[RadixSpline] memory leak and a suggested fix about sosd

[RyanMarcus]

That's a good idea -- I'll tag @andreaskipf and @alexandervanrenen here so they can close this and notify us when it's ready.

[RyanMarcus]

Nice catch, there's probably a ton of these throughout SOSD. One of the reasons we are missing them is because some of the competitors require special AVX instructions that break valgrind. Maybe someone should give this sanitizer a shot?

[RyanMarcus]

Ah, I think the current version stores radix_hint_ as a std::vector member, so that should get cleaned up automatically when the class is dropped. Can you give that a shot?

Hopefully Andreas & Alex will have a better RadixSpline implementation soon.

[jmftrindade]

Cool: yes, if radix_hint_ is no longer a uint32_t* array, that memleak should not be there anymore.

I see there's been a number of recent changes on the code structure. We'll wait for the "standalone" version before we change our setup to use git submodule for it instead of a local copy.

[RyanMarcus]

A standalone version of RS is available here: https://learned.systems/rs