }
    469 
    470 static int ndb_filter_start_field_impl(struct ndb_filter *filter, enum ndb_filter_fieldtype field, char generic)
    471 {
    472 	int i;
    473 	struct ndb_filter_elements *els, *el;
    474 
    475 	if (filter->current) {
    476 		fprintf(stderr, "ndb_filter_start_field: filter field already in progress, did you forget to call ndb_filter_end_field?\n");
    477 		return 0;
    478 	}
    479 
    480 	// you can only start and end fields once
    481 	for (i = 0; i < filter->num_elements; i++) {
    482 		el = filter->elements[i];
    483 		if (el->field.type == field) {
    484 			fprintf(stderr, "ndb_filter_start_field: field '%s' already exists\n",
    485 					ndb_filter_field_name(field));
    486 			return 0;
    487 		}
    488 	}
    489 
    490 	els = (struct ndb_filter_elements *) filter->elem_buf.p ;
    491 	filter->current = els;
    492 
    493 	// advance elem buffer to the variable data section
    494 	if (!cursor_skip(&filter->elem_buf, sizeof(struct ndb_filter_elements))) {
    495 		fprintf(stderr, "ndb_filter_start_field: '%s' oom (todo: realloc?)\n",
    496 				ndb_filter_field_name(field));
    497 		return 0;
    498 	}
    499 
    500 	els->field.type = field;
    501 	els->field.generic = generic;
    502 	els->field.elem_type = 0;
    503 	els->count = 0;
    504 
    505 	return 1;
    506 }
    507 
    508 int ndb_filter_start_field(struct ndb_filter *filter, enum ndb_filter_fieldtype field)
    509 {
    510 	return ndb_filter_start_field_impl(filter, field, 0);
    511 }
    512 
    513 int ndb_filter_start_generic_field(struct ndb_filter *filter, char tag)
    514 {
    515 	return ndb_filter_start_field_impl(filter, NDB_FILTER_GENERIC, tag);
    516 }
    517 
    518 static int ndb_filter_add_element(struct ndb_filter *filter, union ndb_filter_element el)
    519 {
    520 	unsigned char *data;
    521 	const char *str;
    522 
    523 	if (!filter->current)
    524 		return 0;
    525 
    526 	data = filter->data_buf.p;
    527 
    528 	switch (filter->current->field.type) {
    529 	case NDB_FILTER_IDS:
    530 	case NDB_FILTER_AUTHORS:
    531 		if (!cursor_push(&filter->data_buf, (unsigned char *)el.id, 32))
    532 			return 0;
    533 		el.id = data;
    534 		break;
    535 	case NDB_FILTER_KINDS:
    536 		break;
    537 	case NDB_FILTER_SINCE:
    538 	case NDB_FILTER_UNTIL:
    539 	case NDB_FILTER_LIMIT:
    540 		// only one allowed for since/until
    541 		if (filter->current->count != 0)
    542 			return 0;
    543 		break;
    544 	case NDB_FILTER_GENERIC:
    545 		str = (const char *)filter->data_buf.p;
    546 		if (!cursor_push_c_str(&filter->data_buf, el.string))
    547 			return 0;
    548 		// push a pointer of the string in the databuf as an element
    549 		el.string = str;
    550 		break;
    551 	}
    552 
    553 	if (!cursor_push(&filter->elem_buf, (unsigned char*)&el, sizeof(el)))
    554 		return 0;
    555 
    556 	filter->current->count++;
    557 
    558 	return 1;
    559 }
    560 
    561 static int ndb_filter_set_elem_type(struct ndb_filter *filter,
    562 				    enum ndb_generic_element_type elem_type)
    563 {
    564 	enum ndb_generic_element_type current_elem_type;
    565 
    566 	if (!filter->current)
    567 		return 0;
    568 
    569 	current_elem_type = filter->current->field.elem_type;
    570 
    571 	// element types must be uniform
    572 	if (current_elem_type != elem_type && current_elem_type != NDB_ELEMENT_UNKNOWN) {
    573 		fprintf(stderr, "ndb_filter_set_elem_type: element types must be uniform\n");
    574 		return 0;
    575 	}
    576 
    577 	filter->current->field.elem_type = elem_type;
    578 
    579 	return 1;
    580 }
    581 
    582 int ndb_filter_add_str_element(struct ndb_filter *filter, const char *str)
    583 {
    584 	union ndb_filter_element el;
    585 
    586 	if (!filter->current)
    587 		return 0;
    588 
    589 	// only generic queries are allowed to have strings
    590 	switch (filter->current->field.type) {
    591 	case NDB_FILTER_SINCE:
    592 	case NDB_FILTER_UNTIL:
    593 	case NDB_FILTER_LIMIT:
    594 	case NDB_FILTER_IDS:
    595 	case NDB_FILTER_AUTHORS:
    596 	case NDB_FILTER_KINDS:
    597 		return 0;
    598 	case NDB_FILTER_GENERIC:
    599 		break;
    600 	}
    601 
    602 	if (!ndb_filter_set_elem_type(filter, NDB_ELEMENT_STRING))
    603 		return 0;
    604 
    605 	el.string = str;
    606 	return ndb_filter_add_element(filter, el);
    607 }
    608 
    609 int ndb_filter_add_int_element(struct ndb_filter *filter, uint64_t integer)
    610 {
    611 	union ndb_filter_element el;
    612 	if (!filter->current)
    613 		return 0;
    614 
    615 	switch (filter->current->field.type) {
    616 	case NDB_FILTER_IDS:
    617 	case NDB_FILTER_AUTHORS:
    618 	case NDB_FILTER_GENERIC:
    619 		return 0;
    620 	case NDB_FILTER_KINDS:
    621 	case NDB_FILTER_SINCE:
    622 	case NDB_FILTER_UNTIL:
    623 	case NDB_FILTER_LIMIT:
    624 		break;
    625 	}
    626 
    627 	el.integer = integer;
    628 
    629 	return ndb_filter_add_element(filter, el);
    630 }
    631 
    632 int ndb_filter_add_id_element(struct ndb_filter *filter, const unsigned char *id)
    633 {
    634 	union ndb_filter_element el;
    635 
    636 	if (!filter->current)
    637 		return 0;
    638 
    639 	// only certain filter types allow pushing id elements
    640 	switch (filter->current->field.type) {
    641 	case NDB_FILTER_SINCE:
    642 	case NDB_FILTER_UNTIL:
    643 	case NDB_FILTER_LIMIT:
    644 		return 0;
    645 	case NDB_FILTER_IDS:
    646 	case NDB_FILTER_AUTHORS:
    647 	case NDB_FILTER_KINDS:
    648 	case NDB_FILTER_GENERIC:
    649 		break;
    650 	}
    651 
    652 	if (!ndb_filter_set_elem_type(filter, NDB_ELEMENT_ID))
    653 		return 0;
    654 
    655 	// this is needed so that generic filters know its an id
    656 	el.id = id;
    657 
    658 	return ndb_filter_add_element(filter, el);
    659 }
    660 
    661 // TODO: build a hashtable so this is O(1)
    662 static int ndb_generic_filter_matches(struct ndb_filter_elements *els,
    663 				      struct ndb_note *note)
    664 {
    665 	int i;
    666 	union ndb_filter_element el;
    667 	struct ndb_iterator iter, *it = &iter;
    668 	struct ndb_str str;
    669 
    670 	ndb_tags_iterate_start(note, it);
    671 
    672 	while (ndb_tags_iterate_next(it)) {
    673 		// we're looking for tags with 2 or more entries: ["p", id], etc
    674 		if (it->tag->count < 2)
    675 			continue;
    676 
    677 		str = ndb_note_str(note, &it->tag->strs[0]);
    678 
    679 		// we only care about packed strings (single char, etc)
    680 		if (str.flag != NDB_PACKED_STR)
    681 			continue;
    682 
    683 		// do we have #e matching e (or p, etc)
    684 		if (str.str[0] != els->field.generic || str.str[1] != 0)
    685 			continue;
    686 
    687 		str = ndb_note_str(note, &it->tag->strs[1]);
    688 
    689 		switch (els->field.elem_type) {
    690 		case NDB_ELEMENT_ID:
    691 			// if our filter element type is an id, then we
    692 			// expect a packed id in the tag, otherwise skip
    693 			if (str.flag != NDB_PACKED_ID)
    694 				continue;
    695 			break;
    696 		case NDB_ELEMENT_STRING:
    697 			// if our filter element type is a string, then
    698 			// we should not expect an id
    699 			if (str.flag == NDB_PACKED_ID)
    700 				continue;
    701 			break;
    702 		case NDB_ELEMENT_UNKNOWN:
    703 		default:
    704 			// For some reason the element type is not set. It's
    705 			// possible nothing was added to the generic filter?
    706 			// Let's just fail here and log a note for debugging
    707 			fprintf(stderr, "UNUSUAL ndb_generic_filter_matches: have unknown element type %d\n", els->field.elem_type);
    708 			return 0;
    709 		}
    710 
    711 		for (i = 0; i < els->count; i++) {
    712 			el = els->elements[i];
    713 			switch (els->field.elem_type) {
    714 			case NDB_ELEMENT_ID:
    715 				if (!memcmp(el.id, str.id, 32))
    716 					return 1;
    717 				break;
    718 			case NDB_ELEMENT_STRING:
    719 				if (!strcmp(el.string, str.str))
    720 					return 1;
    721 				break;
    722 			case NDB_ELEMENT_UNKNOWN:
    723 				return 0;
    724 			}
    725 		}
    726 	}
    727 
    728 	return 0;
    729 }
    730 
    731 // returns 1 if a filter matches a note
    732 int ndb_filter_matches(struct ndb_filter *filter, struct ndb_note *note)
    733 {
    734 	int i, j;
    735 	struct ndb_filter_elements *els;
    736 
    737 	for (i = 0; i < filter->num_elements; i++) {
    738 		els = filter->elements[i];
    739 
    740 		switch (els->field.type) {
    741 		case NDB_FILTER_KINDS:
    742 			for (j = 0; j < els->count; j++) {
    743 				if ((unsigned int)els->elements[j].integer == note->kind)
    744 					goto cont;
    745 			}
    746 			break;
    747 		// TODO: add filter hashtable for large id lists
    748 		case NDB_FILTER_IDS:
    749 			for (j = 0; j < els->count; j++) {
    750 				if (!memcmp(els->elements[j].id, note->id, 32))
    751 					goto cont;
    752 			}
    753 			break;
    754 		case NDB_FILTER_AUTHORS:
    755 			for (j = 0; j < els->count; j++) {
    756 				if (!memcmp(els->elements[j].id, note->pubkey, 32))
    757 					goto cont;
    758 			}
    759 			break;
    760 		case NDB_FILTER_GENERIC:
    761 			if (ndb_generic_filter_matches(els, note))
    762 				continue;
    763 			break;
    764 		case NDB_FILTER_SINCE:
    765 			assert(els->count == 1);
    766 			if (note->created_at >= els->elements[0].integer)
    767 				continue;
    768 			break;
    769 		case NDB_FILTER_UNTIL:
    770 			assert(els->count == 1);
    771 			if (note->created_at < els->elements[0].integer)
    772 				continue;
    773 		case NDB_FILTER_LIMIT:
    774 cont:
    775 			continue;
    776 		}
    777 
    778 		// all need to match
    779 		return 0;
    780 	}
    781 
    782 	return 1;
    783 }
    784 
    785 void ndb_filter_end_field(struct ndb_filter *filter)
    786 {
    787 	filter->elements[filter->num_elements++] = filter->current;
    788 	filter->current = NULL;
    789 }
    790 
    791 static void ndb_make_search_key(struct ndb_search_key *key, unsigned char *id,
    792 			        uint64_t timestamp, const char *search)
    793 {
    794 	memcpy(key->id, id, 32);
    795 	key->timestamp = timestamp;
    796 	lowercase_strncpy(key->search, search, sizeof(key->search) - 1);
    797 	key->search[sizeof(key->search) - 1] = '\0';
    798 }
    799 
    800 static int ndb_write_profile_search_index(struct ndb_txn *txn,
    801 					  struct ndb_search_key *index_key,
    802 					  uint64_t profile_key)
    803 {
    804 	int rc;
    805 	MDB_val key, val;
    806 	
    807 	key.mv_data = index_key;
    808 	key.mv_size = sizeof(*index_key);
    809 	val.mv_data = &profile_key;
    810 	val.mv_size = sizeof(profile_key);
    811 
    812 	if ((rc = mdb_put(txn->mdb_txn, txn->lmdb->dbs[NDB_DB_PROFILE_SEARCH],
    813 			  &key, &val, 0)))
    814 	{
    815 		ndb_debug("ndb_write_profile_search_index failed: %s\n",
    816 			  mdb_strerror(rc));
    817 		return 0;
    818 	}
    819 
    820 	return 1;
    821 }
    822 
    823 
    824 // map usernames and display names to profile keys for user searching
    825 static int ndb_write_profile_search_indices(struct ndb_txn *txn,
    826 					    struct ndb_note *note,
    827 					    uint64_t profile_key,
    828 					    void *profile_root)
    829 {
    830 	struct ndb_search_key index;
    831 	NdbProfileRecord_table_t profile_record;
    832 	NdbProfile_table_t profile;
    833 
    834 	profile_record = NdbProfileRecord_as_root(profile_root);
    835 	profile = NdbProfileRecord_profile_get(profile_record);
    836 
    837 	const char *name = NdbProfile_name_get(profile);
    838 	const char *display_name = NdbProfile_display_name_get(profile);
    839 
    840 	// words + pubkey + created
    841 	if (name) {
    842 		ndb_make_search_key(&index, note->pubkey, note->created_at,
    843 				    name);
    844 		if (!ndb_write_profile_search_index(txn, &index, profile_key))
    845 			return 0;
    846 	}
    847 
    848 	if (display_name) {
    849 		// don't write the same name/display_name twice
    850 		if (name && !strcmp(display_name, name)) {
    851 			return 1;
    852 		}
    853 		ndb_make_search_key(&index, note->pubkey, note->created_at,
    854 				    display_name);
    855 		if (!ndb_write_profile_search_index(txn, &index, profile_key))
    856 			return 0;
    857 	}
    858 
    859 	return 1;
    860 }
    861 
    862 
    863 static int _ndb_begin_query(struct ndb *ndb, struct ndb_txn *txn, int flags)
    864 {
    865 	txn->lmdb = &ndb->lmdb;
    866 	MDB_txn **mdb_txn = (MDB_txn **)&txn->mdb_txn;
    867     if (!txn->lmdb->env)
    868         return 0;
    869     return mdb_txn_begin(txn->lmdb->env, NULL, flags, mdb_txn) == 0;
    870 }
    871 
    872 int ndb_begin_query(struct ndb *ndb, struct ndb_txn *txn)
    873 {
    874 	return _ndb_begin_query(ndb, txn, MDB_RDONLY);
    875 }
    876 
    877 // this should only be used in migrations, etc
    878 static int ndb_begin_rw_query(struct ndb *ndb, struct ndb_txn *txn)
    879 {
    880 	return _ndb_begin_query(ndb, txn, 0);
    881 }
    882 
    883 
    884 // Migrations
    885 //
    886 
    887 static int ndb_migrate_user_search_indices(struct ndb *ndb)
    888 {
    889 	int rc;
    890 	MDB_cursor *cur;
    891 	MDB_val k, v;
    892 	void *profile_root;
    893 	NdbProfileRecord_table_t record;
    894 	struct ndb_txn txn;
    895 	struct ndb_note *note;
    896 	uint64_t note_key, profile_key;
    897 	size_t len;
    898 	int count;
    899 
    900 	if (!ndb_begin_rw_query(ndb, &txn)) {
    901 		fprintf(stderr, "ndb_migrate_user_search_indices: ndb_begin_rw_query failed\n");
    902 		return 0;
    903 	}
    904 
    905 	if ((rc = mdb_cursor_open(txn.mdb_txn, ndb->lmdb.dbs[NDB_DB_PROFILE], &cur))) {
    906 		fprintf(stderr, "ndb_migrate_user_search_indices: mdb_cursor_open failed, error %d\n", rc);
    907 		return 0;
    908 	}
