Support for 'infinity' timestamps

[yrashk]

Postgres supports infinity timestamps.

=# select 'infinity'::timestamptz;
 timestamptz 
-------------
 infinity
(1 row)

However, this adapter doesn't:

It is, however, a valid and useful range to support, particularly in the context of temporal tables.

[tconbeer]

Interesting. I actually think this is a problem with Harlequin core, or more specifically, the data table implementation. We use Apache Arrow to store the data returned by the query, and I don't think Arrow supports infinity timestamps (they have an Arrow.max constant, I believe). We probably just need to catch this error and replace the infinity value with Arrow.max.

[tconbeer]

Turns out this error is a "feature" of psycopg3: https://www.psycopg.org/psycopg3/docs/basic/from_pg2.html#no-default-infinity-dates-handling

For comparison, duckdb also supports infinity timestamps, and while the fetch_arrow_table() behavior is a little confusing, it looks like it uses the target data type's max value (datetime.max, numpy.max, or arrow.max):

>>> cur  = duckdb.execute("select 'infinity'::timestamp as foo")
>>> cur.fetchall()
[(datetime.datetime(9999, 12, 31, 23, 59, 59, 999999),)]

>>> cur2 = duckdb.execute("select 'infinity'::timestamp as foo")
>>> tab = cur2.fetch_arrow_table()
>> tab
pyarrow.Table
foo: timestamp[us]
----
foo: [[<value out of range: 9223372036854775807>]]
>>> tab[0][0]
<pyarrow.TimestampScalar: '32103-01-10T04:00:54.775807'>

>>> cur3 = duckdb.execute("select 'infinity'::timestamp as foo")
>>> cur3.fetchnumpy()
{'foo': array(['294247-01-10T04:00:54.775807'], dtype='datetime64[us]')}

I think that's the better experience, so I'm going to do that, and then also probably render the max timestamps as infinity or something in the data table.