Indexing via discrimination tree

[FissoreD]

No description provided.

[gares]

Looks much better than this morning. Do the tests pass if we always use the DT index?

[FissoreD]

Most tests pass, but not all yet. I should look deeper to spot the bug.
There are some slow down in performance especially for tests like crypt.
Even if I think this particular test is not a good use case for indexing with trie, at least in the current implementation
where we do not limit the height/depth of the path representing the goal, I suppose
lot of time is spent during the creation of the path corresponding to the goal.
There are 5 timeout:

TIMEOUT  lyp (type checker for λΥP)                elpi
TIMEOUT  map (defined list) (stdlib map reference)   elpi
TIMEOUT  map (rbtree) (stdlib map)                   elpi
TIMEOUT  mu (standard Prolog benchmark)              elpi
TIMEOUT  set (stdlib set)                            elpi

And three failures:

KO       eta_as (eta expansion of as clause)         elpi
KO       trace-browser-chr (trace generation)        elpi
KO       uvar_keyword (uvar kwd status at the meta level) elpi

[gares]

I think you should use the timestamp as a unique id, here you are comparing the clause "code" which can be large.

[FissoreD]

I totally agree