    909 
    910 	count = 0;
    911 
    912 	// loop through all profiles and write search indices
    913 	while (mdb_cursor_get(cur, &k, &v, MDB_NEXT) == 0) {
    914 		profile_root = v.mv_data;
    915 		profile_key = *((uint64_t*)k.mv_data);
    916 		record = NdbProfileRecord_as_root(profile_root);
    917 		note_key = NdbProfileRecord_note_key(record);
    918 		note = ndb_get_note_by_key(&txn, note_key, &len);
    919 
    920 		if (note == NULL) {
    921 			fprintf(stderr, "ndb_migrate_user_search_indices: note lookup failed\n");
    922 			return 0;
    923 		}
    924 
    925 		if (!ndb_write_profile_search_indices(&txn, note, profile_key,
    926 						      profile_root)) {
    927 
    928 			fprintf(stderr, "ndb_migrate_user_search_indices: ndb_write_profile_search_indices failed\n");
    929 			return 0;
    930 		}
    931 
    932 		count++;
    933 	}
    934 
    935 	fprintf(stderr, "migrated %d profiles to include search indices\n", count);
    936 
    937 	mdb_cursor_close(cur);
    938 
    939 	ndb_end_query(&txn);
    940 
    941 	return 1;
    942 }
    943 
    944 static int ndb_migrate_lower_user_search_indices(struct ndb *ndb)
    945 {
    946 	MDB_txn *txn;
    947 
    948 	if (mdb_txn_begin(ndb->lmdb.env, NULL, 0, &txn)) {
    949 		fprintf(stderr, "ndb_migrate_lower_user_search_indices: ndb_txn_begin failed\n");
    950 		return 0;
    951 	}
    952 
    953 	// just drop the search db so we can rebuild it
    954 	if (mdb_drop(txn, ndb->lmdb.dbs[NDB_DB_PROFILE_SEARCH], 0)) {
    955 		fprintf(stderr, "ndb_migrate_lower_user_search_indices: mdb_drop failed\n");
    956 		return 0;
    957 	}
    958 
    959 	mdb_txn_commit(txn);
    960 
    961 	return ndb_migrate_user_search_indices(ndb);
    962 }
    963 
    964 int ndb_process_profile_note(struct ndb_note *note, struct ndb_profile_record_builder *profile);
    965 
    966 
    967 int ndb_db_version(struct ndb *ndb)
    968 {
    969 	int rc;
    970 	uint64_t version, version_key;
    971 	MDB_val k, v;
    972 	MDB_txn *txn;
    973 
    974 	version_key = NDB_META_KEY_VERSION;
    975 	k.mv_data = &version_key;
    976 	k.mv_size = sizeof(version_key);
    977 
    978 	if ((rc = mdb_txn_begin(ndb->lmdb.env, NULL, 0, &txn))) {
    979 		fprintf(stderr, "ndb_db_version: mdb_txn_begin failed, error %d\n", rc);
    980 		return -1;
    981 	}
    982 
    983 	if (mdb_get(txn, ndb->lmdb.dbs[NDB_DB_NDB_META], &k, &v)) {
    984 		version = -1;
    985 	} else {
    986 		if (v.mv_size != 8) {
    987 			fprintf(stderr, "run_migrations: invalid version size?");
    988 			return 0;
    989 		}
    990 		version = *((uint64_t*)v.mv_data);
    991 	}
    992 
    993 	mdb_txn_abort(txn);
    994 	return version;
    995 }
    996 
    997 // custom kind+timestamp comparison function. This is used by lmdb to perform
    998 // b+ tree searches over the kind+timestamp index
    999 static int ndb_u64_tsid_compare(const MDB_val *a, const MDB_val *b)
   1000 {
   1001 	struct ndb_u64_tsid *tsa, *tsb;
   1002 	tsa = a->mv_data;
   1003 	tsb = b->mv_data;
   1004 
   1005 	if (tsa->u64 < tsb->u64)
   1006 		return -1;
   1007 	else if (tsa->u64 > tsb->u64)
   1008 		return 1;
   1009 
   1010 	if (tsa->timestamp < tsb->timestamp)
   1011 		return -1;
   1012 	else if (tsa->timestamp > tsb->timestamp)
   1013 		return 1;
   1014 
   1015 	return 0;
   1016 }
   1017 
   1018 static int ndb_tsid_compare(const MDB_val *a, const MDB_val *b)
   1019 {
   1020 	struct ndb_tsid *tsa, *tsb;
   1021 	MDB_val a2 = *a, b2 = *b;
   1022 
   1023 	a2.mv_size = sizeof(tsa->id);
   1024 	b2.mv_size = sizeof(tsb->id);
   1025 
   1026 	int cmp = mdb_cmp_memn(&a2, &b2);
   1027 	if (cmp) return cmp;
   1028 
   1029 	tsa = a->mv_data;
   1030 	tsb = b->mv_data;
   1031 
   1032 	if (tsa->timestamp < tsb->timestamp)
   1033 		return -1;
   1034 	else if (tsa->timestamp > tsb->timestamp)
   1035 		return 1;
   1036 	return 0;
   1037 }
   1038 
   1039 static inline void ndb_tsid_low(struct ndb_tsid *key, unsigned char *id)
   1040 {
   1041 	memcpy(key->id, id, 32);
   1042 	key->timestamp = 0;
   1043 }
   1044 
   1045 static inline void ndb_tsid_init(struct ndb_tsid *key, unsigned char *id,
   1046 				 uint64_t timestamp)
   1047 {
   1048 	memcpy(key->id, id, 32);
   1049 	key->timestamp = timestamp;
   1050 }
   1051 
   1052 static inline void ndb_u64_tsid_init(struct ndb_u64_tsid *key, uint64_t integer,
   1053 				     uint64_t timestamp)
   1054 {
   1055 	key->u64 = integer;
   1056 	key->timestamp = timestamp;
   1057 }
   1058 
   1059 // useful for range-searching for the latest key with a clustered created_at timen
   1060 static inline void ndb_tsid_high(struct ndb_tsid *key, const unsigned char *id)
   1061 {
   1062 	memcpy(key->id, id, 32);
   1063 	key->timestamp = UINT64_MAX;
   1064 }
   1065 
   1066 enum ndb_ingester_msgtype {
   1067 	NDB_INGEST_EVENT, // write json to the ingester queue for processing 
   1068 	NDB_INGEST_QUIT,  // kill ingester thread immediately
   1069 };
   1070 
   1071 struct ndb_ingester_event {
   1072 	char *json;
   1073 	unsigned client : 1; // ["EVENT", {...}] messages
   1074 	unsigned len : 31;
   1075 };
   1076 
   1077 struct ndb_writer_note {
   1078 	struct ndb_note *note;
   1079 	size_t note_len;
   1080 };
   1081 
   1082 struct ndb_writer_profile {
   1083 	struct ndb_writer_note note;
   1084 	struct ndb_profile_record_builder record;
   1085 };
   1086 
   1087 struct ndb_ingester_msg {
   1088 	enum ndb_ingester_msgtype type;
   1089 	union {
   1090 		struct ndb_ingester_event event;
   1091 	};
   1092 };
   1093 
   1094 struct ndb_writer_ndb_meta {
   1095 	// these are 64 bit because I'm paranoid of db-wide alignment issues
   1096 	uint64_t version;
   1097 };
   1098 
   1099 // Used in the writer thread when writing ndb_profile_fetch_record's
   1100 //   kv = pubkey: recor
   1101 struct ndb_writer_last_fetch {
   1102 	unsigned char pubkey[32];
   1103 	uint64_t fetched_at;
   1104 };
   1105 
   1106 // The different types of messages that the writer thread can write to the
   1107 // database
   1108 struct ndb_writer_msg {
   1109 	enum ndb_writer_msgtype type;
   1110 	union {
   1111 		struct ndb_writer_note note;
   1112 		struct ndb_writer_profile profile;
   1113 		struct ndb_writer_ndb_meta ndb_meta;
   1114 		struct ndb_writer_last_fetch last_fetch;
   1115 	};
   1116 };
   1117 
   1118 static inline int ndb_writer_queue_msg(struct ndb_writer *writer,
   1119 				       struct ndb_writer_msg *msg)
   1120 {
   1121 	return prot_queue_push(&writer->inbox, msg);
   1122 }
   1123 
   1124 static int ndb_migrate_utf8_profile_names(struct ndb *ndb)
   1125 {
   1126 	int rc;
   1127 	MDB_cursor *cur;
   1128 	MDB_val k, v;
   1129 	void *profile_root;
   1130 	NdbProfileRecord_table_t record;
   1131 	struct ndb_txn txn;
   1132 	struct ndb_note *note, *copied_note;
   1133 	uint64_t note_key;
   1134 	size_t len;
   1135 	int count, failed;
   1136 	struct ndb_writer_msg out;
   1137 
   1138 	if (!ndb_begin_rw_query(ndb, &txn)) {
   1139 		fprintf(stderr, "ndb_migrate_utf8_profile_names: ndb_begin_rw_query failed\n");
   1140 		return 0;
   1141 	}
   1142 
   1143 	if ((rc = mdb_cursor_open(txn.mdb_txn, ndb->lmdb.dbs[NDB_DB_PROFILE], &cur))) {
   1144 		fprintf(stderr, "ndb_migrate_utf8_profile_names: mdb_cursor_open failed, error %d\n", rc);
   1145 		return 0;
   1146 	}
   1147 
   1148 	count = 0;
   1149 	failed = 0;
   1150 
   1151 	// loop through all profiles and write search indices
   1152 	while (mdb_cursor_get(cur, &k, &v, MDB_NEXT) == 0) {
   1153 		profile_root = v.mv_data;
   1154 		record = NdbProfileRecord_as_root(profile_root);
   1155 		note_key = NdbProfileRecord_note_key(record);
   1156 		note = ndb_get_note_by_key(&txn, note_key, &len);
   1157 
   1158 		if (note == NULL) {
   1159 			fprintf(stderr, "ndb_migrate_utf8_profile_names: note lookup failed\n");
   1160 			return 0;
   1161 		}
   1162 
   1163 		struct ndb_profile_record_builder *b = &out.profile.record;
   1164 
   1165 		// reprocess profile
   1166 		if (!ndb_process_profile_note(note, b)) {
   1167 			failed++;
   1168 			continue;
   1169 		}
   1170 
   1171 		// the writer needs to own this note, and its expected to free it
   1172 		copied_note = malloc(len);
   1173 		memcpy(copied_note, note, len);
   1174 
   1175 		out.type = NDB_WRITER_PROFILE;
   1176 		out.profile.note.note = copied_note;
   1177 		out.profile.note.note_len = len;
   1178 
   1179 		ndb_writer_queue_msg(&ndb->writer, &out);
   1180 
   1181 		count++;
   1182 	}
   1183 
   1184 	fprintf(stderr, "migrated %d profiles to fix utf8 profile names\n", count);
   1185 
   1186 	if (failed != 0) {
   1187 		fprintf(stderr, "failed to migrate %d profiles to fix utf8 profile names\n", failed);
   1188 	}
   1189 
   1190 	mdb_cursor_close(cur);
   1191 
   1192 	ndb_end_query(&txn);
   1193 
   1194 	return 1;
   1195 }
   1196 
   1197 static struct ndb_migration MIGRATIONS[] = {
   1198 	{ .fn = ndb_migrate_user_search_indices },
   1199 	{ .fn = ndb_migrate_lower_user_search_indices },
   1200 	{ .fn = ndb_migrate_utf8_profile_names }
   1201 };
   1202 
   1203 
   1204 int ndb_end_query(struct ndb_txn *txn)
   1205 {
   1206 	// this works on read or write queries. 
   1207 	return mdb_txn_commit(txn->mdb_txn) == 0;
   1208 }
   1209 
   1210 int ndb_note_verify(void *ctx, unsigned char pubkey[32], unsigned char id[32],
   1211 		    unsigned char sig[64])
   1212 {
   1213 	secp256k1_xonly_pubkey xonly_pubkey;
   1214 	int ok;
   1215 
   1216 	ok = secp256k1_xonly_pubkey_parse((secp256k1_context*)ctx, &xonly_pubkey,
   1217 					  pubkey) != 0;
   1218 	if (!ok) return 0;
   1219 
   1220 	ok = secp256k1_schnorrsig_verify((secp256k1_context*)ctx, sig, id, 32,
   1221 					 &xonly_pubkey) > 0;
   1222 	if (!ok) return 0;
   1223 
   1224 	return 1;
   1225 }
   1226 
   1227 static inline int ndb_writer_queue_msgs(struct ndb_writer *writer,
   1228 					struct ndb_writer_msg *msgs,
   1229 					int num_msgs)
   1230 {
   1231 	return prot_queue_push_all(&writer->inbox, msgs, num_msgs);
   1232 }
   1233 
   1234 static int ndb_writer_queue_note(struct ndb_writer *writer,
   1235 				 struct ndb_note *note, size_t note_len)
   1236 {
   1237 	struct ndb_writer_msg msg;
   1238 	msg.type = NDB_WRITER_NOTE;
   1239 
   1240 	msg.note.note = note;
   1241 	msg.note.note_len = note_len;
   1242 
   1243 	return prot_queue_push(&writer->inbox, &msg);
   1244 }
   1245 
   1246 static void ndb_writer_last_profile_fetch(struct ndb_txn *txn,
   1247 					  const unsigned char *pubkey,
   1248 					  uint64_t fetched_at)
   1249 {
   1250 	int rc;
   1251 	MDB_val key, val;
   1252 	
   1253 	key.mv_data = (unsigned char*)pubkey;
   1254 	key.mv_size = 32;
   1255 	val.mv_data = &fetched_at;
   1256 	val.mv_size = sizeof(fetched_at);
   1257 
   1258 	if ((rc = mdb_put(txn->mdb_txn, txn->lmdb->dbs[NDB_DB_PROFILE_LAST_FETCH],
   1259 			  &key, &val, 0)))
   1260 	{
   1261 		ndb_debug("write version to ndb_meta failed: %s\n",
   1262 				mdb_strerror(rc));
   1263 		return;
   1264 	}
   1265 
   1266 	//fprintf(stderr, "writing version %" PRIu64 "\n", version);
   1267 }
   1268 
   1269 
   1270 // We just received a profile that we haven't processed yet, but it could
   1271 // be an older one! Make sure we only write last fetched profile if it's a new
   1272 // one
   1273 //
   1274 // To do this, we first check the latest profile in the database. If the
   1275 // created_date for this profile note is newer, then we write a
   1276 // last_profile_fetch record, otherwise we do not.
   1277 //
   1278 // WARNING: This function is only valid when called from the writer thread
   1279 static int ndb_maybe_write_last_profile_fetch(struct ndb_txn *txn,
   1280 					       struct ndb_note *note)
   1281 {
   1282 	size_t len;
   1283 	uint64_t profile_key, note_key;
   1284 	void *root;
   1285 	struct ndb_note *last_profile;
   1286 	NdbProfileRecord_table_t record;
   1287 
   1288 	if ((root = ndb_get_profile_by_pubkey(txn, note->pubkey, &len, &profile_key))) {
   1289 		record = NdbProfileRecord_as_root(root);
   1290 		note_key = NdbProfileRecord_note_key(record);
   1291 		last_profile = ndb_get_note_by_key(txn, note_key, &len);
   1292 		if (last_profile == NULL) {
   1293 			return 0;
   1294 		}
   1295 
   1296 		// found profile, let's see if it's newer than ours
   1297 		if (note->created_at > last_profile->created_at) {
   1298 			// this is a new profile note, record last fetched time
   1299 			ndb_writer_last_profile_fetch(txn, note->pubkey, time(NULL));
   1300 		}
   1301 	} else {
   1302 		// couldn't fetch profile. record last fetched time
   1303 		ndb_writer_last_profile_fetch(txn, note->pubkey, time(NULL));
   1304 	}
   1305 
   1306 	return 1;
   1307 }
   1308 
   1309 int ndb_write_last_profile_fetch(struct ndb *ndb, const unsigned char *pubkey,
   1310 				 uint64_t fetched_at)
   1311 {
   1312 	struct ndb_writer_msg msg;
   1313 	msg.type = NDB_WRITER_PROFILE_LAST_FETCH;
   1314 	memcpy(&msg.last_fetch.pubkey[0], pubkey, 32);
   1315 	msg.last_fetch.fetched_at = fetched_at;
   1316 
   1317 	return ndb_writer_queue_msg(&ndb->writer, &msg);
   1318 }
   1319 
   1320 // get some value based on a clustered id key
   1321 int ndb_get_tsid(struct ndb_txn *txn, enum ndb_dbs db, const unsigned char *id,
   1322 		 MDB_val *val)
   1323 {
   1324 	MDB_val k, v;
   1325 	MDB_cursor *cur;
   1326 	int success = 0, rc;
   1327 	struct ndb_tsid tsid;
   1328 
   1329 	// position at the most recent
   1330 	ndb_tsid_high(&tsid, id);
   1331 
   1332 	k.mv_data = &tsid;
   1333 	k.mv_size = sizeof(tsid);
   1334 
   1335 	if ((rc = mdb_cursor_open(txn->mdb_txn, txn->lmdb->dbs[db], &cur))) {
   1336 		ndb_debug("ndb_get_tsid: failed to open cursor: '%s'\n", mdb_strerror(rc));
   1337 		return 0;
   1338 	}
   1339 
   1340 	// Position cursor at the next key greater than or equal to the specified key
   1341 	if (mdb_cursor_get(cur, &k, &v, MDB_SET_RANGE)) {
   1342 		// Failed :(. It could be the last element?
