Path: ...!2.eu.feeder.erje.net!feeder.erje.net!news.mb-net.net!open-news-network.org!.POSTED!not-for-mail
From: Marcel Mueller <news.5.maazl@spamgourmet.org>
Newsgroups: comp.lang.c++
Subject: Re: Why std::vector requires noexcept move constructor?
Date: Mon, 16 Jun 2025 20:50:28 +0200
Organization: MB-NET.NET for Open-News-Network e.V.
Message-ID: <102pp1k$8g7$4@gwaiyur.mb-net.net>
References: <102m01r$3knoc$1@gwaiyur.mb-net.net>
 <102pij2$1op2f$1@dont-email.me>
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Injection-Date: Mon, 16 Jun 2025 18:50:28 -0000 (UTC)
Injection-Info: gwaiyur.mb-net.net;
	logging-data="8711"; mail-complaints-to="abuse@open-news-network.org"
User-Agent: Mozilla Thunderbird
Cancel-Lock: sha1:pQ4Fivypof7EKboBMWr2z/ss18c= sha256:U+yGwQN/S/vUWxXg2LQMf8Ddxr5zx9zWKFDP0UkoPF0=
	sha1:FHhoqydxoSSQLCFyeLIuMbjMGB8= sha256:sMknnjhNeaWp1LQhU9dq865lQ4EgiI3mNgvi/DhMHvk=
Content-Language: en-US
In-Reply-To: <102pij2$1op2f$1@dont-email.me>
Bytes: 2458
Lines: 39

Am 16.06.25 um 19:00 schrieb Paavo Helde:
> On 15.06.2025 11:25, Marcel Mueller wrote:
>> I discovered shortly that a move constructor that is not declared as 
>> noexcept is ignored by std::vector. Instead the objects are moved by 
>> the copy constructor.
> 
> A move constructor is a way to squeeze out max performance from the 
> code.

.... or to return non-copyable objects etc.

> Noexcept also helps in this regard and is usually trivial to add 
> to move operations.

Indeed. But if only one member has a move constructor w/o noexcept any 
auto generated move constructor silently misses noexcept too. I shortly 
run into this pitfall, and this caused excessive copying of objects at 
insert operations (with reallocation).


>> Why?
>> The copy constructor is even more likely not declared as noexcept.
> 
> Depending on the operation, it might be needed to restore the original 
> state of existing data if an exception appears middle-way.

Ah, I wasn't aware that the operation is logically atomic.
This is of course impossible with exceptions in between.

> With copy constructors the original state is not touched at all so 
> recovering from an exception is much simpler.

Indeed.

Maybe a move constructor w/o noexcept should be a warning, because it 
makes no much sense to throw within a move operation.


Marcel