List:Internals« Previous MessageNext Message »
From:Michael Widenius Date:August 30 2001 2:20pm
Subject:Re: understanding the source
View as plain text  

>>>>> "Benjamin" == Benjamin Pflugmann <philemon@stripped> writes:

Benjamin> Hi.
Benjamin> Another question. I have tried to figure this out for over half an
Benjamin> hour now and don't get it. If I start to annoy you, just tell me and I
Benjamin> will search for some longer.

To send an email after figuring half and hour is ok.

Benjamin> The problem is within the intended ha_myisammrg::write_row:

Benjamin> --- mysql-3.23.40/sql/	Wed Jul 18 23:19:10 2001
Benjamin> +++ mysql-3.23.40-philemon/sql/	Thu Aug 30 04:36:49 2001
Benjamin> @@ -67,5 +67,11 @@
Benjamin>  int ha_myisammrg::write_row(byte * buf)
Benjamin>  {
Benjamin> -  return (my_errno=HA_ERR_WRONG_COMMAND);
Benjamin> +  statistic_increment(ha_write_count,&LOCK_status);
Benjamin> +  if (table->time_stamp)
Benjamin> +    update_timestamp(buf+table->time_stamp-1);
Benjamin> +  if (table->next_number_field && buf == table->record[0])
Benjamin> +      return (my_errno=HA_ERR_WRONG_COMMAND);
Benjamin> +  /*  update_auto_increment();   [phi] have to check this before allowing it
> */
Benjamin> +  return myrg_write(file,buf);
Benjamin>  }

Benjamin> This is mainly taken from ha_myisam::write_row and I just assured that
Benjamin> everything still makes sense. My problem with that is, that I cannot
Benjamin> figure out what "buf == table->record[0]" is intended to do/mean. I
Benjamin> understand that table->record[0] seems to be a pointer to a "row".

Benjamin> But what does the comparison mean (i.e. when is
Benjamin> (buf == table->record[0]) true and when false?) and at which place in
Benjamin> the source is table->record[0] set to the relevant value?

Every table object has 3 record object associated with it:

table->record[0]       A changed record or a record with new values
table->record[1]       Records are read into this
table->record[2]       An record with default values

In almost all cases in MySQL we are using table->record[0] to write
new rows.

The only case when we are not is when copying a row from a table to
another (Ie: We are writing an unchanged row (record[1]) to another
identical table).  In this case we should not update the
auto_increment column.

The check in the original code is:

  if (table->next_number_field && buf == table->record[0])

I have now added the following comment to the code:

    If we have an auto_increment column and we are writing a changed row
    or a new row, then update the auto_increment value in the record.


understanding the sourceBenjamin Pflugmann30 Aug
  • bug confirmed (was: Re: understanding the source)Benjamin Pflugmann30 Aug
  • Re: understanding the sourceBenjamin Pflugmann30 Aug
    • Re: understanding the sourceMichael Widenius30 Aug
    • Re: understanding the sourceTimothy Smith30 Aug
      • Re: understanding the sourceMichael Widenius2 Sep
  • understanding the sourceMichael Widenius2 Sep