CodeQL documentation

QL language specification

This is a formal specification for the QL language. It provides a comprehensive reference for terminology, syntax, and other technical details about QL.

Introduction

QL is a query language for CodeQL databases. The data is relational: named relations hold sets of tuples. The query language is a dialect of Datalog, using stratified semantics, and it includes object-oriented classes.

Notation

This section describes the notation used in the specification.

Unicode characters

Unicode characters in this document are described in two ways. One is to supply the character inline in the text, between double quote marks. The other is to write a capital U, followed by a plus sign, followed by a four-digit hexadecimal number representing the character’s code point. As an example of both, the first character in the name QL is “Q” (U+0051).

Grammars

The syntactic forms of QL constructs are specified using a modified Backus-Naur Form (BNF). Syntactic forms, including classes of tokens, are named using bare identifiers. Quoted text denotes a token by its exact sequence of characters in the source code.

BNF derivation rules are written as an identifier naming the syntactic element, followed by ::=, followed by the syntax itself.

In the syntax itself, juxtaposition indicates sequencing. The vertical bar (|, U+007C) indicates alternate syntax. Parentheses indicate grouping. An asterisk (*, U+002A) indicates repetition zero or more times, and a plus sign (+, U+002B) indicates repetition one or more times. Syntax followed by a question mark (?, U+003F) indicates zero or one occurrences of that syntax.

Architecture

A QL program consists of a query module defined in a QL file and a number of library modules defined in QLL files that it imports (see “Import directives”). The module in the QL file includes one or more queries (see “Queries”). A module may also include import directives (see “Import directives”), non-member predicates (see “Non-member predicates”), class definitions (see “Classes”), and module definitions (see “Modules”).

QL programs are interpreted in the context of a database and a library path . The database provides a number of definitions: database types (see “Types”), entities (see “Values”), built-in predicates (see “Built-ins”), and the database content of built-in predicates and external predicates (see “Evaluation”). The library path is a sequence of file-system directories that hold QLL files.

A QL program can be evaluated (see “Evaluation”) to produce a set of tuples of values (see “Values”).

For a QL program to be valid, it must conform to a variety of conditions that are described throughout this specification; otherwise the program is said to be invalid. An implementation of QL must detect all invalid programs and refuse to evaluate them.

Library path

The library path is an ordered list of directory locations. It is used for resolving module imports (see “Module resolution”). The library path is not strictly speaking a core part of the QL language, since different implementations of QL construct it in slightly different ways. Most QL tools also allow you to explicitly specify the library path on the command line for a particular invocation, though that is rarely done, and only useful in very special situations. This section describes the default construction of the library path.

First, determine the query directory of the .ql file being compiled. Starting with the directory containing the .ql file, and walking up the directory structure, each directory is checked for a file called qlpack.yml or codeql-pack.yml. The first directory where such a file is found is the query directory. If there is no such directory, the directory of the .ql file itself is the query directory.

A qlpack.yml file defines a CodeQL pack. The content of a qlpack.yml file is described in the CodeQL CLI documentation. codeql-pack.yml is an alias for qlpack.yml.

The CodeQL CLI and tools based on it (such as, GitHub code scanning and the CodeQL extension for Visual Studio Code) construct a library path using CodeQL packs. For each CodeQL pack added to the library path, the CodeQL packs named in its dependencies will be subsequently added to the library path, and the process continues until all packs have been resolved. The actual library path consists of the root directories of the selected CodeQL packs. This process depends on a mechanism for finding CodeQL packs by pack name and version, as described in the CodeQL CLI documentation.

When the query directory contains neither a qlpack.yml nor codeql-pack.yml file, it is considered to be a CodeQL pack with no name and no library dependencies. This causes the library path to consist of only the query directory itself. This is not generally useful, but it suffices for running toy examples of QL code that don’t use information from the database.

Name resolution

All modules have six environments that dictate name resolution. These are multimaps of keys to declarations.

The environments are:

  • The module environment, whose keys are module names and whose values are modules.
  • The type environment, whose keys are type names and whose values are types.
  • The predicate environment, whose keys are pairs of predicate names and arities and whose values are predicates.
  • The module signature environment, whose keys are module signature names and whose values are module signatures.
  • The type signature environment, whose keys are type signature names and whose values are type signatures.
  • The predicate signature environment, whose keys are pairs of predicate signature names and arities and whose values are predicate signatures.

For each module, some namespaces are enforced to be disjoint:

  • No keys may be shared between the module namespace and the module signature namespace.
  • No keys may be shared between the type namespace and the type signature namespace.
  • No keys may be shared between the module namespace and the type signature namespace.
  • No keys may be shared between the type namespace and the module signature namespace.
  • No keys may be shared between the predicate namespace and the predicate signature namespace.
  • No keys may be shared between the module signature namespace and the type signature namespace.

If not otherwise specified, then the environment for a piece of syntax is the same as the environment of its enclosing syntax.

When a key is resolved in an environment, if there is no value for that key, then the program is invalid.

Environments may be combined as follows:

  • Union. This takes the union of the entry sets of the two environments.
  • Overriding union. This takes the union of two environments, but if there are entries for a key in the first map, then no additional entries for that key are included from the second map.

A definite environment has only values that are equal modulo weak aliasing for each key.

Global environments

The global module environment has a single entry QlBuiltins.

The global type environment has entries for the primitive types int, float, string, boolean, and date.

The global predicate environment includes all the built-in classless predicates.

The three global signature environments are empty.

Database schema environments

The database schema type environment has entries for types declared in the database schema.

The database schema predicate environment has entries for extensional predicates declared in the database schema.

The program is invalid if any of these environments is not definite.

Module environments

For each of modules, types, predicates, module signatures, type signatures, and predicates signatures, we distinguish four environments: publically declared, privately declared, exported, and visible. These are defined as follows (with X denoting the type of entity we are currently considering):

  • The privately declared X environment of a module is the multimap of X declarations and aliases in the module itself that are annotated as private.

  • The publically declared X environment of a module is the multimap of X declarations and aliases in the module itself that are not annotated as private.

  • The exported X environment of a module is the union of

    1. its publically declared X environment, and
    2. for each module which the current module directly imports (excluding private imports - see “Import directives”): all entries from the exported X environment that have a key not present in the publically declared X environment of the current module, and
    3. if X is predicate, then for each module signature S that is implemented by the current module: an entry for each module signature default predicate in S that does not have the same name and arity as any of the entries in the publically declared predicate environment of the current module.
  • The visible X environment of a module is the union of

    1. its exported X environment, and
    2. its privately declared X environment, and
    3. the global X environment, and
    4. for each module which the current module privately imports: all entries from the exported X environment that have a key not present in the publically declared X environment of the current module, and
    5. if there is an enclosing module: all entries from the visible X environment of the enclosing module that have a key not present in the publically declared X environment of the current module, and
    6. if there is no enclosing module and X is either type or predicate: all entries from the database schema X environment that have a key not present in the publically declared X environment of the current module, and
    7. all parameters of the current module that are of type X.

The program is invalid if any of these environments is not definite.

Module definitions may be recursive, so the module environments are defined as the least fixed point of the operator given by the above definition. Since all the operations involved are monotonic, this fixed point exists and is unique.

Modules

Module definitions

A QL module definition has the following syntax:

module ::= annotation* "module" modulename parameters? implements? "{" moduleBody "}"

parameters ::= "<" signatureExpr parameterName ("," signatureExpr parameterName)* ">"

implements ::= "implements" moduleSignatureExpr ("," moduleSignatureExpr)*

moduleBody ::= (import | predicate | class | module | signature | alias | select)*

A module definition extends the current module’s declared module environment with a mapping from the module name to the module definition.

QL files and QLL files consist of simply a module body without a name and surrounding braces:

ql ::= moduleBody

QL files and QLL files define a module corresponding to the file, whose name is the same as the filename.

Kinds of modules

A module may be:

  • A declared module, if it is defined by the module or ql grammar rules.
  • A non-declared module, if it is not a declared module.

A declared module may be:

  • A file module, if it is defined implicitly by a QL file or a QLL file.
  • A query module, if it is defined implicitly by a QL file.
  • A library module, if it is not a query module.

