dtuple_t*
row_rec_to_index_entry(
/*===================*/
/* out, own: index entry built; see the
NOTE below! */
ulint type, /* in: ROW_COPY_DATA, or ROW_COPY_POINTERS:
the former copies also the data fields to
heap as the latter only places pointers to
data fields on the index page */
dict_index_t* index, /* in: index */
rec_t* rec, /* in: record in the index;
NOTE: in the case ROW_COPY_POINTERS
the data fields in the row will point
directly into this record, therefore,
the buffer page of this record must be
at least s-latched and the latch held
as long as the dtuple is used! */
mem_heap_t* heap) /* in: memory heap from which the memory
needed is allocated */
{
dtuple_t* entry;
dfield_t* dfield;
ulint i;
byte* field;
ulint len;
ulint rec_len;
byte* buf;
ut_ad(rec && heap && index);
if (type == ROW_COPY_DATA) {
/* Take a copy of rec to heap */
buf = mem_heap_alloc(heap, rec_get_size(rec));
rec = rec_copy(buf, rec);
}
rec_len = rec_get_n_fields(rec);
entry = dtuple_create(heap, rec_len);
dtuple_set_n_fields_cmp(entry,
dict_index_get_n_unique_in_tree(index));
ut_ad(rec_len == dict_index_get_n_fields(index));
dict_index_copy_types(entry, index, rec_len);
dtuple_set_info_bits(entry, rec_get_info_bits(rec));
for (i = 0; i < rec_len; i++) {
dfield = dtuple_get_nth_field(entry, i);
field = rec_get_nth_field(rec, i, &len);
dfield_set_data(dfield, field, len);
}
ut_ad(dtuple_check_typed(entry));
return(entry);
}
void
rec_copy_prefix_to_dtuple(
/*======================*/
dtuple_t* tuple, /* in: data tuple */
rec_t* rec, /* in: physical record */
ulint n_fields, /* in: number of fields to copy */
mem_heap_t* heap) /* in: memory heap */
{
dfield_t* field;
byte* data;
ulint len;
byte* buf = NULL;
ulint i;
ut_ad(rec_validate(rec));
ut_ad(dtuple_check_typed(tuple));
dtuple_set_info_bits(tuple, rec_get_info_bits(rec));
for (i = 0; i < n_fields; i++) {
field = dtuple_get_nth_field(tuple, i);
data = rec_get_nth_field(rec, i, &len);
if (len != UNIV_SQL_NULL) {
buf = mem_heap_alloc(heap, len);
ut_memcpy(buf, data, len);
}
dfield_set_data(field, buf, len);
}
}
ulint
rec_sprintf(
/*========*/
/* out: printed length in bytes */
char* buf, /* in: buffer to print to */
ulint buf_len,/* in: buffer length */
rec_t* rec) /* in: physical record */
{
byte* data;
ulint len;
ulint k;
ulint n;
ulint i;
ut_ad(rec);
n = rec_get_n_fields(rec);
k = 0;
if (k + 30 > buf_len) {
return(k);
}
k += sprintf(buf + k, "RECORD: info bits %lu", rec_get_info_bits(rec));
for (i = 0; i < n; i++) {
if (k + 30 > buf_len) {
return(k);
}
data = rec_get_nth_field(rec, i, &len);
k += sprintf(buf + k, " %lu:", i);
if (len != UNIV_SQL_NULL) {
if (k + 30 + 5 * len > buf_len) {
return(k);
}
k += ut_sprintf_buf(buf + k, data, len);
} else {
k += sprintf(buf + k, " SQL NULL");
}
k += sprintf(buf + k, ";");
}
return(k);
}
void
rec_print(
/*======*/
rec_t* rec) /* in: physical record */
{
byte* data;
ulint len;
char* offs;
ulint n;
ulint i;
ut_ad(rec);
if (rec_get_1byte_offs_flag(rec)) {
offs = "TRUE";
} else {
offs = "FALSE";
}
n = rec_get_n_fields(rec);
printf(
"PHYSICAL RECORD: n_fields %lu; 1-byte offs %s; info bits %lu\n",
n, offs, rec_get_info_bits(rec));
for (i = 0; i < n; i++) {
data = rec_get_nth_field(rec, i, &len);
printf(" %lu:", i);
if (len != UNIV_SQL_NULL) {
if (len <= 30) {
ut_print_buf(data, len);
} else {
ut_print_buf(data, 30);
printf("...(truncated)");
}
} else {
printf(" SQL NULL, size %lu ",
rec_get_nth_field_size(rec, i));
}
printf(";");
}
printf("\n");
rec_validate(rec);
}
/*******************************************************************//**
Converts an index record to a typed data tuple. NOTE that externally
stored (often big) fields are NOT copied to heap.
@return own: index entry built; see the NOTE below! */
UNIV_INTERN
dtuple_t*
row_rec_to_index_entry(
/*===================*/
ulint type, /*!< in: ROW_COPY_DATA, or
ROW_COPY_POINTERS: the former
copies also the data fields to
heap as the latter only places
pointers to data fields on the
index page */
const rec_t* rec, /*!< in: record in the index;
NOTE: in the case
ROW_COPY_POINTERS the data
fields in the row will point
directly into this record,
therefore, the buffer page of
this record must be at least
s-latched and the latch held
as long as the dtuple is used! */
const dict_index_t* index, /*!< in: index */
ulint* offsets,/*!< in/out: rec_get_offsets(rec) */
ulint* n_ext, /*!< out: number of externally
stored columns */
mem_heap_t* heap) /*!< in: memory heap from which
the memory needed is allocated */
{
dtuple_t* entry;
byte* buf;
ut_ad(rec && heap && index);
ut_ad(rec_offs_validate(rec, index, offsets));
if (type == ROW_COPY_DATA) {
/* Take a copy of rec to heap */
buf = mem_heap_alloc(heap, rec_offs_size(offsets));
rec = rec_copy(buf, rec, offsets);
/* Avoid a debug assertion in rec_offs_validate(). */
rec_offs_make_valid(rec, index, offsets);
#if defined UNIV_DEBUG || defined UNIV_BLOB_LIGHT_DEBUG
} else {
ut_a(!rec_offs_any_null_extern(rec, offsets));
#endif /* UNIV_DEBUG || UNIV_BLOB_LIGHT_DEBUG */
}
entry = row_rec_to_index_entry_low(rec, index, offsets, n_ext, heap);
dtuple_set_info_bits(entry,
rec_get_info_bits(rec, rec_offs_comp(offsets)));
return(entry);
}
void
row_build_to_tuple(
/*===============*/
dtuple_t* row, /* in/out: row built; see the NOTE below! */
dict_index_t* index, /* in: clustered index */
rec_t* rec) /* in: record in the clustered index;
NOTE: the data fields in the row will point
directly into this record, therefore,
the buffer page of this record must be
at least s-latched and the latch held
as long as the row dtuple is used!
