From: Jonathan Wakely <jwakely.gcc@gmail.com>
To: Csaba Raduly <rcsaba@gmail.com>
Cc: "gcc@gcc.gnu.org" <gcc@gcc.gnu.org>
Subject: Re: Wdeprecated-copy on system type
Date: Thu, 28 Jun 2018 11:50:00 -0000 [thread overview]
Message-ID: <CAH6eHdRpWneDKO+jXQzM15+d4Ns1jOfK5yE2VDE_goU10B3jtg@mail.gmail.com> (raw)
In-Reply-To: <CAEhDDbBTd2gG3UQJfTr0jddV9Y-oUGnVV7dzVOV-ktcnavaiaQ@mail.gmail.com>
On Thu, 28 Jun 2018 at 12:32, Csaba Raduly wrote:
>
> On Wed, Jun 27, 2018 at 1:28 PM, Jonathan Wakely wrote:
> > Your mail is off-topic on this list, it would be appropriate on the
> > gcc-help list, or in Bugzilla if you want to report a bug.
> >
> >
> > On Wed, 27 Jun 2018 at 11:27, Csaba Raduly wrote:
> (snip)
> >> 1. Why is a warning emitted only for the const srange?
> >
> > Because line 16 doesn't use the sub_range copy constructor, it calls
> > this instead:
> >
> > template< class ForwardRange2 >
> > sub_range(
> > ForwardRange2& r,
> > BOOST_DEDUCED_TYPENAME ::boost::enable_if<
> > is_compatible_range<ForwardRange2>
> > >::type* = 0
> > )
> > : base(impl::adl_begin(r), impl::adl_end(r))
> > {
> > }
> >
> > This might be a Boost bug, albeit a harmless one. IMHO the template
> > constructor should be constrained so it isn't used for copying
> > non-const lvalues.
>
> So the fact that there's no warning for the non-const version is in
> fact a bug in Boost, right?
I think that's for the Boost maintainers to decide. The constructor
template still gives the right behaviour, even if it's probably not
what was intended (a sub_range<T>& is a compatible range of a
sub_range<T>, and so can be "converted" to it by that constructor). It
probably makes more sense to constrain that constructor so it's not
used for "conversions" from sub_range<T>& to sub_range<T>, and the
copy constructor gets used instead.
> (It's still like that in latest, not yet released Boost 1.68).
>
> This sounds like a "greedy template" you spoke about at ACCUcon 2018.
Yes, exactly.
There are other cases in Boost which cause bigger problems e.g.
https://wandbox.org/permlink/9Py2Nu8ni6qnSU7O which crashes because
trying to copy-construct a non-const lvalue will keep trying to wrap
itself recursively until it overflows the stack.
> >> 2. Shouldn't boost be exempt from this warning, given that it's
> >> included from a system directory (/usr/include) ?
> >
> > Yes, it probably should, but see below.
> >
> >> Shouldn't the "implicitly declared copy constructor" error be located
> >> at the class definition (boost header), rather than where it's being
> >> called (the call which does pass-by-value) ?
> >
> > The warning is only issued at the point where the implicit definition
> > is needed, to avoid issuing warnings when the object is never copied.
> > It's needed at the call site in your code, not in the system header
> > (well, it might also be needed in the boost headers, but those
> > warnings are probably suppressed).
>
> Of course the warning should only be emitted when such a
> constructor/assignment is used.
>
> Still, in my naive opinion, such a constructor is declared
> (implicitly) in the class (sub_range).
> The message at the point of use should be something similar to the
> "instantiated from" message when there's an error inside a template.
>
> Something like:
>
> /usr/include/boost/range/sub_range.hpp:???: implicitly-declared ‘constexpr
> boost::sub_range<const
> std::__cxx11::basic_string<char>>::sub_range(const
> boost::sub_range<const std::__cxx11::basic_string<char> >&)’ is
> deprecated
> [-Werror=deprecated-copy]
>
> /usr/include/boost/range/sub_range.hpp:259:20: note: because REASONS
>
> subra.cpp:7:6: note: used when initializing argument 1 of ‘void func(srange)’
Yes, please report it to our bugzilla.
> >
> > But I agree that if the class is defined in a system header you can't
> > change the code, so the warning is not very helpful. Please report a
> > bug, https://gcc.gnu.org/bugs/
next prev parent reply other threads:[~2018-06-28 11:49 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-06-27 11:28 Csaba Raduly
2018-06-27 15:53 ` Jonathan Wakely
2018-06-28 11:49 ` Csaba Raduly
2018-06-28 11:50 ` Jonathan Wakely [this message]
2018-06-28 12:31 ` Jonathan Wakely
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=CAH6eHdRpWneDKO+jXQzM15+d4Ns1jOfK5yE2VDE_goU10B3jtg@mail.gmail.com \
--to=jwakely.gcc@gmail.com \
--cc=gcc@gcc.gnu.org \
--cc=rcsaba@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).