libknot refactoring

RRSet refactoring

• Add a possibility to use custom allocators ⌛
• Rewrite unittests
• Refactor functions
  • The free functions need a bit of work (i.e. get rid of the 1,0 parameters)
• Remove, in this order:
  • rrset->rrsigs - This bloats the code big time, and is a source of endless stream of bugs. Hopefully, performance drop will be negligible. Amount of work to do this: a lot.
  • Do not store rrset->owner in knot_rrset_t structure, or create a new structure knot_node_rr_t without the owner field.
    • I'd like to see node = (owner, meta, rdataset_t) and rrset = (owner, rdataset_t) we can use rrsets when we parse packets and then take only the rdataset_t part if you want to keep libknot tidy by using rrset_t as it is now. @mvavrusa
  • Get rid of RDATA helper fields, store RDATA indices into rrset->rdata field, end with '\0' - this gets rid of rrset->rdata_count, since we usually (but not always, e.g. when deleting particular RR) want to access all RRs in RRSet.
• move rrset-dump.c from libknot to knot - it is used only by the server @jkadlec : I think this is a neat functionality to have in the library, if nothing else, our future resolver will use it (if we're talking about actual text dump, if we're talking about a debug function, that one can be deleted without looking back twice) rrset-dump is used in knot utils!
• Move rrset->additionals out of libknot (be it node or the refactored structure) The field is still there, but the actual information is stored elsewhere, the field is only used for passing the information. Should still drop if we want clean libknot.

Node refactoring

• Add a possibility to use custom allocators
• Rewrite unittests
• Remove:
  • node->wildcard_child, store it as a flag. @lslovak: I don't really get how this could be stored as a flag. The reference to the wildcard child is used in several places. Or this should be replaced by searching in zone for * ancestor of the node? Then even the flag is not necessary. Will drop the reference. The flag will mark a node whose wildcard child exists, thus should be searched for. This reduces the amount of searches to minimum (only in case there actually is a wildcard child). That's reasonable. (see issue #216 (closed))
  • node->rrset_count - node->rrset_tree will be ended by NULL pointer, iterations will need to change. @lslovak OK, sounds reasonable. It seems that only values which are ever checked are 0, 1 and > 0, and all would be easily checked also after this change. Due to alignment issues this will not reduce memory footprint.
    • On the other hand, if the rrset_tree is ordered by type and we know the count, we could search it using binary search if that is of any help. @lslovak: IMHO not very useful, there aren't many types of RRSets in each node.
  • node->parent + node->children_count - this is questionable, but it might be possible.
    • I wouldn't bother with this so much, you can use refcounting for this purpose and a refcount > 1 => has children @lslovak: I don't get it. References may be anywhere in the code, so > 1 doesn't necessarily mean that besides reference from zone tree there is also a reference from child. I don't see an easy way to get rid of these two fields.

Zone refactoring

• Remove zone->contents->zone->contents ... 791d0da8 @jkadlec: 💌
• Following stuff only valid if we remove zones from libknot: ✅
  • flags field not needed in zone contents. Some relicts from XFR are stored there, and ANY on/off - can be extracted from config. a067337d
  • Remove zone->name, stored in config. Remove zone->data and destructor, zonedata_t can be a direct part of the structure. !158 (merged)
• Zone lookup provides too much data and too many functions
  • Zone should have NO concept of encloser or previous authoritative, as they are needed only for denial of existence proofs
  • Zone should find an exact match or canonical/lexicographical previous node (this is what every answer needs)
  • Proof function should look up encloser or previous authoritative, should it be needed (this is what some answers need)
  • Merge zone_t and contents_t - but only after new zone API is done.