-
Type:
Bug
-
Resolution: Unresolved
-
Priority:
Major - P3
-
None
-
Affects Version/s: None
-
Component/s: None
-
None
-
Query Execution
-
ALL
-
For collations similar to:
{'locale': 'en', 'strength': 1, 'alternate': 'shifted', 'maxVariable': 'space'}
Transitivity is not guaranteed.
For three example values:
a = '\u0020' // Space b = '\u0359' // Combining Asterisk Below c = '\u00a0' // No-Break Space
> coll.find()
{ "_id" : ObjectId("6712505b1fd232a2bd6bfb94"), "value" : " ", "name" : "a" }
{ "_id" : ObjectId("6712505b1fd232a2bd6bfb95"), "value" : "͙", "name" : "b" }
{ "_id" : ObjectId("6712505b1fd232a2bd6bfb96"), "value" : " ", "name" : "c" }
> coll.find({value:{"$eq":a}}).collation(collation)
{ "_id" : ObjectId("6712505b1fd232a2bd6bfb94"), "value" : " ", "name" : "a" }
{ "_id" : ObjectId("6712505b1fd232a2bd6bfb95"), "value" : "͙", "name" : "b" }
> coll.find({value:{"$eq":b}}).collation(collation)
{ "_id" : ObjectId("6712505b1fd232a2bd6bfb94"), "value" : " ", "name" : "a" }
{ "_id" : ObjectId("6712505b1fd232a2bd6bfb95"), "value" : "͙", "name" : "b" }
{ "_id" : ObjectId("6712505b1fd232a2bd6bfb96"), "value" : " ", "name" : "c" }
> coll.find({value:{"$eq":c}}).collation(collation)
{ "_id" : ObjectId("6712505b1fd232a2bd6bfb95"), "value" : "͙", "name" : "b" }
{ "_id" : ObjectId("6712505b1fd232a2bd6bfb96"), "value" : " ", "name" : "c" }
a == b && b == c but a != c.
InListData relies on sort and unique to sort and dedupe values, using the provided collator. As the collator does not meet the requirements for an equivalence relation or a strict weak ordering, these do not result in the desired behaviour - it depends on the order of encountered values.
A such the order of elements in $in can change results.
e.g.,
> coll.find({value:{"$in":[a, b, c]}}).collation(collation)
{ "_id" : ObjectId("6712505b1fd232a2bd6bfb94"), "value" : " ", "name" : "a" }
{ "_id" : ObjectId("6712505b1fd232a2bd6bfb95"), "value" : "͙", "name" : "b" }
> coll.find({value:{"$in":[c, a, b]}}).collation(collation)
{ "_id" : ObjectId("6712505b1fd232a2bd6bfb94"), "value" : " ", "name" : "a" }
{ "_id" : ObjectId("6712505b1fd232a2bd6bfb95"), "value" : "͙", "name" : "b" }
{ "_id" : ObjectId("6712505b1fd232a2bd6bfb96"), "value" : " ", "name" : "c" }
gives a different winning plan:
"winningPlan" : {
"isCached" : false,
"stage" : "COLLSCAN",
"filter" : {
"value" : {
"$eq" : " " /// value of a
}
},
"direction" : "forward"
},
vs
"winningPlan" : {
"isCached" : false,
"stage" : "COLLSCAN",
"filter" : {
"value" : {
"$in" : [
" ", /// value of a
" " /// value of c
]
}
},
"direction" : "forward"
},
Affected versions not yet narrowed down; at least current master around ebd06e3f122.