GitLab

And make it slightly stronger (it now checks something that should be
mostly obvious too, but who knows, with software).

They don't need the mutable versions to work, but they used them anyway.

the tests now depend on libdatasrc, while libdatasrc needs memory.la.
so we need to build any tests after building all libraries.

This will be required for finding whether a hash is less than,
or greater than all nodes in the origin's subtree in an NSEC3 tree.

Note that if a previous find() results in a PARTIALMATCH, the next
find() (when passed a populated node chain and a non-absolute label
sequence) starts at the down-node of the partial match's top of chain
(deepest matched node).  The caller must pass the correctly formed
non-absolute label sequence for that sub-tree-of-tree.

we need to it for TreeNodeRRset::toWire() (actually this method was
introduced exactly for that purpose).

added some test cases to confirm the behavior; update the doc.

...instead of give its type as a class template parameter and instantiate
it within the tree.  it won't work (well) for ZoneData, because we need
to know the RR class to delete RdataSet (which is the data type for
ZoneData tree), but there's no way to tell the RR class about it.
A good side effect of this change is that we now don't have to link
the memory segment for the tree and its nodes with the zone data
(although in practice it's quite likely to use the same segment for
all of them).

I also updated setData() (further from #2100); it now doesn't even delete
any existing data.  This is partly relaetd to the above change on
deleter - we cannot instantiate the deleter within this method.  We could
pass the deleter as a parameter, but, actually, in the case of ZoneData
we wouldn't have to delete the data; when setData() encounters existing
data, it's more likely that ZoneData tries to extend the RdataSet list.
It makes more sense to let ZoneData manage all data, including when to
delete them.

(although not absolutely necessary) I also changed the type of deleter
in the domain tree test to highlight we could use a plain old function
instead of a functor object.

change size is big, but most of the changes are trivial conversion
from something like <T, DT> to <T>.

Some compilers seem to expect it to be present.

It was only used in the memory datasrc, which has been modified to use
the LabelSequence variant.

Also unify some tests.

The Name find() variants already test the implementation well as they
wrap around it, but one test was added for checking the LabelSequence
variant API.

Also change the dot dumper to use the offset_ptr getters.

It should be more accurate notation.  Also, we should eventually get rid of
getName() (we can't right now as it's used in node path).  This change is
one step toward that goal.

Also made some editorial/docuementation cleanups.

i.e., instead of holding them as Name objects, hold them as serialized
opaque data generated by the corresponding LabelSequence, following the
node structure.  the size of the storage for the labels data is encoded
as part of node "flags".

insert/find now internally uses LabelSequence, and compare() can now return
NONE.  It requires some adjustments to other part of the code and tests, but
they should be generally straightforward.

in tests, note especially that we now have a separeate "." level.  that's
the main reason for the big change to dumpTree().