   1343 		if (mdb_cursor_get(cur, &k, &v, MDB_LAST))
   1344 			goto cleanup;
   1345 	} else {
   1346 		// if set range worked and our key exists, it should be
   1347 		// the one right before this one
   1348 		if (mdb_cursor_get(cur, &k, &v, MDB_PREV))
   1349 			goto cleanup;
   1350 	}
   1351 
   1352 	if (memcmp(k.mv_data, id, 32) == 0) {
   1353 		*val = v;
   1354 		success = 1;
   1355 	}
   1356 
   1357 cleanup:
   1358 	mdb_cursor_close(cur);
   1359 	return success;
   1360 }
   1361 
   1362 static void *ndb_lookup_by_key(struct ndb_txn *txn, uint64_t key,
   1363 			       enum ndb_dbs store, size_t *len)
   1364 {
   1365 	MDB_val k, v;
   1366 
   1367 	k.mv_data = &key;
   1368 	k.mv_size = sizeof(key);
   1369 
   1370 	if (mdb_get(txn->mdb_txn, txn->lmdb->dbs[store], &k, &v)) {
   1371 		ndb_debug("ndb_get_profile_by_pubkey: mdb_get note failed\n");
   1372 		return NULL;
   1373 	}
   1374 
   1375 	if (len)
   1376 		*len = v.mv_size;
   1377 
   1378 	return v.mv_data;
   1379 }
   1380 
   1381 static void *ndb_lookup_tsid(struct ndb_txn *txn, enum ndb_dbs ind,
   1382 			     enum ndb_dbs store, const unsigned char *pk,
   1383 			     size_t *len, uint64_t *primkey)
   1384 {
   1385 	MDB_val k, v;
   1386 	void *res = NULL;
   1387 	if (len)
   1388 		*len = 0;
   1389 
   1390 	if (!ndb_get_tsid(txn, ind, pk, &k)) {
   1391 		//ndb_debug("ndb_get_profile_by_pubkey: ndb_get_tsid failed\n");
   1392 		return 0;
   1393 	}
   1394 
   1395 	if (primkey)
   1396 		*primkey = *(uint64_t*)k.mv_data;
   1397 
   1398 	if (mdb_get(txn->mdb_txn, txn->lmdb->dbs[store], &k, &v)) {
   1399 		ndb_debug("ndb_get_profile_by_pubkey: mdb_get note failed\n");
   1400 		return 0;
   1401 	}
   1402 
   1403 	res = v.mv_data;
   1404 	assert(((uint64_t)res % 4) == 0);
   1405 	if (len)
   1406 		*len = v.mv_size;
   1407 	return res;
   1408 }
   1409 
   1410 void *ndb_get_profile_by_pubkey(struct ndb_txn *txn, const unsigned char *pk, size_t *len, uint64_t *key)
   1411 {
   1412 	return ndb_lookup_tsid(txn, NDB_DB_PROFILE_PK, NDB_DB_PROFILE, pk, len, key);
   1413 }
   1414 
   1415 struct ndb_note *ndb_get_note_by_id(struct ndb_txn *txn, const unsigned char *id, size_t *len, uint64_t *key)
   1416 {
   1417 	return ndb_lookup_tsid(txn, NDB_DB_NOTE_ID, NDB_DB_NOTE, id, len, key);
   1418 }
   1419 
   1420 static inline uint64_t ndb_get_indexkey_by_id(struct ndb_txn *txn,
   1421 					      enum ndb_dbs db,
   1422 					      const unsigned char *id)
   1423 {
   1424 	MDB_val k;
   1425 
   1426 	if (!ndb_get_tsid(txn, db, id, &k))
   1427 		return 0;
   1428 
   1429 	return *(uint32_t*)k.mv_data;
   1430 }
   1431 
   1432 uint64_t ndb_get_notekey_by_id(struct ndb_txn *txn, const unsigned char *id)
   1433 {
   1434 	return ndb_get_indexkey_by_id(txn, NDB_DB_NOTE_ID, id);
   1435 }
   1436 
   1437 uint64_t ndb_get_profilekey_by_pubkey(struct ndb_txn *txn, const unsigned char *id)
   1438 {
   1439 	return ndb_get_indexkey_by_id(txn, NDB_DB_PROFILE_PK, id);
   1440 }
   1441 
   1442 struct ndb_note *ndb_get_note_by_key(struct ndb_txn *txn, uint64_t key, size_t *len)
   1443 {
   1444 	return ndb_lookup_by_key(txn, key, NDB_DB_NOTE, len);
   1445 }
   1446 
   1447 void *ndb_get_profile_by_key(struct ndb_txn *txn, uint64_t key, size_t *len)
   1448 {
   1449 	return ndb_lookup_by_key(txn, key, NDB_DB_PROFILE, len);
   1450 }
   1451 
   1452 uint64_t
   1453 ndb_read_last_profile_fetch(struct ndb_txn *txn, const unsigned char *pubkey)
   1454 {
   1455 	MDB_val k, v;
   1456 
   1457 	k.mv_data = (unsigned char*)pubkey;
   1458 	k.mv_size = 32;
   1459 
   1460 	if (mdb_get(txn->mdb_txn, txn->lmdb->dbs[NDB_DB_PROFILE_LAST_FETCH], &k, &v)) {
   1461 		//ndb_debug("ndb_read_last_profile_fetch: mdb_get note failed\n");
   1462 		return 0;
   1463 	}
   1464 
   1465 	return *((uint64_t*)v.mv_data);
   1466 }
   1467 
   1468 
   1469 static int ndb_has_note(struct ndb_txn *txn, const unsigned char *id)
   1470 {
   1471 	MDB_val val;
   1472 
   1473 	if (!ndb_get_tsid(txn, NDB_DB_NOTE_ID, id, &val))
   1474 		return 0;
   1475 
   1476 	return 1;
   1477 }
   1478 
   1479 static void ndb_txn_from_mdb(struct ndb_txn *txn, struct ndb_lmdb *lmdb,
   1480 			     MDB_txn *mdb_txn)
   1481 {
   1482 	txn->lmdb = lmdb;
   1483 	txn->mdb_txn = mdb_txn;
   1484 }
   1485 
   1486 static enum ndb_idres ndb_ingester_json_controller(void *data, const char *hexid)
   1487 {
   1488 	unsigned char id[32];
   1489 	struct ndb_ingest_controller *c = data;
   1490 	struct ndb_txn txn;
   1491 
   1492 	hex_decode(hexid, 64, id, sizeof(id));
   1493 
   1494 	// let's see if we already have it
   1495 
   1496 	ndb_txn_from_mdb(&txn, c->lmdb, c->read_txn);
   1497 	if (!ndb_has_note(&txn, id))
   1498 		return NDB_IDRES_CONT;
   1499 
   1500 	return NDB_IDRES_STOP;
   1501 }
   1502 
   1503 static int ndbprofile_parse_json(flatcc_builder_t *B,
   1504         const char *buf, size_t bufsiz, int flags, NdbProfile_ref_t *profile)
   1505 {
   1506 	flatcc_json_parser_t parser, *ctx = &parser;
   1507 	flatcc_json_parser_init(ctx, B, buf, buf + bufsiz, flags);
   1508 
   1509 	if (flatcc_builder_start_buffer(B, 0, 0, 0))
   1510 		return 0;
   1511 
   1512 	NdbProfile_parse_json_table(ctx, buf, buf + bufsiz, profile);
   1513 	if (ctx->error)
   1514 		return 0;
   1515  
   1516 	if (!flatcc_builder_end_buffer(B, *profile))
   1517 		return 0;
   1518 
   1519 	ctx->end_loc = buf;
   1520 
   1521 
   1522 	return 1;
   1523 }
   1524 
   1525 void ndb_profile_record_builder_init(struct ndb_profile_record_builder *b)
   1526 {
   1527 	b->builder = malloc(sizeof(*b->builder));
   1528 	b->flatbuf = NULL;
   1529 }
   1530 
   1531 void ndb_profile_record_builder_free(struct ndb_profile_record_builder *b)
   1532 {
   1533 	if (b->builder)
   1534 		free(b->builder);
   1535 	if (b->flatbuf)
   1536 		free(b->flatbuf);
   1537 
   1538 	b->builder = NULL;
   1539 	b->flatbuf = NULL;
   1540 }
   1541 
   1542 int ndb_process_profile_note(struct ndb_note *note,
   1543 			     struct ndb_profile_record_builder *profile)
   1544 {
   1545 	int res;
   1546 
   1547 	NdbProfile_ref_t profile_table;
   1548 	flatcc_builder_t *builder;
   1549 
   1550 	ndb_profile_record_builder_init(profile);
   1551 	builder = profile->builder;
   1552 	flatcc_builder_init(builder);
   1553 
   1554 	NdbProfileRecord_start_as_root(builder);
   1555 
   1556 	//printf("parsing profile '%.*s'\n", note->content_length, ndb_note_content(note));
   1557 	if (!(res = ndbprofile_parse_json(builder, ndb_note_content(note),
   1558 					  note->content_length,
   1559 					  flatcc_json_parser_f_skip_unknown,
   1560 					  &profile_table)))
   1561 	{
   1562 		ndb_debug("profile_parse_json failed %d '%.*s'\n", res,
   1563 			  note->content_length, ndb_note_content(note));
   1564 		ndb_profile_record_builder_free(profile);
   1565 		return 0;
   1566 	}
   1567 
   1568 	uint64_t received_at = time(NULL);
   1569 	const char *lnurl = "fixme";
   1570 
   1571 	NdbProfileRecord_profile_add(builder, profile_table);
   1572 	NdbProfileRecord_received_at_add(builder, received_at);
   1573 
   1574 	flatcc_builder_ref_t lnurl_off;
   1575 	lnurl_off = flatcc_builder_create_string_str(builder, lnurl);
   1576 
   1577 	NdbProfileRecord_lnurl_add(builder, lnurl_off);
   1578 
   1579 	//*profile = flatcc_builder_finalize_aligned_buffer(builder, profile_len);
   1580 	return 1;
   1581 }
   1582 
   1583 static int ndb_ingester_process_note(secp256k1_context *ctx,
   1584 				     struct ndb_note *note,
   1585 				     size_t note_size,
   1586 				     struct ndb_writer_msg *out,
   1587 				     struct ndb_ingester *ingester)
   1588 {
   1589 	enum ndb_ingest_filter_action action;
   1590 	action = NDB_INGEST_ACCEPT;
   1591 
   1592 	if (ingester->filter)
   1593 		action = ingester->filter(ingester->filter_context, note);
   1594 
   1595 	if (action == NDB_INGEST_REJECT)
   1596 		return 0;
   1597 
   1598 	// some special situations we might want to skip sig validation,
   1599 	// like during large imports
   1600 	if (action == NDB_INGEST_SKIP_VALIDATION || (ingester->flags & NDB_FLAG_SKIP_NOTE_VERIFY)) {
   1601 		// if we're skipping validation we don't need to verify
   1602 	} else {
   1603 		// verify! If it's an invalid note we don't need to
   1604 		// bother writing it to the database
   1605 		if (!ndb_note_verify(ctx, note->pubkey, note->id, note->sig)) {
   1606 			ndb_debug("signature verification failed\n");
   1607 			return 0;
   1608 		}
   1609 	}
   1610 
   1611 	// we didn't find anything. let's send it
   1612 	// to the writer thread
   1613 	note = realloc(note, note_size);
   1614 	assert(((uint64_t)note % 4) == 0);
   1615 
   1616 	if (note->kind == 0) {
   1617 		struct ndb_profile_record_builder *b = 
   1618 			&out->profile.record;
   1619 
   1620 		ndb_process_profile_note(note, b);
   1621 
   1622 		out->type = NDB_WRITER_PROFILE;
   1623 		out->profile.note.note = note;
   1624 		out->profile.note.note_len = note_size;
   1625 	} else {
   1626 		out->type = NDB_WRITER_NOTE;
   1627 		out->note.note = note;
   1628 		out->note.note_len = note_size;
   1629 	}
   1630 
   1631 	return 1;
   1632 }
   1633 
   1634 
   1635 static int ndb_ingester_process_event(secp256k1_context *ctx,
   1636 				      struct ndb_ingester *ingester,
   1637 				      struct ndb_ingester_event *ev,
   1638 				      struct ndb_writer_msg *out,
   1639 				      MDB_txn *read_txn
   1640 				      )
   1641 {
   1642 	struct ndb_tce tce;
   1643 	struct ndb_fce fce;
   1644 	struct ndb_note *note;
   1645 	struct ndb_ingest_controller controller;
   1646 	struct ndb_id_cb cb;
   1647 	void *buf;
   1648 	int ok;
   1649 	size_t bufsize, note_size;
   1650 
   1651 	ok = 0;
   1652 
   1653 	// we will use this to check if we already have it in the DB during
   1654 	// ID parsing
   1655 	controller.read_txn = read_txn;
   1656 	controller.lmdb = ingester->writer->lmdb;
   1657 	cb.fn = ndb_ingester_json_controller;
   1658 	cb.data = &controller;
   1659 
   1660 	// since we're going to be passing this allocated note to a different
   1661 	// thread, we can't use thread-local buffers. just allocate a block
   1662         bufsize = max(ev->len * 8.0, 4096);
   1663 	buf = malloc(bufsize);
   1664 	if (!buf) {
   1665 		ndb_debug("couldn't malloc buf\n");
   1666 		return 0;
   1667 	}
   1668 
   1669 	note_size =
   1670 		ev->client ? 
   1671 		ndb_client_event_from_json(ev->json, ev->len, &fce, buf, bufsize, &cb) :
   1672 		ndb_ws_event_from_json(ev->json, ev->len, &tce, buf, bufsize, &cb);
   1673 
   1674 	if (note_size == -42) {
   1675 		// we already have this!
   1676 		//ndb_debug("already have id??\n");
   1677 		goto cleanup;
   1678 	} else if (note_size == 0) {
   1679 		ndb_debug("failed to parse '%.*s'\n", ev->len, ev->json);
   1680 		goto cleanup;
   1681 	}
   1682 
   1683 	//ndb_debug("parsed evtype:%d '%.*s'\n", tce.evtype, ev->len, ev->json);
   1684 
   1685 	if (ev->client) {
   1686 		switch (fce.evtype) {
   1687 		case NDB_FCE_EVENT:
   1688 			note = fce.event.note;
   1689 			if (note != buf) {
   1690 				ndb_debug("note buffer not equal to malloc'd buffer\n");
   1691 				goto cleanup;
   1692 			}
   1693 
   1694 			if (!ndb_ingester_process_note(ctx, note, note_size,
   1695 						       out, ingester)) {
   1696 				goto cleanup;
   1697 			} else {
   1698 				// we're done with the original json, free it
   1699 				free(ev->json);
   1700 				return 1;
   1701 			}
   1702 		}
   1703 	} else {
   1704 		switch (tce.evtype) {
   1705 		case NDB_TCE_NOTICE: goto cleanup;
   1706 		case NDB_TCE_EOSE:   goto cleanup;
   1707 		case NDB_TCE_OK:     goto cleanup;
   1708 		case NDB_TCE_EVENT:
   1709 			note = tce.event.note;
   1710 			if (note != buf) {
   1711 				ndb_debug("note buffer not equal to malloc'd buffer\n");
   1712 				goto cleanup;
   1713 			}
   1714 
   1715 			if (!ndb_ingester_process_note(ctx, note, note_size,
   1716 						       out, ingester)) {
   1717 				goto cleanup;
   1718 			} else {
   1719 				// we're done with the original json, free it
   1720 				free(ev->json);
   1721 				return 1;
   1722 			}
   1723 		}
   1724 	}
   1725 
   1726 
   1727 cleanup:
   1728 	free(ev->json);
   1729 	free(buf);
   1730 
   1731 	return ok;
   1732 }
   1733 
   1734 static uint64_t ndb_get_last_key(MDB_txn *txn, MDB_dbi db)
   1735 {
   1736 	MDB_cursor *mc;
   1737 	MDB_val key, val;
   1738 
   1739 	if (mdb_cursor_open(txn, db, &mc))
   1740 		return 0;
   1741 
   1742 	if (mdb_cursor_get(mc, &key, &val, MDB_LAST)) {
   1743 		mdb_cursor_close(mc);
   1744 		return 0;
   1745 	}
   1746 
   1747 	mdb_cursor_close(mc);
   1748 
   1749 	assert(key.mv_size == 8);
   1750         return *((uint64_t*)key.mv_data);
   1751 }
   1752 
   1753 //
   1754 // make a search key meant for user queries without any other note info
   1755 static void ndb_make_search_key_low(struct ndb_search_key *key, const char *search)
   1756 {
   1757 	memset(key->id, 0, sizeof(key->id));
   1758 	key->timestamp = 0;
   1759 	lowercase_strncpy(key->search, search, sizeof(key->search) - 1);
   1760 	key->search[sizeof(key->search) - 1] = '\0';
   1761 }
   1762 
   1763 int ndb_search_profile(struct ndb_txn *txn, struct ndb_search *search, const char *query)
   1764 {
   1765 	int rc;
   1766 	struct ndb_search_key s;
   1767 	MDB_val k, v;
   1768 	search->cursor = NULL;
   1769 
   1770 	MDB_cursor **cursor = (MDB_cursor **)&search->cursor;
   1771 
   1772 	ndb_make_search_key_low(&s, query);
   1773 
   1774 	k.mv_data = &s;
   1775 	k.mv_size = sizeof(s);
   1776 
   1777 	if ((rc = mdb_cursor_open(txn->mdb_txn,
   1778 				  txn->lmdb->dbs[NDB_DB_PROFILE_SEARCH],
   1779 				  cursor))) {
   1780 		printf("search_profile: cursor opened failed: %s\n",
   1781 				mdb_strerror(rc));
   1782 		return 0;
   1783 	}
   1784 
   1785 	// Position cursor at the next key greater than or equal to the specified key
   1786 	if (mdb_cursor_get(search->cursor, &k, &v, MDB_SET_RANGE)) {
   1787 		printf("search_profile: cursor get failed\n");
   1788 		goto cleanup;
   1789 	} else {
   1790 		search->key = k.mv_data;
   1791 		assert(v.mv_size == 8);
   1792 		search->profile_key = *((uint64_t*)v.mv_data);
   1793 		return 1;
   1794 	}
   1795 
   1796 cleanup:
   1797 	mdb_cursor_close(search->cursor);
   1798 	search->cursor = NULL;
   1799 	return 0;
   1800 }
   1801 
   1802 void ndb_search_profile_end(struct ndb_search *search)
   1803 {
   1804 	if (search->cursor)
   1805 		mdb_cursor_close(search->cursor);
   1806 }
   1807 
   1808 int ndb_search_profile_next(struct ndb_search *search)
   1809 {
   1810 	int rc;
   1811 	MDB_val k, v;
   1812 	unsigned char *init_id;
   1813 
   1814 	init_id = search->key->id;
   1815 	k.mv_data = search->key;
   1816 	k.mv_size = sizeof(*search->key);
   1817 
   1818 retry:
   1819 	if ((rc = mdb_cursor_get(search->cursor, &k, &v, MDB_NEXT))) {
   1820 		ndb_debug("ndb_search_profile_next: %s\n",
   1821 				mdb_strerror(rc));
   1822 		return 0;
   1823 	} else {
   1824 		search->key = k.mv_data;
   1825 		assert(v.mv_size == 8);
   1826 		search->profile_key = *((uint64_t*)v.mv_data);
   1827 
   1828 		// skip duplicate pubkeys
   1829 		if (!memcmp(init_id, search->key->id, 32))
   1830 			goto retry;
   1831 	}
   1832 
   1833 	return 1;
   1834 }
   1835 
   1836 static int ndb_search_key_cmp(const MDB_val *a, const MDB_val *b)
   1837 {
   1838 	int cmp;
   1839 	struct ndb_search_key *ska, *skb;
   1840 
   1841 	ska = a->mv_data;
   1842 	skb = b->mv_data;
   1843 
   1844 	MDB_val a2 = *a;
   1845 	MDB_val b2 = *b;
   1846 
   1847 	a2.mv_data = ska->search;
   1848 	a2.mv_size = sizeof(ska->search) + sizeof(ska->id);
   1849 
   1850 	cmp = mdb_cmp_memn(&a2, &b2);
   1851 	if (cmp) return cmp;
   1852 
   1853 	if (ska->timestamp < skb->timestamp)
   1854 		return -1;
   1855 	else if (ska->timestamp > skb->timestamp)
   1856 		return 1;
   1857 
   1858 	return 0;
   1859 }
   1860 
   1861 static int ndb_write_profile_pk_index(struct ndb_txn *txn, struct ndb_note *note, uint64_t profile_key)
   1862 	
   1863 {
   1864 	MDB_val key, val;
   1865 	int rc;
   1866 	struct ndb_tsid tsid;
   1867 	MDB_dbi pk_db;
   1868 
   1869 	pk_db = txn->lmdb->dbs[NDB_DB_PROFILE_PK];
   1870 
   1871 	// write profile_pk + created_at index
   1872 	ndb_tsid_init(&tsid, note->pubkey, note->created_at);
   1873 
   1874 	key.mv_data = &tsid;
   1875 	key.mv_size = sizeof(tsid);
   1876 	val.mv_data = &profile_key;
   1877 	val.mv_size = sizeof(profile_key);
   1878 
   1879 	if ((rc = mdb_put(txn->mdb_txn, pk_db, &key, &val, 0))) {
   1880 		ndb_debug("write profile_pk(%" PRIu64 ") to db failed: %s\n",
   1881 				profile_key, mdb_strerror(rc));
   1882 		return 0;
   1883 	}
   1884 
   1885 	return 1;
   1886 }
   1887 
   1888 static int ndb_write_profile(struct ndb_txn *txn,
   1889 			     struct ndb_writer_profile *profile,
   1890 			     uint64_t note_key)
   1891 {
   1892 	uint64_t profile_key;
   1893 	struct ndb_note *note;
   1894 	void *flatbuf;
   1895 	size_t flatbuf_len;
   1896 	int rc;
   1897 
   1898 	MDB_val key, val;
   1899 	MDB_dbi profile_db;
   1900 	
   1901 	note = profile->note.note;
   1902 
   1903 	// add note_key to profile record
   1904 	NdbProfileRecord_note_key_add(profile->record.builder, note_key);
   1905 	NdbProfileRecord_end_as_root(profile->record.builder);
   1906 
   1907 	flatbuf = profile->record.flatbuf =
   1908 		flatcc_builder_finalize_aligned_buffer(profile->record.builder, &flatbuf_len);
   1909 
   1910 	assert(((uint64_t)flatbuf % 8) == 0);
   1911 
   1912 	// TODO: this may not be safe!?
