Re: gcc7 warnings

[John W. Eaton]

On 04/28/2017 12:08 AM, Dmitri A. Sergatskov wrote:
> I have been compiling octave dev on Fedora-26 (alpha) that has gcc7 (and
> few other newer tools), and while it works just fine there are few extra
> warning which i do not see
> on the Fedora 25 (current release gcc 6.3.1):
>
> * Quite a few of type:
> <<<<
> ../liboctave/operators/mx-inlines.cc:582:1: warning: mangled name for
> ‘bool do_mx_check(const Array<T>&, bool (*)(size_t, const T*) throw ())
> [with T = octave_int<short unsigned int>]’ will change in C++17 because
> the exception specification is part of a function type [-Wnoexcept-type]
> > > > >
> those are, I assume, bening

Yes, the throw() should either be replaced by a noexcept attribute or 
just removed.

> * Few of -Wimplicit-fallthrough which look more serious
> (
> https://developers.redhat.com/blog/2017/03/10/wimplicit-fallthrough-in-gcc-7/
> ) E.g.:
>
> <<<<
> ../libinterp/octave-value/ov-classdef.cc: In member function ‘virtual
> octave_value_list cdef_object_array::subsref(const string&, const
> std::__cxx11::list<octave_value_list>&, int, size_t&, const cdef_class&,
> bool)’:
> ../libinterp/octave-value/ov-classdef.cc:1552:9: warning: this statement
> may fall through [-Wimplicit-fallthrough=]
>          }
>          ^
> ../libinterp/octave-value/ov-classdef.cc:1555:5: note: here
>      default:
>      ^~~~~~~

These should probably be checked to see whether we really intend for the 
code to fall through to the next switch case, and if so, tag them with 
the [[fallthrough]] attribute.  But since we don't assume C++17 yet, I 
guess this should be done with a macro and a configure test to see 
whether [[fallthrough]] is supported.

> * Few  of "-Wint-in-bool-context" like (also look bening to me):
>
> <<<<
> ../libinterp/octave-value/ov-typeinfo.cc: In static member function
> ‘static bool
> octave_value_typeinfo::register_assign_op(octave_value::assign_op, int,
> int, octave_value_typeinfo::assign_op_fcn)’:
> ../libinterp/octave-value/ov-typeinfo.cc:142:10: warning: ?: using
> integer constants in boolean context [-Wint-in-bool-context]
>    return (instance_ok ())
>           ~~~~~~~~~~~~~~~~
>           ? instance->do_register_assign_op (op, t_lhs, t_rhs, f) : -1;
>           ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> ../libinterp/octave-value/ov-typeinfo.cc: In static member function
> ‘static bool
> octave_value_typeinfo::register_assignany_op(octave_value::assign_op,
> int, octave_value_typeinfo::assignany_op_fcn)’:
> ../libinterp/octave-value/ov-typeinfo.cc:150:10: warning: ?: using
> integer constants in boolean context [-Wint-in-bool-context]
>    return (instance_ok ())
>           ~~~~~~~~~~~~~~~~
>           ? instance->do_register_assignany_op (op, t_lhs, f) : -1;
>           ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

I checked in the following changeset for these warnings:

  http://hg.savannah.gnu.org/hgweb/octave/rev/e4fca7cc471c

Are there more?

> If there is an interest, I can post all of them.

Yes, please.

> On related subject, if there is an interest, I should be able to update
> Fedora buildbot
> to F26 sooner rather than later.

OK.  Should we continue testing on the previous version?  If so, I could 
maybe set up a VM to do it.

jwe