A non-declared module may be:

A query module must contain one or more queries.

Import directives

An import directive refers to a module expression:

import ::= annotations "import" importModuleExpr ("as" modulename)?

importModuleExpr ::= importModuleId arguments?

importModuleId ::= qualId | importModuleExpr "::" modulename

qualId ::= simpleId | qualId "." simpleId

arguments ::= "<" argument ("," argument)* ">"

argument ::= moduleExpr | type | predicateRef "/" int

An import directive may optionally name the imported module using an as declaration. If a name is defined, then the import directive adds to the declared module environment of the current module a mapping from the name to the declaration of the imported module. Otherwise, the current module directly imports the imported module.

Module resolution

Module identifiers are resolved to modules as follows.

For simple identifiers:

  • First, the identifier is resolved as a one-segment qualified identifier (see below).
  • If this fails, the identifier is resolved in the current module’s visible module environment.

For selection identifiers (a::b):

  • The qualifier of the selection (a) is resolved as a module, and then the name (b) is resolved in the exported module environment of the qualifier module.

For qualified identifiers (a.b):

  • Build up a list of candidate search paths, consisting of the current file’s directory, then the query directory of the current file, and finally each of the directories on the library path (in order).
  • Determine the first candidate search path that has a matching QLL file for the import directive’s qualified name. A QLL file in a candidate search path is said to match a qualified name if, starting from the candidate search path, there is a subdirectory for each successive qualifier in the qualified name, and the directory named by the final qualifier contains a file whose base name matches the qualified name’s base name, with the addition of the file extension .qll. The file and directory names are matched case-sensitively, regardless of whether the filesystem is case-sensitive or not.
  • The resolved module is the module defined by the selected candidate search path.

A qualified module identifier is only valid within an import.

Module expressions contain a module identifier and optional arguments. If arguments are present, the module expression instantiates the module that the identifier resolves to (see Parameterized modules).

Module expressions cannot refer to Parameterized modules. Instead, parameterized modules must always be fully instantiated when they are referenced.

Parameterized modules

Modules with parameters are called parameterized modules. A declared module has parameters if and only if it is a library module and its declaration uses the parameters syntax.

Parameterized modules can be instantiated with arguments that match the signatures of the parameters:

  • Given a type signature, the corresponding argument must be a type that transitively extends the specified extends types and is a transitive subtype of the specified instanceof types.
  • Given a predicate signature, the corresponding argument must be a predicate with exactly matching relational types.
  • Given a module signature, the corresponding argument must be a module that exports all the specified type and predicate members. Furthermore, the module must be declared as matching the module signature via the implements syntax.

The result of instantiating a parameterized module is an instantiated module. The parameterized module is called the underlying module of the instantiated module.

Instantiation-relative and instantiation-nested entities

Given an instantiated module, every entity has a corresponding entity called the instantiation-relative entity, which is determined as follows:

  • If the entity is the underlying module, its instantiation-relative entity is the instantiated module.
  • If the entity is a parameter of the underlying module, its instantiation-relative entity is the corresponding argument.
  • If the entity is declared inside the underlying module or its nested modules, its instantiation-relative entity is an instantiation-nested entity that is generated by the module instantiation. Parameters of any modules that are nested inside the underlying module are considered declared inside the module for this purpose.
  • Otherwise, the entity’s instantiation-relative entity is the initial entity itself.

When the instantiation-relative entity of an entity is an instantiation-nested entity, then the initial entity is called the underlying nested entity of the instantiation-nested entity*, the instantiated module is called the instantiation root of the instantiation-nested entity, and the underlying module is called the underlying root of the instantiation-nested entity.

The components of an instantiation-nested entity are the instantiation-relative entities of the components of its underlying nested entity. Among other things, this applies to:

  • values in the exported environments of instantiation-nested modules,
  • relational types of instantiation-nested predicates and predicate signatures,
  • required signatures of instantiation-nested parameters,
  • parameters of an instantiation-nested parameterized module,
  • fields and member predicates of instantiation-nested dataclasses.

Given an instantiated module, any alias in the program has a corresponding alias called the instantiation-relative alias, which targets the instantiation-relative entity.

Applicative instantiation

Every entity has an underlying completely uninstantiated entity that is determined as follows:

  • If the entity is an instantiated module, its underlying completely uninstantiated entity is the underlying completely uninstantiated entity of the underlying module.
  • If the entity is an instantiation-nested entity, its underlying completely uninstantiated entity is the underlying completely uninstantiated entity of the underlying nested entity.
  • Otherwise, its underlying completely uninstantiated entity is the entity itself.

An entity is called completely uninstantiated entity if it is its own underlying completely uninstantiated entity.

Every completely uninstantiated entity has a relevant set of parameters, which is the set of all parameters of all the modules that the entity is transitively nested inside. For entities that are not nested inside any modules, the relevant set of parameters is empty.

Note that the relevant set of parameters by construction contains only completely uninstantiated parameters.

For a completely uninstantiated parameter, the bottom-up instantiation-resolution relative to an entity is defined as:

  • If the entity is an instantiated module or an instantiation-nested entity, the bottom-up instantiation-resolution is the instantiation-relative entity of the bottom-up instantiation-resolution relative to the underlying module.
  • Otherwise, the bottom-up instantiation-resolution is the parameter itself.

An entity is called fully instantiated if none of the bottom-up instantiation-resolutions of the parameters in the relevant set of parameters of the entity’s underlying completely uninstantiated entity are parameters.

Two instantiated modules or two instantiation-nested entities are considered equivalent if they have the same underlying completely uninstantiated entity and each parameter in its relevant set of parameters has bottom-up instantiation-resolution*s relative both *instantiated module*s that are *equivalent modulo weak aliases.

Module instantiation is applicative, meaning that equivalent instantiated modules and equivalent instantiation-nested entities are indistinguishable.

Module references and active modules

A module M references another module N if any of the following holds:

  • M imports N.
  • M defines N.
  • N is M’s enclosing module.

In a QL program, the active modules are the modules which are referenced transitively by the query module.

Types

QL is a typed language. This section specifies the kinds of types available, their attributes, and the syntax for referring to them.

Kinds of types

Types in QL are either primitive types, database types, class types, character types, class domain types, type parameters, or instantiation-nested types.

The primitive types are boolean, date, float, int, and string.

Database types are supplied as part of the database. Each database type has a name, which is an identifier starting with an at sign (@, U+0040) followed by lower-case letter. Database types have some number of base types, which are other database types. In a valid database, the base types relation is non-cyclic.

Class types are defined in QL, in a way specified later in this document (see “Classes”). Each class type has a name that is an identifier starting with an upper-case letter. Each class type has one or more base types, which can be any kind of type except a class domain type. A class type may be declared abstract.

Any class in QL has an associated class domain type and an associated character type.

Within the specification the class type for C is written C.class, the character type is written C.C and the domain type is written C.extends. However the class type is still named C.

Type references

With the exception of class domain types and character types (which cannot be referenced explicitly in QL source), a reference to a type is written as the name of the type. In the case of database types, the name includes the at sign (@, U+0040).

type ::= (moduleExpr "::")? classname | dbasetype | "boolean" | "date" | "float" | "int" | "string"

moduleExpr ::= moduleId arguments?

moduleId ::= modulename | moduleExpr "::" modulename

A type reference is resolved to a type as follows:

  • If it is a selection identifier (for example, a::B), then the qualifier (a) is resolved as a module (see “Module resolution”). The identifier (B) is then resolved in the exported type environment of the qualifier module.
  • Otherwise, the identifier is resolved in the current module’s visible type environment.

Relations among types

Types are in a subtype relationship with each other. Type A is a subtype of type B if one of the following is true:

  • A and B are the same type.
  • There is some type C, where A is a subtype of C and C is a subtype of B.
  • A and B are database types, and B is a base type of A.
  • A is the character type of C, and B is the class domain type of C.
  • A is a class type, and B is the character type of A.
  • A is a class domain type, and B is a base type of the associated class type.
  • A is int and B is float.

Supertypes are the converse of subtypes: A is a supertype of B if B is a subtype of A.

