void
MatchPost_Add_Inversion_To_Pool_IMP(MatchPosting *self,
PostingPool *post_pool,
Inversion *inversion, FieldType *type,
int32_t doc_id, float doc_boost,
float length_norm) {
MemoryPool *mem_pool = PostPool_Get_Mem_Pool(post_pool);
const size_t base_size = VTable_Get_Obj_Alloc_Size(RAWPOSTING);
Token **tokens;
uint32_t freq;
UNUSED_VAR(self);
UNUSED_VAR(type);
UNUSED_VAR(doc_boost);
UNUSED_VAR(length_norm);
Inversion_Reset(inversion);
while ((tokens = Inversion_Next_Cluster(inversion, &freq)) != NULL) {
TokenIVARS *const token_ivars = Token_IVARS(*tokens);
uint32_t raw_post_bytes
= MAX_RAW_POSTING_LEN(base_size, token_ivars->len);
RawPosting *raw_posting
= RawPost_new(MemPool_Grab(mem_pool, raw_post_bytes), doc_id,
freq, token_ivars->text, token_ivars->len);
PostPool_Feed(post_pool, &raw_posting);
}
}
void
MatchPost_add_inversion_to_pool(MatchPosting *self, PostingPool *post_pool,
Inversion *inversion, FieldType *type,
i32_t doc_id, float doc_boost,
float length_norm)
{
MemoryPool *mem_pool = post_pool->mem_pool;
Token **tokens;
u32_t freq;
UNUSED_VAR(self);
UNUSED_VAR(type);
UNUSED_VAR(doc_boost);
UNUSED_VAR(length_norm);
Inversion_Reset(inversion);
while ( (tokens = Inversion_Next_Cluster(inversion, &freq)) != NULL ) {
Token *token = *tokens;
u32_t raw_post_bytes = MAX_RAW_POSTING_LEN(token->len);
RawPosting *raw_posting = RawPost_new(
MemPool_Grab(mem_pool, raw_post_bytes), doc_id, freq,
token->text, token->len
);
PostPool_Add_Elem(post_pool, (Obj*)raw_posting);
}
}
Inversion*
Normalizer_transform(Normalizer *self, Inversion *inversion) {
// allocate additional space because utf8proc_reencode adds a
// terminating null char
int32_t static_buffer[INITIAL_BUFSIZE + 1];
int32_t *buffer = static_buffer;
ssize_t bufsize = INITIAL_BUFSIZE;
Token *token;
while (NULL != (token = Inversion_Next(inversion))) {
ssize_t len = utf8proc_decompose((uint8_t*)token->text, token->len,
buffer, bufsize, self->options);
if (len > bufsize) {
// buffer too small, (re)allocate
if (buffer != static_buffer) {
FREEMEM(buffer);
}
// allocate additional INITIAL_BUFSIZE items
bufsize = len + INITIAL_BUFSIZE;
buffer = (int32_t*)MALLOCATE((bufsize + 1) * sizeof(int32_t));
len = utf8proc_decompose((uint8_t*)token->text, token->len,
buffer, bufsize, self->options);
}
if (len < 0) {
continue;
}
len = utf8proc_reencode(buffer, len, self->options);
if (len >= 0) {
if (len > (ssize_t)token->len) {
FREEMEM(token->text);
token->text = (char*)MALLOCATE(len + 1);
}
memcpy(token->text, buffer, len + 1);
token->len = len;
}
}
if (buffer != static_buffer) {
FREEMEM(buffer);
}
Inversion_Reset(inversion);
return (Inversion*)INCREF(inversion);
}
void
ScorePost_Add_Inversion_To_Pool_IMP(ScorePosting *self,
PostingPool *post_pool,
Inversion *inversion, FieldType *type,
int32_t doc_id, float doc_boost,
float length_norm) {
ScorePostingIVARS *const ivars = ScorePost_IVARS(self);
MemoryPool *mem_pool = PostPool_Get_Mem_Pool(post_pool);
Similarity *sim = ivars->sim;
float field_boost = doc_boost * FType_Get_Boost(type) * length_norm;
const uint8_t field_boost_byte = Sim_Encode_Norm(sim, field_boost);
const size_t base_size = Class_Get_Obj_Alloc_Size(RAWPOSTING);
Token **tokens;
uint32_t freq;
Inversion_Reset(inversion);
while ((tokens = Inversion_Next_Cluster(inversion, &freq)) != NULL) {
TokenIVARS *const token_ivars = Token_IVARS(*tokens);
uint32_t raw_post_bytes
= MAX_RAW_POSTING_LEN(base_size, token_ivars->len, freq);
RawPosting *raw_posting
= RawPost_new(MemPool_Grab(mem_pool, raw_post_bytes), doc_id,
freq, token_ivars->text, token_ivars->len);
RawPostingIVARS *const raw_post_ivars = RawPost_IVARS(raw_posting);
char *const start = raw_post_ivars->blob + token_ivars->len;
char *dest = start;
uint32_t last_prox = 0;
// Field_boost.