   1913 	flatbuf_len = (flatbuf_len + 7) & ~7;
   1914 
   1915 	//assert(NdbProfileRecord_verify_as_root(flatbuf, flatbuf_len) == 0);
   1916 
   1917 	// get dbs
   1918 	profile_db = txn->lmdb->dbs[NDB_DB_PROFILE];
   1919 
   1920 	// get new key
   1921 	profile_key = ndb_get_last_key(txn->mdb_txn, profile_db) + 1;
   1922 
   1923 	// write profile to profile store
   1924 	key.mv_data = &profile_key;
   1925 	key.mv_size = sizeof(profile_key);
   1926 	val.mv_data = flatbuf;
   1927 	val.mv_size = flatbuf_len;
   1928 
   1929 	if ((rc = mdb_put(txn->mdb_txn, profile_db, &key, &val, 0))) {
   1930 		ndb_debug("write profile to db failed: %s\n", mdb_strerror(rc));
   1931 		return 0;
   1932 	}
   1933 
   1934 	// write last fetched record
   1935 	if (!ndb_maybe_write_last_profile_fetch(txn, note)) {
   1936 		ndb_debug("failed to write last profile fetched record\n");
   1937 	}
   1938 
   1939 	// write profile pubkey index
   1940 	if (!ndb_write_profile_pk_index(txn, note, profile_key)) {
   1941 		ndb_debug("failed to write profile pubkey index\n");
   1942 		return 0;
   1943 	}
   1944 
   1945 	// write name, display_name profile search indices
   1946 	if (!ndb_write_profile_search_indices(txn, note, profile_key,
   1947 					      flatbuf)) {
   1948 		ndb_debug("failed to write profile search indices\n");
   1949 		return 0;
   1950 	}
   1951 
   1952 	return 1;
   1953 }
   1954 
   1955 // find the last id tag in a note (e, p, etc)
   1956 static unsigned char *ndb_note_last_id_tag(struct ndb_note *note, char type)
   1957 {
   1958 	unsigned char *last = NULL;
   1959 	struct ndb_iterator iter;
   1960 	struct ndb_str str;
   1961 
   1962 	// get the liked event id (last id)
   1963 	ndb_tags_iterate_start(note, &iter);
   1964 
   1965 	while (ndb_tags_iterate_next(&iter)) {
   1966 		if (iter.tag->count < 2)
   1967 			continue;
   1968 
   1969 		str = ndb_note_str(note, &iter.tag->strs[0]);
   1970 
   1971 		// assign liked to the last e tag
   1972 		if (str.flag == NDB_PACKED_STR && str.str[0] == type) {
   1973 			str = ndb_note_str(note, &iter.tag->strs[1]);
   1974 			if (str.flag == NDB_PACKED_ID)
   1975 				last = str.id;
   1976 		}
   1977 	}
   1978 
   1979 	return last;
   1980 }
   1981 
   1982 void *ndb_get_note_meta(struct ndb_txn *txn, const unsigned char *id, size_t *len)
   1983 {
   1984 	MDB_val k, v;
   1985 
   1986 	k.mv_data = (unsigned char*)id;
   1987 	k.mv_size = 32;
   1988 
   1989 	if (mdb_get(txn->mdb_txn, txn->lmdb->dbs[NDB_DB_META], &k, &v)) {
   1990 		ndb_debug("ndb_get_note_meta: mdb_get note failed\n");
   1991 		return NULL;
   1992 	}
   1993 
   1994 	if (len)
   1995 		*len = v.mv_size;
   1996 
   1997 	return v.mv_data;
   1998 }
   1999 
   2000 // When receiving a reaction note, look for the liked id and increase the
   2001 // reaction counter in the note metadata database
   2002 static int ndb_write_reaction_stats(struct ndb_txn *txn, struct ndb_note *note)
   2003 {
   2004 	size_t len;
   2005 	void *root;
   2006 	int reactions, rc;
   2007 	MDB_val key, val;
   2008 	NdbEventMeta_table_t meta;
   2009 	unsigned char *liked = ndb_note_last_id_tag(note, 'e');
   2010 
   2011 	if (liked == NULL)
   2012 		return 0;
   2013 
   2014 	root = ndb_get_note_meta(txn, liked, &len);
   2015 
   2016 	flatcc_builder_t builder;
   2017 	flatcc_builder_init(&builder);
   2018 	NdbEventMeta_start_as_root(&builder);
   2019 
   2020 	// no meta record, let's make one
   2021 	if (root == NULL) {
   2022 		NdbEventMeta_reactions_add(&builder, 1);
   2023 	} else {
   2024 		// clone existing and add to it
   2025 		meta = NdbEventMeta_as_root(root);
   2026 	
   2027 		reactions = NdbEventMeta_reactions_get(meta);
   2028 		NdbEventMeta_clone(&builder, meta);
   2029 		NdbEventMeta_reactions_add(&builder, reactions + 1);
   2030 	}
   2031 
   2032 	NdbProfileRecord_end_as_root(&builder);
   2033 	root = flatcc_builder_finalize_aligned_buffer(&builder, &len);
   2034 	assert(((uint64_t)root % 8) == 0);
   2035 
   2036 	if (root == NULL) {
   2037 		ndb_debug("failed to create note metadata record\n");
   2038 		return 0;
   2039 	}
   2040 
   2041 	// metadata is keyed on id because we want to collect stats regardless
   2042 	// if we have the note yet or not
   2043 	key.mv_data = liked;
   2044 	key.mv_size = 32;
   2045 	
   2046 	val.mv_data = root;
   2047 	val.mv_size = len;
   2048 
   2049 	// write the new meta record
   2050 	//ndb_debug("writing stats record for ");
   2051 	//print_hex(liked, 32);
   2052 	//ndb_debug("\n");
   2053 
   2054 	if ((rc = mdb_put(txn->mdb_txn, txn->lmdb->dbs[NDB_DB_META], &key, &val, 0))) {
   2055 		ndb_debug("write reaction stats to db failed: %s\n", mdb_strerror(rc));
   2056 		return 0;
   2057 	}
   2058 
   2059 	free(root);
   2060 
   2061 	return 1;
   2062 }
   2063 
   2064 
   2065 static int ndb_write_note_id_index(struct ndb_txn *txn, struct ndb_note *note,
   2066 				   uint64_t note_key)
   2067 	
   2068 {
   2069 	struct ndb_tsid tsid;
   2070 	int rc;
   2071 	MDB_val key, val;
   2072 	MDB_dbi id_db;
   2073 
   2074 	ndb_tsid_init(&tsid, note->id, note->created_at);
   2075 
   2076 	key.mv_data = &tsid;
   2077 	key.mv_size = sizeof(tsid);
   2078 	val.mv_data = &note_key;
   2079 	val.mv_size = sizeof(note_key);
   2080 
   2081 	id_db = txn->lmdb->dbs[NDB_DB_NOTE_ID];
   2082 
   2083 	if ((rc = mdb_put(txn->mdb_txn, id_db, &key, &val, 0))) {
   2084 		ndb_debug("write note id index to db failed: %s\n",
   2085 				mdb_strerror(rc));
   2086 		return 0;
   2087 	}
   2088 
   2089 	return 1;
   2090 }
   2091 
   2092 static int ndb_write_note_kind_index(struct ndb_txn *txn, struct ndb_note *note,
   2093 				     uint64_t note_key)
   2094 {
   2095 	struct ndb_u64_tsid tsid;
   2096 	int rc;
   2097 	MDB_val key, val;
   2098 	MDB_dbi kind_db;
   2099 
   2100 	ndb_u64_tsid_init(&tsid, note->kind, note->created_at);
   2101 
   2102 	key.mv_data = &tsid;
   2103 	key.mv_size = sizeof(tsid);
   2104 	val.mv_data = &note_key;
   2105 	val.mv_size = sizeof(note_key);
   2106 
   2107 	kind_db = txn->lmdb->dbs[NDB_DB_NOTE_KIND];
   2108 
   2109 	if ((rc = mdb_put(txn->mdb_txn, kind_db, &key, &val, 0))) {
   2110 		ndb_debug("write note kind index to db failed: %s\n",
   2111 				mdb_strerror(rc));
   2112 		return 0;
   2113 	}
   2114 
   2115 	return 1;
   2116 }
   2117 
   2118 static void consume_whitespace_or_punctuation(struct cursor *cur)
   2119 {
   2120 	while (cur->p < cur->end) {
   2121 		if (!is_right_boundary(*cur->p))
   2122 			return;
   2123 		cur->p++;
   2124 	}
   2125 }
   2126 
   2127 static int ndb_write_word_to_index(struct ndb_txn *txn, const char *word,
   2128 				   int word_len, int word_index,
   2129 				   uint64_t timestamp, uint64_t note_id)
   2130 {
   2131 	// cap to some reasonable key size
   2132 	unsigned char buffer[1024];
   2133 	int keysize, rc;
   2134 	MDB_val k, v;
   2135 	MDB_dbi text_db;
   2136 
   2137 	// build our compressed text index key
   2138 	if (!ndb_make_text_search_key(buffer, sizeof(buffer), word_index,
   2139 				      word_len, word, timestamp, note_id,
   2140 				      &keysize)) {
   2141 		// probably too big
   2142 
   2143 		return 0;
   2144 	}
   2145 
   2146 	k.mv_data = buffer;
   2147 	k.mv_size = keysize;
   2148 
   2149 	v.mv_data = NULL;
   2150 	v.mv_size = 0;
   2151 
   2152 	text_db = txn->lmdb->dbs[NDB_DB_NOTE_TEXT];
   2153 
   2154 	if ((rc = mdb_put(txn->mdb_txn, text_db, &k, &v, 0))) {
   2155 		ndb_debug("write note text index to db failed: %s\n",
   2156 				mdb_strerror(rc));
   2157 		return 0;
   2158 	}
   2159 
   2160 	return 1;
   2161 }
   2162 
   2163 
   2164 
   2165 // break a string into individual words for querying or for building the
   2166 // fulltext search index. This is callback based so we don't need to
   2167 // build up an intermediate structure
   2168 static int ndb_parse_words(struct cursor *cur, void *ctx, ndb_word_parser_fn fn)
   2169 {
   2170 	int word_len, words;
   2171 	const char *word;
   2172 
   2173 	words = 0;
   2174 
   2175 	while (cur->p < cur->end) {
   2176 		consume_whitespace_or_punctuation(cur);
   2177 		if (cur->p >= cur->end)
   2178 			break;
   2179 		word = (const char *)cur->p;
   2180 
   2181 		if (!consume_until_boundary(cur))
   2182 			break;
   2183 
   2184 		// start of word or end
   2185 		word_len = cur->p - (unsigned char *)word;
   2186 		if (word_len == 0 && cur->p >= cur->end)
   2187 			break;
   2188 
   2189 		if (word_len == 0) {
   2190 			if (!cursor_skip(cur, 1))
   2191 				break;
   2192 			continue;
   2193 		}
   2194 
   2195 		//ndb_debug("writing word index '%.*s'\n", word_len, word);
   2196 
   2197 		if (!fn(ctx, word, word_len, words))
   2198 			continue;
   2199 
   2200 		words++;
   2201 	}
   2202 
   2203 	return 1;
   2204 }
   2205 
   2206 struct ndb_word_writer_ctx
   2207 {
   2208 	struct ndb_txn *txn;
   2209 	struct ndb_note *note;
   2210 	uint64_t note_id;
   2211 };
   2212 
   2213 static int ndb_fulltext_word_writer(void *ctx,
   2214 		const char *word, int word_len, int words)
   2215 {
   2216 	struct ndb_word_writer_ctx *wctx = ctx;
   2217 
   2218 	if (!ndb_write_word_to_index(wctx->txn, word, word_len, words,
   2219 				     wctx->note->created_at, wctx->note_id)) {
   2220 		// too big to write this one, just skip it
   2221 		ndb_debug("failed to write word '%.*s' to index\n", word_len, word);
   2222 
   2223 		return 0;
   2224 	}
   2225 
   2226 	//fprintf(stderr, "wrote '%.*s' to note text index\n", word_len, word);
   2227 	return 1;
   2228 }
   2229 
   2230 static int ndb_write_note_fulltext_index(struct ndb_txn *txn,
   2231 					 struct ndb_note *note,
   2232 					 uint64_t note_id)
   2233 {
   2234 	struct cursor cur;
   2235 	unsigned char *content;
   2236 	struct ndb_str str;
   2237 	struct ndb_word_writer_ctx ctx;
   2238 
   2239 	str = ndb_note_str(note, &note->content);
   2240 	// I don't think this should happen?
   2241 	if (unlikely(str.flag == NDB_PACKED_ID))
   2242 		return 0;
   2243 
   2244 	content = (unsigned char *)str.str;
   2245 
   2246 	make_cursor(content, content + note->content_length, &cur);
   2247 
   2248 	ctx.txn = txn;
   2249 	ctx.note = note;
   2250 	ctx.note_id = note_id;
   2251 
   2252 	ndb_parse_words(&cur, &ctx, ndb_fulltext_word_writer);
   2253 
   2254 	return 1;
   2255 }
   2256 
   2257 static int ndb_parse_search_words(void *ctx, const char *word_str, int word_len, int word_index)
   2258 {
   2259 	struct ndb_search_words *words = ctx;
   2260 	struct ndb_word *word;
   2261 
   2262 	if (words->num_words + 1 > MAX_TEXT_SEARCH_WORDS)
   2263 		return 0;
   2264 
   2265 	word = &words->words[words->num_words++];
   2266 	word->word = word_str;
   2267 	word->word_len = word_len;
   2268 
   2269 	return 1;
   2270 }
   2271 
   2272 static void ndb_search_words_init(struct ndb_search_words *words)
   2273 {
   2274 	words->num_words = 0;
   2275 }
   2276 
   2277 static int prefix_count(const char *str1, int len1, const char *str2, int len2) {
   2278 	int i, count = 0;
   2279 	int min_len = len1 < len2 ? len1 : len2;
   2280 
   2281 	for (i = 0; i < min_len; i++) {
   2282 		// case insensitive
   2283 		if (tolower(str1[i]) == tolower(str2[i]))
   2284 			count++;
   2285 		else
   2286 			break;
   2287 	}
   2288 
   2289 	return count;
   2290 }
   2291 
   2292 static void ndb_print_text_search_key(struct ndb_text_search_key *key)
   2293 {
   2294 	printf("K<'%.*s' %d %" PRIu64 " note_id:%" PRIu64 ">", key->str_len, key->str,
   2295 						    key->word_index,
   2296 						    key->timestamp,
   2297 						    key->note_id);
   2298 }
   2299 
   2300 static int ndb_prefix_matches(struct ndb_text_search_result *result,
   2301 			      struct ndb_word *search_word)
   2302 {
   2303 	// Empty strings shouldn't happen but let's
   2304 	if (result->key.str_len < 2 || search_word->word_len < 2)
   2305 		return 0;
   2306 
   2307 	// make sure we at least have two matching prefix characters. exact
   2308 	// matches are nice but range searches allow us to match prefixes as
   2309 	// well. A double-char prefix is suffient, but maybe we could up this
   2310 	// in the future.
   2311 	// 
   2312 	// TODO: How are we handling utf-8 prefix matches like
   2313 	// japanese?
   2314 	//
   2315 	if (   result->key.str[0] != tolower(search_word->word[0])
   2316 	    && result->key.str[1] != tolower(search_word->word[1])
   2317 	    )
   2318 		return 0;
   2319 
   2320 	// count the number of prefix-matched characters. This will be used
   2321 	// for ranking search results
   2322 	result->prefix_chars = prefix_count(result->key.str,
   2323 					    result->key.str_len,
   2324 					    search_word->word,
   2325 					    search_word->word_len);
   2326 
   2327 	if (result->prefix_chars <= (int)((double)search_word->word_len / 1.5)) 
   2328 		return 0;
   2329 
   2330 	return 1;
   2331 }
   2332 
   2333 // This is called when scanning the full text search index. Scanning stops
   2334 // when we no longer have a prefix match for the word
   2335 static int ndb_text_search_next_word(MDB_cursor *cursor, MDB_cursor_op op,
   2336 	MDB_val *k, struct ndb_word *search_word,
   2337 	struct ndb_text_search_result *last_result,
   2338 	struct ndb_text_search_result *result,
   2339 	MDB_cursor_op order_op)
   2340 {
   2341 	struct cursor key_cursor;
   2342 	//struct ndb_text_search_key search_key;
   2343 	MDB_val v;
   2344 	int retries;
   2345 	retries = -1;
   2346 
   2347 	make_cursor(k->mv_data, k->mv_data + k->mv_size, &key_cursor);
   2348 
   2349 	// When op is MDB_SET_RANGE, this initializes the search. Position
   2350 	// the cursor at the next key greater than or equal to the specified
   2351 	// key.
   2352 	//
   2353 	// Subsequent searches should use MDB_NEXT
   2354 	if (mdb_cursor_get(cursor, k, &v, op)) {
   2355 		// we should only do this if we're going in reverse
   2356 		if (op == MDB_SET_RANGE && order_op == MDB_PREV) {
   2357 			// if set range worked and our key exists, it should be
   2358 			// the one right before this one
   2359 			if (mdb_cursor_get(cursor, k, &v, MDB_PREV))
   2360 				return 0;
   2361 		} else {
   2362 			return 0;
   2363 		}
   2364 	}
   2365 
   2366 retry:
   2367 	retries++;
   2368 	/*
   2369 	printf("continuing from ");
   2370 	if (ndb_unpack_text_search_key(k->mv_data, k->mv_size, &search_key)) {
   2371 		ndb_print_text_search_key(&search_key);
   2372 	} else { printf("??"); }
   2373 	printf("\n");
   2374 	*/
   2375 
   2376 	make_cursor(k->mv_data, k->mv_data + k->mv_size, &key_cursor);
   2377 
   2378 	if (unlikely(!ndb_unpack_text_search_key_noteid(&key_cursor, &result->key.note_id))) {
   2379 		fprintf(stderr, "UNUSUAL: failed to unpack text search key note_id\n");
   2380 		return 0;
   2381 	}
   2382 
   2383 	if (last_result) {
   2384 		if (last_result->key.note_id != result->key.note_id)
   2385 			return 0;
   2386 	}
   2387 
   2388 	// On success, this could still be not related at all.
   2389 	// It could just be adjacent to the word. Let's check
   2390 	// if we have a matching prefix at least.
   2391 
   2392 	// Before we unpack the entire key, let's quickly
   2393 	// unpack just the string to check the prefix. We don't
   2394 	// need to unpack the entire key if the prefix doesn't
   2395 	// match
   2396 	if (!ndb_unpack_text_search_key_string(&key_cursor,
   2397 					       &result->key.str,
   2398 					       &result->key.str_len)) {
   2399 		// this should never happen
   2400 		fprintf(stderr, "UNUSUAL: failed to unpack text search key string\n");
   2401 		return 0;
   2402 	}
   2403 
   2404 	if (!ndb_prefix_matches(result, search_word)) {
   2405 		/*
   2406 		printf("result prefix '%.*s' didn't match search word '%.*s'\n",
   2407 			result->key.str_len, result->key.str,
   2408 			search_word->word_len, search_word->word);
   2409 			*/
   2410 		// we should only do this if we're going in reverse
   2411 		if (retries == 0 && op == MDB_SET_RANGE && order_op == MDB_PREV) {
   2412 			// if set range worked and our key exists, it should be
   2413 			// the one right before this one
   2414 			mdb_cursor_get(cursor, k, &v, MDB_PREV);
   2415 			goto retry;
   2416 		} else {
   2417 			return 0;
   2418 		}
   2419 	}
   2420 
   2421 	// Unpack the remaining text search key, we will need this information
   2422 	// when building up our search results.
