Feedback-ID: ie0a040ee:Fastmail
Message-ID: <2afbd9c6-51b7-450f-9ae1-61e552368963@eisentraut.org>
Date: Wed, 9 Apr 2025 11:39:49 +0200
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: pgsql: Make cancel request keys longer
To: Heikki Linnakangas <hlinnaka@iki.fi>,
 pgsql-committers@lists.postgresql.org
References: <E1tzyNB-002IXa-0i@gemulon.postgresql.org>
 <61be9e31-7b7d-49d5-bc11-721800d89d64@eisentraut.org>
 <09323e6a-4743-4be2-9d7f-74b86e4dac64@iki.fi>
 <34d5d731-06cc-43af-88cd-b3a4c5c8d9df@iki.fi>
Content-Language: en-US
From: Peter Eisentraut <peter@eisentraut.org>
In-Reply-To: <34d5d731-06cc-43af-88cd-b3a4c5c8d9df@iki.fi>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Archived-At: 
 <https://www.postgresql.org/message-id/2afbd9c6-51b7-450f-9ae1-61e552368963%40eisentraut.org>
Precedence: bulk

On 09.04.25 10:53, Heikki Linnakangas wrote:
> On 08/04/2025 22:41, Heikki Linnakangas wrote:
>> On 08/04/2025 20:06, Peter Eisentraut wrote:
>>> While I was looking at this, I suggest to make the first argument 
>>> void *.  This is consistent for passing binary data.
>>
>> Ok, sure.
> 
> On second thoughts, -1 on that. 'void *' is appropriate for functions 
> like libc's read() or pq_sendbytes(), where the buffer can point to 
> anything. In other words, the caller is expected to have a pointer like 
> 'foobar *', and it gets cast to 'void *' when you call the function. 
> That's not the case with the cancellation key. The cancellation key is 
> just an array of bytes, the caller is expected to pass an array of 
> bytes, not a struct.
> 
> The right precedent for that are e.g. SCRAM functions in scram-common.h, 
> for example. They use "const uint8 *" for the hashes.
> 
> I'll switch to "const uint *" everywhere that deals with cancel keys. 
> There are a few more variables elsewhere in the backend and in libpq.

I was having the same second thoughts overnight.  I agree with your 
conclusion.