Repository navigation
EQL v3 empty JSON arrays break array accessors #978
Description
The EQL v3 JSON domain accepts an empty searchable vector, but ste_vec()
aggregates zero rows to SQL NULL. As a result, jsonb_array_length returns
NULL instead of zero and jsonb_array_elements raises SQLSTATE 22004 because
its FOR loop has a null upper bound.
Acceptance criteria
- The length of an encrypted empty JSON array is zero.
jsonb_array_elementsreturns zero rows without error.- Populated-array behaviour is unchanged.
- Tests exercise a real encrypted empty-array payload as well as the SQL helper
boundary. - The implementation composes with the umbrella array aggregation design.
Activity
coderdan commentedon Oct 21, 2026
How the EQL v4
json_treefamily (#1288, #1325, #1329) handles the empty array, for comparison, plus one gap it has of its own.In v4 an empty array is a node whose encrypted plaintext says kind array, child count zero.
eql_v4.json_tree_child_countreturns 0 andjson_tree_childrenreturns no rows, with no null loop bound anywhere; a walk that reaches the empty array through gates (json_tree_subtree_at) returns the node itself at depth 0 and the client opens it to[].The gap: an empty container that is a chain head (a top-level field under the sealed root).
json_tree_subtree(doc, token)returns a node's descendants, not the node, so an empty$.tagsyields no rows and the client reports "a node with children" instead of[]; and->on its selector returns the node, which the client refuses to open because a container holds no value of its own. The fix is small and client-side: an extracted container node with zero children should open to its empty value. Not filed separately yet; I will unless someone objects to tracking it under this issue.v3's
ste_vec()aggregation is unchanged by the v4 work.