Message-ID: <88dadcc1-6652-ff7d-264d-73906a53dee9@illuminatedcomputing.com>
Date: Fri, 1 Sep 2023 12:56:40 -0700
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.13.0
Subject: Re: SQL:2011 application time
Content-Language: en-US
To: Vik Fearing <vik@postgresfriends.org>,
 Peter Eisentraut <peter@eisentraut.org>
Cc: Corey Huinker <corey.huinker@gmail.com>,
 PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
References: <CA+renyUApHgSZF9-nd-a0+OPGharLQLO=mDHcY4_qQ0+noCUVg@mail.gmail.com>
 <CA+renyUMKg4-MyVbAi8Kj12Srd9BkU5Gt3FdbGdNr-ezHpL6cw@mail.gmail.com>
 <5650d77a-1d73-fd8d-4410-8e97dbeff572@illuminatedcomputing.com>
 <CA+renyWqpNdsi9OaPmXheeQL72adi_iz36VgVsqyyGpQ20FDAg@mail.gmail.com>
 <21964319-46e5-3e47-217b-6ac4169bdf13@enterprisedb.com>
 <CA+renyX0ipvY6A_jUOHeB1q9mL4bEYfAZ5FBB7G7jUo5bykjrA@mail.gmail.com>
 <f42e5be5-cd3e-bc4f-83e4-04a1e0270b17@enterprisedb.com>
 <27361388-f5ab-ea36-ea35-41d68a90e60d@illuminatedcomputing.com>
 <152e4c2c-36b4-bdc0-1b62-e9f9e8b68adc@enterprisedb.com>
 <251B44D1-D3E0-47E4-AB9D-4A848EDB495E@yesql.se>
 <1c674adf-4114-e8d8-cec7-ac10e2f424c7@enterprisedb.com>
 <CA+renyVyMOb+t6QNoq=PBK1CY_oa9wdJrObjj6S1-RK9CxYwKQ@mail.gmail.com>
 <831f17ba-ff36-1380-a475-a7cd2c65a89c@enterprisedb.com>
 <ab4cf49c-9839-3426-5114-f596fa9ee76b@illuminatedcomputing.com>
 <6f010a6e-8e20-658b-dc05-dc9033a694da@eisentraut.org>
 <7d44dade-ff31-5eef-80ac-78ed838066f6@postgresfriends.org>
From: Paul Jungwirth <pj@illuminatedcomputing.com>
In-Reply-To: <7d44dade-ff31-5eef-80ac-78ed838066f6@postgresfriends.org>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Archived-At: <https://www.postgresql.org/message-id/88dadcc1-6652-ff7d-264d-73906a53dee9%40illuminatedcomputing.com>
Precedence: bulk

On 9/1/23 03:50, Vik Fearing wrote:
> On 9/1/23 11:30, Peter Eisentraut wrote:
>> 1) If I write UNIQUE (a, b, c WITHOUT OVERLAPS), does the WITHOUT 
>> OVERLAPS clause attach to the last column, or to the whole column 
>> list? In the SQL standard, you can only have one period and it has to 
>> be listed last, so this question does not arise.  But here we are 
>> building a more general facility to then build the SQL facility on top 
>> of.  So I think it doesn't make sense that the range column must be 
>> last or that there can only be one.  Also, your implementation 
>> requires at least one non-overlaps column, which also seems like a 
>> confusing restriction.
>>
>> I think the WITHOUT OVERLAPS clause should be per-column, so that 
>> something like UNIQUE (a WITHOUT OVERLAPS, b, c WITHOUT OVERLAPS) 
>> would be possible.  Then the WITHOUT OVERLAPS clause would directly 
>> correspond to the choice between equality or overlaps operator per 
>> column.
>>
>> An alternative interpretation would be that WITHOUT OVERLAPS applies 
>> to the whole column list, and we would take it to mean, for any range 
>> column, use the overlaps operator, for any non-range column, use the 
>> equals operator.  But I think this would be confusing and would 
>> prevent the case of using the equality operator for some ranges and 
>> the overlaps operator for some other ranges in the same key.
> 
> I prefer the first option.  That is: WITHOUT OVERLAPS applies only to 
> the column or expression it is attached to, and need not be last in line.

I agree. The second option seems confusing and is more restrictive.

I think allowing multiple uses of `WITHOUT OVERLAPS` (and in any 
position) is a great recommendation that enables a lot of new 
functionality. Several books[1,2] about temporal databases describe a 
multi-dimensional temporal space (even beyond application time vs. 
system time), and the standard is pretty disappointing here. It's not a 
weird idea.

But I just want to be explicit that this isn't something the standard 
describes. (I think everyone in the conversation so far understands 
that.) So far I've tried to be pretty scrupulous about following 
SQL:2011, although personally I'd rather see Postgres support this 
functionality. And it's not like it goes *against* what the standard 
says. But if there are any objections, I'd love to hear them before 
putting in the work. :-)

If we allow multiple+anywhere WITHOUT OVERLAPS in PRIMARY KEY & UNIQUE 
constraints, then surely we also allow multiple+anywhere PERIOD in 
FOREIGN KEY constraints too. (I guess the standard switched keywords 
because a FK is more like "MUST OVERLAPS". :-)

Also if you have multiple application-time dimensions we probably need 
to allow multiple FOR PORTION OF clauses. I think the syntax would be:

UPDATE t
   FOR PORTION OF valid_at FROM ... TO ...
   FOR PORTION OF asserted_at FROM ... TO ...
   [...]
   SET foo = bar

Does that sound okay?

I don't quite understand this part:

 >> Also, your implementation
 >> requires at least one non-overlaps column, which also seems like a
 >> confusing restriction.

That's just a regular non-temporal constraint. Right? If I'm missing 
something let me know.

[1] C. J. Date, Hugh Darwen, Nikos Lorentzos. Time and Relational 
Theory, Second Edition: Temporal Databases in the Relational Model and 
SQL. 2nd edition, 2014.
[2] Tom Johnston. Bitemporal Data: Theory and Practice. 2014.

-- 
Paul              ~{:-)
pj@illuminatedcomputing.com