xmlelement AND timestamps.

On 02/13/2017 02:56 PM, Lynn Dobbs wrote:

I just migrated from 9.2.4 to 9.6.1 and had several user created
functions fail.

Recreating the failure with "SELECT xmlelement(name foo,
'infinity'::timestamp)
ERROR: timestamp out of range
DETAIL: XML does not support infinite timestamp values.

I don't find anything in the documentation that explains this. I
consider this a regression.

All I could find was this thread from 2009:

/messages/by-id/41F26B729B014C3E8F20F5B7@teje

which indicated it was fixed at that time.

I have many tables that have a "starting" and "ending" timestamp that
default to "-infinity"
and "infinity" respectively. Any function I have that outputs xml
containing those columns
have to have those values cast to text.

Lynn Dobbs
--
Chief Technical Office
CreditLink Corporation
858-496-1000 x 103

--
Adrian Klaver
adrian.klaver@aklaver.com

--
Sent via pgsql-general mailing list (pgsql-general@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-general

On Mon, Feb 13, 2017 at 6:36 PM, Adrian Klaver <adrian.klaver@aklaver.com>
wrote:

On 02/13/2017 02:56 PM, Lynn Dobbs wrote:

I just migrated from 9.2.4 to 9.6.1 and had several user created
functions fail.

Recreating the failure with "SELECT xmlelement(name foo,
'infinity'::timestamp)
ERROR: timestamp out of range
DETAIL: XML does not support infinite timestamp values.

I don't find anything in the documentation that explains this. I
consider this a regression.

All I could find was this thread from 2009:

/messages/by-id/41F26B729B014C3E8F20F5B7@teje

which indicated it was fixed at that time.

​Actually, the cause of said commit (circa 2009) is that xmlelements were
failing but xmlattributes were not​...which makes me wonder how a 9.2 era
release (circa 2012) supposedly worked with the submitted expression.

The basic answer is that the XML data type is defined by standard and we
attempt to conform to the standard. In order to do so we must disallow
infinite timestamps even though we accept them in SQL.

I just tested the OP query on 9.0.x and 9.4.x and get the same error on
both.

There'd be a bit more sympathy if the OP were complaining about a patch
release changing behavior - bug fix or not - but since the complaint
involves going from 9.2 to 9.6 on its face this is an allowed behavior
change regardless of the history.

However, feel free to make a straight argument for accepting infinite
timestamps and thus go above-and-beyond the relevant standards. Personally
this seems like a not-so-useful restriction on our implementation. Let the
user decide whether they want to deviate from the standard and risk
cross-system incompatibilities. XML itself is textual and we don't have
any internal support for DTD or Schema as it is so I'm not sure what
material benefit we gain by restraining ourselves here.

David J.

Lynn Dobbs <lynn.dobbs@creditlink.com> writes:

I just migrated from 9.2.4 to 9.6.1 and had several user created
functions fail.

Recreating the failure with "SELECT xmlelement(name foo,
'infinity'::timestamp)
ERROR: timestamp out of range
DETAIL: XML does not support infinite timestamp values.

I don't find anything in the documentation that explains this. I
consider this a regression.

So far as I can tell, Postgres has rejected converting infinite timestamps
to XML since 8.3 (cf commit 7b76bfbe1). Certainly the above example fails
exactly like that in 9.2. If you think there's a regression here, you
need to show us a case that actually behaves differently in 9.2 and 9.6.

(Speculating wildly, I imagine that your problem has something to do with
9.6 trying to fold a subexpression to a constant in a case where 9.2
didn't, and in fact didn't evaluate the subexpression at all. But you'd
need a much larger example to demonstrate such a behavior.)

regards, tom lane

--
Sent via pgsql-general mailing list (pgsql-general@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-general

On Mon, Feb 13, 2017 at 7:10 PM, David G. Johnston <
david.g.johnston@gmail.com> wrote:

XML itself is textual and we don't have any internal support for DTD or
Schema as it is so I'm not sure what material benefit we gain by
restraining ourselves here.

​This apparently isn't true - the XML output representations include
schema-verifiable namespaces and ​thus have a well-defined allowable set of
output values and formats. While not every use case may have or need these
at present their presence is embedded into the implementation - making any
desired change non-trivial to evaluate.

David J.

Well, I couldn't reproduce allowing infinite timestamps in 9.2.4 either.

While fixing some other minor issues that came up in the migration to
9.6.1, I copied a replaced good functions with bad ones.

The good functions called another function when putting my
starting,ending columns in xml. That function dates the timestamp or
date and formats it and converts to text.

The documentation does say, "The particular behavior for individual data
types is expected to evolve in order to align the SQL and PostgreSQL
data types with the XML Schema specification, ...." I didn't go chasing
that down. My bad.

My bad functions were written using the xml2 extension that treated the
xml as text rather than "real" xml.

Lynn Dobbs
Chief Technical Office
CreditLink Corporation
858-496-1000 x 103