Message-ID: <bce1da9a-2975-462f-8946-c58c878ba82c@iki.fi>
Date: Mon, 11 Nov 2024 15:25:29 +0200
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: Support LIKE with nondeterministic collations
To: Peter Eisentraut <peter@eisentraut.org>,
 Jacob Champion <jacob.champion@enterprisedb.com>
Cc: pgsql-hackers <pgsql-hackers@postgresql.org>,
 Daniel Verite <daniel@manitou-mail.org>,
 Paul A Jungwirth <pj@illuminatedcomputing.com>
References: <a827cf68-608e-45cf-9b94-e46f5e9d4171@manitou-mail.org>
 <fdc379ce-4b88-4357-b622-2ae0ef08e295@eisentraut.org>
 <CA+renyWd-_sAj3YqBRaQVOOMr5uQoeBcA3tjCSyQFzvnbGrMYA@mail.gmail.com>
 <ba4dc9f5-fb14-433f-8d53-3dd291585161@eisentraut.org>
 <3104729e-1fbf-4368-ac21-1f670062de28@eisentraut.org>
 <CAOYmi+nqr4xCe9-g4BAupnu2rZmvLy1T3qq3ejOUWOCsoJ4ZdA@mail.gmail.com>
 <0ca761b5-7b62-42a1-bffd-8bedefad48dd@eisentraut.org>
Content-Language: en-US
From: Heikki Linnakangas <hlinnaka@iki.fi>
In-Reply-To: <0ca761b5-7b62-42a1-bffd-8bedefad48dd@eisentraut.org>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Archived-At: <https://www.postgresql.org/message-id/bce1da9a-2975-462f-8946-c58c878ba82c%40iki.fi>
Precedence: bulk

On 04/11/2024 10:26, Peter Eisentraut wrote:
> On 29.10.24 18:15, Jacob Champion wrote:
>> libfuzzer is unhappy about the following code in MatchText:
>>
>>> +            while (p1len > 0)
>>> +            {
>>> +                if (*p1 == '\\')
>>> +                {
>>> +                    found_escape = true;
>>> +                    NextByte(p1, p1len);
>>> +                }
>>> +                else if (*p1 == '_' || *p1 == '%')
>>> +                    break;
>>> +                NextByte(p1, p1len);
>>> +            }
>>
>> If the pattern ends with a backslash, we'll call NextByte() twice,
>> p1len will wrap around to INT_MAX, and we'll walk off the end of the
>> buffer. (I fixed it locally by duplicating the ERROR case that's
>> directly above this.)
> 
> Thanks.  Here is an updated patch with that fixed.

Sadly the algorithm is O(n^2) with non-deterministic collations.Is there 
any way this could be optimized? We make no claims on how expensive any 
functions or operators are, so I suppose a slow implementation is 
nevertheless better than throwing an error.

Let's at least add some CHECK_FOR_INTERRUPTS(). For example, this takes 
a very long time and is uninterruptible:

  SELECT repeat('x', 100000) LIKE '%xxxy%' COLLATE ignore_accents;

-- 
Heikki Linnakangas
Neon (https://neon.tech)