Solver manager

Because of #142 (closed) I can't write a proper markdown-based description thus I upload a screenshort ...

• remove former StateBlock notification list
• StateBlock::getState returns by copy
• remove StateBlock::getPtr

marked as a Work In Progress

added enhancement label

assigned to @jsola

@jsola @joanvallve This is wip and I'm looking forward to your feedback !

Some tests are crashing, other failing, but you should get the intend here.

changed the description

Since the solver (Ceres) is holding a raw pointer to the StateBlock data vector, there are basically no way of ensuring there won't be any data-race. One could add a mutex to each and every StateBlock and lock it the time the solver is executing but that would prevent any other access to the StateBlock during that time.

added 1 commit

• 1433cc69 - fix StateBlock notification mechanism

Compare with previous version

Tests are fixed in the sense that that are all running, but 3 of them are failing. They also fail on master branch.
Looks like it is precision related, my machine is a mess \crying.
@jsola can you please try to run the tests on master then on this branch and let me know if there are differences please.

added 1 commit

• d1d80ce5 - former solver_manager messing arround

Compare with previous version

@artivis make_unique.h was a soft-link, fix dat.

added 1 commit

• 0ba48005 - former solver_manager messing arround

Compare with previous version

added 11 commits

• 55ef83e9 - add make_unique.h
• 9752a540 - use unique_ptr instead of raw ptr
• 589ff6ee - embed notification within StateBlock class
• 57927e1d - StateBlock func constness
• 5267ac22 - update StateBlock notificationList
• 49dde8c4 - add solver_manager
• 4af7703f - ceres_manager inherits from new solver_manager
• fb00557a - fix tests -> change solver_manager
• 3cd5d50a - fix StateBlock notification mechanism
• a7fe6ac8 - former solver_manager messing arround
• 9ab1f53b - ceres_manager do not have to call update

Compare with previous version

changed the description

added 44 commits

• 9ab1f53b...9d5dafde - 27 commits from branch master
• a4d5d353 - mv solver -> solver_suitesparse
• 6382cffa - ceres_manager indentation
• 176da6e7 - add make_unique.h
• 27ccb967 - use unique_ptr instead of raw ptr
• 57ac7c40 - embed notification within StateBlock class
• 0951653c - StateBlock func constness
• 2a8100c5 - update StateBlock notificationList
• 546200e4 - add solver_manager
• bc67170c - ceres_manager inherits from new solver_manager
• 6769b925 - fix tests -> change solver_manager
• 584996ab - fix StateBlock notification mechanism
• 3f885435 - former solver_manager messing arround
• ab69e89e - ceres_manager do not have to call update
• 2c9ea7de - rename Notification::UPDATE -> Notification::STATE_UPDATE
• 928494d5 - removed formed stateblock notif list
• 7553ba1b - add mutex to ProcessorBase and use it in ProcessorMotion::keyFrameCallback
• a08ccfc2 - ProcessorMotion::keyFrameCallback for time tolerance

Compare with previous version

added 4 commits

• b9ed5e06 - add mutex to ProcessorBase and use it in ProcessorMotion::keyFrameCallback
• bb99c3c9 - ProcessorMotion::keyFrameCallback for time tolerance
• 599aa099 - StateBlock getState return by copy
• 30c82b25 - rm StateBlock::getPtr

Compare with previous version

changed the description

Jeremie: several things, in order of importance.

1. I could not compile your branch as-is. In particular, I cannot compile gtest_constraint_autodiff.cpp. Issues with getPtr() indicate me that your CMakeFiles or installation configuration is not even trying to compile it in your machine. Please have a look -- you probably want to discuss it with jvallve: maybe he just convinces you to leave getPtr() alive...
2. I could not merge to master without quite a few conflicts. This is normal since master has evolved since your MR -- sorry for the delay, I do not catch up.

Please address issue 1, we will see after what to do with 2.

PS: I include error log in snippet below.