NOTE 2: does not work with externally
stored fields! */
{
dict_table_t* table;
ulint n_fields;
ulint i;
dfield_t* dfield;
byte* field;
ulint len;
ulint row_len;
dict_col_t* col;
ut_ad(index && rec);
ut_ad(index->type & DICT_CLUSTERED);
table = index->table;
row_len = dict_table_get_n_cols(table);
dtuple_set_info_bits(row, rec_get_info_bits(rec));
n_fields = dict_index_get_n_fields(index);
ut_ad(n_fields == rec_get_n_fields(rec));
dict_table_copy_types(row, table);
for (i = 0; i < n_fields; i++) {
col = dict_field_get_col(dict_index_get_nth_field(index, i));
dfield = dtuple_get_nth_field(row, dict_col_get_no(col));
field = rec_get_nth_field(rec, i, &len);
dfield_set_data(dfield, field, len);
}
ut_ad(dtuple_check_typed(row));
}
void
rec_print_new(
/*==========*/
FILE* file, /* in: file where to print */
rec_t* rec, /* in: physical record */
const ulint* offsets)/* in: array returned by rec_get_offsets() */
{
const byte* data;
ulint len;
ulint i;
ut_ad(rec_offs_validate(rec, NULL, offsets));
ut_ad(rec);
fprintf(file, "PHYSICAL RECORD: n_fields %lu;"
" compact format; info bits %lu\n",
(ulong) rec_offs_n_fields(offsets),
(ulong) rec_get_info_bits(rec, TRUE));
for (i = 0; i < rec_offs_n_fields(offsets); i++) {
data = rec_get_nth_field(rec, offsets, i, &len);
fprintf(file, " %lu:", (ulong) i);
if (len != UNIV_SQL_NULL) {
if (len <= 30) {
ut_print_buf(file, data, len);
} else {
ut_print_buf(file, data, 30);
fputs("...(truncated)", file);
}
} else {
fputs(" SQL NULL", file);
}
putc(';', file);
}
putc('\n', file);
}
/**********************************************************************//**
Reports in the undo log of an update or delete marking of a clustered index
record.
@return byte offset of the inserted undo log entry on the page if
succeed, 0 if fail */
static
ulint
trx_undo_page_report_modify(
/*========================*/
page_t* undo_page, /*!< in: undo log page */
trx_t* trx, /*!< in: transaction */
dict_index_t* index, /*!< in: clustered index where update or
delete marking is done */
const rec_t* rec, /*!< in: clustered index record which
has NOT yet been modified */
const ulint* offsets, /*!< in: rec_get_offsets(rec, index) */
const upd_t* update, /*!< in: update vector which tells the
columns to be updated; in the case of
a delete, this should be set to NULL */
ulint cmpl_info, /*!< in: compiler info on secondary
index updates */
mtr_t* mtr) /*!< in: mtr */
{
dict_table_t* table;
ulint first_free;
byte* ptr;
const byte* field;
ulint flen;
ulint col_no;
ulint type_cmpl;
byte* type_cmpl_ptr;
ulint i;
trx_id_t trx_id;
ibool ignore_prefix = FALSE;
byte ext_buf[REC_MAX_INDEX_COL_LEN
+ BTR_EXTERN_FIELD_REF_SIZE];
ut_a(dict_index_is_clust(index));
ut_ad(rec_offs_validate(rec, index, offsets));
ut_ad(mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
+ TRX_UNDO_PAGE_TYPE) == TRX_UNDO_UPDATE);
table = index->table;
first_free = mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
+ TRX_UNDO_PAGE_FREE);
ptr = undo_page + first_free;
ut_ad(first_free <= UNIV_PAGE_SIZE);
if (trx_undo_left(undo_page, ptr) < 50) {
/* NOTE: the value 50 must be big enough so that the general
fields written below fit on the undo log page */
return(0);
}
/* Reserve 2 bytes for the pointer to the next undo log record */
ptr += 2;
/* Store first some general parameters to the undo log */
if (!update) {
type_cmpl = TRX_UNDO_DEL_MARK_REC;
} else if (rec_get_deleted_flag(rec, dict_table_is_comp(table))) {
type_cmpl = TRX_UNDO_UPD_DEL_REC;
/* We are about to update a delete marked record.
We don't typically need the prefix in this case unless
the delete marking is done by the same transaction
(which we check below). */
ignore_prefix = TRUE;
} else {
type_cmpl = TRX_UNDO_UPD_EXIST_REC;
}
type_cmpl |= cmpl_info * TRX_UNDO_CMPL_INFO_MULT;
type_cmpl_ptr = ptr;
*ptr++ = (byte) type_cmpl;
ptr += mach_dulint_write_much_compressed(ptr, trx->undo_no);
ptr += mach_dulint_write_much_compressed(ptr, table->id);
/*----------------------------------------*/
/* Store the state of the info bits */
*ptr++ = (byte) rec_get_info_bits(rec, dict_table_is_comp(table));
/* Store the values of the system columns */
field = rec_get_nth_field(rec, offsets,
dict_index_get_sys_col_pos(
index, DATA_TRX_ID), &flen);
ut_ad(flen == DATA_TRX_ID_LEN);
trx_id = trx_read_trx_id(field);
/* If it is an update of a delete marked record, then we are
allowed to ignore blob prefixes if the delete marking was done
by some other trx as it must have committed by now for us to
allow an over-write. */
if (ignore_prefix) {
ignore_prefix = ut_dulint_cmp(trx_id, trx->id) != 0;
}
ptr += mach_dulint_write_compressed(ptr, trx_id);
field = rec_get_nth_field(rec, offsets,
dict_index_get_sys_col_pos(
index, DATA_ROLL_PTR), &flen);
ut_ad(flen == DATA_ROLL_PTR_LEN);
ptr += mach_dulint_write_compressed(ptr, trx_read_roll_ptr(field));
/*----------------------------------------*/
/* Store then the fields required to uniquely determine the
record which will be modified in the clustered index */
for (i = 0; i < dict_index_get_n_unique(index); i++) {
field = rec_get_nth_field(rec, offsets, i, &flen);
/* The ordering columns must not be stored externally. */
ut_ad(!rec_offs_nth_extern(offsets, i));
ut_ad(dict_index_get_nth_col(index, i)->ord_part);
if (trx_undo_left(undo_page, ptr) < 5) {
return(0);
}
ptr += mach_write_compressed(ptr, flen);
if (flen != UNIV_SQL_NULL) {
if (trx_undo_left(undo_page, ptr) < flen) {
return(0);
}
ut_memcpy(ptr, field, flen);
ptr += flen;
}
}
/*----------------------------------------*/
/* Save to the undo log the old values of the columns to be updated. */
if (update) {
if (trx_undo_left(undo_page, ptr) < 5) {
return(0);
}
