Did this file ever worked ?

As it requires ur5.urdf, which not available

Closing the interpreter is bugged.

The idea to remove Py_Finalize comes from https://github.com/numpy/numpy/issues/8097
Those discussions also provide a way to try to clean a bit. Here are
some test results of the memory loss on the python_parser test:

- python3
    - with Py_Finalize

unknown location(0): fatal error in "buildModel": memory access violation at address: 0x000000a0: no mapping at fault address

    - with the workaround

>  HEAP SUMMARY:
>      in use at exit: 6,623,041 bytes in 11,478 blocks
>    total heap usage: 40,988 allocs, 29,510 frees, 35,627,810 bytes allocated
>
>  LEAK SUMMARY:
>     definitely lost: 162,918 bytes in 86 blocks
>     indirectly lost: 0 bytes in 0 blocks
>       possibly lost: 224,284 bytes in 192 blocks
>     still reachable: 6,235,839 bytes in 11,200 blocks
>          suppressed: 0 bytes in 0 blocks
>
>  ERROR SUMMARY: 1470 errors from 115 contexts (suppressed: 0 from 0)

    - with nothing

>  HEAP SUMMARY:
>      in use at exit: 6,652,431 bytes in 11,580 blocks
>    total heap usage: 40,937 allocs, 29,357 frees, 35,604,459 bytes allocated
>
>  LEAK SUMMARY:
>     definitely lost: 164,417 bytes in 87 blocks
>     indirectly lost: 0 bytes in 0 blocks
>       possibly lost: 223,187 bytes in 191 blocks
>     still reachable: 6,264,827 bytes in 11,302 blocks
>          suppressed: 0 bytes in 0 blocks
>
>  ERROR SUMMARY: 1438 errors from 107 contexts (suppressed: 0 from 0)

 - python 2
    - with Py_Finalize:

> HEAP SUMMARY:
>     in use at exit: 3,209,728 bytes in 5,514 blocks
>   total heap usage: 23,363 allocs, 17,849 frees, 27,441,910 bytes allocated
>
> LEAK SUMMARY:
>    definitely lost: 0 bytes in 0 blocks
>    indirectly lost: 0 bytes in 0 blocks
>      possibly lost: 161,830 bytes in 89 blocks
>    still reachable: 3,047,898 bytes in 5,425 blocks
>         suppressed: 0 bytes in 0 blocks
>
> ERROR SUMMARY: 1902 errors from 95 contexts (suppressed: 0 from 0)

    - with the workaround:

> HEAP SUMMARY:
>     in use at exit: 5,437,170 bytes in 7,454 blocks
>   total heap usage: 23,363 allocs, 15,909 frees, 27,441,795 bytes allocated
>
> LEAK SUMMARY:
>    definitely lost: 0 bytes in 0 blocks
>    indirectly lost: 0 bytes in 0 blocks
>      possibly lost: 177,798 bytes in 115 blocks
>    still reachable: 5,259,372 bytes in 7,339 blocks
>         suppressed: 0 bytes in 0 blocks
>
> ERROR SUMMARY: 1149 errors from 69 contexts (suppressed: 0 from 0)

    - with nothing:
>
> HEAP SUMMARY:
>     in use at exit: 5,472,144 bytes in 7,554 blocks
>   total heap usage: 23,361 allocs, 15,807 frees, 27,440,943 bytes allocated
>
> LEAK SUMMARY:
>    definitely lost: 0 bytes in 0 blocks
>    indirectly lost: 0 bytes in 0 blocks
>      possibly lost: 178,318 bytes in 116 blocks
>    still reachable: 5,293,826 bytes in 7,438 blocks
>         suppressed: 0 bytes in 0 blocks
>
> ERROR SUMMARY: 1150 errors from 67 contexts (suppressed: 0 from 0)

So I think we should keep Py_Finalize as long as it works (ie. for
Python 2), and use the workaround otherwise.

3 are not passing, they are commented for now

This commit allows the "python_parser" test to pass, even if pinocchio
is not installed, by compiling all python scripts into
CMAKE_BUILD_DIR/bindings/python/pinocchio, where the .so already is, and
add this directory to the PYTHONPATH for this test.

Compiling python files also allows to detect errors in them, instead of
blindly installing them on PYTHONPATH

We could discuss about installing those compiled versions, as it could
speed up the library use, as normal users usually can't write on the
system's PYTHONPATH, and therefore the interpreter can't
compile-and-install the source files on the fly, so it has to compile
them each and every time.

[bindings][loadReferenceConfig] fix bug when deprecated function is called

Fix python unit tests, fix #677

Fix center of mass derivatives debug compilation error

Create std map referenceConfigurations in model, and load from srdf

Remove redundant semicolons

submodules: use relative url

[bindings][python] expose referenceConfigurations and loadReferenceConfigurations. Make previous functions deprecated

[binding][std::map] Create Python visitors for std::map for aligned allocator. (and also for non-aligned maps, for the sake of completion)

Sync submodule and use https url

get the inverse of MJtJ matrix using the matrix decompositions performed in forwardDynamics or impulsedynamics (#665)

* [dynamics] use matrix decompositions done in forwarddynamics to return the inverse of MJtJ matrix. getMJtJInverse()

* [getMJtJInverse][unittest]

* [getMJtJInverse] python binding, unittest and impulse dynamics and name change to getKKTContactDynamicMatrixInverse

Algorithm to compute the derivatives of the COM velocity

* Added PINOCCHIO_ prefix to Pinocchio EIGEN_ macros

* Converted (most) SE3_ macros to PINOCCHIO_

* Removed assert.hpp

Support for catkinized hpp-fcl