Received: from malur.postgresql.org ([217.196.149.56]) by arkaria.postgresql.org with esmtps (TLS1.3:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.92) (envelope-from ) id 1odzI9-0008H5-MF for pgsql-hackers@arkaria.postgresql.org; Thu, 29 Sep 2022 19:33:53 +0000 Received: from localhost ([127.0.0.1] helo=malur.postgresql.org) by malur.postgresql.org with esmtp (Exim 4.92) (envelope-from ) id 1odzI8-0003hd-6V for pgsql-hackers@arkaria.postgresql.org; Thu, 29 Sep 2022 19:33:52 +0000 Received: from magus.postgresql.org ([2a02:c0:301:0:ffff::29]) by malur.postgresql.org with esmtps (TLS1.3:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.92) (envelope-from ) id 1odzI7-0003hT-TC for pgsql-hackers@lists.postgresql.org; Thu, 29 Sep 2022 19:33:51 +0000 Received: from sss.pgh.pa.us ([66.207.139.130]) by magus.postgresql.org with esmtps (TLS1.3:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.92) (envelope-from ) id 1odzI5-0006rm-Vf for pgsql-hackers@lists.postgresql.org; Thu, 29 Sep 2022 19:33:51 +0000 Received: from sss1.sss.pgh.pa.us (localhost [127.0.0.1]) by sss.pgh.pa.us (8.15.2/8.15.2) with ESMTP id 28TJXQr5344668; Thu, 29 Sep 2022 15:33:26 -0400 From: Tom Lane To: Martin Kalcher cc: Fabien COELHO , Andres Freund , Dean Rasheed , Robert Haas , Andrew Dunstan , John Naylor , Thomas Munro , PostgreSQL Hackers Subject: Re: [PATCH] Introduce array_shuffle() and array_sample() In-reply-to: References: <459f6008-90bc-bc52-6f11-e68570e03ac0@aboutsource.net> <583b6409-df52-86d1-8c0a-ce8e74863ad7@aboutsource.net> <036ba529-1f25-d7b5-c5ad-e6c852d928df@aboutsource.net> <8fabf9e7-3422-58b0-93f6-d0976d4679df@aboutsource.net> <20220922152320.ukx6teo7wa7cuwkv@awork3.anarazel.de> <193caac5-941e-58b8-65e6-351f777ebd25@aboutsource.net> <3502301.1664223385@sss.pgh.pa.us> <9cd0925c-9839-a136-c21d-87de9e8a3aa4@aboutsource.net> <3943591.1664374725@sss.pgh.pa.us> Comments: In-reply-to Martin Kalcher message dated "Thu, 29 Sep 2022 11:39:28 +0200" MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-ID: <344666.1664480006.1@sss.pgh.pa.us> Content-Transfer-Encoding: quoted-printable Date: Thu, 29 Sep 2022 15:33:26 -0400 Message-ID: <344667.1664480006@sss.pgh.pa.us> List-Id: List-Help: List-Subscribe: List-Post: List-Owner: List-Archive: Archived-At: Precedence: bulk Martin Kalcher writes: > New patch: array_shuffle() and array_sample() use pg_global_prng_state n= ow. I took a closer look at the patch today. I find this behavior a bit surprising: +SELECT array_dims(array_sample('[-1:2][2:3]=3D{{1,2},{3,NULL},{5,6},{7,8}= }'::int[], 3)); + array_dims = +------------- + [-1:1][2:3] +(1 row) I can buy preserving the lower bound in array_shuffle(), but array_sample() is not preserving the first-dimension indexes of the array, so ISTM it ought to reset the first lower bound to 1. Some other comments: + Returns n randomly chosen elements from array in selection order. What's "selection order"? And this probably shouldn't just rely on the example to describe what happens with multi-D arrays. Writing "elements" seems somewhere between confusing and wrong. * Personally I think I'd pass the TypeCacheEntry pointer to array_shuffle_= n, and let it pull out what it needs. Less duplicate code that way. * I find array_shuffle_n drastically undercommented, and what comments it has are pretty misleading, eg + /* Swap all elements in item (i) with elements in item (j). */ j is *not* the index of the second item to be swapped. You could make it so, and that might be more readable: j =3D (int) pg_prng_uint64_range(&pg_global_prng_state, i, nitem - 1); jelms =3D elms + j * nelm; jnuls =3D nuls + j * nelm; But if you want the code to stay as it is, this comment needs work. * I think good style for SQL-callable C functions is to make the arguments clear at the top: +array_sample(PG_FUNCTION_ARGS) +{ + ArrayType *array =3D PG_GETARG_ARRAYTYPE_P(0); + int n =3D PG_GETARG_INT32(1); + ArrayType *result; + ... other declarations as needed ... We can't quite make normal C declaration style work, but that's a poor excuse for not making the argument list visible as directly as possible. * I wouldn't bother with the PG_FREE_IF_COPY calls. Those are generally only used in btree comparison functions, in which there's a need to not leak memory. * I wonder if we really need these to be proparallel =3D 'r'. If we let a worker process execute them, they will take numbers from the worker's pg_global_prng_state seeding not the parent process's seeding, but why is that a problem? In the prior version where you were tying them to the parent's drandom() sequence, proparallel =3D 'r' made sense; but now I think it's unnecessary. regards, tom lane