ptr += mach_write_compressed(ptr, upd_get_n_fields(update));
for (i = 0; i < upd_get_n_fields(update); i++) {
ulint pos = upd_get_nth_field(update, i)->field_no;
/* Write field number to undo log */
if (trx_undo_left(undo_page, ptr) < 5) {
return(0);
}
ptr += mach_write_compressed(ptr, pos);
/* Save the old value of field */
field = rec_get_nth_field(rec, offsets, pos, &flen);
if (trx_undo_left(undo_page, ptr) < 15) {
return(0);
}
if (rec_offs_nth_extern(offsets, pos)) {
ptr = trx_undo_page_report_modify_ext(
ptr,
dict_index_get_nth_col(index, pos)
->ord_part
&& !ignore_prefix
&& flen < REC_MAX_INDEX_COL_LEN
? ext_buf : NULL,
dict_table_zip_size(table),
&field, &flen);
/* Notify purge that it eventually has to
free the old externally stored field */
trx->update_undo->del_marks = TRUE;
*type_cmpl_ptr |= TRX_UNDO_UPD_EXTERN;
} else {
ptr += mach_write_compressed(ptr, flen);
}
if (flen != UNIV_SQL_NULL) {
if (trx_undo_left(undo_page, ptr) < flen) {
return(0);
}
ut_memcpy(ptr, field, flen);
ptr += flen;
}
}
}
/*----------------------------------------*/
/* In the case of a delete marking, and also in the case of an update
where any ordering field of any index changes, store the values of all
columns which occur as ordering fields in any index. This info is used
in the purge of old versions where we use it to build and search the
delete marked index records, to look if we can remove them from the
index tree. Note that starting from 4.0.14 also externally stored
fields can be ordering in some index. Starting from 5.2, we no longer
store REC_MAX_INDEX_COL_LEN first bytes to the undo log record,
but we can construct the column prefix fields in the index by
fetching the first page of the BLOB that is pointed to by the
clustered index. This works also in crash recovery, because all pages
(including BLOBs) are recovered before anything is rolled back. */
if (!update || !(cmpl_info & UPD_NODE_NO_ORD_CHANGE)) {
byte* old_ptr = ptr;
trx->update_undo->del_marks = TRUE;
if (trx_undo_left(undo_page, ptr) < 5) {
return(0);
}
/* Reserve 2 bytes to write the number of bytes the stored
fields take in this undo record */
ptr += 2;
for (col_no = 0; col_no < dict_table_get_n_cols(table);
col_no++) {
const dict_col_t* col
= dict_table_get_nth_col(table, col_no);
if (col->ord_part) {
ulint pos;
/* Write field number to undo log */
if (trx_undo_left(undo_page, ptr) < 5 + 15) {
return(0);
}
pos = dict_index_get_nth_col_pos(index,
col_no);
ptr += mach_write_compressed(ptr, pos);
/* Save the old value of field */
field = rec_get_nth_field(rec, offsets, pos,
&flen);
if (rec_offs_nth_extern(offsets, pos)) {
ptr = trx_undo_page_report_modify_ext(
ptr,
flen < REC_MAX_INDEX_COL_LEN
&& !ignore_prefix
? ext_buf : NULL,
dict_table_zip_size(table),
&field, &flen);
} else {
ptr += mach_write_compressed(
ptr, flen);
}
if (flen != UNIV_SQL_NULL) {
if (trx_undo_left(undo_page, ptr)
< flen) {
return(0);
}
ut_memcpy(ptr, field, flen);
ptr += flen;
}
}
}
mach_write_to_2(old_ptr, ptr - old_ptr);
}
/*----------------------------------------*/
/* Write pointers to the previous and the next undo log records */
if (trx_undo_left(undo_page, ptr) < 2) {
return(0);
}
mach_write_to_2(ptr, first_free);
ptr += 2;
mach_write_to_2(undo_page + first_free, ptr - undo_page);
mach_write_to_2(undo_page + TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_FREE,
ptr - undo_page);
/* Write to the REDO log about this change in the UNDO log */
trx_undof_page_add_undo_rec_log(undo_page, first_free,
ptr - undo_page, mtr);
return(first_free);
}
/*******************************************************************//**
An inverse function to row_build_index_entry. Builds a row from a
record in a clustered index.
@return own: row built; see the NOTE below! */
UNIV_INTERN
dtuple_t*
row_build(
/*======*/
ulint type, /*!< in: ROW_COPY_POINTERS or
ROW_COPY_DATA; the latter
copies also the data fields to
heap while the first only
places pointers to data fields
on the index page, and thus is
more efficient */
const dict_index_t* index, /*!< in: clustered index */
const rec_t* rec, /*!< in: record in the clustered
index; NOTE: in the case
ROW_COPY_POINTERS the data
fields in the row will point
directly into this record,
therefore, the buffer page of
this record must be at least
s-latched and the latch held
as long as the row dtuple is used! */
const ulint* offsets,/*!< in: rec_get_offsets(rec,index)
or NULL, in which case this function
will invoke rec_get_offsets() */
const dict_table_t* col_table,
/*!< in: table, to check which
externally stored columns
occur in the ordering columns
of an index, or NULL if
index->table should be
consulted instead */
row_ext_t** ext, /*!< out, own: cache of
externally stored column
prefixes, or NULL */
mem_heap_t* heap) /*!< in: memory heap from which
the memory needed is allocated */
{
dtuple_t* row;
const dict_table_t* table;
ulint n_fields;
ulint n_ext_cols;
ulint* ext_cols = NULL; /* remove warning */
ulint len;
ulint row_len;
byte* buf;
ulint i;
ulint j;
mem_heap_t* tmp_heap = NULL;
ulint offsets_[REC_OFFS_NORMAL_SIZE];
rec_offs_init(offsets_);
ut_ad(index && rec && heap);
ut_ad(dict_index_is_clust(index));
ut_ad(!mutex_own(&kernel_mutex));
if (!offsets) {
offsets = rec_get_offsets(rec, index, offsets_,
ULINT_UNDEFINED, &tmp_heap);
} else {
ut_ad(rec_offs_validate(rec, index, offsets));
}
#if defined UNIV_DEBUG || defined UNIV_BLOB_LIGHT_DEBUG
if (rec_offs_any_null_extern(rec, offsets)) {
/* This condition can occur during crash recovery
before trx_rollback_active() has completed execution,
or when a concurrently executing
row_ins_index_entry_low() has committed the B-tree
mini-transaction but has not yet managed to restore
