In the HISTORY file comming with the source code there are several
modifications noted related to characterset handling in general and to
unicode more specificaly in the versions before and 7.4.13 it self. The
bottom line is that postgresql in earlier versions did allow incorrect
UNICODE sequences. The recent releases do handle UNICODE stricly according to
the rouls and will not allow incorrect sequences. When you now do have a
backup with incorrect sequences in, you will get this error messages. You can
fix this by fixing the wrong sequences within the dump manually or by a tool
like recode.
Best regards
Ivo Rossacher
Am Mittwoch, 21. Juni 2006 12.57 schrieb Thusitha Kodikara:
> Hello,
>
> We did some more testing and managed to get the dump restored on 7.4.10.
> Then took a backup and tried to restore on to 7.4.13, but if failed again
> giving the same errors.
>
> In addition to this we took a dump of an existing DB on 7.4.13 and tried to
> restore it onto 7.4.13 itself. This also failed.
>
> Could it be that our data contain some invalid characters (even though we
> never came across any problem before installing the patch 7.4.13) ? In all
> these instances our encoding was UNICODE.
>
> Does anyone have a clue how to solve this?
>
> Thanks.
> -Thusitha
>
> TK <kthusi@yahoo.com> wrote: Hi,
>
> We took a dump of a database running PostgreSQL 7.3.10 (on Gentoo Linux).
>
> We tried to restore it onto a server running PostgreSQL 7.4.13 (on Gentoo
> Linux). This resulted in the following errors:
>
> psql:<stdin>:13856: ERROR: invalid byte sequence for encoding "UNICODE":
> 0x85 ....
> psql:<stdin>:537916: ERROR: invalid byte sequence for encoding "UNICODE":
> 0xa0 ...
> psql:<stdin>:1009873: ERROR: invalid byte sequence for encoding "UNICODE":
> 0x92 ...
> etc.
>
> We tried to restore the same dump on a server running PostgreSQL 7.3.9 (on
> Gentoo Linux) and the restoration was successful. It also worked without
> any problem on a server running PostgreSQL 7.4.5 (on Gentoo Linux).
>
> All the databases in this scenario were of UNICODE encoding type.
>
> What could be the reson for this, and how can we avoid this?
>
> Thanks in advance,
>
> -Thusitha
