SQL for JSON Rationalization Part 11: JSON Specific Predicates

JSON SQL cannot assume collections containing only homogeneous documents; instead, it must support schemaless documents. This requires JSON specific predicates that are discussed in this blog.

Example Collection

The documents of the following collection “predColl” are used in this blog as examples.

{"a":"b"}
{"a":{"c":1,"d":2},"e":[77,{"x":"eightyeight"}]}
{"a":{"c":1,"d":2},"e":["seventyseven",{"x":88}]}

Definition: Path

A (complete) path in a JSON document is the list of property names and array indexes from the root all the way to a leaf separated by “.”. For example, a complete path is

e.[1].x

A partial path in a JSON document is the list of property names and array indexes from the root to an intermediary property (that is not a leaf property) separated by “.”. For example, a partial path is

e.[1]

It is possible that a path is a complete path in one document and a partial path in another document. For example, a path that is complete and partial at the same time is

a

Each path leads up to a property that is of a specific JSON type. For example,

  • the type of e.[1].x is JSON_STRING in one of the example documents and JSON_NUMBER in another one.
  • the type of e.[1] is JSON_OBJECT in both cases.

Document: Set of Paths

Based on the above definition, a JSON document can be represented as the set of its paths.

The document

{"a":{"c":1,"d":2},"e":["seventyseven",{"x":88}]}

can be represented as

a
a.c
a.d
e.[0]
e.[1]
e.[1].x

The corresponding types are

JSON_OBJECT
JSON_NUMBER
JSON_NUMBER
JSON_STRING
JSON_OBJECT
JSON_NUMBER

Schemaless Documents / Schema-per-Document

Traditionally, databases enforced the definition of a schema and many continue to do so. This means that data can only be stored successfully in a database if the data complies with the schema at that point in time.

Some database systems do not enforce a schema and as a consequence the data is not constrained by a schema. In context of JSON documents, if a document is not constrained by a schema, then it can have any valid structure (in terms of the rules the JSON standard defines).

These documents are termed “schemaless” documents. In reality there is a schema, an implicit  (intentional) one, for each document, termed “schema-per-document”.

In concrete terms “schemaless” in context of a database system means:

  • Any pair of documents within a collection of a database can potentially have different sets of paths
  • The same (complete or partial) path in two different documents can lead to a value of a different JSON type

As a consequence, queries must be able to test for the existence of paths as well as for the existence of a specific JSON type as these cannot be assumed (in contrast to a database that enforces schemas).

JSON Specific Predicates: exists_path and is_of_type

Two predicates are needed in order to check for the existence of paths or JSON types of values.

  • exists_path <path>
  • <path> is_of_type <JSON type>

The query

select {*} from predColl where exists_path a.d

returns

{"a":{"c":1,"d":2},"e":[77,{"x":"eightyeight"}]}
{"a":{"c":1,"d":2},"e":["seventyseven",{"x":88}]}

The query

select {*} from predColl where e.[1].x is_of_type JSON_number

returns

{"a":{"c":1,"d":2},"e":["seventyseven",{"x":88}]}

Negation

Negation of the predicates is useful to determine if there are document that do not have specific paths or paths do not refer to properties of specific JSON types.

select {*} from predColl where not exists_path e.[1].x

returns

{"a":"b"}

The query

select {*} from predColl where not a is_of_type JSON_object

returns

{"a":"b"}

Definition: (Partial) Homogeneity

Queries using the predicates exists_path or is_of_type can be used to determine if all documents have the required or expected paths or if there are documents that are missing specific paths or have paths leading to unexpected JSON types.

With these predicates it is possible to determine if the documents in a collection are homogeneous wrt. given set of paths.

Dynamic Schema Check

A developer can now implement first checking path existence or JSON type compliance before executing business logic related queries. If the required paths or JSON types are missing in specific documents, appropriate error handling can be implemented for these.

From a different viewpoint, a developer now has the tools to dynamically check or select all documents complying to a schema that is “imagined” at the type of querying.

Summary

While schemaless documents are convenient from a development perspective, this requires in general the ability to check for the existence or absence of paths as well as JSON type compliance.

The predicates exists_path and is_of_type provide the necessary querying tools in order to test for the variations in schemaless JSON documents.

Go [ JSON | Relational ] SQL!

Disclaimer

The views expressed on this blog are my own and do not necessarily reflect the views of Oracle.

Advertisements

SQL for JSON Rationalization Part 9: Restriction – Arrays

This installment reviews restriction in JSON SQL based on JSON array literals (all other JSON types except JSON array have been discussed in previous blogs).