the cursor position for writing the big_rec. */
ut_a(trx_undo_roll_ptr_is_insert(
row_get_rec_roll_ptr(rec, index, offsets)));
}
#endif /* UNIV_DEBUG || UNIV_BLOB_LIGHT_DEBUG */
if (type != ROW_COPY_POINTERS) {
/* Take a copy of rec to heap */
buf = mem_heap_alloc(heap, rec_offs_size(offsets));
rec = rec_copy(buf, rec, offsets);
/* Avoid a debug assertion in rec_offs_validate(). */
rec_offs_make_valid(rec, index, (ulint*) offsets);
}
table = index->table;
row_len = dict_table_get_n_cols(table);
row = dtuple_create(heap, row_len);
dict_table_copy_types(row, table);
dtuple_set_info_bits(row, rec_get_info_bits(
rec, dict_table_is_comp(table)));
n_fields = rec_offs_n_fields(offsets);
n_ext_cols = rec_offs_n_extern(offsets);
if (n_ext_cols) {
ext_cols = mem_heap_alloc(heap, n_ext_cols * sizeof *ext_cols);
}
for (i = j = 0; i < n_fields; i++) {
dict_field_t* ind_field
= dict_index_get_nth_field(index, i);
const dict_col_t* col
= dict_field_get_col(ind_field);
ulint col_no
= dict_col_get_no(col);
dfield_t* dfield
= dtuple_get_nth_field(row, col_no);
if (ind_field->prefix_len == 0) {
const byte* field = rec_get_nth_field(
rec, offsets, i, &len);
dfield_set_data(dfield, field, len);
}
if (rec_offs_nth_extern(offsets, i)) {
dfield_set_ext(dfield);
if (UNIV_LIKELY_NULL(col_table)) {
ut_a(col_no
< dict_table_get_n_cols(col_table));
col = dict_table_get_nth_col(
col_table, col_no);
}
if (col->ord_part) {
/* We will have to fetch prefixes of
externally stored columns that are
referenced by column prefixes. */
ext_cols[j++] = col_no;
}
}
}
ut_ad(dtuple_check_typed(row));
if (!ext) {
/* REDUNDANT and COMPACT formats store a local
768-byte prefix of each externally stored
column. No cache is needed. */
ut_ad(dict_table_get_format(index->table)
< DICT_TF_FORMAT_ZIP);
} else if (j) {
*ext = row_ext_create(j, ext_cols, row,
dict_table_zip_size(index->table),
heap);
} else {
*ext = NULL;
}
if (tmp_heap) {
mem_heap_free(tmp_heap);
}
return(row);
}
/*************************************************************//**
This function is used to compare two physical records. Only the common
first fields are compared, and if an externally stored field is
encountered, then 0 is returned.
@return 1, 0, -1 if rec1 is greater, equal, less, respectively */
UNIV_INTERN
int
cmp_rec_rec_with_match(
/*===================*/
const rec_t* rec1, /*!< in: physical record */
const rec_t* rec2, /*!< in: physical record */
const ulint* offsets1,/*!< in: rec_get_offsets(rec1, index) */
const ulint* offsets2,/*!< in: rec_get_offsets(rec2, index) */
dict_index_t* index, /*!< in: data dictionary index */
ulint* matched_fields, /*!< in/out: number of already completely
matched fields; when the function returns,
contains the value the for current
comparison */
ulint* matched_bytes) /*!< in/out: number of already matched
bytes within the first field not completely
matched; when the function returns, contains
the value for the current comparison */
{
ulint rec1_n_fields; /* the number of fields in rec */
ulint rec1_f_len; /* length of current field in rec */
const byte* rec1_b_ptr; /* pointer to the current byte
in rec field */
ulint rec1_byte; /* value of current byte to be
compared in rec */
ulint rec2_n_fields; /* the number of fields in rec */
ulint rec2_f_len; /* length of current field in rec */
const byte* rec2_b_ptr; /* pointer to the current byte
in rec field */
ulint rec2_byte; /* value of current byte to be
compared in rec */
ulint cur_field; /* current field number */
ulint cur_bytes; /* number of already matched
bytes in current field */
int ret = 0; /* return value */
ulint comp;
ut_ad(rec1 && rec2 && index);
ut_ad(rec_offs_validate(rec1, index, offsets1));
ut_ad(rec_offs_validate(rec2, index, offsets2));
ut_ad(rec_offs_comp(offsets1) == rec_offs_comp(offsets2));
comp = rec_offs_comp(offsets1);
rec1_n_fields = rec_offs_n_fields(offsets1);
rec2_n_fields = rec_offs_n_fields(offsets2);
cur_field = *matched_fields;
cur_bytes = *matched_bytes;
/* Match fields in a loop */
while ((cur_field < rec1_n_fields) && (cur_field < rec2_n_fields)) {
ulint mtype;
ulint prtype;
if (UNIV_UNLIKELY(index->type & DICT_UNIVERSAL)) {
/* This is for the insert buffer B-tree. */
mtype = DATA_BINARY;
prtype = 0;
} else {
const dict_col_t* col
= dict_index_get_nth_col(index, cur_field);
mtype = col->mtype;
prtype = col->prtype;
}
rec1_b_ptr = rec_get_nth_field(rec1, offsets1,
cur_field, &rec1_f_len);
rec2_b_ptr = rec_get_nth_field(rec2, offsets2,
cur_field, &rec2_f_len);
if (cur_bytes == 0) {
if (cur_field == 0) {
/* Test if rec is the predefined minimum
record */
if (UNIV_UNLIKELY(rec_get_info_bits(rec1, comp)
& REC_INFO_MIN_REC_FLAG)) {
if (!(rec_get_info_bits(rec2, comp)
& REC_INFO_MIN_REC_FLAG)) {
ret = -1;
}
goto order_resolved;
} else if (UNIV_UNLIKELY
(rec_get_info_bits(rec2, comp)
& REC_INFO_MIN_REC_FLAG)) {
ret = 1;
goto order_resolved;
}
}
if (rec_offs_nth_extern(offsets1, cur_field)
|| rec_offs_nth_extern(offsets2, cur_field)) {
/* We do not compare to an externally
stored field */
goto order_resolved;
}
if (rec1_f_len == UNIV_SQL_NULL
|| rec2_f_len == UNIV_SQL_NULL) {
if (rec1_f_len == rec2_f_len) {
goto next_field;
} else if (rec2_f_len == UNIV_SQL_NULL) {
/* We define the SQL null to be the
smallest possible value of a field
in the alphabetical order */
ret = 1;
} else {
ret = -1;
}
goto order_resolved;
}
}
if (mtype >= DATA_FLOAT
|| (mtype == DATA_BLOB
&& 0 == (prtype & DATA_BINARY_TYPE)
&& dtype_get_charset_coll(prtype)
!= DATA_MYSQL_LATIN1_SWEDISH_CHARSET_COLL)) {
ret = cmp_whole_field(mtype, prtype,
rec1_b_ptr,
(unsigned) rec1_f_len,
rec2_b_ptr,
(unsigned) rec2_f_len);
if (ret != 0) {
cur_bytes = 0;
