Re: bzr commit into mysql-6.0 branch (alik:2728) Bug#39519

[Konstantin Osipov]

* Alexander Nozdrin <alik@stripped> [08/11/12 14:51]:
> per-file messages:
>   include/mysql.h
>     Add a new flag into MYSQL structure (flush_all_delete).
>     This flag determines if all results should be flushed
>     or only the first one:
>     
>       - if MYSQL::flush_all_delete is TRUE, then
>         cli_flush_use_result() will read/flush all pending
>         results. I.e. it will read all packets while server
>         status attribute indicates that there are more results.
>         This is a new semantic, required to fix the bug.
>     
>       - if MYSQL::flush_all_results is FALSE, the old sematic
>         is preserved -- i.e. cli_flush_use_result() reads data
>         untile first EOF-packet.

Can't extend MYSQL with more members, this breaks the ABI.
And this is not necessary, see below.

> === modified file 'libmysql/libmysql.c'
> --- a/libmysql/libmysql.c	2008-08-07 17:52:43 +0000
> +++ b/libmysql/libmysql.c	2008-11-12 11:22:58 +0000
> @@ -4692,7 +4692,9 @@ my_bool STDCALL mysql_stmt_close(MYSQL_S
>            Flush result set of the connection. If it does not belong
>            to this statement, set a warning.
>          */
> +        mysql->flush_all_results= TRUE;
>          (*mysql->methods->flush_use_result)(mysql);
> +        mysql->flush_all_results= FALSE;
>          if (mysql->unbuffered_fetch_owner)
>            *mysql->unbuffered_fetch_owner= TRUE;
>          mysql->status= MYSQL_STATUS_READY;

This is the only place where you use this flag.
Why not use an own function instead of the flag?


> +    {
> +      /* Analyze OK-packet. */
> +
> +      uchar *pos= mysql->net.read_pos + 1;
> +
> +      net_field_length_ll(&pos); /* affected rows */
> +      net_field_length_ll(&pos); /* insert id */
> +
> +      mysql->server_status=uint2korr(pos);
> +      pos+=2;
> +
>        if (protocol_41(mysql))
>        {
> -        char *pos= (char*) mysql->net.read_pos + 1;
> -        mysql->warning_count=uint2korr(pos); pos+=2;
> -        mysql->server_status=uint2korr(pos); pos+=2;
> +        mysql->warning_count=uint2korr(pos);
> +        pos+=2;
>        }
> -      break;                            /* End of data */
> +
> +      /* message is ignored. */
> +
> +      if (mysql->server_status & SERVER_MORE_RESULTS_EXISTS)
> +        continue;

In prepared statements, you can't get server_status &
SERVER_MORE_RESULTS_EXISTS after an OK packet. 
When does this code work?


--