Incompatible word-length error

[wouterbeek]

After the first merge, RocksDB gives the following error upon enumeration:

PL_recorded_external(): Incompatible word-length (64)

Here is a reproducible case (reproducible on the LOD machine):

:- use_module(library(filesex)).
:- use_module(library(ordsets)).
:- use_module(library(rocksdb)).

test :-
  absolute_file_name(tmp, Dir, [access(write),file_type(directory)]),
  delete_directory_contents(Dir),
  setup_call_cleanup(
    rocks_open(Dir, Db, [key(atom),merge(merge0),value(term)]),
    (
      rocks_merge(Db, x, [x]),
      forall(
	rocks_enum(Db, A, B),
	writeln(A-B)
      )
    ),
    rocks_close(Db)
  ).

merge0(partial, _, A, B, C) :- ord_union(A, B, C).

[JanWielemaker]

You have no merge rule for full. If we add (as the example states):

merge0(full, Key, Initial, Additions, Result) :-
  append([Initial|Additions], List),
  sort(List, Result).

All seems to work fine. Unfortunately (AFAIK), the merger cannot forward errors to RocksDB.

[wouterbeek]

@JanWielemaker Is it always required to define both? IIUC rocks_merge(Db, x, [x]) should only use the partial variant?

[JanWielemaker]

On 06/14/2017 10:16 AM, Wouter Beek wrote: @JanWielemaker <https://github.com/janwielemaker> Is it always required to define both? IIUC |rocks_merge(Db, x, [x])| should only use the |partial| variant?

AFAIK, yes. A simple writes says it calls the `full` one. I forgot the rationale behind this in RocksDB.

…

--- Jan

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#3 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AC7cqrhqABfDuczW2b0RymP4GgKG9mVLks5sD5bggaJpZM4N4tS2>.

[wouterbeek]

Thanks! That clears things up for me. I'll update the docs to reflect RocksDB's requirement.