Few more new features for LHCbMath

1. Extend operations with polynomials
2. Allow change/extend/shrink domain for Bernstein polynomials
3. Allow math-ooperations for (Bernstein) polynomials with different domains
4. Collect interpolation functions into separate header file
5. Add Neville interpolation algorithm
6. Add few more signatures for interpolation functions
7. Add optional range for splines/interpolations
8. Reduce code duplication
9. Add more protection for invalid covariances in math-operations for ValueWithError objects

const auto ny = std::distance(  ybegin, yend );

is much less verbose, and thus easier to read (at least for me ;-) (and the lost information isn't relevant IMHO)

I suspect you want:

[] ( const auto& i ) { return i.first; }

to avoid copying the pair. And note that this returns i.first by value, not reference...

for the zeroth there were many auto in the code, and C++ part compiles nicely. But when I've tried to build the dictionaries for the derived (non-templated) functions, I've got problems form included headers with lambdas and autos.. It loosk like dictionary-buiding machinery puts some constraints.
I've replaced all autos with the exact types and everything is fine now. Probably is would be enough to replace only part of autos... but since it works, it does nto matter too much...

@graven This is exactly that reflex(?) does not like :-)

@graven Sorry - I mean reflex does not like when such construction appears in header files.. For this *.cpp file it is fine. I've reshuffles lines several times between *.h and * .cpp, cleaning *.h from stuff that Refles does not like..

given the repeat occurrences, why not add

namespace {
      template <typename F, typename S> 
      constexpr struct select1st_t {
             F& operator()(std::pair<F,S>& p) const { return p.first; }
             const F& operator()( const std::pair<F,S>& p ) const { return p.first; }
             F& operator() (std::pair<F,S>&& )  const = delete;  // avoid returning a dangling reference
     } select1st {};
}

to the top of the file, and then use select1st instead of repeating the lambda ...

counting number of lines and occurances of this cconstruction , I see no obvious gain :-) It is almost the same.

If you also count all other occurrences in other bits of code throughout LHCb, I'm sure there is a gain if we put select1st and select2nd in a common header file (and the presence 3rd operator() = delete has the advantage of refusing to compile in case of a silly mistake -- could even make it do the right thing by returning a value instead of a reference in that case ;-)

please note that std::unary_function has been deprecated in C++11, and will be removed in C++17... (see eg. http://en.cppreference.com/w/cpp/utility/functional/unary_function)

@graven Thank you.

Re reflex: I thought that only I/O still needed dictionaries, and that for the rest one could just get cling (i.e. clang) to parse the header... hence we should be able to considerably reduce the entries in dictionaries....

@graven I think the dictionaries are also needed for the python bindings?

Status changed to merged

mentioned in commit 366e0980

@cattanem : I was under the impression the python bindings relied entirely on cling, and thus that it is sufficient to get cling to parse the headers. But @clemenci surely knows the real answer!

@graven is correct: dictionaries are needed only for I/O. See GAUDI-1098.

mentioned in merge request !174 (merged)

mentioned in commit 96cfa093

Mentioned in commit 366e0980

Mentioned in commit 96cfa093