Types A and B are compatible with each other if they either have a common supertype, or they each have some supertype that is a database type.

Typing environments

A typing environment is a finite map of variables to types. Each variable in the map is either an identifier or one of two special symbols: this, and result.

Most forms of QL syntax have a typing environment that applies to them. That typing environment is determined by the context the syntax appears in.

Note that this is distinct from the type environment, which is a map from type names to types.

Active types

In a QL program, the active types are those defined in active modules. In the remainder of this specification, any reference to the types in the program refers only to the active types.

Values

Values are the fundamental data that QL programs compute over. This section specifies the kinds of values available in QL, specifies the sorting order for them, and describes how values can be combined into tuples.

Kinds of values

There are six kinds of values in QL: one kind for each of the five primitive types, and entities. Each value has a type.

A boolean value is of type boolean, and may have one of two distinct values: true or false.

A date value is of type date. It encodes a time and a date in the Gregorian calendar. Specifically, it includes a year, a month, a day, an hour, a minute, a second, and a millisecond, each of which are integers. The year ranges from -16777216 to 16777215, the month from 0 to 11, the day from 1 to 31, the hour from 0 to 23, the minutes from 0 to 59, the seconds from 0 to 59, and the milliseconds from 0 to 999.

A float value is of type float. Each float value is a binary 64-bit floating-point value as specified in IEEE 754.

An integer value is of type int. Each value is a 32-bit two’s complement integer.

A string is a finite sequence of 16-bit characters. The characters are interpreted as Unicode code points.

The database includes a number of opaque entity values. Each such value has a type that is one of the database types, and an identifying integer. An entity value is written as the name of its database type followed by its identifying integer in parentheses. For example, @tree(12), @person(16), and @location(38132) are entity values. The identifying integers are left opaque to programmers in this specification, so an implementation of QL is free to use some other set of countable labels to identify its entities.

Ordering

Values in general do not have a specified ordering. In particular, entity values have no specified ordering with entities or any other values. Primitive values, however, have a total ordering with other primitive values in the same type. Primitives types and their subtypes are said to be orderable.

For booleans, false is ordered before true.

For dates, the ordering is chronological.

For floats, the ordering is as specified in IEEE 754 when one exists, except that NaN is considered equal to itself and is ordered after all other floats, and negative zero is considered to be strictly less than positive zero.

For integers, the ordering is as for two’s complement integers.

For strings, the ordering is lexicographic.

Tuples

Values can be grouped into tuples in two different ways.

An ordered tuple is a finite, ordered sequence of values. For example, (1, 2, "three") is an ordered sequence of two integers and a string.

A named tuple is a finite map of variables to values. Each variable in a named tuple is either an identifier, this, or result.

A variable declaration list provides a sequence of variables and a type for each one:

var_decls ::= (var_decl ("," var_decl)*)?
var_decl ::= type lowerId

A valid variable declaration list must not include two declarations with the same variable name. Moreover, if the declaration has a typing environment that applies, it must not use a variable name that is already present in that typing environment.

An extension of a named tuple for a given variable declaration list is a named tuple that additionally maps each variable in the list to a value. The value mapped by each new variable must be in the type that is associated with that variable in the given list; see “The store” for the definition of a value being in a type.

The store

QL programs evaluate in the context of a store. This section specifies several definitions related to the store.

A fact is a predicate or type along with a named tuple. A fact is written as the predicate name or type name followed immediately by the tuple. Here are some examples of facts:

successor(fst: 0, snd:1)
Tree.toString(this:@method_tree(12), result:"def println")
Location.class(this:@location(43))
Location.getURL(this: @location(43), result:"file:///etc/hosts:2:0:2:12")

A store is a mutable set of facts. The store can be mutated by adding more facts to it.

An named tuple directly satisfies a predicate or type with a given tuple if there is a fact in the store with the given tuple and predicate or type.

A value v is in a type t under any of the following conditions:

  • The type of v is t and t is a primitive type.
  • There is a tuple with this component v that directly satisfies t.

An ordered tuple v directly satisfies a predicate with a given tuple if there is a fact in the store with the given predicate and a named tuple v' such that taking the ordered tuple formed by the this component of v' followed by the component for each argument equals the ordered tuple.

An ordered tuple satisfies a predicate p under the following circumstances. If p is not a member predicate, then the tuple satisfies the predicate whenever the named tuple satisfies the tuple.

Otherwise, the tuple must be the tuple of a fact in the store with predicate q, where q shares a root definition with p. The first element of the tuple must be in the type before the dot in q, and there must be no other predicate that overrides q such that this is true (see “Classes” for details on overriding and root definitions).

An ordered tuple (a0, an) satisfies the + closure of a predicate if there is a sequence of binary tuples (a0, a1), (a1, a2), …, (an-1, an) that all satisfy the predicate. An ordered tuple (a, b) satisfies the * closure of a predicate if it either satisfies the + closure, or if a and b are the same, and if moreover they are in each argument type of the predicate.

Lexical syntax

QL and QLL files contain a sequence of tokens that are encoded as Unicode text. This section describes the tokenization algorithm, the kinds of available tokens, and their representation in Unicode.

Some kinds of tokens have an identifier given in parentheses in the section title. That identifier, if present, is a terminal used in grammar productions later in the specification. Additionally, the “Identifiers” section gives several kinds of identifiers, each of which has its own grammar terminal.

Tokenization

Source files are interpreted as a sequence of tokens according to the following algorithm. First, the longest-match rule, described below, is applied starting at the beginning of the file. Second, all whitespace tokens and comments are discarded from the sequence.

The longest-match rule is applied as follows. The first token in the file is the longest token consisting of a contiguous sequence of characters at the beginning of the file. The next token after any other token is the longest token consisting of contiguous characters that immediately follow any previous token.

If the file cannot be tokenized in its entirety, then the file is invalid.

Whitespace

A whitespace token is a sequence of spaces (U+0020), tabs (U+0009), carriage returns (U+000D), and line feeds (U+000A).

Comments

There are two kinds of comments in QL: one-line and multiline.

A one-line comment is two slash characters (/, U+002F) followed by any sequence of characters other than line feeds (U+000A) and carriage returns (U+000D). Here is an example of a one-line comment:

// This is a comment

A multiline comment is a comment start, followed by a comment body, followed by a comment end. A comment start is a slash (/, U+002F) followed by an asterisk (*, U+002A), and a comment end is an asterisk followed by a slash. A comment body is any sequence of characters that does not include a comment end and does not start with an asterisk. Here is an example multiline comment:

/*
  It was the best of code.
  It was the worst of code.
  It had a multiline comment.
*/

QLDoc (qldoc)

A QLDoc comment is a qldoc comment start, followed by a qldoc comment body, followed by a qldoc comment end. A comment start is a slash (/, U+002F) followed by two asterisks (*, U+002A), and a qldoc comment end is an asterisk followed by a slash. A qldoc comment body is any sequence of characters that does not include a comment end. Here is an example QLDoc comment:

/**
  It was the best of code.
  It was the worst of code.
  It had a qldoc comment.
*/

The “content” of a QLDoc comment is the comment body of the comment, omitting the initial /**, the trailing */, and the leading whitespace followed by * on each internal line.

For more information about how the content is interpreted, see “QLDoc” below.

Keywords

The following sequences of characters are keyword tokens:

and
any
as
asc
avg
boolean
by
class
concat
count
date
desc
else
exists
extends
false
float
forall
forex
from
if
implies
import
in
instanceof
int
max
min
module
newtype
none
not
or
order
predicate
rank
result
select
strictconcat
strictcount
strictsum
string
sum
super
then
this
true
unique
where

Operators

The following sequences of characters are operator tokens:

<
<=
=
>
>=
_
-
,
;
!=
/
.
..
(
)
[
]
{
}
*
%
+
|

Identifiers

An identifier is an optional “@” sign (U+0040) followed by a sequence of identifier characters. Identifier characters are lower-case ASCII letters (a through z, U+0061 through U+007A), upper-case ASCII letters (A through Z, U+0041 through U+005A), decimal digits (0 through 9, U+0030 through U+0039), and underscores (_, U+005F). The first character of an identifier other than any “@” sign must be a letter.

