Re: [boost] Constructing string_ref from rvalue string

Subject: Re: [boost] Constructing string_ref from rvalue string
From: Olaf van der Spek (ml_at_[hidden])
Date: 2017-03-24 08:10:07

On Fri, Mar 24, 2017 at 4:39 AM, Marshall Clow via Boost
<boost_at_[hidden]> wrote:
> On Fri, Mar 17, 2017 at 12:29 AM, Simon Sasburg via Boost <
> boost_at_[hidden]> wrote:
>
>> I've just tested our codebase with the 1.64.0 master snapshot, and i've
>> encountered an issue which i've tracked down to this commit:
>>
>> https://github.com/boostorg/utility/commit/9960d9f395b79ee860e39064cd4696
>> 1f76d2cb55
>>
>> Basically this removes the constructor which allows constructing a
>> boost::string_ref from an rvalue std::string. I am wondering if this is
>> really that dangerous. In our codebase we almost exlusively use
>> boost::string_ref as function parameters, with the understanding that
>> passed string_refs parameters are only valid within the function body. (so
>> you need to copy it into an std::string if you need to access it
>> afterwards)
>> With these constraints i think it is OK to construct a string_ref from an
>> rvalue std::string because the temporary should live long enough. (but
>> please correct me if i'm wrong here)
>>
>> Is the safety (?) added by this commit worth it if it breaks working code?
>>
>>
> I'll also point out that that commit was in response to this bug report:
>     https://svn.boost.org/trac/boost/ticket/11740

The regression breaks perfectly valid code. Could it be reverted?

-- 
Olaf