Shouldn't hwloc_topology_get_default_nodeset take an hwloc_const_topology_t?

[HadrienG2]

The hwloc_topology_get_default_nodeset() function that was introduced in hwloc v2.12.0 takes an hwloc_topology_t, which allows it to modify the topology. But to the best of my knowledge, it has no reason to do so, and indeed if I read the implementation code right it currently doesn't. Shouldn't it take an hwloc_const_topology_t instead?

[bgoglin]

Hello. In practice yes. In theory, we could have some sort of internal cache that would be updated inside function when needed before returning the nodeset. Not sure this will ever happen.
There are lots of get() function that would deserve a const too, for instance hwloc_topology_get_topology_cpuset() but it looks we don't even define hwloc_const_topology_t yet?

[HadrienG2]

Oh my, it looks like I just made up this typedef on the Rust bindings side long ago (or maybe one of the abandoned predecessor projects that hwlocality took over like hwloc-rs did) and then I forgot about it since I had it around for so long ^^'

[HadrienG2]

I don't remember if mutating from const T* is UB in C these days (I have hazy memories of it being considered bad taste but legal at the language level in old standard revisions). If it is UB, the lazy initialization that hwloc_topology_refresh() tries to make eager is probably going to make it harder to use a prospective hwloc_const_topology_t typedef in places where it logically should be used.