Repository navigation
Conversation
efc55d1 to
4d9b7cb
Compare
Signed-off-by: Muhammad Aaqil <aaqilcs102@gmail.com>
77041f3 to
c453e6c
Compare
samarpanB
left a comment
There was a problem hiding this comment.
Thanks for tackling JSONB membership support. The proposed SQL translation is promising for the reported scenario, but there are correctness and compatibility issues around native arrays, input typing, empty filters, cross-connector semantics, and missing database-backed regression coverage. I've left specific inline comments; I'd like these addressed before merging.
| if (!arrayItemType) { | ||
| throw new Error(`Unsupported array item type: ${itemType ? itemType.name : 'unknown'}`); | ||
| } | ||
| const isArray = propertyDefinition.postgresql.dataType === 'ARRAY'; |
There was a problem hiding this comment.
This native-array detection looks incorrect. The connector already supports native PostgreSQL mappings such as varchar[] (see toColumnValue()), but dataType === 'ARRAY' would not recognize them. They would instead be sent through the JSONB branch, generating an incompatible @> operand type. Can we distinguish actual native array types from JSONB columns and reuse the existing array operators (including containsAnyOf for overlap) where appropriate? Please add native-array execution tests.
| let values = operatorValue; | ||
| if (typeof values === 'string') { | ||
| try { | ||
| values = JSON.parse(values); |
There was a problem hiding this comment.
Parsing every string with JSON.parse() can silently change identifier types. For example, a legitimate string ID '123' becomes the number 123, and JSONB distinguishes ['123'] from [123], so an existing reference may no longer match. Can we normalize values using the declared element type instead of guessing from string contents? Please add a numeric-looking string ID regression test.
| params.push(JSON.stringify([value])); | ||
| } | ||
| }); | ||
| return new ParameterizedSQL(`(${conditions.join(' OR ')})`, params); |
There was a problem hiding this comment.
For jsonContains: [], conditions stays empty and this returns (), which is invalid SQL. Empty candidate lists can occur in dynamically generated inq filters. Please define explicit empty-input behavior (typically a false predicate for match-any) and test it.
| } | ||
| if (!Array.isArray(values)) { values = [values]; } | ||
|
|
||
| const itemType = propertyDefinition.type[0]; |
There was a problem hiding this comment.
This path assumes propertyDefinition.type is an array and propertyDefinition.postgresql exists. A model array property does not necessarily declare PostgreSQL-specific metadata, so the later .dataType access can throw a TypeError rather than handling the filter deliberately. Can we validate the supported property types/storage mappings and return a clear error or supported fallback?
| return new ParameterizedSQL(`to_tsvector(${columnName}) @@ to_tsquery(?)`, [operatorValue]); | ||
| // Map the abstract 'jsonContains' operator to PostgreSQL's JSONB containment operator (@>), | ||
| // parsing or normalizing string/array inputs to build parameterized clauses connected with OR. | ||
| case 'jsonContains': |
There was a problem hiding this comment.
We need to define the public semantics of jsonContains before adding this as a connector operator. Here an array of candidates becomes OR/match-any, while JSON containment of an array in other connectors can mean match-all. Since datasource-juggler #2615 proposes rewriting inq into jsonContains, this risks different results across connectors for the same filter. Can we preserve the inq contract explicitly and establish consistent operator semantics across the relevant connectors?
| db.automigrate(['ItemWithJson', 'ItemWithIntegerArray', 'ItemWithStringArray'], done); | ||
| }); | ||
|
|
||
| beforeEach(function(done) { |
There was a problem hiding this comment.
The added tests inspect buildWhere() output but do not persist records and execute a find() against PostgreSQL, so they cannot catch type/cast failures or incorrect results. Please add data-backed cases for JSONB, native arrays, multiple candidates, numeric-looking string IDs, empty inputs, and the actual referencesMany flow. Also, this hook only waits for the final destroyAll(done); the preceding two destroyAll() calls are unawaited and may overlap with subsequent tests. Could we await all three operations?
jsonContainsoperator mapping inside PostgreSQL'sbuildExpressionto support array/containment queries (such as those required forreferencesManyrelations).col @> CAST(? AS JSONB)) connected withOR.This PR depends on PR #11883 (loopbackio/loopback-next#11883) to successfully access property metadata after relation creation.
This PR also depends on loopbackio/loopback-next#11146. The find method never resolves the options parameter. This PR adds that.
This PR also depends on loopbackio/loopback-datasource-juggler#2615. The datasource-juggler identifies the
inqfilter with referencesMany and converts the filter tojsonContains.Closes Issue loopbackio/loopback-next#11169
Checklist
npm testpasses on your machine