An identifier cannot have the same sequence of characters as a keyword, nor can it be an “@” sign followed by a keyword.

Here are some examples of identifiers:

width
Window_width
window5000_mark_II
@expr

There are several kinds of identifiers:

  • lowerId: an identifier that starts with a lower-case letter.
  • upperId: an identifier that starts with an upper-case letter.
  • atLowerId: an identifier that starts with an “@” sign and then a lower-case letter.

Identifiers are used in following syntactic constructs:

simpleId            ::= lowerId | upperId
modulename          ::= simpleId
moduleSignatureName ::= upperId
classname           ::= upperId
dbasetype           ::= atLowerId
predicateRef        ::= (moduleExpr "::")? literalId
signatureExpr       ::= (moduleExpr "::")? simpleId ("/" Integer | arguments)?;
predicateName       ::= lowerId
parameterName       ::= simpleId
varname             ::= lowerId
literalId           ::= lowerId | atLowerId

Integer literals (int)

An integer literal is a possibly negated sequence of decimal digits (0 through 9, U+0030 through U+0039). Here are some examples of integer literals:

0
42
123
-2147483648

Float literals (float)

A floating-point literals is a possibly negated two non-negative integers literals separated by a dot (., U+002E). Here are some examples of float literals:

0.5
2.0
123.456
-100.5

String literals (string)

A string literal denotes a sequence of characters. It begins and ends with a double quote character (U+0022). In between the double quotes are a sequence of string character indicators, each of which indicates one character that should be included in the string. The string character indicators are as follows.

  • Any character other than a double quote (U+0022), backslash (U+005C), line feed (U+000A), carriage return (U+000D), or tab (U+0009). Such a character indicates itself.
  • A backslash (U+005C) followed by one of the following characters:
    • Another backslash (U+005C), in which case a backslash character is indicated.
    • A double quote (U+0022), in which case a double quote is indicated.
    • The letter “n” (U+006E), in which case a line feed (U+000A) is indicated.
    • The letter “r” (U+0072), in which case a carriage return (U+000D) is indicated.
    • The letter “t” (U+0074), in which case a tab (U+0009) is indicated.

Here are some examples of string literals:

"hello"
"He said, \"Logic clearly dictates that the needs of the many...\""

Annotations

Various kinds of syntax can have annotations applied to them. Annotations are as follows:

annotations ::= annotation*

annotation ::= simpleAnnotation | argsAnnotation

simpleAnnotation ::= "abstract"
                 |   "cached"
                 |   "external"
                 |   "extensible"
                 |   "final"
                 |   "transient"
                 |   "library"
                 |   "private"
                 |   "deprecated"
                 |   "override"
                 |   "additional"
                 |   "query"

argsAnnotation ::= "pragma" "[" ("inline" | "inline_late" | "noinline" | "nomagic" | "noopt" | "assume_small_delta") "]"
               |   "language" "[" "monotonicAggregates" "]"
               |   "bindingset" "[" (variable ( "," variable)*)? "]"

Each simple annotation adds a same-named attribute to the syntactic entity it precedes. For example, if a class is preceded by the abstract annotation, then the class is said to be abstract.

A valid annotation list may not include the same simple annotation more than once, or the same parameterized annotation more than once with the same arguments. However, it may include the same parameterized annotation more than once with different arguments.

Simple annotations

The following table summarizes the syntactic constructs which can be marked with each annotation in a valid program; for example, an abstract annotation preceding a character is invalid.

Annotation Classes Characters Member predicates Non-member predicates Imports Fields Modules Aliases Signatures
abstract yes   yes            
cached yes yes yes yes     yes    
external       yes          
extensible       yes          
final yes   yes     yes   (yes)  
transient       yes          
library (yes)                
private yes   yes yes yes yes yes yes yes
deprecated yes   yes yes yes yes yes yes yes
override     yes     yes      
additional yes     yes     yes yes yes
query       yes       yes  

The library annotation is only usable within a QLL file, not a QL file. The final annotation is usable on type aliases, but not on module aliases and predicate aliases.

Annotations on aliases apply to the name introduced by the alias. An alias may, for example, have different privacy to the name it aliases.

Parameterized annotations

Parameterized annotations take some additional arguments.

The parameterized annotation pragma supplies compiler pragmas, and may be applied in various contexts depending on the pragma in question.

Pragma Classes Characters Member predicates Non-member predicates Imports Fields Modules Aliases
inline   yes yes yes        
inline_late   yes yes yes        
noinline   yes yes yes        
nomagic   yes yes yes        
noopt   yes yes yes        
assume_small_delta   yes yes yes        

The parameterized annotation language supplies language pragmas which change the behavior of the language. Language pragmas apply at the scope level, and are inherited by nested scopes.

Pragma Classes Characters Member predicates Non-member predicates Imports Fields Modules Aliases
monotonicAggregates yes yes yes yes     yes  

A binding set for a predicate is a subset of the predicate’s arguments such that if those arguments are bound (restricted to a finite range of values), then all of the predicate’s arguments are bound.

The parameterized annotation bindingset can be applied to a predicate (see “Non-member predicates” and “Members”) to specify a binding set.

This annotation accepts a (possibly empty) list of variable names as parameters. The named variables must all be arguments of the predicate, possibly including this for characteristic predicates and member predicates, and result for predicates that yield a result.

In the default case where no binding sets are specified by the user, then it is assumed that there is precisely one, empty binding set - that is, the body of the predicate must bind all the arguments.

Binding sets are checked by the QL compiler in the following way:

  1. It assumes that all variables mentioned in the binding set are bound.
  2. It checks that, under this assumption, all the remaining argument variables are bound by the predicate body.

A predicate may have several different binding sets, which can be stated by using multiple bindingset annotations on the same predicate.

Annotation Classes Characters Member predicates Non-member predicates Imports Fields Modules Aliases Signatures
bindingset   yes yes yes         (yes)

The bindingset pragma is usable with type signatures and predicate signatures, but not with module signatures.

QLDoc

QLDoc is used for documenting QL entities and bindings. QLDoc that is used as part of the declaration is said to be declared.

Ambiguous QLDoc

If QLDoc can be parsed as part of a file module or as part of the first declaration in the file then it is parsed as part of the first declaration.

Inheriting QLDoc

If no QLDoc is provided then it may be inherited.

In the case of an alias then it may be inherited from the right-hand side of the alias.

In the case of a member predicate we collect all member predicates that it overrides with declared QLDoc. If there is a member predicate in that collection that overrides every other member predicate in that collection, then the QLDoc of that member predicate is used as the QLDoc.

In the case of a field we collect all fields that it overrides with declared QLDoc. If there is a field in that collection that overrides every other field in that collection, then the QLDoc of that field is used as the QLDoc.

Content

The content of a QLDoc comment is interpreted as CommonMark, with the following extensions:

  • Automatic interpretation of links and email addresses.
  • Use of appropriate characters for ellipses, dashes, apostrophes, and quotes.

The content of a QLDoc comment may contain metadata tags as follows:

The tag begins with any number of whitespace characters, followed by an @ sign. At this point there may be any number of non-whitespace characters, which form the key of the tag. Then, a single whitespace character which separates the key from the value. The value of the tag is formed by the remainder of the line, and any subsequent lines until another @ tag is seen, or the end of the content is reached. Any sequence of consecutive whitespace characters in the value are replaced by a single space.

Metadata

If the query file starts with whitespace followed by a QLDoc comment, then the tags from that QLDoc comment form the query metadata.

Top-level entities

Modules include five kinds of top-level entity: predicates, classes, modules, aliases, signatures, and select clauses.

Non-member predicates

A predicate is declared as a sequence of annotations, a head, and an optional body:

predicate ::= qldoc? annotations head optbody

A predicate definition adds a mapping from the predicate name and arity to the predicate declaration to the current module’s declared predicate environment.

When a predicate is a top-level clause in a module, it is called a non-member predicate. See below for “Member predicates.”

A valid non-member predicate can be annotated with cached, deprecated, external, transient, private, and query. Note, the transient annotation can only be applied if the non-member predicate is also annotated with external.