   2423 	if (!ndb_unpack_remaining_text_search_key(&key_cursor, &result->key)) {
   2424 		// This should never happen
   2425 		fprintf(stderr, "UNUSUAL: failed to unpack text search key\n");
   2426 		return 0;
   2427 	}
   2428 
   2429 	if (last_result) {
   2430 		if (result->key.word_index < last_result->key.word_index) {
   2431 			/*
   2432 			fprintf(stderr, "skipping '%.*s' because it is before last result '%.*s'\n",
   2433 					result->key.str_len, result->key.str,
   2434 					last_result->key.str_len, last_result->key.str);
   2435 					*/
   2436 			return 0;
   2437 		}
   2438 	}
   2439 
   2440 	return 1;
   2441 }
   2442 
   2443 static void ndb_text_search_results_init(
   2444 		struct ndb_text_search_results *results) {
   2445 	results->num_results = 0;
   2446 }
   2447 
   2448 void ndb_default_text_search_config(struct ndb_text_search_config *cfg)
   2449 {
   2450 	cfg->order = NDB_ORDER_DESCENDING;
   2451 	cfg->limit = MAX_TEXT_SEARCH_RESULTS;
   2452 }
   2453 
   2454 void ndb_text_search_config_set_order(struct ndb_text_search_config *cfg,
   2455 				     enum ndb_search_order order)
   2456 {
   2457 	cfg->order = order;
   2458 }
   2459 
   2460 void ndb_text_search_config_set_limit(struct ndb_text_search_config *cfg, int limit)
   2461 {
   2462 	cfg->limit = limit;
   2463 }
   2464 
   2465 int ndb_text_search(struct ndb_txn *txn, const char *query,
   2466 		    struct ndb_text_search_results *results,
   2467 		    struct ndb_text_search_config *config)
   2468 {
   2469 	unsigned char buffer[1024], *buf;
   2470 	unsigned char saved_buf[1024], *saved;
   2471 	struct ndb_text_search_result *result, *last_result;
   2472 	struct ndb_text_search_result candidate, last_candidate;
   2473 	struct ndb_search_words search_words;
   2474 	//struct ndb_text_search_key search_key;
   2475 	struct ndb_word *search_word;
   2476 	struct cursor cur;
   2477 	ndb_text_search_key_order_fn key_order_fn;
   2478 	MDB_dbi text_db;
   2479 	MDB_cursor *cursor;
   2480 	MDB_val k, v;
   2481 	int i, j, keysize, saved_size, limit;
   2482 	MDB_cursor_op op, order_op;
   2483 	//int num_note_ids;
   2484 
   2485 	saved = NULL;
   2486 	ndb_text_search_results_init(results);
   2487 	ndb_search_words_init(&search_words);
   2488 
   2489 	// search config
   2490 	limit = MAX_TEXT_SEARCH_RESULTS;
   2491 	order_op = MDB_PREV;
   2492 	key_order_fn = ndb_make_text_search_key_high;
   2493 	if (config) {
   2494 		if (config->order == NDB_ORDER_ASCENDING) {
   2495 			order_op = MDB_NEXT;
   2496 			key_order_fn = ndb_make_text_search_key_low;
   2497 		}
   2498 		limit = min(limit, config->limit);
   2499 	}
   2500 	// end search config
   2501 	
   2502 	text_db = txn->lmdb->dbs[NDB_DB_NOTE_TEXT];
   2503 	make_cursor((unsigned char *)query, (unsigned char *)query + strlen(query), &cur);
   2504 
   2505 	ndb_parse_words(&cur, &search_words, ndb_parse_search_words);
   2506 	if (search_words.num_words == 0)
   2507 		return 0;
   2508 
   2509 	if ((i = mdb_cursor_open(txn->mdb_txn, text_db, &cursor))) {
   2510 		fprintf(stderr, "nd_text_search: mdb_cursor_open failed, error %d\n", i);
   2511 		return 0;
   2512 	}
   2513 
   2514 	// for each word, we recursively find all of the submatches
   2515 	while (results->num_results < limit) {
   2516 		last_result = NULL;
   2517 		result = &results->results[results->num_results];
   2518 
   2519 		// if we have saved, then we continue from the last root search
   2520 		// sequence
   2521 		if (saved) {
   2522 			buf = saved_buf;
   2523 			saved = NULL;
   2524 			keysize = saved_size;
   2525 
   2526 			k.mv_data = buf;
   2527 			k.mv_size = saved_size;
   2528 
   2529 			// reposition the cursor so we can continue
   2530 			if (mdb_cursor_get(cursor, &k, &v, MDB_SET_RANGE))
   2531 				break;
   2532 
   2533 			op = order_op;
   2534 		} else {
   2535 			// construct a packed fulltext search key using this
   2536 			// word this key doesn't contain any timestamp or index
   2537 			// info, so it should range match instead of exact
   2538 			// match
   2539 			if (!key_order_fn(buffer, sizeof(buffer),
   2540 					  search_words.words[0].word_len,
   2541 					  search_words.words[0].word, &keysize))
   2542 			{
   2543 				// word is too big to fit in 1024-sized key
   2544 				continue;
   2545 			}
   2546 
   2547 			buf = buffer;
   2548 			op = MDB_SET_RANGE;
   2549 		}
   2550 
   2551 		for (j = 0; j < search_words.num_words; j++) {
   2552 			search_word = &search_words.words[j];
   2553 
   2554 			// shouldn't happen but let's be defensive a bit
   2555 			if (search_word->word_len == 0)
   2556 				continue;
   2557 
   2558 			// if we already matched a note in this phrase, make
   2559 			// sure we're including the note id in the query
   2560 			if (last_result) {
   2561 				// we are narrowing down a search.
   2562 				// if we already have this note id, just continue
   2563 				for (i = 0; i < results->num_results; i++) {
   2564 					if (results->results[i].key.note_id == last_result->key.note_id)
   2565 						goto cont;
   2566 				}
   2567 
   2568 				if (!ndb_make_noted_text_search_key(
   2569 					buffer, sizeof(buffer),
   2570 					search_word->word_len,
   2571 					search_word->word,
   2572 					last_result->key.timestamp,
   2573 					last_result->key.note_id,
   2574 					&keysize))
   2575 				{
   2576 					continue;
   2577 				}
   2578 
   2579 				buf = buffer;
   2580 			}
   2581 
   2582 			k.mv_data = buf;
   2583 			k.mv_size = keysize;
   2584 
   2585 			if (!ndb_text_search_next_word(cursor, op, &k,
   2586 						       search_word,
   2587 						       last_result,
   2588 						       &candidate,
   2589 						       order_op)) {
   2590 				break;
   2591 			}
   2592 
   2593 			*result = candidate;
   2594 			op = MDB_SET_RANGE;
   2595 
   2596 			// save the first key match, since we will continue from
   2597 			// this on the next root word result
   2598 			if (j == 0 && !saved) {
   2599 				memcpy(saved_buf, k.mv_data, k.mv_size);
   2600 				saved = saved_buf;
   2601 				saved_size = k.mv_size;
   2602 			}
   2603 
   2604 			last_candidate = *result;
   2605 			last_result = &last_candidate;
   2606 		}
   2607 
   2608 cont:
   2609 		// we matched all of the queries!
   2610 		if (j == search_words.num_words) {
   2611 			results->num_results++;
   2612 		} else if (j == 0) {
   2613 			break;
   2614 		}
   2615 
   2616 	}
   2617 
   2618 	mdb_cursor_close(cursor);
   2619 
   2620 	return 1;
   2621 }
   2622 
   2623 static uint64_t ndb_write_note(struct ndb_txn *txn,
   2624 			       struct ndb_writer_note *note)
   2625 {
   2626 	int rc;
   2627 	uint64_t note_key;
   2628 	MDB_dbi note_db;
   2629 	MDB_val key, val;
   2630 
   2631 	// let's quickly sanity check if we already have this note
   2632 	if (ndb_get_notekey_by_id(txn, note->note->id))
   2633 		return 0;
   2634 	
   2635 	// get dbs
   2636 	note_db = txn->lmdb->dbs[NDB_DB_NOTE];
   2637 
   2638 	// get new key
   2639 	note_key = ndb_get_last_key(txn->mdb_txn, note_db) + 1;
   2640 
   2641 	// write note to event store
   2642 	key.mv_data = &note_key;
   2643 	key.mv_size = sizeof(note_key);
   2644 	val.mv_data = note->note;
   2645 	val.mv_size = note->note_len;
   2646 
   2647 	if ((rc = mdb_put(txn->mdb_txn, note_db, &key, &val, 0))) {
   2648 		ndb_debug("write note to db failed: %s\n", mdb_strerror(rc));
   2649 		return 0;
   2650 	}
   2651 
   2652 	// write id index key clustered with created_at
   2653 	if (!ndb_write_note_id_index(txn, note->note, note_key))
   2654 		return 0;
   2655 
   2656 	// write note kind index
   2657 	if (!ndb_write_note_kind_index(txn, note->note, note_key))
   2658 		return 0;
   2659 
   2660 	// only do fulltext index on text and longform notes
   2661 	if (note->note->kind == 1 || note->note->kind == 30023) {
   2662 		if (!ndb_write_note_fulltext_index(txn, note->note, note_key))
   2663 			return 0;
   2664 	}
   2665 
   2666 	if (note->note->kind == 7) {
   2667 		ndb_write_reaction_stats(txn, note->note);
   2668 	}
   2669 
   2670 	return note_key;
   2671 }
   2672 
   2673 // only to be called from the writer thread
   2674 static void ndb_write_version(struct ndb_txn *txn, uint64_t version)
   2675 {
   2676 	int rc;
   2677 	MDB_val key, val;
   2678 	uint64_t version_key;
   2679 
   2680 	version_key = NDB_META_KEY_VERSION;
   2681 	
   2682 	key.mv_data = &version_key;
   2683 	key.mv_size = sizeof(version_key);
   2684 	val.mv_data = &version;
   2685 	val.mv_size = sizeof(version);
   2686 
   2687 	if ((rc = mdb_put(txn->mdb_txn, txn->lmdb->dbs[NDB_DB_NDB_META], &key, &val, 0))) {
   2688 		ndb_debug("write version to ndb_meta failed: %s\n",
   2689 				mdb_strerror(rc));
   2690 		return;
   2691 	}
   2692 
   2693 	//fprintf(stderr, "writing version %" PRIu64 "\n", version);
   2694 }
   2695 
   2696 static void *ndb_writer_thread(void *data)
   2697 {
   2698 	struct ndb_writer *writer = data;
   2699 	struct ndb_writer_msg msgs[THREAD_QUEUE_BATCH], *msg;
   2700 	int i, popped, done, any_note;
   2701 	uint64_t note_nkey;
   2702 	MDB_txn *mdb_txn = NULL;
   2703 	struct ndb_txn txn;
   2704 	ndb_txn_from_mdb(&txn, writer->lmdb, mdb_txn);
   2705 
   2706 	done = 0;
   2707 	while (!done) {
   2708 		txn.mdb_txn = NULL;
   2709 		popped = prot_queue_pop_all(&writer->inbox, msgs, THREAD_QUEUE_BATCH);
   2710 		//ndb_debug("writer popped %d items\n", popped);
   2711 
   2712 		any_note = 0;
   2713 		for (i = 0 ; i < popped; i++) {
   2714 			msg = &msgs[i];
   2715 			switch (msg->type) {
   2716 			case NDB_WRITER_NOTE: any_note = 1; break;
   2717 			case NDB_WRITER_PROFILE: any_note = 1; break;
   2718 			case NDB_WRITER_DBMETA: any_note = 1; break;
   2719 			case NDB_WRITER_PROFILE_LAST_FETCH: any_note = 1; break;
   2720 			case NDB_WRITER_QUIT: break;
   2721 			}
   2722 		}
   2723 
   2724 		if (any_note && mdb_txn_begin(txn.lmdb->env, NULL, 0, (MDB_txn **)&txn.mdb_txn))
   2725 		{
   2726 			fprintf(stderr, "writer thread txn_begin failed");
   2727 			// should definitely not happen unless DB is full
   2728 			// or something ?
   2729 			assert(false);
   2730 		}
   2731 
   2732 		for (i = 0; i < popped; i++) {
   2733 			msg = &msgs[i];
   2734 
   2735 			switch (msg->type) {
   2736 			case NDB_WRITER_QUIT:
   2737 				// quits are handled before this
   2738 				done = 1;
   2739 				continue;
   2740 			case NDB_WRITER_PROFILE:
   2741 				note_nkey = 
   2742 					ndb_write_note(&txn, &msg->note);
   2743 				if (msg->profile.record.builder) {
   2744 					// only write if parsing didn't fail
   2745 					ndb_write_profile(&txn, &msg->profile,
   2746 							  note_nkey);
   2747 				}
   2748 				break;
   2749 			case NDB_WRITER_NOTE:
   2750 				ndb_write_note(&txn, &msg->note);
   2751 				//printf("wrote note ");
   2752 				//print_hex(msg->note.note->id, 32);
   2753 				//printf("\n");
   2754 				break;
   2755 			case NDB_WRITER_DBMETA:
   2756 				ndb_write_version(&txn, msg->ndb_meta.version);
   2757 				break;
   2758 			case NDB_WRITER_PROFILE_LAST_FETCH:
   2759 				ndb_writer_last_profile_fetch(&txn,
   2760 						msg->last_fetch.pubkey,
   2761 						msg->last_fetch.fetched_at
   2762 						);
   2763 				break;
   2764 			}
   2765 		}
   2766 
   2767 		// commit writes
   2768 		if (any_note && !ndb_end_query(&txn)) {
   2769 			fprintf(stderr, "writer thread txn commit failed");
   2770 			assert(false);
   2771 		}
   2772 
   2773 		// free notes
   2774 		for (i = 0; i < popped; i++) {
   2775 			msg = &msgs[i];
   2776 			if (msg->type == NDB_WRITER_NOTE)
   2777 				free(msg->note.note);
   2778 			else if (msg->type == NDB_WRITER_PROFILE) {
   2779 				free(msg->profile.note.note);
   2780 				ndb_profile_record_builder_free(&msg->profile.record);
   2781 			}
   2782 		}
   2783 	}
   2784 
   2785 	ndb_debug("quitting writer thread\n");
   2786 	return NULL;
   2787 }
   2788 
   2789 static void *ndb_ingester_thread(void *data)
   2790 {
   2791 	secp256k1_context *ctx;
   2792 	struct thread *thread = data;
   2793 	struct ndb_ingester *ingester = (struct ndb_ingester *)thread->ctx;
   2794 	struct ndb_lmdb *lmdb = ingester->writer->lmdb;
   2795 	struct ndb_ingester_msg msgs[THREAD_QUEUE_BATCH], *msg;
   2796 	struct ndb_writer_msg outs[THREAD_QUEUE_BATCH], *out;
   2797 	int i, to_write, popped, done, any_event;
   2798 	MDB_txn *read_txn = NULL;
   2799 	int rc;
   2800 
   2801 	ctx = secp256k1_context_create(SECP256K1_CONTEXT_VERIFY);
   2802 	ndb_debug("started ingester thread\n");
   2803 
   2804 	done = 0;
   2805 	while (!done) {
   2806 		to_write = 0;
   2807 		any_event = 0;
   2808 
   2809 		popped = prot_queue_pop_all(&thread->inbox, msgs, THREAD_QUEUE_BATCH);
   2810 		//ndb_debug("ingester popped %d items\n", popped);
   2811 
   2812 		for (i = 0; i < popped; i++) {
   2813 			msg = &msgs[i];
   2814 			if (msg->type == NDB_INGEST_EVENT) {
   2815 				any_event = 1;
   2816 				break;
   2817 			}
   2818 		}
   2819 
   2820 		if (any_event && (rc = mdb_txn_begin(lmdb->env, NULL, MDB_RDONLY, &read_txn))) {
   2821 			// this is bad
   2822 			fprintf(stderr, "UNUSUAL ndb_ingester: mdb_txn_begin failed: '%s'\n",
   2823 					mdb_strerror(rc));
   2824 			continue;
   2825 		}
   2826 
   2827 		for (i = 0; i < popped; i++) {
   2828 			msg = &msgs[i];
   2829 			switch (msg->type) {
   2830 			case NDB_INGEST_QUIT:
   2831 				done = 1;
   2832 				break;
   2833 
   2834 			case NDB_INGEST_EVENT:
   2835 				out = &outs[to_write];
   2836 				if (ndb_ingester_process_event(ctx, ingester,
   2837 							       &msg->event, out,
   2838 							       read_txn)) {
   2839 					to_write++;
   2840 				}
   2841 			}
   2842 		}
   2843 
   2844 		if (any_event)
   2845 			mdb_txn_abort(read_txn);
   2846 
   2847 		if (to_write > 0) {
   2848 			//ndb_debug("pushing %d events to write queue\n", to_write); 
   2849 			if (!ndb_writer_queue_msgs(ingester->writer, outs, to_write)) {
   2850 				ndb_debug("failed pushing %d events to write queue\n", to_write); 
   2851 			}
   2852 		}
   2853 	}
   2854 
   2855 	ndb_debug("quitting ingester thread\n");
   2856 	secp256k1_context_destroy(ctx);
   2857 	return NULL;
   2858 }
   2859 
   2860 
   2861 static int ndb_writer_init(struct ndb_writer *writer, struct ndb_lmdb *lmdb)
   2862 {
   2863 	writer->lmdb = lmdb;
   2864 	writer->queue_buflen = sizeof(struct ndb_writer_msg) * DEFAULT_QUEUE_SIZE;
   2865 	writer->queue_buf = malloc(writer->queue_buflen);
   2866 	if (writer->queue_buf == NULL) {
   2867 		fprintf(stderr, "ndb: failed to allocate space for writer queue");
   2868 		return 0;
   2869 	}
   2870 
   2871 	// init the writer queue.
   2872 	prot_queue_init(&writer->inbox, writer->queue_buf,
   2873 			writer->queue_buflen, sizeof(struct ndb_writer_msg));
   2874 
   2875 	// spin up the writer thread
   2876 	if (pthread_create(&writer->thread_id, NULL, ndb_writer_thread, writer))
   2877 	{
   2878 		fprintf(stderr, "ndb writer thread failed to create\n");
   2879 		return 0;
   2880 	}
   2881 	
   2882 	return 1;
   2883 }
   2884 
   2885 // initialize the ingester queue and then spawn the thread
   2886 static int ndb_ingester_init(struct ndb_ingester *ingester,
   2887 			     struct ndb_writer *writer,
   2888 			     struct ndb_config *config)
   2889 {
   2890 	int elem_size, num_elems;
   2891 	static struct ndb_ingester_msg quit_msg = { .type = NDB_INGEST_QUIT };
   2892 
   2893 	// TODO: configurable queue sizes
   2894 	elem_size = sizeof(struct ndb_ingester_msg);
   2895 	num_elems = DEFAULT_QUEUE_SIZE;
   2896 
   2897 	ingester->writer = writer;
   2898 	ingester->flags = config->flags;
   2899 	ingester->filter = config->ingest_filter;
   2900 	ingester->filter_context = config->filter_context;
   2901 
   2902 	if (!threadpool_init(&ingester->tp, config->ingester_threads,
   2903 			     elem_size, num_elems, &quit_msg, ingester,
   2904 			     ndb_ingester_thread))
   2905 	{
   2906 		fprintf(stderr, "ndb ingester threadpool failed to init\n");
   2907 		return 0;
   2908 	}
   2909 
   2910 	return 1;
   2911 }
   2912 
   2913 static int ndb_writer_destroy(struct ndb_writer *writer)
   2914 {
   2915 	struct ndb_writer_msg msg;
   2916 
   2917 	// kill thread
   2918 	msg.type = NDB_WRITER_QUIT;
   2919 	if (!prot_queue_push(&writer->inbox, &msg)) {
   2920 		// queue is too full to push quit message. just kill it.
