Subject: Re: BUG #15623: Inconsistent use of default for updatable view
References: <15623-5d67a46788ec8b7f@postgresql.org>
 <952caeff-2cc7-6db9-1f2b-418442b51045@lab.ntt.co.jp>
 <CAEZATCUmSp3-8nLOpgGcPkpUEXK9TJGM=iA6q4E2Sn=+bwkKNA@mail.gmail.com>
 <CA+HiwqGyuNo7O-G0LUz-=1f4ZSSST4-9sLhP+Twx-PFJSPRfug@mail.gmail.com>
 <CAEZATCXTg_gxzf86+eTyLzJCY3OACQ71Wo_TPaeZyNAuo7frOA@mail.gmail.com>
 <CAEZATCX1cN4Lcd80ZJXHf7qVRrMReOSOUF_0Z6YFhk89NNcGbw@mail.gmail.com>
 <CAEZATCVGid6C7v6aMdA_7ZKPJPGYpwQ2bQhYrZ0=E5RDJQp7yw@mail.gmail.com>
 <CAEZATCUDzfG_dLMjSkwd+2R=GUFtYoT9NWc8XebM0FUf1BDddQ@mail.gmail.com>
From: Amit Langote <Langote_Amit_f8@lab.ntt.co.jp>
Message-ID: <a1e31ca8-717a-094e-b701-0502f7cfb9b7@lab.ntt.co.jp>
Date: Thu, 28 Feb 2019 16:46:57 +0900
User-Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:60.0) Gecko/20100101
 Thunderbird/60.5.1
MIME-Version: 1.0
In-Reply-To: 
 <CAEZATCUDzfG_dLMjSkwd+2R=GUFtYoT9NWc8XebM0FUf1BDddQ@mail.gmail.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
To: Dean Rasheed <dean.a.rasheed@gmail.com>,
        Amit Langote <amitlangote09@gmail.com>
Cc: Roger Curley <rocurley@gmail.com>, pgsql-bugs@lists.postgresql.org,
        PG Bug reporting form <noreply@postgresql.org>,
        PostgreSQL Hackers <pgsql-hackers@postgresql.org>
Precedence: bulk

On 2019/02/27 18:37, Dean Rasheed wrote:
> On Tue, 12 Feb 2019 at 10:33, Dean Rasheed <dean.a.rasheed@gmail.com> wrote:
>> Here's an updated patch ...
> 
> So I pushed that. However, ...
> 
> Playing around with it some more, I realised that whilst this appeared
> to fix the reported problem, it exposes another issue which is down to
> the interaction between rewriteTargetListIU() and rewriteValuesRTE()
> --- for an INSERT with a VALUES RTE, rewriteTargetListIU() computes a
> list of target attribute numbers (attrno_list) from the targetlist in
> its original order, which rewriteValuesRTE() then relies on being the
> same length (and in the same order) as each of the lists in the VALUES
> RTE. That's OK for the initial invocation of those functions, but it
> breaks down when they're recursively invoked for auto-updatable views.
>> So actually, I think the right way to fix this is to give up trying to
> compute attrno_list in rewriteTargetListIU(), and have
> rewriteValuesRTE() work out the attribute assignments itself for each
> column of the VALUES RTE by examining the rewritten targetlist. That
> looks to be quite straightforward, and results in a cleaner separation
> of logic between the 2 functions, per the attached patch.

+1.  Only rewriteValuesRTE needs to use that information, so it's better
to teach it to figure it by itself.

> I think that rewriteValuesRTE() should only ever see DEFAULT items for
> columns that are simple assignments to columns of the target relation,
> so it only needs to work out the target attribute numbers for TLEs
> that contain simple Vars referring to the VALUES RTE. Any DEFAULT seen
> in a column not matching that would be an error, but actually I think
> such a thing ought to be a "can't happen" error because of the prior
> checks during parse analysis, so I've used elog() to report if this
> does happen.

+                if (attrno == 0)
+                    elog(ERROR, "Cannot set value in column %d to
DEFAULT", i);

Maybe: s/Cannot/cannot/g

+        Assert(list_length(sublist) == numattrs);

Isn't this Assert useless, because we're setting numattrs to
list_length(<first-sublist>) and transformInsertStmt ensures that all
sublists are same length?


Thanks,
Amit