*((uint8_t*)dest) = field_boost_byte;
dest++;
// Positions.
for (uint32_t i = 0; i < freq; i++) {
TokenIVARS *const t_ivars = Token_IVARS(tokens[i]);
const uint32_t prox_delta = t_ivars->pos - last_prox;
NumUtil_encode_c32(prox_delta, &dest);
last_prox = t_ivars->pos;
}
// Resize raw posting memory allocation.
raw_post_ivars->aux_len = dest - start;
raw_post_bytes = dest - (char*)raw_posting;
MemPool_Resize(mem_pool, raw_posting, raw_post_bytes);
PostPool_Feed(post_pool, (Obj*)raw_posting);
}
}
void
RichPost_add_inversion_to_pool(RichPosting *self, PostingPool *post_pool,
Inversion *inversion, FieldType *type,
int32_t doc_id, float doc_boost,
float length_norm)
{
MemoryPool *mem_pool = PostPool_Get_Mem_Pool(post_pool);
Similarity *sim = self->sim;
float field_boost = doc_boost * FType_Get_Boost(type) * length_norm;
Token **tokens;
uint32_t freq;
Inversion_Reset(inversion);
while ( (tokens = Inversion_Next_Cluster(inversion, &freq)) != NULL ) {
Token *token = *tokens;
uint32_t raw_post_bytes = MAX_RAW_POSTING_LEN(token->len, freq);
RawPosting *raw_posting = RawPost_new(
MemPool_Grab(mem_pool, raw_post_bytes), doc_id, freq,
token->text, token->len
);
char *const start = raw_posting->blob + token->len;
char *dest = start;
uint32_t last_prox = 0;
uint32_t i;
// Positions and boosts.
for (i = 0; i < freq; i++) {
Token *const t = tokens[i];
const uint32_t prox_delta = t->pos - last_prox;
const float boost = field_boost * t->boost;
NumUtil_encode_c32(prox_delta, &dest);
last_prox = t->pos;
*((uint8_t*)dest) = Sim_Encode_Norm(sim, boost);
dest++;
}
// Resize raw posting memory allocation.
raw_posting->aux_len = dest - start;
raw_post_bytes = dest - (char*)raw_posting;
MemPool_Resize(mem_pool, raw_posting, raw_post_bytes);
PostPool_Feed(post_pool, &raw_posting);
}
}
Inversion*
SnowStemmer_Transform_IMP(SnowballStemmer *self, Inversion *inversion) {
Token *token;
SnowballStemmerIVARS *const ivars = SnowStemmer_IVARS(self);
struct sb_stemmer *const snowstemmer
= (struct sb_stemmer*)ivars->snowstemmer;
while (NULL != (token = Inversion_Next(inversion))) {
TokenIVARS *const token_ivars = Token_IVARS(token);
const sb_symbol *stemmed_text
= sb_stemmer_stem(snowstemmer, (sb_symbol*)token_ivars->text,
token_ivars->len);
size_t len = sb_stemmer_length(snowstemmer);
if (len > token_ivars->len) {
FREEMEM(token_ivars->text);
token_ivars->text = (char*)MALLOCATE(len + 1);
}
memcpy(token_ivars->text, stemmed_text, len + 1);
token_ivars->len = len;
}
Inversion_Reset(inversion);
return (Inversion*)INCREF(inversion);
}
ByteBuf*
HLWriter_TV_Buf_IMP(HighlightWriter *self, Inversion *inversion) {
const char *last_text = "";
size_t last_len = 0;
ByteBuf *tv_buf = BB_new(20 + Inversion_Get_Size(inversion) * 8);
uint32_t num_postings = 0;
Token **tokens;
uint32_t freq;
UNUSED_VAR(self);
// Leave space for a c32 indicating the number of postings.
