Skip to content

Latest commit

 

History

History
199 lines (149 loc) · 5.97 KB

cxn.md

File metadata and controls

199 lines (149 loc) · 5.97 KB
Raw
layout shorty synopsis status uacp
cds-ref
Expressions
Specification of the Core Expression Notation (CXN) used to capture expressions as plain JavaScript objects.
released

Expression Notation (CXN) { #expressions}

Expressions in CDS definitions and queries can be one of:

[[toc]]

expr = // one of...
  val    |   // [literal values]: #literal-values
  ref    |   // references or functions
  xpr    |   // operator expressions
  func   |   // function calls
  list   |   // lists/tupels
  param  |   // binding parameters
  sym    |   // enum symbol
  SELECT     // subqueries

Literal Values

Literal values are represented as {val:...} with property val holding the actual literal value as specified in JSON.

val = {val:literal}
literal = string | number | true | false | null

Examples:

cds.parse.expr(`'a string'`)  == {val:'a string'}
cds.parse.expr(`11`)  == {val:11}
cds.parse.expr(`true`)  == {val:true}
cds.parse.expr(`null`)  == {val:null}
cds.parse.expr(`date'2023-04-15'`)  == {val: '2023-04-15', literal: 'date'}
cds.parse.expr(`time'13:05:23Z'`)  == {val: '13:05:23Z', literal: 'time'}
cds.parse.expr(`timestamp'2023-04-15T13:05:23Z'`)  == {val: '2023-04-15T13:05:23Z', literal: 'timestamp'}

References

A reference is represented as { ref: … } with property ref. This property holds an array of reference segments as plain identifier strings. Only in case of infix filters and/or arguments, the property holds an object { id: 'identifier', … } and all properties except id are optional, as shown in the following snippet:

ref = {ref:[..._segment]}
_segment = string | { id: string, args: _named, where: _xpr,
                      groupBy: [ ...expr ], having: _xpr,
                      orderBy: [ ...ordering_term ], limit: { rows: expr, offset: expr } }
_named = { ... <name>:expr }

Examples:

let cqn4 = cds.parse.expr
cqn4(`![keyword]`) == {ref:['keyword']}
cqn4(`foo.bar`) == {ref:['foo','bar']}
cqn4(`foo[9].bar`) == {ref:[{ id:'foo', where:[{val:9}] }, 'bar' ]}
cqn4(`foo(p:x).bar`) == {ref:[{ id:'foo', args:{p:{ref:['x']}} }, 'bar' ]}
cqn4(`foo[where a=1 group by b having b>2 order by c limit 7].bar`)
  == {ref:[{ id:'foo', where:[{ref:['a']}, '=', {val:9}],
                       groupBy: [{ref: ['b']}], having: [{ref: ['b']}, '>',  {val:2}],
                       orderBy: [{ref: ['c']}], limit: {rows: {val: 7}} },
           'bar' ]}

Function Calls

Function calls are represented as follows:

func = { func:string, args: _positional | _named, xpr:_xpr }
_positional = [ ...expr ]
_named = { ... <name>:expr }

The optional attribute xpr is used for the over clause of SQL window functions.

Examples:

let cqn4 = cds.parse.expr
cqn4(`foo(p=>x)`) == {func:'foo', args:{p:{ref:['x']}}}
cqn4(`sum(x)`)   == {func:'sum', args:[{ref:['x']}]}
cqn4(`count(*)`) == {func:'count', args:['*']}
cqn4(`rank() over (...)`) == {func:'rank', args:[], xpr:['over', {xpr:[...]}]}

Method style function calls and instantiation syntax for spatial functions are represented as {xpr:...} with . and new as operators:

cqn4(`shape.ST_Area()`)
  == {xpr: [{ref: ['shape']}, '.', {func: 'ST_Area', 'args': []}]}
cqn4(`new ST_Point(2, 3)`)
  == {xpr: ['new', {func: 'ST_Point', args: [{val: 2}, {val: 3}]}]}

Lists { #lists}

Lists or tupels are represented as {list:...}, with property list holding an array of the list entries.

Examples:

cds.parse.expr(`(1, 2, 3)`) == {list: [{val: 1}, {val: 2}, {val: 3}]}
cds.parse.expr(`(foo, bar)`) == {list: [{ref: ['foo']}, {ref: ['bar']}]}

Operator Expressions { #operators}

Operators join one or more expressions into complex ones, represented as {xpr:...}. The property xpr holds a sequence of operators and operands.

xpr = {xpr:_xpr}
_xpr = [...( _operand | _operator )]
_operand = expr
_operator = string
  • Operands can be any kind of expression
  • Operators are represented as plain strings, like '=' or 'and'
  • Parentheses ( ... ) around sub-expressions are represented as nested xpr

Examples:

cds.parse.expr(`x<9`)  ==//> returns:
{xpr:[ {ref:['x']}, '<', {val:9} ]}

cds.parse.expr(`x<9 and (y=1 or z=2)`)  ==//> returns:
{xpr:[
  {ref:['x']}, '<', {val:9}, 'and', {xpr:[
    {ref:['y']}, '=', {val:1}, 'or', {ref:['z']}, '=', {val:2}
  ]}
]}

cds.parse.expr(`exists books[year = 2000]`)  ==//> returns:
{xpr:[
  'exists',
  {ref: [ {id:'books', where:[ {'ref':['year']}, '=', {'val': 2000} ]}]}
]}

CQN intentionally doesn’t aim to understand the individual operators and related expressions. It captures them as arbitrary sequences, in the same lexical structure and order they’re written in the source. This 'ignorance' allows us to stay open to any kind of operators and keywords. For example, we can easily express native extensions of underlying database dialects.

As an exception to that rule, CDS supports the ternary conditional operator on source level, but immediately converts it to the corresponding CASE expression in CXN:

cds.parse.expr(`x<10 ? y : z`)  ==//> returns:
{xpr:['case', 'when', {ref:['x']}, '<', {val:10},
      'then', {ref:['y']}, 'else', {ref:['z']}, 'end']}

Binding Parameters

Binding parameters for prepared statements are represented as {ref:..., param:true} with values for ref as follows.

param = { ref:[ '?' | number | name ], param:true }

Examples:

cds.parse.expr(`x=:1`) == [{ref:['x']}, '=', {ref:[1], param:true}]
cds.parse.expr(`x=:y`) == [{ref:['x']}, '=', {ref:['y'], param:true}]
cds.parse.expr(`x=?`)  == [{ref:['x']}, '=', {ref:['?'], param:true}]

Sub Queries

See CQN{.learn-more}