JSON Object Notation

JSON SQL follows the JSON object notation as defined in the JSON standard. An empty JSON array is denoted as [] and a non-empty JSON array has one or more comma separated values, including JSON array.

A JSON array literal is either an empty JSON array or a non-empty JSON array. A JSON array literal is not enclosed in quotes. The only JSON literal enclosed in quotes is JSON string. If a JSON array is enclosed in quotes then it is not a JSON array, but a JSON string.

Sample Document Set

The following document set is used in this blog and the documents are stored in a collection called “arrayColl”.

select {*} from arrayColl

results in

{"one":[{"a":1},{"b":2}]}
{"one":"[{\"a\": 1}, {\"b\": 2}]"}
{"three":[{"b":[{"c":null},{"d":true}]}]}
{"four":[{"x":8,"y":9}]}
{"five":[]}

Restriction based on JSON Object Literal

Starting with the empty JSON array literal, the following two queries product the same result.

select {*} from arrayColl where five = []

and

select {*} from arrayColl where [] = five

result in

{"five":[]}

In the following, queries show the JSON array literal on the right side of the operator, however, it can be on either side.

Operators = And <>

The operators = and <> are defined for a JSON array literal. JSON SQL regards two JSON arrays as equal if both have the same (equal) values in the same order; and not equal otherwise.

The query (restriction using JSON array literal)

select {*} from arrayColl where one = [{"a": 1}, {"b": 2}]

returns

{"one":[{"a":1},{"b":2}]}

The query (restriction using JSON string literal(!))

select {*} from arrayColl where one = '[{"a": 1}, {"b": 2}]'

returns

{"one":"[{\"a\": 1}, {\"b\": 2}]"}

A restriction can reach into the JSON array as well using the path notation. The query

select {*}
from arrayColl
where three.[0].b = [{"c": null}, {"d": true}]

returns

{"three":[{"b":[{"c":null},{"d":true}]}]}

Operators <, >, <= And >=

The operators <, >, <= and >= could be defined recursively for convenience with some restrictions. For example, a JSON array could be considered less than another JSON array if both have the same values and if the corresponding values are less than another.

However, JSON true, JSON false and JSON null would not be able to participate in the operator <, >, <= or >=, only JSON object, JSON array, JSON number and JSON string.

Those four operators are currently not directly implemented in JSON SQL since it is possible to achieve the same by writing a complex conjunctive restriction (details on this approach will be discussed in a subsequent blog as well as strategies of what to do if any of JSON true, JSON false or JSON null are present).

Canonical Interpretation

The order of the values inside a JSON array is significant, but not within a JSON object. The query

select {*} from arrayColl where four = [{"y": 9, "x": 8}]

therefore results in

{"four":[{"x":8,"y":9}]}

Summary

Restriction by JSON array is provided by JSON SQL without problem and the syntax extends the Relational SQL syntax naturally.

Go [ JSON | Relational ] SQL!

Disclaimer

The views expressed on this blog are my own and do not necessarily reflect the views of Oracle.

SQL for JSON Rationalization Part 8: Restriction – Objects

This installment reviews restriction in JSON SQL based on JSON object literals (all other JSON types except JSON array have been discussed in previous blogs).

JSON Object Notation

JSON SQL follows the JSON object notation as defined in the JSON standard. An empty JSON object is denoted as {} and a non-empty JSON object has one or more comma separated pairs (a pair is a tuple of string and JSON type separated by a colon – also referred to as property).

A JSON object literal is either an empty JSON object or a non-empty JSON object. A JSON object literal is not enclosed in quotes. The only JSON literal enclosed in quotes is JSON string. If a JSON object is enclosed in quotes then it is not a JSON object, but a JSON string.

Sample Document Set

The following document set is used in this blog and the documents are stored in a collection called “objectColl”.

select {*} from objectColl

results in

{"one": {"a": 1}}
{"one": "{\"a\": 1}"}
{"three": {"b": {"c": null}}}
{"four": {"x": 8, "y": 9}}
{"five": {}}

Restriction based on JSON Object Literal

Starting with the empty JSON object literal, the following two queries product the same result.

select {*} from objectColl where five = {}

and

select {*} from objectColl where {} = five

result in

{"five": {}}

In the following, queries show the JSON object literal on the right side of the operator, however, it can be on either side.

Operators = And <>

The operators = and <> are defined for a JSON object literal. JSON SQL regards two JSON objects as equal if both have the same pairs (recursively), in any order; and not equal otherwise.