   2921 		pthread_exit(&writer->thread_id);
   2922 	} else {
   2923 		pthread_join(writer->thread_id, NULL);
   2924 	}
   2925 
   2926 	// cleanup
   2927 	prot_queue_destroy(&writer->inbox);
   2928 
   2929 	free(writer->queue_buf);
   2930 
   2931 	return 1;
   2932 }
   2933 
   2934 static int ndb_ingester_destroy(struct ndb_ingester *ingester)
   2935 {
   2936 	threadpool_destroy(&ingester->tp);
   2937 	return 1;
   2938 }
   2939 
   2940 static int ndb_ingester_queue_event(struct ndb_ingester *ingester,
   2941 				    char *json, unsigned len, unsigned client)
   2942 {
   2943 	struct ndb_ingester_msg msg;
   2944 	msg.type = NDB_INGEST_EVENT;
   2945 
   2946 	msg.event.json = json;
   2947 	msg.event.len = len;
   2948 	msg.event.client = client;
   2949 
   2950 	return threadpool_dispatch(&ingester->tp, &msg);
   2951 }
   2952 
   2953 static int ndb_init_lmdb(const char *filename, struct ndb_lmdb *lmdb, size_t mapsize)
   2954 {
   2955 	int rc;
   2956 	MDB_txn *txn;
   2957 
   2958 	if ((rc = mdb_env_create(&lmdb->env))) {
   2959 		fprintf(stderr, "mdb_env_create failed, error %d\n", rc);
   2960 		return 0;
   2961 	}
   2962 
   2963 	if ((rc = mdb_env_set_mapsize(lmdb->env, mapsize))) {
   2964 		fprintf(stderr, "mdb_env_set_mapsize failed, error %d\n", rc);
   2965 		return 0;
   2966 	}
   2967 
   2968 	if ((rc = mdb_env_set_maxdbs(lmdb->env, NDB_DBS))) {
   2969 		fprintf(stderr, "mdb_env_set_maxdbs failed, error %d\n", rc);
   2970 		return 0;
   2971 	}
   2972 
   2973 	if ((rc = mdb_env_open(lmdb->env, filename, 0, 0664))) {
   2974 		fprintf(stderr, "mdb_env_open failed, error %d\n", rc);
   2975 		return 0;
   2976 	}
   2977 
   2978 	// Initialize DBs
   2979 	if ((rc = mdb_txn_begin(lmdb->env, NULL, 0, &txn))) {
   2980 		fprintf(stderr, "mdb_txn_begin failed, error %d\n", rc);
   2981 		return 0;
   2982 	}
   2983 
   2984 	// note flatbuffer db
   2985 	if ((rc = mdb_dbi_open(txn, "note", MDB_CREATE | MDB_INTEGERKEY, &lmdb->dbs[NDB_DB_NOTE]))) {
   2986 		fprintf(stderr, "mdb_dbi_open event failed, error %d\n", rc);
   2987 		return 0;
   2988 	}
   2989 
   2990 	// note metadata db
   2991 	if ((rc = mdb_dbi_open(txn, "meta", MDB_CREATE, &lmdb->dbs[NDB_DB_META]))) {
   2992 		fprintf(stderr, "mdb_dbi_open meta failed, error %d\n", rc);
   2993 		return 0;
   2994 	}
   2995 
   2996 	// profile flatbuffer db
   2997 	if ((rc = mdb_dbi_open(txn, "profile", MDB_CREATE | MDB_INTEGERKEY, &lmdb->dbs[NDB_DB_PROFILE]))) {
   2998 		fprintf(stderr, "mdb_dbi_open profile failed, error %d\n", rc);
   2999 		return 0;
   3000 	}
   3001 
   3002 	// profile search db
   3003 	if ((rc = mdb_dbi_open(txn, "profile_search", MDB_CREATE, &lmdb->dbs[NDB_DB_PROFILE_SEARCH]))) {
   3004 		fprintf(stderr, "mdb_dbi_open profile_search failed, error %d\n", rc);
   3005 		return 0;
   3006 	}
   3007 	mdb_set_compare(txn, lmdb->dbs[NDB_DB_PROFILE_SEARCH], ndb_search_key_cmp);
   3008 
   3009 	// ndb metadata (db version, etc)
   3010 	if ((rc = mdb_dbi_open(txn, "ndb_meta", MDB_CREATE | MDB_INTEGERKEY, &lmdb->dbs[NDB_DB_NDB_META]))) {
   3011 		fprintf(stderr, "mdb_dbi_open ndb_meta failed, error %d\n", rc);
   3012 		return 0;
   3013 	}
   3014 
   3015 	// profile last fetches
   3016 	if ((rc = mdb_dbi_open(txn, "profile_last_fetch", MDB_CREATE, &lmdb->dbs[NDB_DB_PROFILE_LAST_FETCH]))) {
   3017 		fprintf(stderr, "mdb_dbi_open profile last fetch, error %d\n", rc);
   3018 		return 0;
   3019 	}
   3020 
   3021 	// id+ts index flags
   3022 	unsigned int tsid_flags = MDB_CREATE | MDB_DUPSORT | MDB_DUPFIXED;
   3023 
   3024 	// index dbs
   3025 	if ((rc = mdb_dbi_open(txn, "note_id", tsid_flags, &lmdb->dbs[NDB_DB_NOTE_ID]))) {
   3026 		fprintf(stderr, "mdb_dbi_open id failed: %s\n", mdb_strerror(rc));
   3027 		return 0;
   3028 	}
   3029 	mdb_set_compare(txn, lmdb->dbs[NDB_DB_NOTE_ID], ndb_tsid_compare);
   3030 
   3031 	if ((rc = mdb_dbi_open(txn, "profile_pk", tsid_flags, &lmdb->dbs[NDB_DB_PROFILE_PK]))) {
   3032 		fprintf(stderr, "mdb_dbi_open id failed: %s\n", mdb_strerror(rc));
   3033 		return 0;
   3034 	}
   3035 	mdb_set_compare(txn, lmdb->dbs[NDB_DB_PROFILE_PK], ndb_tsid_compare);
   3036 
   3037 	if ((rc = mdb_dbi_open(txn, "note_kind", tsid_flags, &lmdb->dbs[NDB_DB_NOTE_KIND]))) {
   3038 		fprintf(stderr, "mdb_dbi_open id failed: %s\n", mdb_strerror(rc));
   3039 		return 0;
   3040 	}
   3041 	mdb_set_compare(txn, lmdb->dbs[NDB_DB_NOTE_KIND], ndb_u64_tsid_compare);
   3042 
   3043 	if ((rc = mdb_dbi_open(txn, "note_text", MDB_CREATE | MDB_DUPSORT,
   3044 			       &lmdb->dbs[NDB_DB_NOTE_TEXT]))) {
   3045 		fprintf(stderr, "mdb_dbi_open id failed: %s\n", mdb_strerror(rc));
   3046 		return 0;
   3047 	}
   3048 	mdb_set_compare(txn, lmdb->dbs[NDB_DB_NOTE_TEXT], ndb_text_search_key_compare);
   3049 
   3050 	// Commit the transaction
   3051 	if ((rc = mdb_txn_commit(txn))) {
   3052 		fprintf(stderr, "mdb_txn_commit failed, error %d\n", rc);
   3053 		return 0;
   3054 	}
   3055 
   3056 	return 1;
   3057 }
   3058 
   3059 static int ndb_queue_write_version(struct ndb *ndb, uint64_t version)
   3060 {
   3061 	struct ndb_writer_msg msg;
   3062 	msg.type = NDB_WRITER_DBMETA;
   3063 	msg.ndb_meta.version = version;
   3064 	return ndb_writer_queue_msg(&ndb->writer, &msg);
   3065 }
   3066 
   3067 static int ndb_run_migrations(struct ndb *ndb)
   3068 {
   3069 	uint64_t version, latest_version, i;
   3070 	
   3071 	latest_version = sizeof(MIGRATIONS) / sizeof(MIGRATIONS[0]);
   3072 
   3073 	if ((version = ndb_db_version(ndb)) == -1) {
   3074 		ndb_debug("run_migrations: no version found, assuming new db\n");
   3075 		version = latest_version;
   3076 
   3077 		// no version found. fresh db?
   3078 		if (!ndb_queue_write_version(ndb, version)) {
   3079 			fprintf(stderr, "run_migrations: failed writing db version");
   3080 			return 0;
   3081 		}
   3082 
   3083 		return 1;
   3084 	} else {
   3085 		ndb_debug("ndb: version %" PRIu64 " found\n", version);
   3086 	}
   3087 
   3088 	if (version < latest_version)
   3089 		ndb_debug("nostrdb: migrating v%d -> v%d\n",
   3090 				(int)version, (int)latest_version);
   3091 
   3092 	for (i = version; i < latest_version; i++) {
   3093 		if (!MIGRATIONS[i].fn(ndb)) {
   3094 			fprintf(stderr, "run_migrations: migration v%d -> v%d failed\n", (int)i, (int)(i+1));
   3095 			return 0;
   3096 		}
   3097 
   3098 		if (!ndb_queue_write_version(ndb, i+1)) {
   3099 			fprintf(stderr, "run_migrations: failed writing db version");
   3100 			return 0;
   3101 		}
   3102 
   3103 		version = i+1;
   3104 	}
   3105 
   3106 	ndb->version = version;
   3107 
   3108 	return 1;
   3109 }
   3110 
   3111 int ndb_init(struct ndb **pndb, const char *filename, struct ndb_config *config)
   3112 {
   3113 	struct ndb *ndb;
   3114 	//MDB_dbi ind_id; // TODO: ind_pk, etc
   3115 
   3116 	ndb = *pndb = calloc(1, sizeof(struct ndb));
   3117 	ndb->flags = config->flags;
   3118 
   3119 	if (ndb == NULL) {
   3120 		fprintf(stderr, "ndb_init: malloc failed\n");
   3121 		return 0;
   3122 	}
   3123 
   3124 	if (!ndb_init_lmdb(filename, &ndb->lmdb, config->mapsize))
   3125 		return 0;
   3126 
   3127 	if (!ndb_writer_init(&ndb->writer, &ndb->lmdb)) {
   3128 		fprintf(stderr, "ndb_writer_init failed\n");
   3129 		return 0;
   3130 	}
   3131 
   3132 	if (!ndb_ingester_init(&ndb->ingester, &ndb->writer, config)) {
   3133 		fprintf(stderr, "failed to initialize %d ingester thread(s)\n",
   3134 				config->ingester_threads);
   3135 		return 0;
   3136 	}
   3137 
   3138 	if (!ndb_flag_set(config->flags, NDB_FLAG_NOMIGRATE) &&
   3139 			  !ndb_run_migrations(ndb)) {
   3140 		fprintf(stderr, "failed to run migrations\n");
   3141 		return 0;
   3142 	}
   3143 
   3144 	// Initialize LMDB environment and spin up threads
   3145 	return 1;
   3146 }
   3147 
   3148 void ndb_destroy(struct ndb *ndb)
   3149 {
   3150 	if (ndb == NULL)
   3151 		return;
   3152 
   3153 	// ingester depends on writer and must be destroyed first
   3154 	ndb_ingester_destroy(&ndb->ingester);
   3155 	ndb_writer_destroy(&ndb->writer);
   3156 
   3157 	mdb_env_close(ndb->lmdb.env);
   3158 
   3159 	free(ndb);
   3160 }
   3161 
   3162 // Process a nostr event from a client
   3163 //
   3164 // ie: ["EVENT", {"content":"..."} ...]
   3165 //
   3166 // The client-sent variation of ndb_process_event
   3167 int ndb_process_client_event(struct ndb *ndb, const char *json, int len)
   3168 {
   3169 	// Since we need to return as soon as possible, and we're not
   3170 	// making any assumptions about the lifetime of the string, we
   3171 	// definitely need to copy the json here. In the future once we
   3172 	// have our thread that manages a websocket connection, we can
   3173 	// avoid the copy and just use the buffer we get from that
   3174 	// thread.
   3175 	char *json_copy = strdupn(json, len);
   3176 	if (json_copy == NULL)
   3177 		return 0;
   3178 
   3179 	return ndb_ingester_queue_event(&ndb->ingester, json_copy, len, 1);
   3180 }
   3181 
   3182 // Process anostr event from a relay,
   3183 //
   3184 // ie: ["EVENT", "subid", {"content":"..."}...]
   3185 // 
   3186 // This function returns as soon as possible, first copying the passed
   3187 // json and then queueing it up for processing. Worker threads then take
   3188 // the json and process it.
   3189 //
   3190 // Processing:
   3191 //
   3192 // 1. The event is parsed into ndb_notes and the signature is validated
   3193 // 2. A quick lookup is made on the database to see if we already have
   3194 //    the note id, if we do we don't need to waste time on json parsing
   3195 //    or note validation.
   3196 // 3. Once validation is done we pass it to the writer queue for writing
   3197 //    to LMDB.
   3198 //
   3199 int ndb_process_event(struct ndb *ndb, const char *json, int json_len)
   3200 {
   3201 	// Since we need to return as soon as possible, and we're not
   3202 	// making any assumptions about the lifetime of the string, we
   3203 	// definitely need to copy the json here. In the future once we
   3204 	// have our thread that manages a websocket connection, we can
   3205 	// avoid the copy and just use the buffer we get from that
   3206 	// thread.
   3207 	char *json_copy = strdupn(json, json_len);
   3208 	if (json_copy == NULL)
   3209 		return 0;
   3210 
   3211 	return ndb_ingester_queue_event(&ndb->ingester, json_copy, json_len, 0);
   3212 }
   3213 
   3214 
   3215 int _ndb_process_events(struct ndb *ndb, const char *ldjson, size_t json_len, int client)
   3216 {
   3217 	const char *start, *end, *very_end;
   3218 	start = ldjson;
   3219 	end = start + json_len;
   3220 	very_end = ldjson + json_len;
   3221 	int (* process)(struct ndb *, const char *, int);
   3222 #if DEBUG
   3223 	int processed = 0;
   3224 #endif
   3225 	process = client ? ndb_process_client_event : ndb_process_event;
   3226 
   3227 	while ((end = fast_strchr(start, '\n', very_end - start))) {
   3228 		//printf("processing '%.*s'\n", (int)(end-start), start);
   3229 		if (!process(ndb, start, end - start)) {
   3230 			ndb_debug("ndb_process_client_event failed\n");
   3231 			return 0;
   3232 		}
   3233 		start = end + 1;
   3234 #if DEBUG
   3235 		processed++;
   3236 #endif
   3237 	}
   3238 
   3239 #if DEBUG
   3240 	ndb_debug("ndb_process_events: processed %d events\n", processed);
   3241 #endif
   3242 
   3243 	return 1;
   3244 }
   3245 
   3246 int ndb_process_client_events(struct ndb *ndb, const char *ldjson, size_t json_len)
   3247 {
   3248 	return _ndb_process_events(ndb, ldjson, json_len, 1);
   3249 }
   3250 
   3251 int ndb_process_events(struct ndb *ndb, const char *ldjson, size_t json_len)
   3252 {
   3253 	return _ndb_process_events(ndb, ldjson, json_len, 0);
   3254 }
   3255 
   3256 static inline int cursor_push_tag(struct cursor *cur, struct ndb_tag *tag)
   3257 {
   3258 	return cursor_push_u16(cur, tag->count);
   3259 }
   3260 
   3261 int ndb_builder_init(struct ndb_builder *builder, unsigned char *buf,
   3262 		     int bufsize)
   3263 {
   3264 	struct ndb_note *note;
   3265 	int half, size, str_indices_size;
   3266 
   3267 	// come on bruh
   3268 	if (bufsize < sizeof(struct ndb_note) * 2)
   3269 		return 0;
   3270 
   3271 	str_indices_size = bufsize / 32;
   3272 	size = bufsize - str_indices_size;
   3273 	half = size / 2;
   3274 
   3275 	//debug("size %d half %d str_indices %d\n", size, half, str_indices_size);
   3276 
   3277 	// make a safe cursor of our available memory
   3278 	make_cursor(buf, buf + bufsize, &builder->mem);
   3279 
   3280 	note = builder->note = (struct ndb_note *)buf;
   3281 
   3282 	// take slices of the memory into subcursors
   3283 	if (!(cursor_slice(&builder->mem, &builder->note_cur, half) &&
   3284 	      cursor_slice(&builder->mem, &builder->strings, half) &&
   3285 	      cursor_slice(&builder->mem, &builder->str_indices, str_indices_size))) {
   3286 		return 0;
   3287 	}
   3288 
   3289 	memset(note, 0, sizeof(*note));
   3290 	builder->note_cur.p += sizeof(*note);
   3291 
   3292 	note->strings = builder->strings.start - buf;
   3293 	note->version = 1;
   3294 
   3295 	return 1;
   3296 }
   3297 
   3298 
   3299 
   3300 static inline int ndb_json_parser_init(struct ndb_json_parser *p,
   3301 				       const char *json, int json_len,
   3302 				       unsigned char *buf, int bufsize)
   3303 {
   3304 	int half = bufsize / 2;
   3305 
   3306 	unsigned char *tok_start = buf + half;
   3307 	unsigned char *tok_end = buf + bufsize;
   3308 
   3309 	p->toks = (jsmntok_t*)tok_start;
   3310 	p->toks_end = (jsmntok_t*)tok_end;
   3311 	p->num_tokens = 0;
   3312 	p->json = json;
   3313 	p->json_len = json_len;
   3314 
   3315 	// ndb_builder gets the first half of the buffer, and jsmn gets the
   3316 	// second half. I like this way of alloating memory (without actually
   3317 	// dynamically allocating memory). You get one big chunk upfront and
   3318 	// then submodules can recursively subdivide it. Maybe you could do
   3319 	// something even more clever like golden-ratio style subdivision where
   3320 	// the more important stuff gets a larger chunk and then it spirals
   3321 	// downward into smaller chunks. Thanks for coming to my TED talk.
   3322 
   3323 	if (!ndb_builder_init(&p->builder, buf, half))
   3324 		return 0;
   3325 
   3326 	jsmn_init(&p->json_parser);
   3327 
   3328 	return 1;
   3329 }
   3330 
   3331 static inline int ndb_json_parser_parse(struct ndb_json_parser *p,
   3332 					struct ndb_id_cb *cb)
   3333 {
   3334 	jsmntok_t *tok;
   3335 	int cap = ((unsigned char *)p->toks_end - (unsigned char*)p->toks)/sizeof(*p->toks);
   3336 	int res =
   3337 		jsmn_parse(&p->json_parser, p->json, p->json_len, p->toks, cap, cb != NULL);
   3338 
   3339 	// got an ID!