goto order_resolved;
} else {
goto next_field;
}
}
/* Set the pointers at the current byte */
rec1_b_ptr = rec1_b_ptr + cur_bytes;
rec2_b_ptr = rec2_b_ptr + cur_bytes;
/* Compare then the fields */
for (;;) {
if (rec2_f_len <= cur_bytes) {
if (rec1_f_len <= cur_bytes) {
goto next_field;
}
rec2_byte = dtype_get_pad_char(mtype, prtype);
if (rec2_byte == ULINT_UNDEFINED) {
ret = 1;
goto order_resolved;
}
} else {
rec2_byte = *rec2_b_ptr;
}
if (rec1_f_len <= cur_bytes) {
rec1_byte = dtype_get_pad_char(mtype, prtype);
if (rec1_byte == ULINT_UNDEFINED) {
ret = -1;
goto order_resolved;
}
} else {
rec1_byte = *rec1_b_ptr;
}
if (rec1_byte == rec2_byte) {
/* If the bytes are equal, they will remain
such even after the collation transformation
below */
goto next_byte;
}
if (mtype <= DATA_CHAR
|| (mtype == DATA_BLOB
&& !(prtype & DATA_BINARY_TYPE))) {
rec1_byte = cmp_collate(rec1_byte);
rec2_byte = cmp_collate(rec2_byte);
}
if (rec1_byte < rec2_byte) {
ret = -1;
goto order_resolved;
} else if (rec1_byte > rec2_byte) {
ret = 1;
goto order_resolved;
}
next_byte:
/* Next byte */
cur_bytes++;
rec1_b_ptr++;
rec2_b_ptr++;
}
next_field:
cur_field++;
cur_bytes = 0;
}
ut_ad(cur_bytes == 0);
/* If we ran out of fields, rec1 was equal to rec2 up
to the common fields */
ut_ad(ret == 0);
order_resolved:
ut_ad((ret >= - 1) && (ret <= 1));
*matched_fields = cur_field;
*matched_bytes = cur_bytes;
return(ret);
}
/*************************************************************//**
This function is used to compare a data tuple to a physical record.
Only dtuple->n_fields_cmp first fields are taken into account for
the data tuple! If we denote by n = n_fields_cmp, then rec must
have either m >= n fields, or it must differ from dtuple in some of
the m fields rec has. If rec has an externally stored field we do not
compare it but return with value 0 if such a comparison should be
made.
@return 1, 0, -1, if dtuple is greater, equal, less than rec,
respectively, when only the common first fields are compared, or until
the first externally stored field in rec */
UNIV_INTERN
int
cmp_dtuple_rec_with_match(
/*======================*/
const dtuple_t* dtuple, /*!< in: data tuple */
const rec_t* rec, /*!< in: physical record which differs from
dtuple in some of the common fields, or which
has an equal number or more fields than
dtuple */
const ulint* offsets,/*!< in: array returned by rec_get_offsets() */
ulint* matched_fields, /*!< in/out: number of already completely
matched fields; when function returns,
contains the value for current comparison */
ulint* matched_bytes) /*!< in/out: number of already matched
bytes within the first field not completely
matched; when function returns, contains the
value for current comparison */
{
const dfield_t* dtuple_field; /* current field in logical record */
ulint dtuple_f_len; /* the length of the current field
in the logical record */
const byte* dtuple_b_ptr; /* pointer to the current byte in
logical field data */
ulint dtuple_byte; /* value of current byte to be compared
in dtuple*/
ulint rec_f_len; /* length of current field in rec */
const byte* rec_b_ptr; /* pointer to the current byte in
rec field */
ulint rec_byte; /* value of current byte to be
compared in rec */
ulint cur_field; /* current field number */
ulint cur_bytes; /* number of already matched bytes
in current field */
int ret = 3333; /* return value */
ut_ad(dtuple && rec && matched_fields && matched_bytes);
ut_ad(dtuple_check_typed(dtuple));
ut_ad(rec_offs_validate(rec, NULL, offsets));
cur_field = *matched_fields;
cur_bytes = *matched_bytes;
ut_ad(cur_field <= dtuple_get_n_fields_cmp(dtuple));
ut_ad(cur_field <= rec_offs_n_fields(offsets));
if (cur_bytes == 0 && cur_field == 0) {
ulint rec_info = rec_get_info_bits(rec,
rec_offs_comp(offsets));
ulint tup_info = dtuple_get_info_bits(dtuple);
if (UNIV_UNLIKELY(rec_info & REC_INFO_MIN_REC_FLAG)) {
ret = !(tup_info & REC_INFO_MIN_REC_FLAG);
goto order_resolved;
} else if (UNIV_UNLIKELY(tup_info & REC_INFO_MIN_REC_FLAG)) {
ret = -1;
goto order_resolved;
}
}
/* Match fields in a loop; stop if we run out of fields in dtuple
or find an externally stored field */
while (cur_field < dtuple_get_n_fields_cmp(dtuple)) {
ulint mtype;
ulint prtype;
dtuple_field = dtuple_get_nth_field(dtuple, cur_field);
{
const dtype_t* type
= dfield_get_type(dtuple_field);
mtype = type->mtype;
prtype = type->prtype;
}
dtuple_f_len = dfield_get_len(dtuple_field);
rec_b_ptr = rec_get_nth_field(rec, offsets,
cur_field, &rec_f_len);
/* If we have matched yet 0 bytes, it may be that one or
both the fields are SQL null, or the record or dtuple may be
the predefined minimum record, or the field is externally
stored */
if (UNIV_LIKELY(cur_bytes == 0)) {
if (rec_offs_nth_extern(offsets, cur_field)) {
/* We do not compare to an externally
stored field */
ret = 0;
goto order_resolved;
}
if (dtuple_f_len == UNIV_SQL_NULL) {
if (rec_f_len == UNIV_SQL_NULL) {
goto next_field;
}
ret = -1;
goto order_resolved;
} else if (rec_f_len == UNIV_SQL_NULL) {
/* We define the SQL null to be the
smallest possible value of a field
in the alphabetical order */
ret = 1;
goto order_resolved;
}
}
if (mtype >= DATA_FLOAT
|| (mtype == DATA_BLOB
&& 0 == (prtype & DATA_BINARY_TYPE)
&& dtype_get_charset_coll(prtype)
!= DATA_MYSQL_LATIN1_SWEDISH_CHARSET_COLL)) {
ret = cmp_whole_field(mtype, prtype,
dfield_get_data(dtuple_field),
(unsigned) dtuple_f_len,
rec_b_ptr, (unsigned) rec_f_len);
if (ret != 0) {
cur_bytes = 0;
goto order_resolved;
} else {
goto next_field;
}
}
/* Set the pointers at the current byte */
rec_b_ptr = rec_b_ptr + cur_bytes;
dtuple_b_ptr = (byte*)dfield_get_data(dtuple_field)
+ cur_bytes;
