Query Bug: False positive results

[panterarocks49]

The following query produces results when I would expect it to produce none. (I'm on 0.9.20)

(require '[datalevin.core :as dl])

(def conn (dl/create-conn "/tmp/dl/test" {:block/uid  {:db/unique :db.unique/identity}
                                           :node/title {:db/unique :db.unique/identity}
                                           :block/refs {:db/valueType   :db.type/ref
                                                        :db/cardinality :db.cardinality/many}}))

(dl/transact! conn [{:db/id     -1
                     :block/uid "test"}
                    {:db/id      -2
                     :block/refs -1}])

(dl/q
 '[:find ?SBlock
   :where
   [?ParentPage :block/uid "test"]
   [?SPage :node/title "Opposed By"]
   [?ParentPage :block/uid ?some-uid]
   [?SBlock :block/refs ?SPage]
   ]
 @conn)
;; => #{[2]}

If I swap the clause order it does return what I would expect

(dl/q
 '[:find ?SBlock
   :where
   [?ParentPage :block/uid "test"]
   [?ParentPage :block/uid ?some-uid]
   [?SPage :node/title "Opposed By"]
   [?SBlock :block/refs ?SPage]
   ]
 @conn)
  ;; => #{}

A similar issue happens if I remove the :block/refs clause, but this time it returns a nil id

(dl/q
 '[:find ?SPage
   :where
   [?ParentPage :block/uid "test"]
   [?SPage :node/title "Opposed By"]
   [?ParentPage :block/uid ?some-uid]
   ]
 @conn)
;; => #{[nil]}

Thanks in advance for looking into this!

[huahaiy]

These queries do not seem to make sense. :block/uid is unique identity, so how is it possible to have two clauses about them?

[?ParentPage :block/uid "test"]
[?ParentPage :block/uid ?some-uid]

So ?some-uid is just "test", isn't it?

This is a pathological case, and I would consider this undefined behavior. Of course, we could check for this and make the handling of these more robust. But it would be an enhancement, not a bug fix.

[panterarocks49]

This query is part of a larger query which is generated by a higher level query language so it's definitely an edge case

The full query is a bit harder to follow so I just posted the simplest case I could find which produced similar behavior

This is the full query, but it won't reproduce on this test data

[:find
 (pull ?Evidence [:block/string :node/title :block/uid])
 (pull ?Evidence [:block/uid])
 :where
 [?gLLeASJNc-ParentPage :block/uid "test"]
 [?gLLeASJNc-SPage :node/title "Opposed By"]
 [?gLLeASJNc-ParentPage :block/uid ?gLLeASJNc-v6ruZRoob-uid]
 [?v6ruZRoob :block/uid ?gLLeASJNc-v6ruZRoob-uid]
 [?gLLeASJNc-SBlock :block/refs ?gLLeASJNc-SPage]
 [?gLLeASJNc-SBlock :block/children ?gLLeASJNc-Block]
 [?gLLeASJNc-Block :block/refs ?gLLeASJNc-Page]
 [?gLLeASJNc-Page :node/title ?gLLeASJNc-Page-Title]
 [?gLLeASJNc-Page :block/uid ?gLLeASJNc-Evidence-uid]
 [?Evidence :block/uid ?gLLeASJNc-Evidence-uid]
 [?gLLeASJNc-PBlock :block/refs ?gLLeASJNc-ParentPage]
 [?gLLeASJNc-PBlock :block/children ?gLLeASJNc-SBlock]
 [(re-pattern "^\[\[EVD\]\] - (.*?) - (.*?)$") ?gLLeASJNc-EVD---regex]
 [(re-find ?gLLeASJNc-EVD---regex ?gLLeASJNc-Page-Title)]
 ]

I can observe the same behavior with a non unique identity as well like this, to me it seems to have to do with the non unifying clauses.

(dl/q
 '[:find ?SBlock
   :where
   [?ParentPage :block/refs 1]
   [?SPage :node/title "Opposed By"]
   [?ParentPage :block/refs ?some-uid]
   [?SBlock :block/refs ?SPage]
   ]
 @conn)

Anyway, if this is expected I can try to detect it in my code

[huahaiy]

It is an edge case that I have not considered, so it needs to be detected and handled properly. The next release should have a fix.