DigitalSetByAssociativeContainer

[dcoeurjo]

Introduce a new model of digital sets using any associative container of the STL.
Basically, std::unordered_set (c++11) or boost::unordered_set (hash function based containers) can be used to store points of a digital set. Hence, just using this model of digital set speeds up the processing up to 40% (see benchmark results).

This PR also specializes std::hash and boost::hash with hash functions on DGtal points/vectors.

Once accepted, we could also remove DigitalSetBySTLSet and derive from DigitalSetByAssociativeContainer on specialized std::set container type.
Furthermore, I would suggest to change the default digital set type in StdDefs.h to a hash-map based container to get huge speed-ups in many DGtal and DGtaltools processes.

[dcoeurjo]

Can someone have a look to this PR ?

[JacquesOlivierLachaud]

I review it.

[JacquesOlivierLachaud]

"digital set points" => "digital sets"

[JacquesOlivierLachaud]

Both compile errors are related to the concept check:
BOOST_STATIC_ASSERT(( boost::is_same<Iterator,ConstIterator>::value ));
This concept check is indeed valid for sorted associative containers, but I wonder why we require it.

[JacquesOlivierLachaud]

Otherwise there are unrelated errors in Board3D. Was there a problem in the last PR merged ?

[rolanddenis]

The error in testBoard3D.cpp is due to the insertion of a point outside the domain, line 67 (p6 point). As DigitalSetByAssociativeContainer is now the default DigitalSet for Z2i and Z3i, this class now checks that inserted point are inside the domain, unlike DigitalSetBySTLSet.

[dcoeurjo]

Ok.. I've switched the default DigitalSet as suggested and got plenty of issues which I'm fixing.
So the PR is still a WIP.

BTW, many related issues has been raised because some DigitalSet adapters require more things than expected in the concept. For instance, ImageContainerMap< DigitalSetDomain < DigitalSet > > requires that the set has a reverse iterator which is not the case for DigitalSet models (as described in the concept).

So yes i'm working on that and the PR is more bogus than it was.

[dcoeurjo]

@JacquesOlivierLachaud for your issues, was C++11 activated ?

[dcoeurjo]

(everything compiled perfectly on travis bot --ubuntu-gcc / macos-clang--)

[dcoeurjo]

(when you have backtrace like that, please create a gist and add the link, otherwise the markdown interpretation pollutes everything)

[JacquesOlivierLachaud]

Yes cpp11 was activated.

[rolanddenis]

I have the same compilation issues with clang 3.5.0 and C++11 enabled, see https://gist.github.com/rolanddenis/18310385c470a3ac2827

[dcoeurjo]

okay... plenty of concept checks are missing or commented in DGtal... bunch of errors are raised changing the default DigitalSet model. Still working on it

[dcoeurjo]

seems to be ok... @JacquesOlivierLachaud & @rolanddenis can you please check your compilers ?

(FYI, iterator == const_iterator on associative containers is compiler version dependent.. 😢 )

[rolanddenis]

Everything is fine with my compilers ;)

[dcoeurjo]

cool.. waiting for @JacquesOlivierLachaud 's go and I merge

[JacquesOlivierLachaud]

It works now on my computer both with clang and g++, both in DEBUG and Release. Good PR since manything will get faster. I merge.