The head of the predicate gives a name, an optional result type, and a sequence of variables declarations that are arguments:

head ::= ("predicate" | type) predicateName "(" var_decls ")"

The body of a predicate is of one of three forms:

optbody ::= ";"
        |  "{" formula "}"
        |  "=" literalId "(" (predicateRef "/" int ("," predicateRef "/" int)*)? ")" "(" (exprs)? ")"

In the first form, with just a semicolon, the predicate is said to not have a body. In the second form, the body of the predicate is the given formula (see “Formulas”). In the third form, the body is a higher-order relation.

A valid non-member predicate must have a body, either a formula or a higher-order relation, unless it is external, in which case it must not have a body.

The typing environment for the body of the formula, if present, maps the variables in the head of the predicate to their associated types. If the predicate has a result type, then the typing environment also maps result to the result type.

Classes

A class definition has the following syntax:

class ::= qldoc? annotations "class" classname ("extends" type ("," type)*)? ("instanceof" type ("," type)*)? "{" member* "}"

The identifier following the class keyword is the name of the class.

The types specified after the extends keyword are the base types of the class.

The types specified after the instanceof keyword are the instanceof types of the class.

A class type is said to final inherit from base types that are final or referenced through final aliases, and a class type is said to inherit from its other base types. In addition, inheritance is transitive:

  • If a type A inherits from a type B, and B inherits from a type C, then A inherits from C.
  • If a type A final inherits from a type B, and B inherits from a type C, then A final inherits from C.
  • If a type A inherits from a type B, and B final inherits from a type C, then A final inherits from C.
  • If a type A final inherits from a type B, and B final inherits from a type C, then A final inherits from C.

A class adds a mapping from the class name to the class declaration to the current module’s declared type environment.

A valid class can be annotated with abstract, final, library, and private. Any other annotation renders the class invalid.

A valid class may not inherit from itself, or from more than one primitive type. The set of types that a valid class inherits from must be disjoint from the set of types that it final inherits from.

A valid class must have at least one base type or instanceof type.

Class dependencies

The program is invalid if there is a cycle of class dependencies.

The following are class dependencies:

  • C depends on C.C
  • C.C depends on C.extends
  • If C is abstract then it depends on all classes D such that C is a base type of D and D inherits from C.
  • C.extends depends on D.D for each base type D of C.
  • C.extends depends on D for each instanceof type D of C.

Class environments

For each of member predicates and fields a class inherits and declares, and exports an environment. These are defined as follows (with X denoting the type of entity we are currently considering):

  • The inherited X environment of a class is the union of the exported X environments of types it inherits from, excluding any elements that are overridden by another element.
  • The declared X environment of a class is the multimap of X declarations in the class itself.
  • The exported X environment of a class is the overriding union of its declared X environment (excluding private declaration entries) with its inherited X environment.
  • The visible X environment is the overriding union of the declared X environment and the inherited X environment.

The program is invalid if any of these environments is not definite.

For each of member predicates and fields a domain type exports an environment. We say the exported X extends environment is the union of the exported X environments of types the class inherits from, excluding any elements that are overridden by another element. We say the exported X instanceof environement is the union of the exported X environments of types that a instanceof type inherits from, excluding any elements that are overridden by another element. The exported X environment of the domain type is the union of the exported X extends environment and the exported X instanceof environement.

Members

Each member of a class is either a character, a predicate, or a field:

member ::= character | predicate | field
character ::= qldoc? annotations classname "(" ")" "{" formula "}"
field ::= qldoc? annotations var_decl ";"

Characters

A valid character must have the same name as the name of the class. A valid class has at most one character provided in the source code.

A valid character can be annotated with cached. Any other annotation renders the character invalid.

Member predicates

A predicate that is a member of a class is called a member predicate. The name of the predicate is the identifier just before the open parenthesis.

A member predicate adds a mapping from the predicate name and arity to the predicate declaration in the class’s declared member predicate environment.

A valid member predicate can be annotated with abstract, cached, final, private, deprecated, and override.

If a type is provided before the name of the member predicate, then that type is the result type of the predicate. Otherwise, the predicate has no result type. The types of the variables in the var_decls are called the predicate’s argument types.

A member predicate p with enclosing class C overrides a member predicate p' with enclosing class D when p is annotated overrride, C inherits from D, p' is visible in C, p' is not final, and both p and p' have the same name and the same arity. An overriding predicate must have the same sequence of argument types as any predicates which it overrides, otherwise the program is invalid.

A member predicate p with enclosing class C shadows a member predicate p' with enclosing class D when C final inherits from D, p' is visible in C, and both p and p' have the same name and the same arity. Additionally, a member predicate p with enclosing class C shadows a member predicate p' with enclosing class D when C inherits from D, p' is visible in C, p' is final, and both p and p' have the same name and the same arity.

Member predicates have one or more root definitions. If a member predicate overrides no other member predicate, then it is its own root definition. Otherwise, its root definitions are those of any member predicate that it overrides.

A valid member predicate must have a body unless it is abstract or external, in which case it must not have a body.

A valid member predicate must override another member predicate if it is annotated override.

When member predicate p overrides member predicate q, either p and q must both have a result type, or neither of them may have a result type. If they do have result types, then the result type of p must be a subtype of the result type of q. q may not be a final predicate. If p is abstract, then q must be as well.

A class may not inherit from a class with an abstract member predicate unless it either includes a member predicate overriding that abstract predicate, or it inherits from another class that does.

A valid class must include a non-private predicate named toString with no arguments and a result type of string, or it must inherit from a class that does.

A valid class may not inherit from two different classes that include a predicate with the same name and number of arguments, unless either one of the predicates overrides or shadows the other, or the class defines a predicate that overrides or shadows both of them.

A valid class may not final inherit from two different classes that include a predicate with the same name and number of arguments, unless either one of the predicates overrides or shadows the other, or the class defines a predicate that shadows both of them.

The typing environment for a member predicate or character is the same as if it were a non-member predicate, except that it additionally maps this to a type and also maps any fields on a class to a type. If the member is a character, then the typing environment maps this to the class domain type of the class. Otherwise, it maps this to the class type of the class itself. The typing environment also maps any field to the type of the field.

Fields

A field declaration introduces a mapping from the field name to the field declaration in the class’s declared field environment.

A field f with enclosing class C overrides a field f' with enclosing class D when f is annotated override, C inherits from D, p' is visible in C, p' is not final, and both p and p' have the same name.

A field f with enclosing class C shadows a field f' with enclosing class D when C final inherits from D, p' is visible in C, and both p and p' have the same name. Additionally, a field f with enclosing class C shadows a field f' with enclosing class D when C inherits from D, p' is visible in C, p' is final, and both p and p' have the same name.

A valid class may not inherit from two different classes that include a field with the same name, unless either one of the fields overrides or shadows the other, or the class defines a field that overrides or shadows both of them.

A valid class may not final inherit from two different classes that include a field with the same name, unless either one of the fields overrides or shadows the other, or the class defines a field that shadows both of them.

A valid field must override another field if it is annotated override.

When field f overrides field g the type of f must be a subtype of the type of g. f may not be a final field.

Signatures

A signature definition has the following syntax:

signature ::= predicateSignature | typeSignature | moduleSignature

predicateSignature ::= qldoc? annotations "signature" head ";"

typeSignature ::= qldoc? annotations "signature" "class" classname ("extends" type ("," type)*)? (";" | "{" signaturePredicate* "}")

moduleSignature ::= qldoc? annotation* "signature" "module" moduleSignatureName parameters? "{" moduleSignatureBody "}"

moduleSignatureBody ::= (signaturePredicate | defaultPredicate | signatureType)*

signaturePredicate ::= qldoc? annotations head ";"

defaultPredicate ::= qldoc? annotations "default" head "{" formula "}"

signatureType ::= qldoc? annotations "class" classname ("extends" type ("," type)*)? "{" signaturePredicate* "}"

A predicate signature definition extends the current module’s declared predicate signature environment with a mapping from the predicate signature name and arity to the predicate signature definition.

A type signature definition extends the current module’s declared type signature environment with a mapping from the type signature name to the type signature definition.

A module signature definition extends the current module’s declared module signature environment with a mapping from the module signature name to the module signature definition.

