If a resultset in binary protpcpl doesn't contain data
(fields equal zero) and there are no more pending
result sets we need to change the statement status to
FETCH_DONE, otherwise it will be impossible to execute
other commands until the statement was closed.
If prepare step failed in mariadb_stmt_execute_direct now both
mysql_stmt_error and mysql_error return the error message from
prepare step instead of error message of execute.
- ER() macro now checks if the error code is known, if not it will return
"Unknown or undefined error code" (instead of crashing)
- SET_CLIENT_STMT_ERROR now maps to stmt_set_error and accepts variadic
arguments
Since Connector/C is not able to retrieve
error strings for server error codes, the following
error codes were replaced:
- ER_NET_PACKET_TOO_LARFE by CR_NET_PACKET_TOO_LARGE
- ER_OUT_OF_RESOURCES by CR_OUT_OF_MEMORY
- ER_NET_WRITE_ERROR by CR_ERR_NET_WRITE (new constant)
- ER_NET_UNCOMPRESS_ERROR by CR_ERR_NET_UNCOMPRESS (new constant)
If mariadb_stmt_execute_direct fails, we need to set the number
of parameters (obtained by response packet of mysql_stmt_prepare)
back to the number of prebinded parameters to avoid memory
overrun.
This fix also includes several adress-sanitizer bugs in unit tests
of Connector/C.
When calculating the length of buffer for prepared statement
executiong a maximum length of 5 for length encoded types was
used. This should be 9, since length encoded types starting
with 0xFE are followed by length of 8 bytes.
If cursor type is set to read only, and SERVER_STATUS_CURSOR_EXISTS
flag was not set in the server_status of EOF packet, the assumption
was made that a single result set will follow which will be stored
by a call to mysql_stmt_store_result().
As the example in the bug report shows, this is not always the case,
since a stored procedure call might also unset the CURSOR flag and
will always return multiple result sets. In this case it will be
handled now as "normal" unbuffered result set.
- removed unused methods from MYSQL_STMT handle,
(left over from PHP's mysqlnd)
- Added execute_generate_request method: This will allow
Connector/Python to prefill the execute buffer without
numerous GIL acquire/release calls.
Added a new option MARIADB_OPT_SKIP_READ_RESPONSE which skips automatic
reading of server response after sending a command to the server.
Server packets have to be retrieved by calling the corresponding methods,
e.g:
Send command Read method
mysql_real_query/mysql_send_query db_read_query_result
mysql_stmt_prepare db_read_prepare_response
mysql_stmt_execute,
mariadb_stmt_execute_direct db_read_execute_response
unittests always reset the default connection (and depending statements)
which might end up in an error when trying to close a statement which was
not prepared before.
In this case mysql_stmt_reset only needs to set statement status to
MYSQL_STMT_PREPARE if the statement id was set before.
While in text protocol the number of rows is resetted in mysql_store/use_result
in binary protocol we need to explicitly reset it when switching to next result set.
- If prepare failed, we need to ignore errormessage from mysql_stmt_execute()
(which is always an error packet, mentioning invalid sttmt_id)
- Added additional check in mysql_stmt_reset for stmt_id == -1.
1) In case a cursor is used in a stored procedure server sends a SERVER_STATUS_CURSOR_EXISTS flag,
even if the client didn't open a cursor via mysql_stmt_attr_set. This ends in a "comands out of sync" error,
since the client is sending a COM_STMT_FETCH command while the result sets were not read before.
We check now in case server sends a SERVER_STATUS_CURSOR_EXIST flag, if the client opened a cursor before.
2) The stmt error codes weren't update for all COM_STMT* commands correctly, e.g. COM_STMT_FETCH didn't show an error even if it failed.
