SortCache*
SortCache_init(SortCache *self, Schema *schema, Folder *folder,
Segment *segment, i32_t field_num)
{
CharBuf *field = Seg_Field_Name(segment, field_num);
CharBuf *seg_name = Seg_Get_Name(segment);
CharBuf *ord_file = CB_newf("%o/sort-%i32.ord", seg_name, field_num);
CharBuf *ix_file = CB_newf("%o/sort-%i32.ix", seg_name, field_num);
CharBuf *dat_file = CB_newf("%o/sort-%i32.dat", seg_name, field_num);
i64_t ord_len, ix_len, dat_len;
/* Derive. */
self->doc_max = Seg_Get_Count(segment);
self->type = Schema_Fetch_Type(schema, field);
if (!self->type || !FType_Sortable(self->type)) {
THROW("'%o' isn't a sortable field", field);
}
/* Open instreams. */
self->ord_in = Folder_Open_In(folder, ord_file);
self->ix_in = Folder_Open_In(folder, ix_file);
self->dat_in = Folder_Open_In(folder, dat_file);
if (!self->ix_in || !self->dat_in || !self->ord_in) {
CharBuf *mess = MAKE_MESS("Can't open either %o, %o or %o", ord_file,
ix_file, dat_file);
DECREF(ord_file);
DECREF(ix_file);
DECREF(dat_file);
Err_throw_mess(mess);
}
ord_len = InStream_Length(self->ord_in);
ix_len = InStream_Length(self->ix_in);
dat_len = InStream_Length(self->dat_in);
/* Calculate the number of unique values and derive the ord bit width. */
self->num_uniq = (i32_t)(ix_len / 8) - 1;
self->width = S_calc_width(self->num_uniq);
/* Validate file lengths. */
{
double bytes_per_doc = self->width / 8.0;
double max_ords = ord_len / bytes_per_doc;
if (max_ords < self->doc_max + 1) {
THROW("Conflict between ord count max %f64 and doc_max %i32",
max_ords, self->doc_max);
}
}
/* Mmap ords, offsets and character data. */
self->ords = InStream_Buf(self->ord_in, (size_t)ord_len);
self->offsets = (i64_t*)InStream_Buf(self->ix_in, (size_t)ix_len);
self->char_data = InStream_Buf(self->dat_in, dat_len);
{
char *offs = (char*)self->offsets;
self->offsets_limit = (i64_t*)(offs + ix_len);
self->char_data_limit = self->char_data + dat_len;
}
DECREF(ord_file);
DECREF(ix_file);
DECREF(dat_file);
return self;
}
static int32_t
S_write_files(SortFieldWriter *self, OutStream *ord_out, OutStream *ix_out,
OutStream *dat_out) {
SortFieldWriterIVARS *const ivars = SortFieldWriter_IVARS(self);
int8_t prim_id = ivars->prim_id;
int32_t doc_max = (int32_t)Seg_Get_Count(ivars->segment);
bool has_nulls = ivars->count == doc_max ? false : true;
size_t size = (doc_max + 1) * sizeof(int32_t);
int32_t *ords = (int32_t*)MALLOCATE(size);
int32_t ord = 0;
int64_t dat_start = OutStream_Tell(dat_out);
// Assign -1 as a stand-in for the NULL ord.
for (int32_t i = 0; i <= doc_max; i++) {
ords[i] = -1;
}
// Grab the first item and record its ord. Add a dummy ord for invalid
// doc id 0.
SFWriterElem *elem = (SFWriterElem*)SortFieldWriter_Fetch(self);
SFWriterElemIVARS *elem_ivars = SFWriterElem_IVARS(elem);
if (elem_ivars->doc_id > doc_max) {
THROW(ERR, "doc_id %i32 greater than doc_max %i32",
elem_ivars->doc_id, doc_max);
}
ords[elem_ivars->doc_id] = ord;
ords[0] = 0;
// Build array of ords, write non-NULL sorted values.
Obj *last_val = INCREF(elem_ivars->value);
S_write_val(elem_ivars->value, prim_id, ix_out, dat_out, dat_start);
DECREF(elem);
while (NULL != (elem = (SFWriterElem*)SortFieldWriter_Fetch(self))) {
elem_ivars = SFWriterElem_IVARS(elem);
if (elem_ivars->value != last_val) {
int32_t comparison
= FType_Compare_Values(ivars->type, elem_ivars->value,
last_val);
if (comparison != 0) {
ord++;
S_write_val(elem_ivars->value, prim_id, ix_out, dat_out,
dat_start);
}
DECREF(last_val);
last_val = INCREF(elem_ivars->value);
}
if (elem_ivars->doc_id > doc_max) {
THROW(ERR, "doc_id %i32 greater than doc_max %i32",
elem_ivars->doc_id, doc_max);
}
ords[elem_ivars->doc_id] = ord;
DECREF(elem);
}
DECREF(last_val);
// If there are NULL values, write one now and record the NULL ord.
if (has_nulls) {
S_write_val(NULL, prim_id, ix_out, dat_out, dat_start);
ord++;
ivars->null_ord = ord;
}
int32_t null_ord = ivars->null_ord;
// Write one extra file pointer so that we can always derive length.
if (ivars->var_width) {
OutStream_Write_I64(ix_out, OutStream_Tell(dat_out) - dat_start);
}
// Calculate cardinality and ord width.
int32_t cardinality = ord + 1;
ivars->ord_width = S_calc_width(cardinality);
int32_t ord_width = ivars->ord_width;
// Write ords.
const double BITS_PER_BYTE = 8.0;
double bytes_per_doc = ord_width / BITS_PER_BYTE;
double byte_count = ceil((doc_max + 1) * bytes_per_doc);
char *compressed_ords
= (char*)CALLOCATE((size_t)byte_count, sizeof(char));
for (int32_t i = 0; i <= doc_max; i++) {
int32_t real_ord = ords[i] == -1 ? null_ord : ords[i];
S_write_ord(compressed_ords, ord_width, i, real_ord);
}
OutStream_Write_Bytes(ord_out, compressed_ords, (size_t)byte_count);
FREEMEM(compressed_ords);
FREEMEM(ords);
return cardinality;
}