/* Compare then the fields */
for (;;) {
if (UNIV_UNLIKELY(rec_f_len <= cur_bytes)) {
if (dtuple_f_len <= cur_bytes) {
goto next_field;
}
rec_byte = dtype_get_pad_char(mtype, prtype);
if (rec_byte == ULINT_UNDEFINED) {
ret = 1;
goto order_resolved;
}
} else {
rec_byte = *rec_b_ptr;
}
if (UNIV_UNLIKELY(dtuple_f_len <= cur_bytes)) {
dtuple_byte = dtype_get_pad_char(mtype,
prtype);
if (dtuple_byte == ULINT_UNDEFINED) {
ret = -1;
goto order_resolved;
}
} else {
dtuple_byte = *dtuple_b_ptr;
}
if (dtuple_byte == rec_byte) {
/* If the bytes are equal, they will
remain such even after the collation
transformation below */
goto next_byte;
}
if (mtype <= DATA_CHAR
|| (mtype == DATA_BLOB
&& !(prtype & DATA_BINARY_TYPE))) {
rec_byte = cmp_collate(rec_byte);
dtuple_byte = cmp_collate(dtuple_byte);
}
ret = (int) (dtuple_byte - rec_byte);
if (UNIV_LIKELY(ret)) {
if (ret < 0) {
ret = -1;
goto order_resolved;
} else {
ret = 1;
goto order_resolved;
}
}
next_byte:
/* Next byte */
cur_bytes++;
rec_b_ptr++;
dtuple_b_ptr++;
}
next_field:
cur_field++;
cur_bytes = 0;
}
ut_ad(cur_bytes == 0);
ret = 0; /* If we ran out of fields, dtuple was equal to rec
up to the common fields */
order_resolved:
ut_ad((ret >= - 1) && (ret <= 1));
ut_ad(ret == cmp_debug_dtuple_rec_with_match(dtuple, rec, offsets,
matched_fields));
ut_ad(*matched_fields == cur_field); /* In the debug version, the
above cmp_debug_... sets
*matched_fields to a value */
*matched_fields = cur_field;
*matched_bytes = cur_bytes;
return(ret);
}
/*************************************************************//**
Used in debug checking of cmp_dtuple_... .
This function is used to compare a data tuple to a physical record. If
dtuple has n fields then rec must have either m >= n fields, or it must
differ from dtuple in some of the m fields rec has. If encounters an
externally stored field, returns 0.
@return 1, 0, -1, if dtuple is greater, equal, less than rec,
respectively, when only the common first fields are compared */
static
int
cmp_debug_dtuple_rec_with_match(
/*============================*/
const dtuple_t* dtuple, /*!< in: data tuple */
const rec_t* rec, /*!< in: physical record which differs from
dtuple in some of the common fields, or which
has an equal number or more fields than
dtuple */
const ulint* offsets,/*!< in: array returned by rec_get_offsets() */
ulint* matched_fields) /*!< in/out: number of already
completely matched fields; when function
returns, contains the value for current
comparison */
{
const dfield_t* dtuple_field; /* current field in logical record */
ulint dtuple_f_len; /* the length of the current field
in the logical record */
const byte* dtuple_f_data; /* pointer to the current logical
field data */
ulint rec_f_len; /* length of current field in rec */
const byte* rec_f_data; /* pointer to the current rec field */
int ret = 3333; /* return value */
ulint cur_field; /* current field number */
ut_ad(dtuple && rec && matched_fields);
ut_ad(dtuple_check_typed(dtuple));
ut_ad(rec_offs_validate(rec, NULL, offsets));
ut_ad(*matched_fields <= dtuple_get_n_fields_cmp(dtuple));
ut_ad(*matched_fields <= rec_offs_n_fields(offsets));
cur_field = *matched_fields;
if (cur_field == 0) {
if (UNIV_UNLIKELY
(rec_get_info_bits(rec, rec_offs_comp(offsets))
& REC_INFO_MIN_REC_FLAG)) {
ret = !(dtuple_get_info_bits(dtuple)
& REC_INFO_MIN_REC_FLAG);
goto order_resolved;
}
if (UNIV_UNLIKELY
(dtuple_get_info_bits(dtuple) & REC_INFO_MIN_REC_FLAG)) {
ret = -1;
goto order_resolved;
}
}
/* Match fields in a loop; stop if we run out of fields in dtuple */
while (cur_field < dtuple_get_n_fields_cmp(dtuple)) {
ulint mtype;
ulint prtype;
dtuple_field = dtuple_get_nth_field(dtuple, cur_field);
{
const dtype_t* type
= dfield_get_type(dtuple_field);
mtype = type->mtype;
prtype = type->prtype;
}
dtuple_f_data = dfield_get_data(dtuple_field);
dtuple_f_len = dfield_get_len(dtuple_field);
rec_f_data = rec_get_nth_field(rec, offsets,
cur_field, &rec_f_len);
if (rec_offs_nth_extern(offsets, cur_field)) {
/* We do not compare to an externally stored field */
ret = 0;
goto order_resolved;
}
ret = cmp_data_data(mtype, prtype, dtuple_f_data, dtuple_f_len,
rec_f_data, rec_f_len);
if (ret != 0) {
goto order_resolved;
}
cur_field++;
}
ret = 0; /* If we ran out of fields, dtuple was equal to rec
up to the common fields */
order_resolved:
ut_ad((ret >= - 1) && (ret <= 1));
*matched_fields = cur_field;
return(ret);
}
/**************************************************************************
Reports in the undo log of an update or delete marking of a clustered index
record. */
static
ulint
trx_undo_page_report_modify(
/*========================*/
/* out: byte offset of the inserted
undo log entry on the page if succeed,
0 if fail */
page_t* undo_page, /* in: undo log page */
trx_t* trx, /* in: transaction */
dict_index_t* index, /* in: clustered index where update or
delete marking is done */
rec_t* rec, /* in: clustered index record which
has NOT yet been modified */
const ulint* offsets, /* in: rec_get_offsets(rec, index) */
upd_t* update, /* in: update vector which tells the
columns to be updated; in the case of
a delete, this should be set to NULL */
ulint cmpl_info, /* in: compiler info on secondary
index updates */
mtr_t* mtr) /* in: mtr */
{
dict_table_t* table;
upd_field_t* upd_field;
ulint first_free;
byte* ptr;
ulint len;
byte* field;
ulint flen;
ulint pos;
dulint roll_ptr;
dulint trx_id;
ulint bits;
ulint col_no;
byte* old_ptr;
ulint type_cmpl;
byte* type_cmpl_ptr;
ulint i;
ut_a(index->type & DICT_CLUSTERED);
ut_ad(rec_offs_validate(rec, index, offsets));
ut_ad(mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
+ TRX_UNDO_PAGE_TYPE) == TRX_UNDO_UPDATE);
