Aurelien Bouteiller

Some good unused ideas here: splitting `parsec_init` from `parsec_context_init`

I have slated this for 4.0 but we are getting really close to our self-imposed deadline. Given that the CI fails randomly without this, I'd like to see it in.

> Moreover, we do have another solution: add the MCA `device_cuda_memory_use` to the tests (specialized by the accelerator). lets see if that works #629

~~As shown in #629 there is something more nefarious going on, we try to allocate way more memory that is possibly available whatever limits we set.~~ As seen in resolution...

For v4 I will reduce the scope of this to only warning about parameters that we used to have but have been removed. Capturing arbitrary wrong params would be good...

this is somewhat rare but can be reproduced for sure with `ctest --repeat until-fail:100` Another variant ``` dtd_test_task_insertion: /home/bouteill/parsec/dplasma/parsec/parsec/mca/termdet/local/termdet_local_module.c:106: parsec_termdet_local_taskpool_state: Assertion `0' failed. ```

I have it on master

Now PR #323

**Original comment by Thomas Herault (Bitbucket: [herault](https://bitbucket.org/herault), GitHub: [therault](https://github.com/therault)).** ---------------------------------------- PR #433 is supposed to look at this issue. It has been taken over in PR #442 ‌

**Original comment by Thomas Herault (Bitbucket: [herault](https://bitbucket.org/herault), GitHub: [therault](https://github.com/therault)).** ---------------------------------------- We will re-enable the empty set feature for local indices by generating the same kind of functions as for the...