Re: EPICS 7.0.1-pre1 (pdbsingle.cpp)

[Dirk Zimoch <dirk.zimoch@psi.ch>]

I could build EPICS 7 for vxWorks 6.6, vxWorks 6.7 and ELDK 4.2 after 
this fix. I checked the resulting libraries and executables for any 
occurrence of (undefined) FieldName(const FieldName&) symbols. There are 
none. So it seems the compiler wants to have the definition but never 
actually uses it.
On 23.11.2017 14:21, Dirk Zimoch wrote:
> This is my fix in pvif.h line 290:
>
> #if defined(__GNUC__)&&__GNUC__*100+__GNUC_MINOR__<403
> // workaround for gcc bug https://gcc.gnu.org/bugs/#cxx_rvalbind
> public:
> #else
> private:
> #endif
>     FieldName(const FieldName&);
>     FieldName& operator=(const FieldName&);
>
>
> On 23.11.2017 13:56, Dirk Zimoch wrote:
> > https://gcc.gnu.org/bugs/#cxx_rvalbind
> >
> >
> > On 23.11.2017 12:03, Dirk Zimoch wrote:
> > > On 23.11.2017 11:19, Dirk Zimoch wrote:
> > > > Clang 3.4.2 gives an interesting hint with its warnings. It seems that
> > > > this call creates a temporary, for which of course a copy is needed. If
> > > > you wanted to detect accidental copies, this is such a case.
> > > I have analyzed the code a bit.
> > >
> > > The message was:
> > > "C++98 requires an accessible copy constructor for class 'FieldName'
> > > when binding a reference to a temporary"
> > >
> > > The line in question is:
> > > pvif(channel->pv->builder->attach(channel->pv->chan, pvf, FieldName()))
> > >
> > > This is the constructor call of pvif in the constructor of PDBSinglePut.
> > > It is true, that a temporary is used here. FieldName() creates an object
> > > that falls out of scope immediately after the attach() call.
> > >
> > > The compiler cannot know what builder->attach() will do with with the
> > > temporary. It has to expect the worst, which is to keep a reference to
> > > the temporary FieldName object after the attach() call has returned.
> > > Since this would then reference an invalid object.
> > >
> > > So maybe the compiler rather implicitly creates a copy of which
> > > builder->attach can keep a reference, like this:   attach(new
> > > FieldName(FieldName()))
> > >
> > > I need to check the C++ standard to see what the documented behavior is
> > > in such a case.