table = index->table;
first_free = mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
+ TRX_UNDO_PAGE_FREE);
ptr = undo_page + first_free;
ut_ad(first_free <= UNIV_PAGE_SIZE);
if (trx_undo_left(undo_page, ptr) < 50) {
/* NOTE: the value 50 must be big enough so that the general
fields written below fit on the undo log page */
return(0);
}
/* Reserve 2 bytes for the pointer to the next undo log record */
ptr += 2;
/* Store first some general parameters to the undo log */
if (update) {
if (rec_get_deleted_flag(rec, dict_table_is_comp(table))) {
type_cmpl = TRX_UNDO_UPD_DEL_REC;
} else {
type_cmpl = TRX_UNDO_UPD_EXIST_REC;
}
} else {
type_cmpl = TRX_UNDO_DEL_MARK_REC;
}
type_cmpl = type_cmpl | (cmpl_info * TRX_UNDO_CMPL_INFO_MULT);
mach_write_to_1(ptr, type_cmpl);
type_cmpl_ptr = ptr;
ptr++;
len = mach_dulint_write_much_compressed(ptr, trx->undo_no);
ptr += len;
len = mach_dulint_write_much_compressed(ptr, table->id);
ptr += len;
/*----------------------------------------*/
/* Store the state of the info bits */
bits = rec_get_info_bits(rec, dict_table_is_comp(table));
mach_write_to_1(ptr, bits);
ptr += 1;
/* Store the values of the system columns */
field = rec_get_nth_field(rec, offsets,
dict_index_get_sys_col_pos(
index, DATA_TRX_ID), &len);
ut_ad(len == DATA_TRX_ID_LEN);
trx_id = trx_read_trx_id(field);
field = rec_get_nth_field(rec, offsets,
dict_index_get_sys_col_pos(
index, DATA_ROLL_PTR), &len);
ut_ad(len == DATA_ROLL_PTR_LEN);
roll_ptr = trx_read_roll_ptr(field);
len = mach_dulint_write_compressed(ptr, trx_id);
ptr += len;
len = mach_dulint_write_compressed(ptr, roll_ptr);
ptr += len;
/*----------------------------------------*/
/* Store then the fields required to uniquely determine the
record which will be modified in the clustered index */
for (i = 0; i < dict_index_get_n_unique(index); i++) {
field = rec_get_nth_field(rec, offsets, i, &flen);
if (trx_undo_left(undo_page, ptr) < 4) {
return(0);
}
len = mach_write_compressed(ptr, flen);
ptr += len;
if (flen != UNIV_SQL_NULL) {
if (trx_undo_left(undo_page, ptr) < flen) {
return(0);
}
ut_memcpy(ptr, field, flen);
ptr += flen;
}
}
/*----------------------------------------*/
/* Save to the undo log the old values of the columns to be updated. */
if (update) {
if (trx_undo_left(undo_page, ptr) < 5) {
return(0);
}
len = mach_write_compressed(ptr, upd_get_n_fields(update));
ptr += len;
for (i = 0; i < upd_get_n_fields(update); i++) {
upd_field = upd_get_nth_field(update, i);
pos = upd_field->field_no;
/* Write field number to undo log */
if (trx_undo_left(undo_page, ptr) < 5) {
return(0);
}
len = mach_write_compressed(ptr, pos);
ptr += len;
/* Save the old value of field */
field = rec_get_nth_field(rec, offsets, pos, &flen);
if (trx_undo_left(undo_page, ptr) < 5) {
return(0);
}
if (rec_offs_nth_extern(offsets, pos)) {
/* If a field has external storage, we add
to flen the flag */
len = mach_write_compressed(
ptr,
UNIV_EXTERN_STORAGE_FIELD + flen);
/* Notify purge that it eventually has to
free the old externally stored field */
trx->update_undo->del_marks = TRUE;
*type_cmpl_ptr = *type_cmpl_ptr
| TRX_UNDO_UPD_EXTERN;
} else {
len = mach_write_compressed(ptr, flen);
}
ptr += len;
if (flen != UNIV_SQL_NULL) {
if (trx_undo_left(undo_page, ptr) < flen) {
return(0);
}
ut_memcpy(ptr, field, flen);
ptr += flen;
}
}
}
/*----------------------------------------*/
/* In the case of a delete marking, and also in the case of an update
where any ordering field of any index changes, store the values of all
columns which occur as ordering fields in any index. This info is used
in the purge of old versions where we use it to build and search the
delete marked index records, to look if we can remove them from the
index tree. Note that starting from 4.0.14 also externally stored
fields can be ordering in some index. But we always store at least
384 first bytes locally to the clustered index record, which means
we can construct the column prefix fields in the index from the
stored data. */
if (!update || !(cmpl_info & UPD_NODE_NO_ORD_CHANGE)) {
trx->update_undo->del_marks = TRUE;
if (trx_undo_left(undo_page, ptr) < 5) {
return(0);
}
old_ptr = ptr;
/* Reserve 2 bytes to write the number of bytes the stored
fields take in this undo record */
ptr += 2;
for (col_no = 0; col_no < dict_table_get_n_cols(table);
col_no++) {
const dict_col_t* col
= dict_table_get_nth_col(table, col_no);
if (col->ord_part > 0) {
pos = dict_index_get_nth_col_pos(index,
col_no);
/* Write field number to undo log */
if (trx_undo_left(undo_page, ptr) < 5) {
return(0);
}
len = mach_write_compressed(ptr, pos);
ptr += len;
/* Save the old value of field */
field = rec_get_nth_field(rec, offsets, pos,
&flen);
if (trx_undo_left(undo_page, ptr) < 5) {
return(0);
}
len = mach_write_compressed(ptr, flen);
ptr += len;
if (flen != UNIV_SQL_NULL) {
if (trx_undo_left(undo_page, ptr)
< flen) {
return(0);
}
ut_memcpy(ptr, field, flen);
ptr += flen;
}
}
}
mach_write_to_2(old_ptr, ptr - old_ptr);
}
/*----------------------------------------*/
/* Write pointers to the previous and the next undo log records */
if (trx_undo_left(undo_page, ptr) < 2) {
return(0);
}
mach_write_to_2(ptr, first_free);
ptr += 2;
mach_write_to_2(undo_page + first_free, ptr - undo_page);
mach_write_to_2(undo_page + TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_FREE,
ptr - undo_page);
/* Write to the REDO log about this change in the UNDO log */
trx_undof_page_add_undo_rec_log(undo_page, first_free,
ptr - undo_page, mtr);
return(first_free);
}
/***************************************************************
Removes a secondary index entry without modifying the index tree,
if possible.