   3340 	if (res == -42) {
   3341 		tok = &p->toks[p->json_parser.toknext-1];
   3342 
   3343 		switch (cb->fn(cb->data, p->json + tok->start)) {
   3344 		case NDB_IDRES_CONT:
   3345 			res = jsmn_parse(&p->json_parser, p->json, p->json_len,
   3346 					 p->toks, cap, 0);
   3347 			break;
   3348 		case NDB_IDRES_STOP:
   3349 			return -42;
   3350 		}
   3351 	} else if (res == 0) {
   3352 		return 0;
   3353 	}
   3354 
   3355 	p->num_tokens = res;
   3356 	p->i = 0;
   3357 
   3358 	return 1;
   3359 }
   3360 
   3361 static inline int toksize(jsmntok_t *tok)
   3362 {
   3363 	return tok->end - tok->start;
   3364 }
   3365 
   3366 
   3367 
   3368 static int cursor_push_unescaped_char(struct cursor *cur, char c1, char c2)
   3369 {
   3370 	switch (c2) {
   3371 	case 't':  return cursor_push_byte(cur, '\t');
   3372 	case 'n':  return cursor_push_byte(cur, '\n');
   3373 	case 'r':  return cursor_push_byte(cur, '\r');
   3374 	case 'b':  return cursor_push_byte(cur, '\b');
   3375 	case 'f':  return cursor_push_byte(cur, '\f');
   3376 	case '\\': return cursor_push_byte(cur, '\\');
   3377 	case '/':  return cursor_push_byte(cur, '/');
   3378 	case '"':  return cursor_push_byte(cur, '"');
   3379 	case 'u':
   3380 		// these aren't handled yet
   3381 		return 0;
   3382 	default:
   3383 		return cursor_push_byte(cur, c1) && cursor_push_byte(cur, c2);
   3384 	}
   3385 }
   3386 
   3387 static int cursor_push_escaped_char(struct cursor *cur, char c)
   3388 {
   3389         switch (c) {
   3390         case '"':  return cursor_push_str(cur, "\\\"");
   3391         case '\\': return cursor_push_str(cur, "\\\\");
   3392         case '\b': return cursor_push_str(cur, "\\b");
   3393         case '\f': return cursor_push_str(cur, "\\f");
   3394         case '\n': return cursor_push_str(cur, "\\n");
   3395         case '\r': return cursor_push_str(cur, "\\r");
   3396         case '\t': return cursor_push_str(cur, "\\t");
   3397         // TODO: \u hex hex hex hex
   3398         }
   3399         return cursor_push_byte(cur, c);
   3400 }
   3401 
   3402 static int cursor_push_hex_str(struct cursor *cur, unsigned char *buf, int len)
   3403 {
   3404 	int i;
   3405 
   3406 	if (len % 2 != 0)
   3407 		return 0;
   3408 
   3409         if (!cursor_push_byte(cur, '"'))
   3410                 return 0;
   3411 
   3412 	for (i = 0; i < len; i++) {
   3413 		unsigned int c = ((const unsigned char *)buf)[i];
   3414 		if (!cursor_push_byte(cur, hexchar(c >> 4)))
   3415 			return 0;
   3416 		if (!cursor_push_byte(cur, hexchar(c & 0xF)))
   3417 			return 0;
   3418 	}
   3419 
   3420         if (!cursor_push_byte(cur, '"'))
   3421                 return 0;
   3422 
   3423 	return 1;
   3424 }
   3425 
   3426 static int cursor_push_jsonstr(struct cursor *cur, const char *str)
   3427 {
   3428 	int i;
   3429         int len;
   3430 
   3431 	len = strlen(str);
   3432 
   3433         if (!cursor_push_byte(cur, '"'))
   3434                 return 0;
   3435 
   3436         for (i = 0; i < len; i++) {
   3437                 if (!cursor_push_escaped_char(cur, str[i]))
   3438                         return 0;
   3439         }
   3440 
   3441         if (!cursor_push_byte(cur, '"'))
   3442                 return 0;
   3443 
   3444         return 1;
   3445 }
   3446 
   3447 
   3448 static inline int cursor_push_json_tag_str(struct cursor *cur, struct ndb_str str)
   3449 {
   3450 	if (str.flag == NDB_PACKED_ID)
   3451 		return cursor_push_hex_str(cur, str.id, 32);
   3452 
   3453 	return cursor_push_jsonstr(cur, str.str);
   3454 }
   3455 
   3456 static int cursor_push_json_tag(struct cursor *cur, struct ndb_note *note,
   3457 				struct ndb_tag *tag)
   3458 {
   3459         int i;
   3460 
   3461         if (!cursor_push_byte(cur, '['))
   3462                 return 0;
   3463 
   3464         for (i = 0; i < tag->count; i++) {
   3465                 if (!cursor_push_json_tag_str(cur, ndb_note_str(note, &tag->strs[i])))
   3466                         return 0;
   3467                 if (i != tag->count-1 && !cursor_push_byte(cur, ','))
   3468 			return 0;
   3469         }
   3470 
   3471         return cursor_push_byte(cur, ']');
   3472 }
   3473 
   3474 static int cursor_push_json_tags(struct cursor *cur, struct ndb_note *note)
   3475 {
   3476 	int i;
   3477 	struct ndb_iterator iter, *it = &iter;
   3478 	ndb_tags_iterate_start(note, it);
   3479 
   3480         if (!cursor_push_byte(cur, '['))
   3481                 return 0;
   3482 
   3483 	i = 0;
   3484 	while (ndb_tags_iterate_next(it)) {
   3485 		if (!cursor_push_json_tag(cur, note, it->tag))
   3486 			return 0;
   3487                 if (i != note->tags.count-1 && !cursor_push_str(cur, ","))
   3488 			return 0;
   3489 		i++;
   3490 	}
   3491 
   3492         if (!cursor_push_byte(cur, ']'))
   3493                 return 0;
   3494 
   3495 	return 1;
   3496 }
   3497 
   3498 static int ndb_event_commitment(struct ndb_note *ev, unsigned char *buf, int buflen)
   3499 {
   3500 	char timebuf[16] = {0};
   3501 	char kindbuf[16] = {0};
   3502 	char pubkey[65];
   3503 	struct cursor cur;
   3504 	int ok;
   3505 
   3506 	if (!hex_encode(ev->pubkey, sizeof(ev->pubkey), pubkey, sizeof(pubkey)))
   3507 		return 0;
   3508 
   3509 	make_cursor(buf, buf + buflen, &cur);
   3510 
   3511 	// TODO: update in 2106 ...
   3512 	snprintf(timebuf, sizeof(timebuf), "%d", (uint32_t)ev->created_at);
   3513 	snprintf(kindbuf, sizeof(kindbuf), "%d", ev->kind);
   3514 
   3515 	ok =
   3516 		cursor_push_str(&cur, "[0,\"") &&
   3517 		cursor_push_str(&cur, pubkey) &&
   3518 		cursor_push_str(&cur, "\",") &&
   3519 		cursor_push_str(&cur, timebuf) &&
   3520 		cursor_push_str(&cur, ",") &&
   3521 		cursor_push_str(&cur, kindbuf) &&
   3522 		cursor_push_str(&cur, ",") &&
   3523 		cursor_push_json_tags(&cur, ev) &&
   3524 		cursor_push_str(&cur, ",") &&
   3525 		cursor_push_jsonstr(&cur, ndb_note_str(ev, &ev->content).str) &&
   3526 		cursor_push_str(&cur, "]");
   3527 
   3528 	if (!ok)
   3529 		return 0;
   3530 
   3531 	return cur.p - cur.start;
   3532 }
   3533 
   3534 int ndb_calculate_id(struct ndb_note *note, unsigned char *buf, int buflen) {
   3535 	int len;
   3536 
   3537 	if (!(len = ndb_event_commitment(note, buf, buflen)))
   3538 		return 0;
   3539 
   3540 	//fprintf(stderr, "%.*s\n", len, buf);
   3541 
   3542 	sha256((struct sha256*)note->id, buf, len);
   3543 
   3544 	return 1;
   3545 }
   3546 
   3547 int ndb_sign_id(struct ndb_keypair *keypair, unsigned char id[32],
   3548 		unsigned char sig[64])
   3549 {
   3550 	unsigned char aux[32];
   3551 	secp256k1_keypair *pair = (secp256k1_keypair*) keypair->pair;
   3552 
   3553 	if (!fill_random(aux, sizeof(aux)))
   3554 		return 0;
   3555 
   3556 	secp256k1_context *ctx =
   3557 		secp256k1_context_create(SECP256K1_CONTEXT_NONE);
   3558 
   3559 	return secp256k1_schnorrsig_sign32(ctx, sig, id, pair, aux);
   3560 }
   3561 
   3562 int ndb_create_keypair(struct ndb_keypair *kp)
   3563 {
   3564 	secp256k1_keypair *keypair = (secp256k1_keypair*)kp->pair;
   3565 	secp256k1_xonly_pubkey pubkey;
   3566 
   3567 	secp256k1_context *ctx =
   3568 		secp256k1_context_create(SECP256K1_CONTEXT_NONE);;
   3569 
   3570 	/* Try to create a keypair with a valid context, it should only
   3571 	 * fail if the secret key is zero or out of range. */
   3572 	if (!secp256k1_keypair_create(ctx, keypair, kp->secret))
   3573 		return 0;
   3574 
   3575 	if (!secp256k1_keypair_xonly_pub(ctx, &pubkey, NULL, keypair))
   3576 		return 0;
   3577 
   3578 	/* Serialize the public key. Should always return 1 for a valid public key. */
   3579 	return secp256k1_xonly_pubkey_serialize(ctx, kp->pubkey, &pubkey);
   3580 }
   3581 
   3582 int ndb_decode_key(const char *secstr, struct ndb_keypair *keypair)
   3583 {
   3584 	if (!hex_decode(secstr, strlen(secstr), keypair->secret, 32)) {
   3585 		fprintf(stderr, "could not hex decode secret key\n");
   3586 		return 0;
   3587 	}
   3588 
   3589 	return ndb_create_keypair(keypair);
   3590 }
   3591 
   3592 int ndb_builder_finalize(struct ndb_builder *builder, struct ndb_note **note,
   3593 			 struct ndb_keypair *keypair)
   3594 {
   3595 	int strings_len = builder->strings.p - builder->strings.start;
   3596 	unsigned char *note_end = builder->note_cur.p + strings_len;
   3597 	int total_size = note_end - builder->note_cur.start;
   3598 
   3599 	// move the strings buffer next to the end of our ndb_note
   3600 	memmove(builder->note_cur.p, builder->strings.start, strings_len);
   3601 
   3602 	// set the strings location
   3603 	builder->note->strings = builder->note_cur.p - builder->note_cur.start;
   3604 
   3605 	// record the total size
   3606 	//builder->note->size = total_size;
   3607 
   3608 	*note = builder->note;
   3609 
   3610 	// generate id and sign if we're building this manually
   3611 	if (keypair) {
   3612 		// use the remaining memory for building our id buffer
   3613 		unsigned char *end   = builder->mem.end;
   3614 		unsigned char *start = (unsigned char*)(*note) + total_size;
   3615 
   3616 		ndb_builder_set_pubkey(builder, keypair->pubkey);
   3617 
   3618 		if (!ndb_calculate_id(builder->note, start, end - start))
   3619 			return 0;
   3620 
   3621 		if (!ndb_sign_id(keypair, (*note)->id, (*note)->sig))
   3622 			return 0;
   3623 	}
   3624 
   3625 	// make sure we're aligned as a whole
   3626 	total_size = (total_size + 7) & ~7;
   3627 	assert((total_size % 8) == 0);
   3628 	return total_size;
   3629 }
   3630 
   3631 struct ndb_note * ndb_builder_note(struct ndb_builder *builder)
   3632 {
   3633 	return builder->note;
   3634 }
   3635 
   3636 /// find an existing string via str_indices. these indices only exist in the
   3637 /// builder phase just for this purpose.
   3638 static inline int ndb_builder_find_str(struct ndb_builder *builder,
   3639 				       const char *str, int len,
   3640 				       union ndb_packed_str *pstr)
   3641 {
   3642 	// find existing matching string to avoid duplicate strings
   3643 	int indices = cursor_count(&builder->str_indices, sizeof(uint32_t));
   3644 	for (int i = 0; i < indices; i++) {
   3645 		uint32_t index = ((uint32_t*)builder->str_indices.start)[i];
   3646 		const char *some_str = (const char*)builder->strings.start + index;
   3647 
   3648 		if (!memcmp(some_str, str, len)) {
   3649 			// found an existing matching str, use that index
   3650 			*pstr = ndb_offset_str(index);
   3651 			return 1;
   3652 		}
   3653 	}
   3654 
   3655 	return 0;
   3656 }
   3657 
   3658 static int ndb_builder_push_str(struct ndb_builder *builder, const char *str,
   3659 				int len, union ndb_packed_str *pstr)
   3660 {
   3661 	uint32_t loc;
   3662 
   3663 	// no string found, push a new one
   3664 	loc = builder->strings.p - builder->strings.start;
   3665 	if (!(cursor_push(&builder->strings, (unsigned char*)str, len) &&
   3666 	      cursor_push_byte(&builder->strings, '\0'))) {
   3667 		return 0;
   3668 	}
   3669 
   3670 	*pstr = ndb_offset_str(loc);
   3671 
   3672 	// record in builder indices. ignore return value, if we can't cache it
   3673 	// then whatever
   3674 	cursor_push_u32(&builder->str_indices, loc);
   3675 
   3676 	return 1;
   3677 }
   3678 
   3679 static int ndb_builder_push_packed_id(struct ndb_builder *builder,
   3680 				      unsigned char *id,
   3681 				      union ndb_packed_str *pstr)
   3682 {
   3683 	// Don't both find id duplicates. very rarely are they duplicated
   3684 	// and it slows things down quite a bit. If we really care about this
   3685 	// We can switch to a hash table.
   3686 	//if (ndb_builder_find_str(builder, (const char*)id, 32, pstr)) {
   3687 	//	pstr->packed.flag = NDB_PACKED_ID;
   3688 	//	return 1;
   3689 	//}
   3690 
   3691 	if (ndb_builder_push_str(builder, (const char*)id, 32, pstr)) {
   3692 		pstr->packed.flag = NDB_PACKED_ID;
   3693 		return 1;
   3694 	}
   3695 
   3696 	return 0;
   3697 }
   3698 
   3699 
   3700 /// Check for small strings to pack
   3701 static inline int ndb_builder_try_compact_str(struct ndb_builder *builder,
   3702 					      const char *str, int len,
   3703 					      union ndb_packed_str *pstr,
   3704 					      int pack_ids)
   3705 {
   3706 	unsigned char id_buf[32];
   3707 
   3708 	if (len == 0) {
   3709 		*pstr = ndb_char_to_packed_str(0);
   3710 		return 1;
   3711 	} else if (len == 1) {
   3712 		*pstr = ndb_char_to_packed_str(str[0]);
   3713 		return 1;
   3714 	} else if (len == 2) {
   3715 		*pstr = ndb_chars_to_packed_str(str[0], str[1]);
   3716 		return 1;
   3717 	} else if (pack_ids && len == 64 && hex_decode(str, 64, id_buf, 32)) {
   3718 		return ndb_builder_push_packed_id(builder, id_buf, pstr);
   3719 	}
   3720 
   3721 	return 0;
   3722 }
   3723 
   3724 
   3725 static int ndb_builder_push_unpacked_str(struct ndb_builder *builder,
   3726 					 const char *str, int len,
   3727 					 union ndb_packed_str *pstr)
   3728 {
   3729 	if (ndb_builder_find_str(builder, str, len, pstr))
   3730 		return 1;
   3731 
   3732 	return ndb_builder_push_str(builder, str, len, pstr);
   3733 }
   3734 
   3735 int ndb_builder_make_str(struct ndb_builder *builder, const char *str, int len,
   3736 			 union ndb_packed_str *pstr, int pack_ids)
   3737 {
   3738 	if (ndb_builder_try_compact_str(builder, str, len, pstr, pack_ids))
   3739 		return 1;
   3740 
   3741 	return ndb_builder_push_unpacked_str(builder, str, len, pstr);
   3742 }
   3743 
   3744 int ndb_builder_set_content(struct ndb_builder *builder, const char *content,
   3745 			    int len)
   3746 {
   3747 	int pack_ids = 0;
   3748 	builder->note->content_length = len;
   3749 	return ndb_builder_make_str(builder, content, len,
   3750 				    &builder->note->content, pack_ids);
   3751 }
   3752 
   3753 
   3754 static inline int jsoneq(const char *json, jsmntok_t *tok, int tok_len,
   3755 			 const char *s)
   3756 {
   3757 	if (tok->type == JSMN_STRING && (int)strlen(s) == tok_len &&
   3758 	    memcmp(json + tok->start, s, tok_len) == 0) {
   3759 		return 1;
   3760 	}
   3761 	return 0;
   3762 }
   3763 
   3764 static int ndb_builder_finalize_tag(struct ndb_builder *builder,
   3765 				    union ndb_packed_str offset)
   3766 {
   3767 	if (!cursor_push_u32(&builder->note_cur, offset.offset))
   3768 		return 0;
   3769 	builder->current_tag->count++;
   3770 	return 1;
   3771 }
   3772 
   3773 /// Unescape and push json strings
   3774 static int ndb_builder_make_json_str(struct ndb_builder *builder,
   3775 				     const char *str, int len,
   3776 				     union ndb_packed_str *pstr,
   3777 				     int *written, int pack_ids)
   3778 {
   3779 	// let's not care about de-duping these. we should just unescape
   3780 	// in-place directly into the strings table. 
   3781 	if (written)
   3782 		*written = len;
   3783 
   3784 	const char *p, *end, *start;
   3785 	unsigned char *builder_start;
   3786 
   3787 	// always try compact strings first
   3788 	if (ndb_builder_try_compact_str(builder, str, len, pstr, pack_ids))
   3789 		return 1;
   3790 
   3791 	end = str + len;
   3792 	start = str; // Initialize start to the beginning of the string
   3793 
   3794 	*pstr = ndb_offset_str(builder->strings.p - builder->strings.start);
   3795 	builder_start = builder->strings.p;
   3796 
   3797 	for (p = str; p < end; p++) {
   3798 		if (*p == '\\' && p+1 < end) {
   3799 			// Push the chunk of unescaped characters before this escape sequence
   3800 			if (start < p && !cursor_push(&builder->strings,
   3801 						(unsigned char *)start,
   3802 						p - start)) {
   3803 				return 0;
   3804 			}
   3805 
   3806 			if (!cursor_push_unescaped_char(&builder->strings, *p, *(p+1)))
   3807 				return 0;
   3808 
   3809 			p++; // Skip the character following the backslash
   3810 			start = p + 1; // Update the start pointer to the next character
   3811 		}
   3812 	}
   3813 
   3814 	// Handle the last chunk after the last escape sequence (or if there are no escape sequences at all)
   3815 	if (start < p && !cursor_push(&builder->strings, (unsigned char *)start,
   3816 				      p - start)) {
   3817 		return 0;
   3818 	}
   3819 
   3820 	if (written)
   3821 		*written = builder->strings.p - builder_start;
   3822 
   3823 	// TODO: dedupe these!?
