On Mon, 15 Jan 2018, Jonathan Wakely wrote:
On 15 January 2018 at 02:09, <TPCgcc@xxxxxxxxxxxxxxxxxxx> wrote:
Dear All,
I am attempting to build this application
http://prdownloads.sourceforge.net/nget/nget-0.27.1.tar.gz?download with gcc
7.2.0. It built faultlessly under GCC 4.2.0 but it fails with these
(sample) errors under 7.2.0.
g++ -g -O2 -Wall -MMD -MP -DHAVE_CONFIG_H -std=c++1z -Wwrite-strings
-Wno-deprecated -c -o etree.o etree.cc
etree.cc: In instantiation of 'pred<ClassType>* comparison(const string&,
getterT, T2) [with ClassType = const c_nntp_file; getterT = long unsigned
int (c_nntp_file::*)() const; T2 = long unsigned int; std::string =
std::basic_string<char>]':
etree.cc:200:81: required from here
etree.cc:87:73: error: no matching function for call to
'new_comparison<template<class T, class T2> struct Op_eq, const
c_nntp_file>(long unsigned int (c_nntp_file::*&)() const, long unsigned
int&)'
if (opstr.compare("==")==0) return
new_comparison<Op_eq,ClassType>(get, v);
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~
etree.cc:78:18: note: candidate: template<template<class A, class B> class
Op, class ClassType, class RetType> pred<ClassType>* new_comparison(RetType
ClassType::*, RetType)
pred<ClassType> *new_comparison(RetType (ClassType::*member), RetType v){
^~~~~~~~~~~~~~
etree.cc:78:18: note: template argument deduction/substitution failed:
etree.cc:87:73: note: deduced conflicting types for parameter 'RetType'
('long unsigned int() const' and 'long unsigned int')
if (opstr.compare("==")==0) return
new_comparison<Op_eq,ClassType>(get, v);
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~
etree.cc:82:18: note: candidate: template<template<class A, class B> class
Op, class ClassType, class RetType> pred<ClassType>* new_comparison(RetType
(ClassType::*)(), RetType)
pred<ClassType> *new_comparison(RetType (ClassType::*memberf)(void),
RetType v){
^~~~~~~~~~~~~~
etree.cc:82:18: note: template argument deduction/substitution failed:
etree.cc:87:73: note: types 'RetType (c_nntp_file::)()' and 'long unsigned
int (c_nntp_file::)() const' have incompatible cv-qualifiers
if (opstr.compare("==")==0) return
new_comparison<Op_eq,ClassType>(get, v);
etc.
I am not a C++ programmer but AIUI it is failing to find the specific
function to execute in the template overload because of disagreements over
typing (function return value vs a scalar, const vs non-cost etc.) and for
that reason what would otherwise be a warning becomes a fatal error. Is
this correct?
I was hoping one of the -std switches would revert the back to the GCC 4.2.0
behaviour and allow me to do the build but none of the -std switches I tried
made any difference.
The etree.{h,cc} files and full (colourised) error report is at
http://www.mklab.rhul.ac.uk/~tom/tmpetree.zip.
Any thoughts?
The code is not valid C++. Presumably GCC 4.2 failed to diagnose it,
but that's not something you can change with -std switches. It's
simply invalid in all versions of C++.
The code tries to call new_comparison with a member function pointer
of type 'long unsigned int (c_nntp_file::*&)() const' (which is a
pointer to a const member function).
There are two overloads of new_comparison, one takes a pointer to
data-member, and one takes a pointer to non-const member function.
That means neither overload is viable.
You should be able to fix the code by adding this to etree.cpp
template <template <class A, class B> class Op, class ClassType, class RetType>
pred<ClassType> *new_comparison(RetType (ClassType::*memberf)(void)
const, RetType v){
return new Comparison<Op, MemfuncGetter, RetType, ClassType>(memberf, v);
}
This adds a third overload that takes a pointer to const member function.
You might be able to simply add "const" to the existing second
overload, which would let line 87 compile, but it would fail if there
are other calls that really do need to use non-const member functions.
Adding a third overload should work, whether or not the second one is
really needed.
Jonathan,
Many thanks for the explantion. Applying your extra overload did
indeed fix those errors. Predictably a bunch of other template related
errors also showed up which I'm unsure how to deal with. I have updated
the above zipfile with the new errors (tmp2c.log) if you are willing to
take a look. I am also CCing the nget developer but am unsure if he is
still maintaining it.
Best regards
Tom Crane.
--
Tom Crane, Dept. Physics, Royal Holloway, University of London, Egham Hill,
Egham, Surrey, TW20 0EX, England.
Email: T.Crane@xxxxxxxxxx