Select clauses

A QL file may include at most one select clause. That select clause has the following syntax:

select ::= ("from" var_decls)? ("where" formula)? "select" select_exprs ("order" "by" orderbys)?

A valid QLL file may not include any select clauses.

A select clause is considered to be a declaration of an anonymous predicate whose arguments correspond to the select expressions of the select clause.

The from keyword, if present, is followed by the variables of the formula. Otherwise, the select clause has no variables.

The where keyword, if present, is followed by the formula of the select clause. Otherwise, the select clause has no formula.

The select keyword is followed by a number of select expressions. Select expressions have the following syntax:

as_exprs ::= as_expr ("," as_expr)*
as_expr ::= expr ("as" lowerId)?

The keyword as gives a label to the select expression it is part of. No two select expressions may have the same label. No expression label may be the same as one of the variables of the select clause.

The order keyword, if present, is followed by a number of ordering directives. Ordering directives have the following syntax:

orderbys ::= orderby ("," orderby)*
orderby ::= lowerId ("asc" | "desc")?

Each identifier in an ordering directive must identify exactly one of the select expressions. It must either be the label of the expression, or it must be a variable expression that is equivalent to exactly one of the select expressions. The type of the designated select expression must be a subtype of a primitive type.

No select expression may be specified by more than one ordering directive. See “Ordering” for more information.

Queries

The queries in a QL module are:

  • The select clause, if any, defined in that module.
  • Any predicates annotated with query which are in scope in that module.

The target predicate of the query is either the select clause or the annotated predicate.

Each argument of the target predicate of the query must be of a type which has a toString() member predicate.

Expressions

Expressions are a form of syntax used to denote values. Every expression has a typing environment that is determined by the context where the expression occurs. Every valid expression has a type, as specified in this section, except if it is a don’t-care expression.

Given a named tuple and a store, each expression has one or more values. This section specifies the values of each kind of expression.

There are several kinds of expressions:

exprs ::= expr ("," expr)*

expr ::= dontcare
     |   unop
     |   binop
     |   cast
     |   primary

primary ::= eparen
        |   literal
        |   variable
        |   super_expr
        |   postfix_cast
        |   callwithresults
        |   aggregation
        |   expression_pragma
        |   any
        |   range
        |   setliteral

Parenthesized expressions

A parenthesized expression is an expression surrounded by parentheses:

eparen ::= "(" expr ")"

The type environment of the nested expression is the same as that of the outer expression. The type and values of the outer expression are the same as those of the nested expression.

Don’t-care expressions

A don’t-care expression is written as a single underscore:

dontcare ::= "_"

All values are values of a don’t-care expression.

Literals

A literal expression is as follows:

literal ::= "false" | "true" | int | float | string

The type of a literal expression is the type of the value denoted by the literal: boolean for false or true, int for an integer literal, float for a floating-point literal, or string for a string literal. The value of a literal expression is the same as the value denoted by the literal.

Unary operations

A unary operation is the application of + or - to another expression:

unop ::= "+" expr
     |   "-" expr

The + or - in the operation is called the operator, and the expression is called the operand. The typing environment of the operand is the same as for the unary operation.

For a valid unary operation, the operand must be of type int or float. The operation has the same type as its operand.

If the operator is +, then the values of the expression are the same as the values of the operand. If the operator is -, then the values of the expression are the arithmetic negations of the values of the operand.

Binary operations

A binary operation is written as a left operand followed by a binary operator, followed by a right operand:

binop ::= expr "+" expr
      |   expr "-" expr
      |   expr "*" expr
      |   expr "/" expr
      |   expr "%" expr

The typing environment for the two environments is the same as for the operation. If the operator is +, then either both operands must be subtypes of int or float, or at least one operand must be a subtype of string. If the operator is anything else, then each operand must be a subtype of int or float.

The type of the operation is string if either operand is a subtype of string. Otherwise, the type of the operation is int if both operands are subtypes of int. Otherwise, the type of the operation is float.

If the result is of type string, then the left values of the operation are the values of a “call with results” expression with the left operand as the receiver, toString as the predicate name, and no arguments (see “Calls with results”). Otherwise the left values are the values of the left operand. Likewise, the right values are either the values from calling toString on the right operand, or the values of the right operand as it is.

The binary operation has one value for each combination of a left value and a right value. That value is determined as follows:

  • If the left and right operand types are subtypes of string, then the operation has a value that is the concatenation of the left and right values.
  • Otherwise, if both operand types are subtypes of int, then the value of the operation is the result of applying the two’s-complement 32-bit integer operation corresponding to the QL binary operator.
  • Otherwise, both operand types must be subtypes of float. If either operand is of type int then they are converted to a float. The value of the operation is then the result of applying the IEEE 754 floating-point operator that corresponds to the QL binary operator: addition for +, subtraction for -, multiplication for *, division for /, or remainder for %.

Variables

A variable has the following syntax:

variable ::= varname | "this" | "result"

A valid variable expression must occur in the typing environment. The type of the variable expression is the same as the type of the variable in the typing environment.

The value of the variable is the value of the variable in the named tuple.

Super

A super expression has the following syntax:

super_expr ::= "super" | type "." "super"

For a super expression to be valid, the this keyword must have a type and value in the typing environment. The type of the expression is the same as the domain type of the type of this in the typing environment.

The value of a super expression is the same as the value of this in the named tuple.

Casts

A cast expression is a type in parentheses followed by another expression:

cast ::= "(" type ")" expr

The typing environment for the nested expression is the same as for the cast expression. The type of the cast expression is the type between parentheses.

The values of the cast expression are those values of the nested expression that are in the type given within parentheses.

For casts between the primitive float and int types, the above rule means that for the cast expression to have a value, it must be representable as both 32-bit two’s complement integers and 64-bit IEEE 754 floats. Other values will not be included in the values of the cast expression.

Postfix casts

A postfix cast is a primary expression followed by a dot and then a class or primitive type in parentheses:

postfix_cast ::= primary "." "(" type ")"

All the rules for ordinary casts apply to postfix casts: a postfix cast is exactly equivalent to a parenthesized ordinary cast.

Calls with results

An expression for a call with results is of one of two forms:

callwithresult ::= predicateRef (closure)? "(" (exprs)? ")"
               |   primary "." predicateName (closure)? "(" (exprs)? ")"
closure        ::= "*" | "+"

The expressions in parentheses are the arguments of the call. The expression before the dot, if there is one, is the receiver of the call.

The type environment for the arguments is the same as for the call.

A valid call with results resolves to a set of predicates. The ways a call can resolve are as follows:

  • If the call has no receiver and the predicate reference is a simple identifier, then the call is resolved by looking up the predicate reference and arity in the visible predicate environment of the enclosing class.
  • If the call has no receiver and the predicate reference is a simple identifier, then the call is resolved by looking up the predicate reference and arity in the visible predicate environment of the enclosing module.
  • If the call has no receiver and the predicate reference is a selection identifier, then the qualifier is resolved as a module (see “Module resolution”) and the call is resolved by looking up the identifier in the exported predicate environment of the qualifier module.
  • If the the call has a receiver and the type of the receiver is the same as the enclosing class, the call is resolved by looking up the predicate name and arity in the visible predicate environment of the enclosing class.
  • If the the call has a receiver and the type of the receiver is not the same as the enclosing class, the call is resolved by looking up the predicate name and arity in the exported predicate environment of the type of the receiver.

If all the predicates that the call resolves to are declared on a primitive type, we then restrict to the set of predicates where each argument of the call is a subtype of the corresponding predicate argument type. Then we find all predicates p from this new set such that there is not another predicate p' where each argument of p' is a subtype of the corresponding argument in p. We then say the call resolves to this set instead.

A valid call must only resolve to a single predicate.

For each argument other than a don’t-care expression, the type of the argument must be compatible with the type of the corresponding argument type of the predicate, otherwise the call is invalid.

A valid call with results must resolve to a predicate that has a result type. That result type is also the type of the call.