   3824 	return cursor_push_byte(&builder->strings, '\0');
   3825 }
   3826 
   3827 static int ndb_builder_push_json_tag(struct ndb_builder *builder,
   3828 				     const char *str, int len)
   3829 {
   3830 	union ndb_packed_str pstr;
   3831 	int pack_ids = 1;
   3832 	if (!ndb_builder_make_json_str(builder, str, len, &pstr, NULL, pack_ids))
   3833 		return 0;
   3834 	return ndb_builder_finalize_tag(builder, pstr);
   3835 }
   3836 
   3837 // Push a json array into an ndb tag ["p", "abcd..."] -> struct ndb_tag
   3838 static int ndb_builder_tag_from_json_array(struct ndb_json_parser *p,
   3839 					   jsmntok_t *array)
   3840 {
   3841 	jsmntok_t *str_tok;
   3842 	const char *str;
   3843 
   3844 	if (array->size == 0)
   3845 		return 0;
   3846 
   3847 	if (!ndb_builder_new_tag(&p->builder))
   3848 		return 0;
   3849 
   3850 	for (int i = 0; i < array->size; i++) {
   3851 		str_tok = &array[i+1];
   3852 		str = p->json + str_tok->start;
   3853 
   3854 		if (!ndb_builder_push_json_tag(&p->builder, str,
   3855 					       toksize(str_tok))) {
   3856 			return 0;
   3857 		}
   3858 	}
   3859 
   3860 	return 1;
   3861 }
   3862 
   3863 // Push json tags into ndb data
   3864 //   [["t", "hashtag"], ["p", "abcde..."]] -> struct ndb_tags
   3865 static inline int ndb_builder_process_json_tags(struct ndb_json_parser *p,
   3866 						jsmntok_t *array)
   3867 {
   3868 	jsmntok_t *tag = array;
   3869 
   3870 	if (array->size == 0)
   3871 		return 1;
   3872 
   3873 	for (int i = 0; i < array->size; i++) {
   3874 		if (!ndb_builder_tag_from_json_array(p, &tag[i+1]))
   3875 			return 0;
   3876 		tag += tag[i+1].size;
   3877 	}
   3878 
   3879 	return 1;
   3880 }
   3881 
   3882 static int parse_unsigned_int(const char *start, int len, unsigned int *num)
   3883 {
   3884 	unsigned int number = 0;
   3885 	const char *p = start, *end = start + len;
   3886 	int digits = 0;
   3887 
   3888 	while (p < end) {
   3889 		char c = *p;
   3890 
   3891 		if (c < '0' || c > '9')
   3892 			break;
   3893 
   3894 		// Check for overflow
   3895 		char digit = c - '0';
   3896 		if (number > (UINT_MAX - digit) / 10)
   3897 			return 0; // Overflow detected
   3898 
   3899 		number = number * 10 + digit;
   3900 
   3901 		p++;
   3902 		digits++;
   3903 	}
   3904 
   3905 	if (digits == 0)
   3906 		return 0;
   3907 
   3908 	*num = number;
   3909 	return 1;
   3910 }
   3911 
   3912 int ndb_client_event_from_json(const char *json, int len, struct ndb_fce *fce,
   3913 			       unsigned char *buf, int bufsize, struct ndb_id_cb *cb)
   3914 {
   3915 	jsmntok_t *tok = NULL;
   3916 	int tok_len, res;
   3917 	struct ndb_json_parser parser;
   3918 
   3919 	ndb_json_parser_init(&parser, json, len, buf, bufsize);
   3920 
   3921 	if ((res = ndb_json_parser_parse(&parser, cb)) < 0)
   3922 		return res;
   3923 
   3924 	if (parser.num_tokens <= 3 || parser.toks[0].type != JSMN_ARRAY)
   3925 		return 0;
   3926 
   3927 	parser.i = 1;
   3928 	tok = &parser.toks[parser.i++];
   3929 	tok_len = toksize(tok);
   3930 	if (tok->type != JSMN_STRING)
   3931 		return 0;
   3932 
   3933 	if (tok_len == 5 && !memcmp("EVENT", json + tok->start, 5)) {
   3934 		fce->evtype = NDB_FCE_EVENT;
   3935 		struct ndb_event *ev = &fce->event;
   3936 		return ndb_parse_json_note(&parser, &ev->note);
   3937 	}
   3938 
   3939 	return 0;
   3940 }
   3941 
   3942 
   3943 int ndb_ws_event_from_json(const char *json, int len, struct ndb_tce *tce,
   3944 			   unsigned char *buf, int bufsize,
   3945 			   struct ndb_id_cb *cb)
   3946 {
   3947 	jsmntok_t *tok = NULL;
   3948 	int tok_len, res;
   3949 	struct ndb_json_parser parser;
   3950 
   3951 	tce->subid_len = 0;
   3952 	tce->subid = "";
   3953 
   3954 	ndb_json_parser_init(&parser, json, len, buf, bufsize);
   3955 
   3956 	if ((res = ndb_json_parser_parse(&parser, cb)) < 0)
   3957 		return res;
   3958 
   3959 	if (parser.num_tokens < 3 || parser.toks[0].type != JSMN_ARRAY)
   3960 		return 0;
   3961 
   3962 	parser.i = 1;
   3963 	tok = &parser.toks[parser.i++];
   3964 	tok_len = toksize(tok);
   3965 	if (tok->type != JSMN_STRING)
   3966 		return 0;
   3967 
   3968 	if (tok_len == 5 && !memcmp("EVENT", json + tok->start, 5)) {
   3969 		tce->evtype = NDB_TCE_EVENT;
   3970 		struct ndb_event *ev = &tce->event;
   3971 
   3972 		tok = &parser.toks[parser.i++];
   3973 		if (tok->type != JSMN_STRING)
   3974 			return 0;
   3975 
   3976 		tce->subid = json + tok->start;
   3977 		tce->subid_len = toksize(tok);
   3978 
   3979 		return ndb_parse_json_note(&parser, &ev->note);
   3980 	} else if (tok_len == 4 && !memcmp("EOSE", json + tok->start, 4)) { 
   3981 		tce->evtype = NDB_TCE_EOSE;
   3982 
   3983 		tok = &parser.toks[parser.i++];
   3984 		if (tok->type != JSMN_STRING)
   3985 			return 0;
   3986 
   3987 		tce->subid = json + tok->start;
   3988 		tce->subid_len = toksize(tok);
   3989 		return 1;
   3990 	} else if (tok_len == 2 && !memcmp("OK", json + tok->start, 2)) {
   3991 		if (parser.num_tokens != 5)
   3992 			return 0;
   3993 
   3994 		struct ndb_command_result *cr = &tce->command_result;
   3995 
   3996 		tce->evtype = NDB_TCE_OK;
   3997 
   3998 		tok = &parser.toks[parser.i++];
   3999 		if (tok->type != JSMN_STRING)
   4000 			return 0;
   4001 
   4002 		tce->subid = json + tok->start;
   4003 		tce->subid_len = toksize(tok);
   4004 
   4005 		tok = &parser.toks[parser.i++];
   4006 		if (tok->type != JSMN_PRIMITIVE || toksize(tok) == 0)
   4007 			return 0;
   4008 
   4009 		cr->ok = (json + tok->start)[0] == 't';
   4010 
   4011 		tok = &parser.toks[parser.i++];
   4012 		if (tok->type != JSMN_STRING)
   4013 			return 0;
   4014 
   4015 		tce->command_result.msg = json + tok->start;
   4016 		tce->command_result.msglen = toksize(tok);
   4017 
   4018 		return 1;
   4019 	} else if (tok_len == 4 && !memcmp("AUTH", json + tok->start, 4)) {
   4020 		tce->evtype = NDB_TCE_AUTH;
   4021 
   4022 		tok = &parser.toks[parser.i++];
   4023 		if (tok->type != JSMN_STRING)
   4024 			return 0;
   4025 
   4026 		tce->subid = json + tok->start;
   4027 		tce->subid_len = toksize(tok);
   4028 
   4029 		return 1;
   4030 	}
   4031 
   4032 	return 0;
   4033 }
   4034 
   4035 int ndb_parse_json_note(struct ndb_json_parser *parser, struct ndb_note **note)
   4036 {
   4037 	jsmntok_t *tok = NULL;
   4038 	unsigned char hexbuf[64];
   4039 	const char *json = parser->json;
   4040 	const char *start;
   4041 	int i, tok_len, parsed;
   4042 
   4043 	parsed = 0;
   4044 
   4045 	if (parser->toks[parser->i].type != JSMN_OBJECT)
   4046 		return 0;
   4047 
   4048 	// TODO: build id buffer and verify at end
   4049 
   4050 	for (i = parser->i + 1; i < parser->num_tokens; i++) {
   4051 		tok = &parser->toks[i];
   4052 		start = json + tok->start;
   4053 		tok_len = toksize(tok);
   4054 
   4055 		//printf("toplevel %.*s %d\n", tok_len, json + tok->start, tok->type);
   4056 		if (tok_len == 0 || i + 1 >= parser->num_tokens)
   4057 			continue;
   4058 
   4059 		if (start[0] == 'p' && jsoneq(json, tok, tok_len, "pubkey")) {
   4060 			// pubkey
   4061 			tok = &parser->toks[i+1];
   4062 			hex_decode(json + tok->start, toksize(tok), hexbuf, sizeof(hexbuf));
   4063 			parsed |= NDB_PARSED_PUBKEY;
   4064 			ndb_builder_set_pubkey(&parser->builder, hexbuf);
   4065 		} else if (tok_len == 2 && start[0] == 'i' && start[1] == 'd') {
   4066 			// id
   4067 			tok = &parser->toks[i+1];
   4068 			hex_decode(json + tok->start, toksize(tok), hexbuf, sizeof(hexbuf));
   4069 			parsed |= NDB_PARSED_ID;
   4070 			ndb_builder_set_id(&parser->builder, hexbuf);
   4071 		} else if (tok_len == 3 && start[0] == 's' && start[1] == 'i' && start[2] == 'g') {
   4072 			// sig
   4073 			tok = &parser->toks[i+1];
   4074 			hex_decode(json + tok->start, toksize(tok), hexbuf, sizeof(hexbuf));
   4075 			parsed |= NDB_PARSED_SIG;
   4076 			ndb_builder_set_sig(&parser->builder, hexbuf);
   4077 		} else if (start[0] == 'k' && jsoneq(json, tok, tok_len, "kind")) {
   4078 			// kind
   4079 			tok = &parser->toks[i+1];
   4080 			start = json + tok->start;
   4081 			if (tok->type != JSMN_PRIMITIVE || tok_len <= 0)
   4082 				return 0;
   4083 			if (!parse_unsigned_int(start, toksize(tok),
   4084 						&parser->builder.note->kind))
   4085 					return 0;
   4086 			parsed |= NDB_PARSED_KIND;
   4087 		} else if (start[0] == 'c') {
   4088 			if (jsoneq(json, tok, tok_len, "created_at")) {
   4089 				// created_at
   4090 				tok = &parser->toks[i+1];
   4091 				start = json + tok->start;
   4092 				if (tok->type != JSMN_PRIMITIVE || tok_len <= 0)
   4093 					return 0;
   4094 				// TODO: update to int64 in 2106 ... xD
   4095 				unsigned int bigi;
   4096 				if (!parse_unsigned_int(start, toksize(tok), &bigi))
   4097 					return 0;
   4098 				parser->builder.note->created_at = bigi;
   4099 				parsed |= NDB_PARSED_CREATED_AT;
   4100 			} else if (jsoneq(json, tok, tok_len, "content")) {
   4101 				// content
   4102 				tok = &parser->toks[i+1];
   4103 				union ndb_packed_str pstr;
   4104 				tok_len = toksize(tok);
   4105 				int written, pack_ids = 0;
   4106 				if (!ndb_builder_make_json_str(&parser->builder,
   4107 							json + tok->start,
   4108 							tok_len, &pstr,
   4109 							&written, pack_ids)) {
   4110 					ndb_debug("ndb_builder_make_json_str failed\n");
   4111 					return 0;
   4112 				}
   4113 				parser->builder.note->content_length = written;
   4114 				parser->builder.note->content = pstr;
   4115 				parsed |= NDB_PARSED_CONTENT;
   4116 			}
   4117 		} else if (start[0] == 't' && jsoneq(json, tok, tok_len, "tags")) {
   4118 			tok = &parser->toks[i+1];
   4119 			ndb_builder_process_json_tags(parser, tok);
   4120 			i += tok->size;
   4121 			parsed |= NDB_PARSED_TAGS;
   4122 		}
   4123 	}
   4124 
   4125 	//ndb_debug("parsed %d = %d, &->%d", parsed, NDB_PARSED_ALL, parsed & NDB_PARSED_ALL);
   4126 	if (parsed != NDB_PARSED_ALL)
   4127 		return 0;
   4128 
   4129 	return ndb_builder_finalize(&parser->builder, note, NULL);
   4130 }
   4131 
   4132 int ndb_note_from_json(const char *json, int len, struct ndb_note **note,
   4133 		       unsigned char *buf, int bufsize)
   4134 {
   4135 	struct ndb_json_parser parser;
   4136 	int res;
   4137 
   4138 	ndb_json_parser_init(&parser, json, len, buf, bufsize);
   4139 	if ((res = ndb_json_parser_parse(&parser, NULL)) < 0)
   4140 		return res;
   4141 
   4142 	if (parser.num_tokens < 1)
   4143 		return 0;
   4144 
   4145 	return ndb_parse_json_note(&parser, note);
   4146 }
   4147 
   4148 void ndb_builder_set_pubkey(struct ndb_builder *builder, unsigned char *pubkey)
   4149 {
   4150 	memcpy(builder->note->pubkey, pubkey, 32);
   4151 }
   4152 
   4153 void ndb_builder_set_id(struct ndb_builder *builder, unsigned char *id)
   4154 {
   4155 	memcpy(builder->note->id, id, 32);
   4156 }
   4157 
   4158 void ndb_builder_set_sig(struct ndb_builder *builder, unsigned char *sig)
   4159 {
   4160 	memcpy(builder->note->sig, sig, 64);
   4161 }
   4162 
   4163 void ndb_builder_set_kind(struct ndb_builder *builder, uint32_t kind)
   4164 {
   4165 	builder->note->kind = kind;
   4166 }
   4167 
   4168 void ndb_builder_set_created_at(struct ndb_builder *builder, uint64_t created_at)
   4169 {
   4170 	builder->note->created_at = created_at;
   4171 }
   4172 
   4173 int ndb_builder_new_tag(struct ndb_builder *builder)
   4174 {
   4175 	builder->note->tags.count++;
   4176 	struct ndb_tag tag = {0};
   4177 	builder->current_tag = (struct ndb_tag *)builder->note_cur.p;
   4178 	return cursor_push_tag(&builder->note_cur, &tag);
   4179 }
   4180 
   4181 void ndb_stat_counts_init(struct ndb_stat_counts *counts)
   4182 {
   4183 	counts->count = 0;
   4184 	counts->key_size = 0;
   4185 	counts->value_size = 0;
   4186 }
   4187 
   4188 static void ndb_stat_init(struct ndb_stat *stat)
   4189 {
   4190 	// init stats
   4191 	int i;
   4192 
   4193 	for (i = 0; i < NDB_CKIND_COUNT; i++) {
   4194 		ndb_stat_counts_init(&stat->common_kinds[i]);
   4195 	}
   4196 
   4197 	for (i = 0; i < NDB_DBS; i++) {
   4198 		ndb_stat_counts_init(&stat->dbs[i]);
   4199 	}
   4200 
   4201 	ndb_stat_counts_init(&stat->other_kinds);
   4202 }
   4203 
   4204 int ndb_stat(struct ndb *ndb, struct ndb_stat *stat)
   4205 {
   4206 	int rc;
   4207 	MDB_cursor *cur;
   4208 	MDB_val k, v;
   4209 	MDB_dbi db;
   4210 	struct ndb_txn txn;
   4211 	struct ndb_note *note;
   4212 	int i;
   4213 	enum ndb_common_kind common_kind;
   4214 
   4215 	// initialize to 0
   4216 	ndb_stat_init(stat);
   4217 
   4218 	if (!ndb_begin_query(ndb, &txn)) {
   4219 		fprintf(stderr, "ndb_stat failed at ndb_begin_query\n");
   4220 		return 0;
   4221 	}
   4222 
   4223 	// stat each dbi in the database
   4224 	for (i = 0; i < NDB_DBS; i++)
   4225 	{
   4226 		db = ndb->lmdb.dbs[i];
   4227 
   4228 		if ((rc = mdb_cursor_open(txn.mdb_txn, db, &cur))) {
   4229 			fprintf(stderr, "ndb_stat: mdb_cursor_open failed, error '%s'\n",
   4230 					mdb_strerror(rc));
   4231 			return 0;
   4232 		}
   4233 
   4234 		// loop over every entry and count kv sizes
   4235 		while (mdb_cursor_get(cur, &k, &v, MDB_NEXT) == 0) {
   4236 			// we gather more detailed per-kind stats if we're in
   4237 			// the notes db
   4238 			if (i == NDB_DB_NOTE) {
   4239 				note = v.mv_data;
   4240 				common_kind = ndb_kind_to_common_kind(note->kind);
   4241 
   4242 				// uncommon kind? just count them in bulk
   4243 				if (common_kind == -1) {
   4244 					stat->other_kinds.count++;
   4245 					stat->other_kinds.key_size += k.mv_size;
   4246 					stat->other_kinds.value_size += v.mv_size;
   4247 				} else {
   4248 					stat->common_kinds[common_kind].count++;
   4249 					stat->common_kinds[common_kind].key_size += k.mv_size;
   4250 					stat->common_kinds[common_kind].value_size += v.mv_size;
   4251 				}
   4252 			}
   4253 
   4254 			stat->dbs[i].count++;
   4255 			stat->dbs[i].key_size += k.mv_size;
   4256 			stat->dbs[i].value_size += v.mv_size;
   4257 		}
   4258 
   4259 		// close the cursor, they are per-dbi
   4260 		mdb_cursor_close(cur);
   4261 	}
   4262 
   4263 	ndb_end_query(&txn);
   4264 
   4265 	return 1;
   4266 }
   4267 
   4268 /// Push an element to the current tag
   4269 /// 
   4270 /// Basic idea is to call ndb_builder_new_tag
   4271 inline int ndb_builder_push_tag_str(struct ndb_builder *builder,
   4272 				    const char *str, int len)
   4273 {
   4274 	union ndb_packed_str pstr;
   4275 	int pack_ids = 1;
   4276 	if (!ndb_builder_make_str(builder, str, len, &pstr, pack_ids))
   4277 		return 0;
   4278 	return ndb_builder_finalize_tag(builder, pstr);
   4279 }
   4280 
   4281 //
   4282 // CONFIG
   4283 // 
   4284 void ndb_default_config(struct ndb_config *config)
   4285 {
   4286 	int cores = get_cpu_cores();
   4287 	config->mapsize = 1024UL * 1024UL * 1024UL * 32UL; // 32 GiB
   4288 	config->ingester_threads = cores == -1 ? 4 : cores;
   4289 	config->flags = 0;
   4290 	config->ingest_filter = NULL;
   4291 	config->filter_context = NULL;
   4292 }
   4293 
   4294 void ndb_config_set_ingest_threads(struct ndb_config *config, int threads)
   4295 {
   4296 	config->ingester_threads = threads;
   4297 }
   4298 
   4299 void ndb_config_set_flags(struct ndb_config *config, int flags)
   4300 {
   4301 	config->flags = flags;
   4302 }
   4303 
   4304 void ndb_config_set_mapsize(struct ndb_config *config, size_t mapsize)
   4305 {
   4306 	config->mapsize = mapsize;
   4307 }
   4308 
   4309 void ndb_config_set_ingest_filter(struct ndb_config *config,
   4310 				  ndb_ingest_filter_fn fn, void *filter_ctx)
   4311 {
   4312 	config->ingest_filter = fn;
   4313 	config->filter_context = filter_ctx;
   4314 }
   4315 
   4316 // used by ndb.c
   4317 int ndb_print_search_keys(struct ndb_txn *txn)
   4318 {
   4319 	MDB_cursor *cur;
   4320 	MDB_val k, v;
   4321 	int i;
   4322 	struct ndb_text_search_key search_key;
   4323 
   4324 	if (mdb_cursor_open(txn->mdb_txn, txn->lmdb->dbs[NDB_DB_NOTE_TEXT], &cur))
   4325 		return 0;
   4326 
   4327 	i = 1;
   4328 	while (mdb_cursor_get(cur, &k, &v, MDB_NEXT) == 0) {
   4329 		if (!ndb_unpack_text_search_key(k.mv_data, k.mv_size, &search_key)) {
   4330 			fprintf(stderr, "error decoding key %d\n", i);
   4331 			continue;
   4332 		}
   4333 
   4334 		ndb_print_text_search_key(&search_key);
   4335 		printf("\n");
   4336 
   4337 		i++;
   4338 	}
   4339 
   4340 	return 1;
   4341 }