BB_Set_Size(tv_buf, C32_MAX_BYTES);
Inversion_Reset(inversion);
while ((tokens = Inversion_Next_Cluster(inversion, &freq)) != NULL) {
Token *token = *tokens;
char *const token_text = Token_Get_Text(token);
const int32_t token_len = Token_Get_Len(token);
int32_t overlap = StrHelp_overlap(last_text, token_text,
last_len, token_len);
char *ptr;
char *orig;
size_t old_size = BB_Get_Size(tv_buf);
size_t new_size = old_size
+ C32_MAX_BYTES // overlap
+ C32_MAX_BYTES // length of string diff
+ (token_len - overlap) // diff char data
+ C32_MAX_BYTES // num prox
+ (C32_MAX_BYTES * freq * 3); // pos data
// Allocate for worst-case scenario.
ptr = BB_Grow(tv_buf, new_size);
orig = ptr;
ptr += old_size;
// Track number of postings.
num_postings += 1;
// Append the string diff to the tv_buf.
NumUtil_encode_c32(overlap, &ptr);
NumUtil_encode_c32((token_len - overlap), &ptr);
memcpy(ptr, (token_text + overlap), (token_len - overlap));
ptr += token_len - overlap;
// Save text and text_len for comparison next loop.
last_text = token_text;
last_len = token_len;
// Append the number of positions for this term.
NumUtil_encode_c32(freq, &ptr);
do {
// Add position, start_offset, and end_offset to tv_buf.
NumUtil_encode_c32(Token_Get_Pos(token), &ptr);
NumUtil_encode_c32(Token_Get_Start_Offset(token), &ptr);
NumUtil_encode_c32(Token_Get_End_Offset(token), &ptr);
} while (--freq && (token = *++tokens));
// Set new byte length.
BB_Set_Size(tv_buf, ptr - orig);
}
// Go back and start the term vector string with the posting count.
char *dest = BB_Get_Buf(tv_buf);
NumUtil_encode_padded_c32(num_postings, &dest);
return tv_buf;
}
ByteBuf*
HLWriter_tv_buf(HighlightWriter *self, Inversion *inversion)
{
char *last_text = "";
size_t last_len = 0;
ByteBuf *tv_buf = BB_new(20 + inversion->size * 8); /* generous */
u32_t num_postings = 0;
char *dest;
Token **tokens;
u32_t freq;
UNUSED_VAR(self); /* heh. */
/* Leave space for a c32 indicating the number of postings. */
BB_Set_Size(tv_buf, C32_MAX_BYTES);
Inversion_Reset(inversion);
while ( (tokens = Inversion_Next_Cluster(inversion, &freq)) != NULL ) {
Token *token = *tokens;
i32_t overlap = StrHelp_string_diff(last_text, token->text,
last_len, token->len);
char *ptr;
size_t new_size = BB_Get_Size(tv_buf)
+ C32_MAX_BYTES /* overlap */
+ C32_MAX_BYTES /* length of string diff */
+ (token->len - overlap) /* diff char data */
+ C32_MAX_BYTES /* num prox */
+ (C32_MAX_BYTES * freq * 3); /* pos data */
/* Allocate for worst-case scenario. */
BB_Grow(tv_buf, new_size);
ptr = BBEND(tv_buf);
/* Track number of postings. */
num_postings += 1;
/* Append the string diff to the tv_buf. */
Math_encode_c32(overlap, &ptr);
Math_encode_c32( (token->len - overlap), &ptr);
memcpy(ptr, (token->text + overlap), (token->len - overlap));
ptr += token->len - overlap;
/* Save text and text_len for comparison next loop. */
last_text = token->text;
last_len = token->len;
/* Append the number of positions for this term. */
Math_encode_c32(freq, &ptr);
do {
/* Add position, start_offset, and end_offset to tv_buf. */
Math_encode_c32(token->pos, &ptr);
Math_encode_c32(token->start_offset, &ptr);
Math_encode_c32(token->end_offset, &ptr);
} while (--freq && (token = *++tokens));
/* Set new byte length. */
BB_Set_Size(tv_buf, ptr - tv_buf->ptr);
}
/* Go back and start the term vector string with the number of postings. */
dest = tv_buf->ptr;
Math_encode_padded_c32(num_postings, &dest);
return tv_buf;
}