The query (restriction using JSON object literal)

select {*} from objectColl where one = {"a": 1}

returns

{"one": {"a": 1}}

The query (restriction using JSON string literal(!))

select {*} from objectColl where one = '{"a": 1}'

returns

{"one": "{\"a\": 1}"}

A restriction can reach into the JSON object as well using the path notation. The query

select {*} from object Coll where three.b = {"c": null}

returns

{"three": {"b": {"c": null}}}

Operators <, >, <= And >=

The operators <, >, <= and >= could be defined recursively for convenience with some restrictions. For example, a JSON object could be considered less than another JSON object if both have the same pairs and if the values of the corresponding pairs are less than another.

However, JSON true, JSON false and JSON null would not be able to participate in the operator <, >, <= or >=, only JSON object, JSON array, JSON number and JSON string.

Those four operators are currently not directly implemented in JSON SQL since it is possible to achieve the same by writing a complex conjunctive restriction (details on this approach will be discussed in a subsequent blog as well as strategies of what to do if any of JSON true, JSON false or JSON null are present).

Canonical Interpretation

The order of the pairs inside a JSON object is not significant (according to the JSON standard). The query

select {*} from objectColl where four = {"y": 9, "x": 8}

therefore results in

{"four": {"x": 8, "y": 9}}

Summary

Restriction by JSON object is provided by JSON SQL without problem and the syntax extends the Relational SQL syntax naturally.

Go [ JSON | Relational ] SQL!

Disclaimer

The views expressed on this blog are my own and do not necessarily reflect the views of Oracle.

SQL for JSON Rationalization Part 7: Restriction – True, False and Null

The last blog (Part 6) introduced the general notion of restriction and focused on JSON String and JSON Number. This blog will extend the discussion covering JSON true, JSON false and JSON null.

Sample Data Set

The following set of documents stored in the collection “boolcoll” is used in this blog:

select {*} from boolcoll

returns

{"a":true}
{"a":false}
{"true":false}
{"true":"null"}

Textual Representation

Even though the JSON standard defines that JSON true, JSON false and JSON null are all lowercase, in context of JSON SQL all combinations of upper and lower case characters are permitted:

select {*} from boolcoll where a = TruE

returns

{"a":true}

Operators

Not all of the default operators are defined on JSON true, JSON false or JSON null. Defined are = and <>, undefined are <, >, <= and >=. If one of the undefined operators is used in conjunction with JSON true, JSON false or JSON null, a semantic query analysis error is returned before the query is executed.

Restriction Syntax

As in the last blog, the property name can be on either side of the operator. The following two queries return the same result:

select {*} from boolcoll where a <> false

and

select {*}  from boolcoll where false <> a

return

{"a":true}

Execution Semantics

If the specified property is present, the restriction is evaluated and the document is added to the result set if it fulfills the restriction. If the property is not present, no evaluation and consequently no inclusion into the result set takes place.

select {*} from boolcoll where null = null

returns all documents as null = null is always true.

Disambiguation

As shown in the example data set for this blog, it is possible that a property name is “true”, “false” or “null”. So far only the short form of property names was used with JSON SQL queries, i.e., property names without being enclosed in double quotes (contrary to the JSON standard definition).

However, as soon as property names can be the same as keywords like JSON true, JSON false or JSON null, disambiguation has to take place. This is accomplished by following the JSON standard: enclosing the property name in double quotes. For example,

select {*} from boolcoll where "true" = false

returns all documents with a property name of “true” that has the value JSON false.

{"true":false}

Since Relational SQL does not use double quotes, there cannot be any confusion:

Along the same lines,

select {*} from boolcoll where "true" = 'null'

returns all documents where the property “true” has the value ‘null’ (string).

{"true":"null"}

This is not ambiguous, either, as Relational SQL uses single quotes to denote String literals.

Needless to say that double quotes can be used outside a disambiguation. For example, one of the above queries could be specified as

select {*} from boolcoll where "a" = TruE

returning

{"a":true}

Double quotes can be used in the projection clause as well.

select "a" from boolcoll where "a" = tRUe

returns

|a                        |
+-------------------------+
|true                     |

Summary

JSON true, JSON false and JSON null can be used in JSON SQL queries without restriction and in a well-defined way. Disambiguation is not interfering with either the syntax as defined by the JSON standard, or the regular Relational SQL syntax. Great!

Go [ JSON | Relational ] SQL!

Disclaimer

The views expressed on this blog are my own and do not necessarily reflect the views of Oracle.