If the resolved predicate is built in, then the call may not include a closure. If the call does have a closure, then it must resolve to a predicate where the relational arity of the predicate is 2. The relational arity of a predicate is the sum of the following numbers:

  • The number of arguments to the predicate.
  • The number 1 if the predicate is a member predicate, otherwise 0.
  • The number 1 if the predicate has a result, otherwise 0.

If the call includes a closure, then all declared predicate arguments, the enclosing type of the declaration (if it exists), and the result type of the declaration (if it exists) must be compatible. If one of those types is a subtype of int, then all the other arguments must be a subtype of int.

A call to a member predicate may be a direct call:
  • If the receiver is not a super expression it is not direct.
  • If the receiver is A.super and A is an instanceof type and not a base type that is inherited from then it is not direct.
  • If the receiver is A.super and A is a base type that is final inherited from then it is not direct.
  • If the receiver is A.super and A is a base type that is inherited from and not an instanceof type then it is direct.
  • If the receiver is A.super and A is a base type that is inherited from and an instanceof type then the call is not valid.
  • If the receiver is super and the member predicate is in the exported member predicate environment of an instanceof type and not in the exported member predicate environment of a base type then it isn’t direct.
  • If the receiver is super and the member predicate is in the exported member predicate environment of a base type and not in the exported member predicate environment of an instanceof type then it is direct.
  • If the receiver is super and the member predicate is in the exported member predicate environment of a base type and in the exported member predicate environment of an instanceof type then the call is not valid.

If the call resolves to a member predicate, then the receiver values are as follows. If the call has a receiver, then the receiver values are the values of that receiver. If the call does not have a receiver, then the single receiver value is the value of this in the contextual named tuple.

The tuple prefixes of a call with results include one value from each of the argument expressions’ values, in the same order as the order of the arguments. If the call resolves to a non-member predicate, then those values are exactly the tuple prefixes of the call. If the call instead resolves to a member predicate, then the tuple prefixes additionally include a receiver value, ordered before the argument values.

The matching tuples of a call with results are all ordered tuples that are one of the tuple prefixes followed by any value of the same type as the call. If the call has no closure, then all matching tuples must additionally satisfy the resolved predicate of the call, unless the call is direct in which case they must directly satisfy the resolved predicate of the call. If the call has a * or + closure, then the matching tuples must satisfy or directly satisfy the associated closure of the resolved predicate.

The values of a call with results are the final elements of each of the call’s matching tuples.

Aggregations

An aggregation can be written in one of two forms:

aggregation ::= aggid ("[" expr "]")? "(" var_decls ("|" (formula)? ("|" as_exprs ("order" "by" aggorderbys)?)?)? ")"
            |   aggid ("[" expr "]")? "(" as_exprs ("order" "by" aggorderbys)? ")"
            |   "unique" "(" var_decls "|" (formula)? ("|" as_exprs)? ")"

aggid ::= "avg" | "concat" | "count" | "max" | "min" | "rank" | "strictconcat" | "strictcount" | "strictsum" | "sum"

aggorderbys ::= aggorderby ("," aggorderby)*

aggorderby ::= expr ("asc" | "desc")?

The expression enclosed in square brackets ([ and ], U+005B and U+005D), if present, is called the rank expression. It must have type int.

The as_exprs, if present, are called the aggregation expressions. If an aggregation expression is of the form expr as v then the expression is said to be named v.

The rank expression must be present if the aggregate id is rank; otherwise it must not be present.

Apart from the presence or absence of the rank variable, all other reduced forms of an aggregation are equivalent to a full form using the following steps:

  • If the formula is omitted, then it is taken to be any().
  • If there are no aggregation expressions, then either:
    • The aggregation id is count or strictcount and the expression is taken to be 1.
    • There must be precisely one variable declaration, and the aggregation expression is taken to be a reference to that variable.
  • If the aggregation id is concat or strictconcat and it has a single expression then the second expression is taken to be "".
  • If the monotonicAggregates language pragma is not enabled, or the original formula and variable declarations are both omitted, then the aggregate is transformed as follows:
    • For each aggregation expression expr_i, a fresh variable v_i is declared with the same type as the expression in addition to the original variable declarations.
    • The new range is the conjunction of the original range and a term v_i = expr_i for each aggregation expression expr_i.
    • Each original aggregation expression expr_i is replaced by a new aggregation expression v_i.

The variables in the variable declarations list must not occur in the typing environment.

The typing environment for the rank expression is the same as for the aggregation.

The typing environment for the formula is obtained by taking the typing environment for the aggregation and adding all the variable types in the given var_decls list.

The typing environment for an aggregation expression is obtained by taking the typing environment for the formula and then, for each named aggregation expression that occurs earlier than the current expression, adding a mapping from the earlier expression’s name to the earlier expression’s type.

The typing environment for ordering directives is obtained by taking the typing environment for the formula and then, for each named aggregation expression in the aggregation, adding a mapping from the expression’s name to the expression’s type.

The number and types of the aggregation expressions are restricted as follows:

  • A max, min, rank or unique aggregation must have a single expression.
  • The type of the expression in a max, min or rank aggregation without an ordering directive expression must be an orderable type.
  • A count or strictcount aggregation must not have an expression.
  • A sum, strictsum or avg aggregation must have a single aggregation expression, which must have a type which is a subtype of float.
  • A concat or strictconcat aggregation must have two expressions. Both expressions must have types which are subtypes of string.

The type of a count, strictcount aggregation is int. The type of an avg aggregation is float. The type of a concat or strictconcat aggregation is string. The type of a sum or strictsum aggregation is int if the aggregation expression is a subtype of int, otherwise it is float. The type of a rank, min or max aggregation is the type of the single expression.

An ordering directive may only be specified for a max, min, rank, concat or strictconcat aggregation. The type of the expression in an ordering directive must be an orderable type.

The values of the aggregation expression are determined as follows. Firstly, the range tuples are extensions of the named tuple that the aggregation is being evaluated in with the variable declarations of the aggregation, and which match the formula (see “Formulas”).

For each range tuple, the aggregation tuples are the extension of the range tuples to aggregation variables and sort variables.

The aggregation variables are given by the aggregation expressions. If an aggregation expression is named, then its aggregation variable is given by its name, otherwise a fresh synthetic variable is created. The value is given by evaluating the expression with the named tuple being the result of the previous expression, or the range tuple if this is the first aggregation expression.

The sort variables are synthetic variables created for each expression in the ordering directive with values given by the values of the expressions within the ordering directive.

If the aggregation id is max, min or rank and there was no ordering directive, then for each aggregation tuple a synthetic sort variable is added with value given by the aggregation variable.

The values of the aggregation expression are given by applying the aggregation function to each set of tuples obtained by picking exactly one aggregation tuple for each range tuple.

  • If the aggregation id is avg, and the set is non-empty, then the resulting value is the average of the value for the aggregation variable in each tuple in the set, weighted by the number of tuples in the set, after converting the value to a floating-point number.
  • If the aggregation id is count, then the resulting value is the number of tuples in the set. If there are no tuples in the set, then the value is the integer 0.
  • If the aggregation id is max, then the values are the those values of the aggregation variable which are associated with a maximal tuple of sort values. If the set is empty, then the aggregation has no value.
  • If the aggregation id is min, then the values are the those values of the aggregation variable which are associated with a minimal tuple of sort values. If the set is empty, then the aggregation has no value.
  • If the aggregation id is rank, then the resulting values are values of the aggregation variable such that the number of aggregation tuples with a strictly smaller tuple of sort variables is exactly one less than an integer bound by the rank expression of the aggregation. If no such values exist, then the aggregation has no values.
  • If the aggregation id is strictcount, then the resulting value is the same as if the aggregation id were count, unless the set of tuples is empty. If the set of tuples is empty, then the aggregation has no value.
  • If the aggregation id is strictsum, then the resulting value is the same as if the aggregation id were sum, unless the set of tuples is empty. If the set of tuples is empty, then the aggregation has no value.
  • If the aggregation id is sum, then the resulting value is the same as the sum of the values of the aggregation variable across the tuples in the set, weighted by the number of times each value occurs in the tuples in the set. If there are no tuples in the set, then the resulting value of the aggregation is the integer 0.
  • If the aggregation id is concat, then there is one value for each value of the second aggregation variable, given by the concatenation of the value of the first aggregation variable of each tuple with the value of the second aggregation variable used as a separator, ordered by the sort variables. If there are multiple aggregation tuples with the same sort variables then the first distinguished value is used to break ties. If there are no tuples in the set, then the single value of the aggregation is the empty string.
  • If the aggregation id is strictconcat, then the result is the same as for concat except in the case where there are no aggregation tuples in which case the aggregation has no value.
  • If the aggregation id is unique, then the result is the value of the aggregation variable if there is precisely one such value. Otherwise, the aggregation has no value.

