From: Sven Sandberg
Date: June 29 2011 8:33am
Subject: Re: BUG#11747577
List-Archive: http://lists.mysql.com/commits/139995
Message-Id: <4E0AE351.9090105@oracle.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

> === modified file 'client/mysql.cc'
> --- client/mysql.cc	2011-06-14 13:34:19 +0000
> +++ client/mysql.cc	2011-06-29 08:13:09 +0000
> @@ -1973,8 +1973,13 @@ static int read_and_execute(bool interac
>    {
>      if (!interactive)
>      {
> +      /*
> +        batch_readline can return 0 on EOF or error.
> +        In that case, we need to double check that we have a valid
> +        line before actually setting line_length to read_length.
> +        */
>        line= batch_readline(status.line_buff, real_binary_mode);
> -      line_length= status.line_buff->read_length;
> +      line_length= line ? status.line_buff->read_length : 0;
>        /*
>          ASCII 0x00 is not allowed appearing in queries if it is not in binary
>          mode.
> @@ -2004,7 +2009,11 @@ static int read_and_execute(bool interac
>             (uchar) line[0] == 0xEF &&
>             (uchar) line[1] == 0xBB &&
>             (uchar) line[2] == 0xBF)
> +      {
>          line+= 3;
> +        // decrease the line length accordingly to the 3 bytes chopped
> +        line_length -=3;
> +      }
>        line_number++;
>        if (!glob_buffer.length())
>  	status.query_start_line=line_number;

This looks good. I would suggest one refactoring here: there are three 
'if' blocks in sequence that test for line!=NULL. To clarify the code, 
can you move that check to an enclosing 'if'? I.e.:

   line= batch_readline
   if (line)
   {
     line_length= status.line_buff->read_length;
     if (!real_binary_mode && strlen(line) != line_length)
     {...}
     if (!line_number &&
         (uchar) line[0] == 0xEF &&
         (uchar) line[1] == 0xBB &&
         (uchar) line[2] == 0xBF)
     {...}
   }


/Sven