Re: BUG #5735: pg_upgrade thinks that it did not start the old server

On 11/10/2010 05:32 AM, Tom Lane wrote:
> Bruce Momjian<bruce(at)momjian(dot)us> writes:
>
>>> On 10/30/2010 7:33 PM, Dave Page wrote:
>>>
>>>> upgrade from a 32bit 8.3 server to a 64 bit 9.0 server, which isn't
>>>> going to work without a dump/restore. With pg_upgrade, the two builds
>>>> need to be from the same platform, same word size, and have the same
>>>> configuration for certain settings like integer_datetimes.
>>>>
>> Can anyone suggest a way pg_upgrade could detect an upgrade from a
>> 32-bit to 64-bit cpu and throw an error?
>>
> Surely it does that already, as a result of comparing pg_control
> contents.
>

The HTML manual might need an update or a small clarification too.
Currently, it does not seem that the manual explicitly states that
``pg_upgrade is not applicable in upgrading 32bit systems to 64bit
ones''. A good place to write such a sentence would be at the beginning
of [1], at the intro of F.32. pg_upgrade. Maybe the documentation
already implicitly states that in F.32.4. Limitations in Migrating from
PostgreSQL 8.3 section of [1] by this description:

``For Windows users, note that due to different integer datetimes
settings used by the one-click installer and the MSI installer, it is
only possible to upgrade from version 8.3 of the one-click distribution
to version 8.4 or later of the one-click distribution. It is not
possible to upgrade from the MSI installer to the one-click installer.''

Unfortunately, I could not understand in full detail the above.

Thanks,
arturas

[1] http://www.postgresql.org/docs/9.0/static/pgupgrade.html