Обсуждение: binary upgade errors

David Modica <davidmo@imaginesoftware.com> writes:
> I have been unsuccessfully trying to use pg_upgrade to upgrade from 9.6 to 10.4.
> we have the uint extension in some of the databases. a combination of that extension
> and probably how we have used it is causing the upgrade to fail. I will include the error msg.

> pg_restore: [archiver (db)] Error while PROCESSING TOC:
> pg_restore: [archiver (db)] Error from TOC entry 4538; 1247 280489 DOMAIN bool_t postgres
> pg_restore: [archiver (db)] could not execute query: ERROR:  cannot cast type integer to public.uint1
>     Command was:
> -- For binary upgrade, must preserve pg_type oid
> SELECT pg_catalog.binary_upgrade_set_next_pg_type_oid('280489'::pg_catalog.oid);

> CREATE DOMAIN "its"."bool_t" AS "public"."uint1" DEFAULT (0)::"public"."uint1";

Hmm ... it looks like this domain is depending on there to be a cast from
integer to uint1, but that hasn't been created yet.  There may be a bug
here; I'm not sure why pg_dump didn't delay dumping the domain to after
it'd dumped the cast.  In the meantime, though, it seems like it would
work (and probably be faster anyway) if you spelled the default value
like '0'::uint1 rather than 0::uint1, as the former isn't depending on
any run-time cast.  So try altering the domain like that and then doing
the upgrade.

            regards, tom lane

tom,

thank you so much for your reply.

isn't 0::uint1 casting an int as uint and '0'::uint1 is casting a text as a uint1 ?

not sure why one doesn't require a run time cast and the other doesn't. would you

be able to explain to me ? I am sorry but this casting is new for me.

thanks,

david



-----Original Message-----
From: Tom Lane <tgl@sss.pgh.pa.us>
Sent: Thursday, June 14, 2018 4:24 PM
To: David Modica <davidmo@imaginesoftware.com>
Cc: pgsql-admin@lists.postgresql.org
Subject: Re: binary upgade errors

David Modica <davidmo@imaginesoftware.com> writes:
> I have been unsuccessfully trying to use pg_upgrade to upgrade from 9.6 to 10.4.
> we have the uint extension in some of the databases. a combination of
> that extension and probably how we have used it is causing the upgrade to fail. I will include the error msg.

> pg_restore: [archiver (db)] Error while PROCESSING TOC:
> pg_restore: [archiver (db)] Error from TOC entry 4538; 1247 280489
> DOMAIN bool_t postgres
> pg_restore: [archiver (db)] could not execute query: ERROR:  cannot cast type integer to public.uint1
>     Command was:
> -- For binary upgrade, must preserve pg_type oid SELECT
> pg_catalog.binary_upgrade_set_next_pg_type_oid('280489'::pg_catalog.oi
> d);

> CREATE DOMAIN "its"."bool_t" AS "public"."uint1" DEFAULT
> (0)::"public"."uint1";

Hmm ... it looks like this domain is depending on there to be a cast from integer to uint1, but that hasn't been
createdyet.  There may be a bug here; I'm not sure why pg_dump didn't delay dumping the domain to after it'd dumped the
cast. In the meantime, though, it seems like it would work (and probably be faster anyway) if you spelled the default
valuelike '0'::uint1 rather than 0::uint1, as the former isn't depending on any run-time cast.  So try altering the
domainlike that and then doing the upgrade. 

            regards, tom lane

David Modica <davidmo@imaginesoftware.com> writes:
> isn't 0::uint1 casting an int as uint and '0'::uint1 is casting a text as a uint1 ?

No.  0 is an integer constant and so the first case requires invoking
an int -> uint1 cast.  The second syntax implies applying uint1's type
input function to the string '0', and arriving at a constant immediately.
It's perhaps unfortunate that similar-looking syntax means two basically
different things ... but it's not the only place in SQL where much turns
on punctuation.

https://www.postgresql.org/docs/current/static/sql-expressions.html#SQL-SYNTAX-TYPE-CASTS

            regards, tom lane

isn't 0::uint1 casting an int as uint and '0'::uint1 is casting a text as a uint1 ?