Repository navigation
feat(postgres): type a json key read by the column's json schema - #95
Merged
Merged
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
Contributor
There was a problem hiding this comment.
ℹ️ One narrow soundness gap inline. Everything else I traced holds up.
Reviewed changes
I reviewed the whole PR at 56457ca: the core schema walk, the Postgres reader and probe changes, the carry-through for derived tables and with queries, the new oracle case, and the docs and skill.
schema_type::at_path: walks a canonical schema one key at a time. It returns the schema found at the path and whether the value is there for certain. It givesNonefornot, and also for an integer key under#>that might land on an array.Expr::JsonPathin the reader: the built-in->,->>,#>and#>>with literal keys over a column (bare or narrowed) now build a path, and chained reads extend it. Non-literal keys, shadowed operators and non-column bases stayExpr::Unknown, which matches the old behaviour, since->was never inPOSTGRES_OPERATORS.json_readand the probe: a top-level read sets the column'sdocument(for a JSON read) or itsvalues(for a text read whose schema allows only strings). Thechecks.valuesloop no longer overwrites those values withNone.ItemColumninitem_columns: a derived table orwithquery now passes its column'sdocumentandvaluesto an outer column that reads it as it is.- Tests and docs: a new
oracle/cases/json_schema_keyscase with 14 statements, unit tests for the walk and the reader, a captured docs card, and the skill line.
Points I checked and found sound:
- Canonical schemas fold
patternPropertiesintoadditionalProperties. anyOfandenumignore sibling keywords, which can only loosen the type.- A text key read with
->on an array returns SQL NULL. #> '{}'returns the whole document.
claude-opus-5-5 ｜ 𝕏
colinhacks
force-pushed
the
pg-json-keys
branch
from
October 23, 2026 23:02
56457ca to
6053c4b
Compare
Contributor
There was a problem hiding this comment.
✅ No new issues found. The new commit fixes the
#>index gap from the prior review.
Reviewed changes
I reviewed the one commit pushed since the prior review, 6053c4b.
- Index keys follow
strtoint: under#>and#>>,at_keynow counts a key as a possible array index ifstrtointwould read the whole key: optional leading C whitespace, one optional+or-, then digits. So-1,+0and1now stop the walk where an array may be. A key with trailing whitespace, whichstrtointrejects, is still read as an object key. A digit key too large to fit is also counted as an index, which is the safe side. - Coverage: the new oracle statement
signed_index.sqlreads a value that is either an array or an object, using signed keys, a space-led key and a plain key. Unit tests inschema_type.rscover the same keys. The doc comments inschema_type.rsandinfer.rsnow describe the rule. - Artifacts: all three
engine.wasmfiles were rebuilt for the core change.
claude-opus-5-5 ｜ 𝕏
colinhacks
force-pushed
the
pg-json-keys
branch
from
October 23, 2026 23:18
6053c4b to
0a8a346
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
On Postgres, a
jsonorjsonbcolumn a JSON SchemaCHECKholds was typed by that schema only when read whole. A key read from it was typed by the operator's result alone:->asjsonb, which isunknown, and->>astext, both nullable, so a value the schema already describes needed a cast, and never came back non-null.Now a read at literal keys, with the built-in
->,->>,#>or#>>, takes what the schema says is there. Each step reads the schema's object at its key, as the server reads the value's:->and#>are typed by the property's schema;->>and#>>aretext, a union of strings where the schema keeps the value to a list of them.#>(digits, after a sign or leading whitespace, a negative one counting from the end) may index an array instead, which is not followed: the read staysunknownwhere a step may land on one.->(hstore defines one), and any other read stay as they were.withquery or a derived table keeps its type when the statement selects its column as it is, as built JSON does.With
docchecked against{ name: string, kind: "a" | "b", size: integer | null, owner: { id: integer, email?: string }, meta?: { note: string } | null }, requiredname,kind,sizeandowner:doc -> 'name'unknown | nullstringdoc ->> 'kind'string | null"a" | "b"doc -> 'size'unknown | nullnumber | null(a JSON null, in a non-nulljsonb)doc ->> 'size'string | nullstring | nulldoc -> 'owner' -> 'id'/doc #> '{owner,id}'unknown | nullnumberdoc -> 'meta' ->> 'note'string | nullstring | null(doc -> 'owner' ->> 'id')::intnumber | nullnumberjsonb_build_object('id', doc #> '{owner,id}'){ id: unknown | null }{ id: number }doc #> '{tags,0}',doc -> :keyunknown | nullunknown | nullThe walk over a schema is core's (
schema_type::at_path), over the canonical schema the engine already reads aCHECKinto, so the MySQL and SQLite readers can use it too; this changes all three artifacts' fingerprints. A key read from a typed JSON parameter (:p::jsonb('…') -> 'x') staysunknown: its schema is applied after the analysis that reads the operator.Tests:
oracle/cases/json_schema_keys: 15 statements over rows with a key left out, a JSON null, a nested object that may be null, a key no property names, an array, a value that is an array or an object read at signed and space-led keys, a tagged union, ajsoncolumn and a nullable column, sound over the seeded edge values.json_schema_checks/expression.sqlnow typesdoc -> 'name'asstring.engine/core/src/schema_type.rs(the walk) andengine/postgres/src/infer.rs(which reads are claimed, the literal keys, and the read's nullability).--strict-coverage, the engine's tests, the workspace build with no warning, the three rebuilt artifacts' fingerprints, the loader's typecheck and tests, the docs site.