Обсуждение: More on updates with first() vs absolute()

Good morning,
Yesterday I reported a crash in updateRow() when preceded by an
absolute(1) positioning call, when preceding it with first() worked. It
now seems the DB update was happening OK; the crash was a null pointer
exception at the line:

     rowBuffer[columnIndex] =
connection.getEncoding().encode(String.valueOf(valueObject));

of updateRowBuffer(). It appears the rowBuffer isn't being initialized
properly, and, sure enough, when you compare first() with absolute(),
the former has:

            rowBuffer = new byte[this_row.length][];
        System.arraycopy(this_row, 0, rowBuffer, 0,
this_row.length);

while the latter does not. Adding the initialization cures the crash.
There might be other locations needing the same patch.

If this is an acceptable patch, I'll be happy to submit it if someone
tells me how.  (It looks like a cvs diff to pgsql-jdbc is the desired
way, but I'm not clear how to do that: do I start with the current
pg-snapshot or something else?)

                               Rich Cullingford
                               System Detection, Inc.
                               5 W 19th St, NYC, NY 10010
                               rculling@sysd.com

Rich,

Thanks for debugging.  The fix is obvious enough from your description
that a patch is unnecessary.  I will check the fix into CVS later today.

thanks,
--Barry


Rich Cullingford wrote:
> Good morning,
> Yesterday I reported a crash in updateRow() when preceded by an
> absolute(1) positioning call, when preceding it with first() worked. It
> now seems the DB update was happening OK; the crash was a null pointer
> exception at the line:
>
>     rowBuffer[columnIndex] =
> connection.getEncoding().encode(String.valueOf(valueObject));
>
> of updateRowBuffer(). It appears the rowBuffer isn't being initialized
> properly, and, sure enough, when you compare first() with absolute(),
> the former has:
>
>            rowBuffer = new byte[this_row.length][];
>         System.arraycopy(this_row, 0, rowBuffer, 0,      this_row.length);
>
> while the latter does not. Adding the initialization cures the crash.
> There might be other locations needing the same patch.
>
> If this is an acceptable patch, I'll be happy to submit it if someone
> tells me how.  (It looks like a cvs diff to pgsql-jdbc is the desired
> way, but I'm not clear how to do that: do I start with the current
> pg-snapshot or something else?)
>
>                               Rich Cullingford
>                               System Detection, Inc.
>                               5 W 19th St, NYC, NY 10010
>                               rculling@sysd.com
>
>
> ---------------------------(end of broadcast)---------------------------
> TIP 2: you can get off all lists at once with the unregister command
>    (send "unregister YourEmailAddressHere" to majordomo@postgresql.org)
>