Re: remaining sql/json patches

On Tue, Mar 19, 2024 at 6:46 PM Amit Langote <amitlangote09(at)gmail(dot)com> wrote:
>
> I intend to commit 0001+0002 after a bit more polishing.
>

V43 is far more intuitive! thanks!

if (isnull ||
(exprType(expr) == JSONBOID &&
btype == default_behavior))
coerce = true;
else
coerced_expr =
coerce_to_target_type(pstate, expr, exprType(expr),
returning->typid, returning->typmod,
COERCION_EXPLICIT, COERCE_EXPLICIT_CAST,
exprLocation((Node *) behavior));

obviously, there are cases where "coerce" is false, and "coerced_expr"
is not null.
so I think the bool "coerce" variable naming is not very intuitive.
maybe we can add some comments or change to a better name.

JsonPathVariableEvalContext
JsonPathVarCallback
JsonItemType
JsonExprPostEvalState
these should remove from src/tools/pgindent/typedefs.list

+/*
+ * Performs JsonPath{Exists|Query|Value}() for a given context_item and
+ * jsonpath.
+ *
+ * Result is set in *op->resvalue and *op->resnull. Return value is the
+ * step address to be performed next.
+ *
+ * On return, JsonExprPostEvalState is populated with the following details:
+ * - error.value: true if an error occurred during JsonPath evaluation
+ * - empty.value: true if JsonPath{Query|Value}() found no matching item
+ *
+ * No return if the ON ERROR/EMPTY behavior is ERROR.
+ */
+int
+ExecEvalJsonExprPath(ExprState *state, ExprEvalStep *op,
+ ExprContext *econtext)

" No return if the ON ERROR/EMPTY behavior is ERROR." is wrong?
counter example:
SELECT JSON_QUERY(jsonb '{"a":[12,2]}', '$.a' RETURNING int4RANGE omit
quotes error on error);
also "JsonExprPostEvalState" does not exist any more.
overall feel like ExecEvalJsonExprPath comments need to be rephrased.