@return TRUE if success or if not found */
static
ibool
row_purge_remove_sec_if_poss_leaf(
/*==============================*/
purge_node_t* node, /*!< in: row purge node */
dict_index_t* index, /*!< in: index */
const dtuple_t* entry) /*!< in: index entry */
{
mtr_t mtr;
btr_pcur_t pcur;
enum row_search_result search_result;
log_free_check();
mtr_start(&mtr);
/* Set the purge node for the call to row_purge_poss_sec(). */
pcur.btr_cur.purge_node = node;
/* Set the query thread, so that ibuf_insert_low() will be
able to invoke thd_get_trx(). */
pcur.btr_cur.thr = que_node_get_parent(node);
search_result = row_search_index_entry(
index, entry, BTR_MODIFY_LEAF | BTR_DELETE, &pcur, &mtr);
switch (search_result) {
ibool success;
case ROW_FOUND:
/* Before attempting to purge a record, check
if it is safe to do so. */
if (row_purge_poss_sec(node, index, entry)) {
btr_cur_t* btr_cur = btr_pcur_get_btr_cur(&pcur);
/* Only delete-marked records should be purged. */
ut_ad(REC_INFO_DELETED_FLAG
& rec_get_info_bits(
btr_cur_get_rec(btr_cur),
dict_table_is_comp(index->table)));
if (!btr_cur_optimistic_delete(btr_cur, &mtr)) {
/* The index entry could not be deleted. */
success = FALSE;
goto func_exit;
}
}
/* fall through (the index entry is still needed,
or the deletion succeeded) */
case ROW_NOT_DELETED_REF:
/* The index entry is still needed. */
case ROW_BUFFERED:
/* The deletion was buffered. */
case ROW_NOT_FOUND:
/* The index entry does not exist, nothing to do. */
success = TRUE;
func_exit:
btr_pcur_close(&pcur);
mtr_commit(&mtr);
return(success);
}
ut_error;
return(FALSE);
}
/***************************************************************
Removes a secondary index entry if possible, by modifying the
index tree. Does not try to buffer the delete.
@return TRUE if success or if not found */
static
ibool
row_purge_remove_sec_if_poss_tree(
/*==============================*/
purge_node_t* node, /*!< in: row purge node */
dict_index_t* index, /*!< in: index */
const dtuple_t* entry) /*!< in: index entry */
{
btr_pcur_t pcur;
btr_cur_t* btr_cur;
ibool success = TRUE;
ulint err;
mtr_t mtr;
enum row_search_result search_result;
log_free_check();
mtr_start(&mtr);
search_result = row_search_index_entry(index, entry, BTR_MODIFY_TREE,
&pcur, &mtr);
switch (search_result) {
case ROW_NOT_FOUND:
/* Not found. This is a legitimate condition. In a
rollback, InnoDB will remove secondary recs that would
be purged anyway. Then the actual purge will not find
the secondary index record. Also, the purge itself is
eager: if it comes to consider a secondary index
record, and notices it does not need to exist in the
index, it will remove it. Then if/when the purge
comes to consider the secondary index record a second
time, it will not exist any more in the index. */
/* fputs("PURGE:........sec entry not found\n", stderr); */
/* dtuple_print(stderr, entry); */
goto func_exit;
case ROW_FOUND:
break;
case ROW_BUFFERED:
case ROW_NOT_DELETED_REF:
/* These are invalid outcomes, because the mode passed
to row_search_index_entry() did not include any of the
flags BTR_INSERT, BTR_DELETE, or BTR_DELETE_MARK. */
ut_error;
}
btr_cur = btr_pcur_get_btr_cur(&pcur);
/* We should remove the index record if no later version of the row,
which cannot be purged yet, requires its existence. If some requires,
we should do nothing. */
if (row_purge_poss_sec(node, index, entry)) {
/* Remove the index record, which should have been
marked for deletion. */
ut_ad(REC_INFO_DELETED_FLAG
& rec_get_info_bits(btr_cur_get_rec(btr_cur),
dict_table_is_comp(index->table)));
btr_cur_pessimistic_delete(&err, FALSE, btr_cur,
RB_NONE, &mtr);
switch (UNIV_EXPECT(err, DB_SUCCESS)) {
case DB_SUCCESS:
break;
case DB_OUT_OF_FILE_SPACE:
success = FALSE;
break;
default:
ut_error;
}
}
func_exit:
btr_pcur_close(&pcur);
mtr_commit(&mtr);
return(success);
}
dtuple_t*
row_build(
/*======*/
/* out, own: row built; see the NOTE below! */
ulint type, /* in: ROW_COPY_POINTERS, ROW_COPY_DATA, or
ROW_COPY_ALSO_EXTERNALS,
the two last copy also the data fields to
heap as the first only places pointers to
data fields on the index page, and thus is
more efficient */
dict_index_t* index, /* in: clustered index */
rec_t* rec, /* in: record in the clustered index;
NOTE: in the case ROW_COPY_POINTERS
the data fields in the row will point
directly into this record, therefore,
the buffer page of this record must be
at least s-latched and the latch held
as long as the row dtuple is used! */
mem_heap_t* heap) /* in: memory heap from which the memory
needed is allocated */
{
dtuple_t* row;
dict_table_t* table;
dict_field_t* ind_field;
dict_col_t* col;
dfield_t* dfield;
ulint n_fields;
byte* field;
ulint len;
ulint row_len;
byte* buf;
ulint i;
ut_ad(index && rec && heap);
ut_ad(index->type & DICT_CLUSTERED);
if (type != ROW_COPY_POINTERS) {
/* Take a copy of rec to heap */
buf = mem_heap_alloc(heap, rec_get_size(rec));
rec = rec_copy(buf, rec);
}
table = index->table;
row_len = dict_table_get_n_cols(table);
row = dtuple_create(heap, row_len);
dtuple_set_info_bits(row, rec_get_info_bits(rec));
n_fields = dict_index_get_n_fields(index);
ut_ad(n_fields == rec_get_n_fields(rec));
dict_table_copy_types(row, table);
for (i = 0; i < n_fields; i++) {
ind_field = dict_index_get_nth_field(index, i);
if (ind_field->prefix_len == 0) {
col = dict_field_get_col(ind_field);
dfield = dtuple_get_nth_field(row,
dict_col_get_no(col));
field = rec_get_nth_field(rec, i, &len);
if (type == ROW_COPY_ALSO_EXTERNALS
&& rec_get_nth_field_extern_bit(rec, i)) {
field = btr_rec_copy_externally_stored_field(
rec, i, &len, heap);
}
dfield_set_data(dfield, field, len);
}
}
ut_ad(dtuple_check_typed(row));
return(row);
}
