Feedback-ID: ie0a040ee:Fastmail
Message-ID: <22c4dd72-66fa-4f76-95f2-757d754e2aad@eisentraut.org>
Date: Mon, 3 Mar 2025 11:05:22 +0100
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: SQL:2011 application time
To: Paul Jungwirth <pj@illuminatedcomputing.com>
Cc: jian he <jian.universality@gmail.com>,
 PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>,
 vignesh C <vignesh21@gmail.com>
References: <CA+renyUApHgSZF9-nd-a0+OPGharLQLO=mDHcY4_qQ0+noCUVg@mail.gmail.com>
 <66e47fc0-77b7-4763-9b70-5c0b7ff15f21@eisentraut.org>
 <a2ee0c1a-5ef0-43ab-897c-b56adb25b34f@illuminatedcomputing.com>
 <333d3886-b737-45c3-93f4-594c96bb405d@eisentraut.org>
 <e8e51e74-10d7-410d-a1d1-8ec681d8436a@eisentraut.org>
 <f4056005-0fb7-417a-bcb3-e704f9aeadc3@illuminatedcomputing.com>
 <6a2247d9-461d-4b0a-b692-f0f3c0fab27f@eisentraut.org>
 <8f995a4c-64e3-40cd-bfed-06b54a280d36@illuminatedcomputing.com>
 <0832f18d-523f-4c15-acb0-338a785150c5@illuminatedcomputing.com>
 <5f0c25de-2c13-4fb5-a4df-a1e03c43cd14@illuminatedcomputing.com>
 <1b45f81d-4e99-4813-a2c6-72bf5fd80980@eisentraut.org>
 <e773bc11-4ac1-40de-bb91-814e02f05b6d@eisentraut.org>
 <3e061027-781d-42c1-b6e8-bafbf3b51033@eisentraut.org>
 <c3af7816-1944-4030-ab73-5b055de8cba3@illuminatedcomputing.com>
 <3b33690e-0d67-47a5-8424-93ac1fed4c26@eisentraut.org>
 <1085d8a3-17a7-4639-afd2-bd6dfa09713d@illuminatedcomputing.com>
Content-Language: en-US
From: Peter Eisentraut <peter@eisentraut.org>
In-Reply-To: <1085d8a3-17a7-4639-afd2-bd6dfa09713d@illuminatedcomputing.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Archived-At: <https://www.postgresql.org/message-id/22c4dd72-66fa-4f76-95f2-757d754e2aad%40eisentraut.org>
Precedence: bulk

On 26.02.25 06:15, Paul Jungwirth wrote:
> I agree with that last principle: it shouldn't matter how the primary 
> keys are split up. But it seems to me that "matches" in the standard 
> should include the period. It does for NO ACTION, so why not RESTRICT? 
> That's why your example of expanding the referenced range succeeds. None 
> of the referenced moments were changed, so there are no referencing 
> moments to match.
> 
>  > I'm not sure what other behavior of RESTRICT there might be that is 
> internally consistent and is
>  > meaningfully different from NO ACTION.
> 
> The difference between RESTRICT and NO ACTION for temporal foreign keys 
> is the same as the difference for ordinary foreign keys: we perform the 
> check prior to applying any "action" or allowing any other commands to 
> provide substitutes for the lost references. There are tests in sql/ 
> without_overlaps.sql showing how their behavior differs.
> 
> Also you haven't yet explained why anyone would *want* to use RESTRICT 
> as you've described it, since the temporal part of the key is just 
> ignored, and they could just define a non-temporal foreign key instead. 
> Or to be precise, it fails *more* than a non-temporal foreign key, 
> because changing the period can violate the constraint, even though we 
> ignore the period when looking for matches.

This is not what I'm aiming for.  (Maybe my patches were wrong about that.)

In the theory of the SQL standard, executing referential actions and 
checking the foreign-key constraint are two separate steps.  So it kind 
of goes like this:

1. run command
2. run any referential actions
3. check that foreign key is still satisfied

This is why the default referential action is called "NO ACTION": It 
just skips the step 2.  But it still does step 3.

This means that under RESTRICT and with my interpretation, the check for 
a RESTRICT violation in step 2 can "ignore" the period part, but the 
step 3 still has to observe the period part.

In the implementation, these steps are mostly combined into one trigger 
function, so it might be a bit tricky to untangle them.

> But since we don't agree on the behavior, it seems best to me to wait to 
> implement RESTRICT. Not much is lost, since NO ACTION is so similar. We 
> can wait for the SQL committee to clarify things, or see what another 
> RDBMS vendor does.

I'm fine with that.