Thomas Haller

> Unit tests are not a problem. As for the equality function, do you know if the current one is not enough? https://github.com/thom311/libnl/blob/main/lib/route/route_obj.c#L395 That's just the compare function. It matters...

the caching of routes clearly has issues. But they go beyond this patch. the major problem is that `route_obj_ops.oo_id_attrs` contains ``` .oo_id_attrs = (ROUTE_ATTR_FAMILY | ROUTE_ATTR_TOS | ROUTE_ATTR_TABLE | ROUTE_ATTR_DST...

As you found out, the objects and data structures of libnl are not thread-safe. What is supposed to work, is that you (for example) create two caches on two threads,...

> Add installing NetworkManager-config-routing-rules package when NM selected as network_provider. The package is renamed to `NetworkManager-dispatcher-routing-rules`. > As I googled this package is mandatory for routing-rules support in NM this...

I find it very ugly to have udev read's NetworkManager configuration (be it keyfiles or ifcfg files). Which is also what initscripts' `/usr/lib/udev/rename_device` does. I wish, .link files or udev...

previously, there was a patch here which dropped the chunking from `nftables.set_restore()`. That doesn't work (yet?). Patch was dropped (for now?).

> previously, there was a patch here which dropped the chunking from `nftables.set_restore()`. That doesn't work (yet?). Patch was dropped (for now?). With patch `improvement(nftables): combine request for entries of...

> Do you have any benchmarks about the improvements from del and clear() commands? No. You brought up how the ipset transaction has to be split due to memory (peak)...

Hi @erig0 `nftables` does: ``` def _set_get_family(self, name): ipset = self._fw.ipset.get_ipset(name) ``` where `get_ipset()` does ``` def get_ipset(self, name, applied=False): self.check_ipset(name) obj = self._ipsets[name] if applied: self.check_applied_obj(obj) return obj ```...

the configuration script is mangled. Would it be possible to fix it?