Remove net_printf_error(). Do not talk to network directly in

check_user()/check_connection()/check_for_max_user_connections().

This is a pre-requisite patch for the fix for Bug#12713 "Error in a stored 
function called from a SELECT doesn't cause ROLLBACK of statem"

Implement review comments.


sql/mysql_priv.h:
  check_for_max_user_connections() is used in one place only, make it static.
sql/mysqld.cc:
  Remove net_printf_error(): a consolidation of error reporting facilities
  is necessary to simplify maintenance of the query cache, the 
  client-server protocol, stored procedure continue handlers. 
  Rewrite the only place where its use is somewhat justified 
  (my_error() can not be used since we need to report an error for the thread 
  that does not exist) with my_snprintf()/net_send_error().
sql/protocol.cc:
  Remove net_printf_error().
sql/protocol.h:
  Remove net_printf_error().
sql/sql_connect.cc:
  Remove net_printf_error(). In check_connection()/check_user()/
  check_for_max_user_connections() do not write directly to the network,
  but use the standard my_error() mechanism to record an error in THD.
  It will be sent to the client by the caller. This was the last place
  in the server that would attempt to send an error directly, mainly left 
  untouched by 5.0 refactoring because it is executed only during 
  thread startup.
sql/sql_parse.cc:
  In the old code, when res was greater than 0, it contained an exact 
  error code, e.g. ER_OUT_OF_RESOURCES or NO SUCH DATABASE, 
  or ER_HANDSHAKE_ERROR. I don't know the reason why this error code was 
  ignored, and instead a generic  ER_UNKNOWN_COM_ERROR was pushed into the
  error stack, but knowing the relaxed attitude towards preserving the error
  codes in the old code, I'm inclinded to think that it was a bug.
  
  After this patch, the most specific error message is already pushed,
  so calling my_message() again is useless.
  
  If res is < 0, the error used to be already sent. This is not done
  by the new code, but will be done later, in the end
  of dispatch_command(). When this is done, clear_error() will be called 
  for us - it is in the first lines of do_command.
  
  To sum up, this change is to remove COM_CHANGE_USER specific error handling
  in favor of the standard one employed for all other COM_* commands.

sql/sql_parse.cc

@@ -915,11 +915,6 @@ bool dispatch_command(enum enum_server_command command, THD *thd,
 
     if (res)
     {
-      /* authentication failure, we shall restore old user */
-      if (res > 0)
-        my_message(ER_UNKNOWN_COM_ERROR, ER(ER_UNKNOWN_COM_ERROR), MYF(0));
-      else
-        thd->clear_error();                     // Error already sent to client
       x_free(thd->security_ctx->user);
       *thd->security_ctx= save_security_ctx;
       thd->user_connect= save_user_connect;