On Fri, Jun 12, 2015 at 10:06 PM, Michael Meskes <meskes@postgresql.org> wrote:
> On Fri, Jun 12, 2015 at 03:53:43PM +0900, Michael Paquier wrote:
>> (Adding Michael Meskes in CC:)
>
> Thanks.
>
>> Nice catch. Regarding your patch, it seems to me that this is not
>> enough. I think that we should return NULL with ECPGget_sqlca in case
>> of an OOM instead of logging a message with ecpg_log and let each code
>> path decide what to do when sqlca is NULL. Some code paths can
>> directly use ecpg_raise with ECPG_OUT_OF_MEMORY. Other code paths,
>> like the ones in error.c will need to show up with appropriate error
>> messages.
>
> Agreed on all accounts.
So, here is a patch implementing those ideas. In code paths where a
line number is available ecpg_raise() is called to report the error.
In other paths ecpg_log() is used to log an "out of memory" message.
Now, the routines of error.c, like ecpg_raise() can fail as well their
malloc() call, hence it seems adapted to me to fallback to ecpg_log()
and report the error to the user.
Thoughts?
--
Michael