Re: [Quantlib-dev] c++0x

[Peter C. <pca...@gm...>]

following up the topic below: I missed to look at the test-suite, where some more
auto_ptr would have to be replaced by unique_ptr.

https://github.com/pcaspers/quantlib/commit/466fbfbc423312eb7c4536ab59d861689cde70a0

Furthermore the header ql/utilities/clone.hpp has to be changed.

https://github.com/pcaspers/quantlib/commit/fe5a87dcfb60c10f1d8600b1511037e200915e4c
https://github.com/pcaspers/quantlib/commit/7a12fc4c62b44d175dc794b487db0582803ef6a5

The test-suite now compiles and runs using g++ (with and without -std=c++0x)
which is the main thing I guess. The whole thing does not look very
nice though. Also I am not c++ expert enough to say that the upgrade of
clone.hpp to unique_ptr should be done like this. Luigi ?

Thank you
 Peter

Peter Caspers <pca...@gm...> writes:

> Hi Luigi, all,
>
> we discussed compiling the core lib under the c++11 standard (e.g. with
> g++ -std=c++0x). This produces some errors which can be fixed easily
> (keeping backward compatibility):
>
> https://github.com/pcaspers/quantlib/commit/5b32b7705530264551e0622d0a2673813067be5b
>
> What is left is a bunch of warnings saying that std::auto_ptr is
> depricated under c++11, the replacement being std::unique_ptr. A
> possible solution is to replace the auto_ptr in the source by unique_ptr
> (thus upgrading the code to c++11 already) and replace the unique_ptr by
> auto_ptr again in case that compilation is done under c++ versions <
> 11. I have done this here
>
> https://github.com/pcaspers/quantlib/commit/3e47a82a936112f12e3bb502292833d860f5aac9
>
> with a #define in qldefines.hpp resetting the pointers to auto_ptr again
> based on the boost macro BOOST_NO_CXX_SMART_PTR. Unfortunately this
> macro is available only in later boost versions so if one uses an older
> one and provides no c++11 support he or she will get errors. Therefore
> the following solution might be better
>
> https://github.com/pcaspers/quantlib/commit/6db965375b02094a7ea0da4e48efd7e0a1b8fd85
>
> using the __cplusplus macro to identify c++11. However since gcc sets
> this macro simply to 1 in versions 4.6 (should be solved starting in
> 4.7), I had to add another criterion for gcc based on
> __GXX_EXPERIMENTAL_CXX0X_.
>
> For msvc 2010 the __cplusplus is also not 201103L but still 199711L. This
> seems also to be the case in 2012 and this specific case was reported as a bug to
> Microsoft,
>
> http://connect.microsoft.com/VisualStudio/feedback/details/763051/a-value-of-predefined-macro-cplusplus-is-still-199711l
>
> though not solved yet. Therefore I added a direct test on msvc 2010 or
> higher versions based on _MSC_VER 
>
> https://github.com/pcaspers/quantlib/commit/796beb89e2a707f951b77869bf869694e67ac769
>
> This final solution should work for all versions of msvc and gcc. I can not check other
> compilers however.
>
> Since the #define seems a bit like a dirty hack we also thought of a
> typedef. But then we would need a template typedef which is available only in
> c++11 again and a metaprogramming-like workaround a la
>
> template<class T> struct PTR {
>     typedef std::unique_ptr<T> Type;
> };
>
> PTR<double>::Type a(new double(0.0));
>
> does not seem to improve the code either.
>
> Do you think we should do the upgrade to c++11 like propsed above or is there
> maybe a better solution (I bet there is...). Or should we defer the
> upgrade until a later release ? I personally would very much like to be
> able to compile under c++11 without warnings already now.
>
> Thank you
>  Peter
>
>
>
>