Received: from localhost (wm.hub.org [200.46.204.128]) by postgresql.org (Postfix) with ESMTP id B97579FB223; Fri, 29 Sep 2006 14:59:33 -0300 (ADT) Received: from postgresql.org ([200.46.204.71]) by localhost (mx1.hub.org [200.46.204.128]) (amavisd-new, port 10024) with ESMTP id 39673-05; Fri, 29 Sep 2006 17:59:29 +0000 (UTC) X-Greylist: from auto-whitelisted by SQLgrey- X-Greylist: from auto-whitelisted by SQLgrey- Received: from sss.pgh.pa.us (sss.pgh.pa.us [66.207.139.130]) by postgresql.org (Postfix) with ESMTP id 6B5B99FB2B8; Fri, 29 Sep 2006 14:59:29 -0300 (ADT) Received: from sss2.sss.pgh.pa.us (tgl@localhost [127.0.0.1]) by sss.pgh.pa.us (8.13.6/8.13.6) with ESMTP id k8THxRbE018111; Fri, 29 Sep 2006 13:59:27 -0400 (EDT) To: "John D. Burger" cc: Postgres General , PostgreSQL-development , pgsql-admin@postgresql.org Subject: Re: [GENERAL] Array assignment behavior (was Re: Stored procedure array limits) In-reply-to: References: <451C48AB.6090508@pnlassociates.com> <9461.1159502232@sss.pgh.pa.us> <451D34FD.3050709@pnlassociates.com> <451D37C8.8060502@pnlassociates.com> <16926.1159546451@sss.pgh.pa.us> <595CA7DC-7EC2-47B6-A894-DCD0ED3825F3@pandora.com> Comments: In-reply-to "John D. Burger" message dated "Fri, 29 Sep 2006 12:40:48 -0400" Date: Fri, 29 Sep 2006 13:59:27 -0400 Message-ID: <18110.1159552767@sss.pgh.pa.us> From: Tom Lane X-Virus-Scanned: Maia Mailguard 1.0.1 X-Archive-Number: 200609/373 X-Sequence-Number: 23287 "John D. Burger" writes: >>> As of 8.2 we could allow assignment to >>> arbitrary positions by filling the intermediate positions with nulls. >>> The code hasn't actually been changed to allow that, but it's >>> something we could consider doing now. >> >> At first blush, this strikes me as a bit too magical/implicit. Are >> there other languages where sequences behave similarly? >>> perl -e '@A = (1, 2, 3); print "@A\n"; $A[10] = 10; print "@A\n";' > 1 2 3 > 1 2 3 10 Actually, now that I look closely, I think the SQL spec demands exactly this. Recall that SQL99 only allows one-dimensional, lower-bound-one arrays. The specification for UPDATE ... SET C[I] = SV ... reads Case: i) If the value of C is null, then an exception condition is raised: data exception - null value in array target. ii) Otherwise: 1) Let N be the maximum cardinality of C. 2) Let M be the cardinality of the value of C. 3) Let I be the value of the immediately contained in . 4) Let EDT be the element type of C. 5) Case: A) If I is greater than zero and less than or equal to M, then the value of C is replaced by an array A with element type EDT and cardinality M derived as follows: I) For j varying from 1 (one) to I-1 and from I+1 to M, the j-th element in A is the value of the j-th element in C. II) The I-th element of A is set to the specified update value, denoted by SV, by applying the General Rules of Subclause 9.2, "Store assignment", to the I-th element of A and SV as TARGET and VALUE, respectively. B) If I is greater than M and less than or equal to N, then the value of C is replaced by an array A with element type EDT and cardinality I derived as follows: I) For j varying from 1 (one) to M, the j-th element in A is the value of the j-th element in C. II) For j varying from M+1 to I-1, the j-th element in A is the null value. III) The I-th element of A is set to the specified update value, denoted by SV, by applying the General Rules of Subclause 9.2, "Store assignment", to the I-th element of A and SV as TARGET and VALUE, respectively. C) Otherwise, an exception condition is raised: data exception - array element error. We currently violate case i by allowing the null array value to be replaced by a single-element array. I'm disinclined to change that, as I think our behavior is more useful than the spec's. But case ii.5.B pretty clearly describes null-fill, so I think we'd better do that, now that we can. regards, tom lane