Обсуждение: COPY problem with bad dates

It looks like when a bad date is entered using COPY the backend connection
gets killed.
I noticed this when having a problem loading a database with several copies
in one transaction. The backend died consistantly at the same point, but
not near where the bad dates were. I eventually found the bad dates
and corrected the problem with the load. However I noticed that the
error message I got indicated the backend died, so I thought I should
try to get a simpler case that shows the same problem.
Below is a short psql example run against a 7.3 database that also
gets the backend to crash, though not in the middle of a large transaction.
bruno=> create table test (col1 timestamp);
CREATE TABLE
bruno=> copy test from stdin;
Enter data to be copied followed by a newline.
End with a backslash and a period on a line by itself.
>> 11/31/02
>> \.
ERROR:  copy: line 1, Bad timestamp external representation '11/31/02'
lost synchronization with server, resetting connection

On Wed, Dec 18, 2002 at 09:14:18 -0600,
  Bruno Wolff III <bruno@wolff.to> wrote:
> It looks like when a bad date is entered using COPY the backend connection
> gets killed.

I just tried it with 7.3.1 and get the same error.

Bruno Wolff III <bruno@wolff.to> writes:
> ERROR:  copy: line 1, Bad timestamp external representation '11/31/02'
> lost synchronization with server, resetting connection

This is not a backend crash, this is libpq's standard operating
procedure for recovering from an error during COPY.

If we had a less brain-dead COPY protocol (ie, one that actually had
provisions for error abort), we'd not have to do this.  See plans for
7.4 protocol revisions ...

            regards, tom lane