Any

The any expression is a special kind of quantified expression.

any ::= "any" "(" var_decls ("|" (formula)? ("|" expr)?)? ")"

The values of an any expression are those values of the expression for which the formula matches.

The abbreviated cases for an any expression are interpreted in the same way as for an aggregation.

Expression Pragma

Expression pragmas can be used to guide optimization.

::

expression_pragma ::= “pragma” “[” expression_pragma_type “]” “(” expr “)”

expression_pragma_type ::= “only_bind_out” | “only_bind_into”

The values of an expression pragma are the values of the contained expression.

The type only_bind_out hints that uses of the result of the expression pragma should not be used to guide the evaluation of the result of the contained expression. When checking to see that all values are bound the compiler does not assume that if the result of the expression pragma is bound then the result of the contained expression is bound.

The type only_bind_into hints that uses of the contained expression should not be used to guide the evaluation of the result of the expression pragma. When checking to see that all values are bound the compiler does not assume that if the result of the contained expression is bound then the result of the expression pragma is bound.

Ranges

Range expressions denote a range of values.

range ::= "[" expr ".." expr "]"

Both expressions must be subtypes of int, float, or date. If either of them are type date, then both of them must be.

If both expressions are subtypes of int then the type of the range is int. If both expressions are subtypes of date then the type of the range is date. Otherwise the type of the range is float.

The values of a range expression are those values which are ordered inclusively between a value of the first expression and a value of the second expression.

Set literals

Set literals denote a choice from a collection of values.

setliteral ::= "[" expr ("," expr)* ","? "]"

Set literals can be of any type, but the types within a set literal have to be consistent according to the following criterion: At least one of the set elements has to be of a type that is a supertype of all the set element types. This supertype is the type of the set literal. For example, float is a supertype of float and int, therefore x = [4, 5.6] is valid. On the other hand, y = [5, "test"] does not adhere to the criterion.

The values of a set literal expression are all the values of all the contained element expressions.

Since release 2.7.1 of the CodeQL CLI, a trailing comma is allowed in a set literal.

Disambiguation of expressions

The grammar given in this section is disambiguated first by precedence, and second by associating left to right. The order of precedence from highest to lowest is:

  • casts
  • unary + and -
  • binary * , / and %
  • binary + and -

Whenever a sequence of tokens can be interpreted either as a call to a predicate with result (with specified closure), or as a binary operation with operator + or *, the syntax is interpreted as a call to a predicate with result.

Whenever a sequence of tokens can be interpreted either as arithmetic with a parenthesized variable or as a prefix cast of a unary operation, the syntax is interpreted as a cast.

Formulas

A formula is a form of syntax used to match a named tuple given a store.

There are several kinds of formulas:

formula ::= fparen
        |   disjunction
        |   conjunction
        |   implies
        |   ifthen
        |   negated
        |   quantified
        |   comparison
        |   instanceof
        |   inrange
        |   call

This section specifies the syntax for each kind of formula and what tuples they match.

Parenthesized formulas

A parenthesized formula is a formula enclosed by a pair of parentheses:

fparen ::= "(" formula ")"

A parenthesized formula matches the same tuples as the nested formula matches.

Disjunctions

A disjunction is two formulas separated by the or keyword:

disjunction ::= formula "or" formula

A disjunction matches any tuple that matches either of the nested formulas.

Conjunctions

A conjunction is two formulas separated by the and keyword:

conjunction ::= formula "and" formula

A conjunction matches any tuple that also matches both of the two nested formulas.

Implications

An implication formula is two formulas separated by the implies keyword:

implies ::= formula "implies" formula

Neither of the two formulas may be another implication.

An implied formula matches if either the second formula matches, or the first formula does not match.

Conditional formulas

A conditional formula has the following syntax:

ifthen ::= "if" formula "then" formula "else" formula

The first formula is called the condition of the conditional formula. The second formula is called the true branch, and the second formula is called the false branch.

The conditional formula matches if the condition and the true branch both match. It also matches if the false branch matches and the condition does not match.

Negations

A negation formula is a formula preceded by the not keyword:

negated ::= "not" formula

A negation formula matches any tuple that does not match the nested formula.

Quantified formulas

A quantified formula has several syntaxes:

quantified ::= "exists" "(" expr ")"
           |   "exists" "(" var_decls ("|" formula)? ("|" formula)? ")"
           |   "forall" "(" var_decls ("|" formula)? "|" formula ")"
           |   "forex"  "(" var_decls ("|" formula)? "|" formula ")"

In all cases, the typing environment for the nested expressions or formulas is the same as the typing environment for the quantified formula, except that it also maps the variables in the variable declaration to their associated types.

The first form matches if the given expression has at least one value.

For the other forms, the extensions of the current named tuple for the given variable declarations are called the quantifier extensions. The nested formulas are called the first quantified formula and, if present, the second quantified formula.

The second exists formula matches if one of the quantifier extensions is such that the quantified formula or formulas all match.

A forall formula that has one quantified formula matches if that quantified formula matches all of the quantifier extensions. A forall with two quantified formulas matches if the second formula matches all extensions where the first formula matches.

A forex formula with one quantified formula matches under the same conditions as a forall formula matching, except that there must be at least one quantifier extension where that first quantified formula matches.

Comparisons

A comparison formula is two expressions separated by a comparison operator:

comparison ::= expr compop expr
compop ::= "=" | "!=" | "<" | ">" | "<=" | ">="

A comparison formula matches if there is one value of the left expression that is in the given ordering with one of the values of the right expression. The ordering used is specified in “Ordering.” If one of the values is an integer and the other is a float value, then the integer is converted to a float value before the comparison.

If the operator is =, then at least one of the left and right expressions must have a type; if they both have a type, those types must be compatible.

If the operator is !=, then both expressions must have a type, and those types must be compatible.

If the operator is any other operator, then both expressions must have a type. Those types must be compatible with each other. Each of those types must be orderable.

Type checks

A type check formula has the following syntax:

instanceof ::= expr "instanceof" type

The type to the right of instanceof is called the type-check type.

The type of the expression must be compatible with the type-check type.

The formula matches if one of the values of the expression is in the type-check type.

Range checks

A range check has the following syntax:

inrange ::= expr "in" (range | setliteral)

The formula is equivalent to expr "=" range or expr "=" setliteral.

Calls

A call has the following syntax:

call ::= predicateRef (closure)? "(" (exprs)? ")"
     |   primary "." predicateName (closure)? "(" (exprs)? ")"

The identifier is called the predicate name of the call.

A call must resolve to a predicate, using the same definition of resolve as for calls with results (see “Calls with results”).

A call may be direct using the same definition of direct as for calls with results (see “Calls with results”).

The resolved predicate must not have a result type.

If the resolved predicate is a built-in member predicate of a primitive type, then the call may not include a closure. If the call does have a closure, then the call must resolve to a predicate with relational arity of 2.

The candidate tuples of a call are the ordered tuples formed by selecting a value from each of the arguments of the call.

If the call has no closure, then it matches whenever one of the candidate tuples satisfies the resolved predicate of the call, unless the call is direct, in which case the candidate tuple must directly satisfy the resolved predicate. If the call has * or + closure, then the call matches whenever one of the candidate tuples satisfies or directly satisfies the associated closure of the resolved predicate.

Disambiguation of formulas

The grammar given in this section is disambiguated first by precedence, and second by associating left to right, except for implication which is non-associative. The order of precedence from highest to lowest is:

  • Negation
  • Conditional formulas
  • Conjunction
  • Disjunction
  • Implication

Aliases

Aliases define new names for existing QL bindings.

alias ::= qldoc? annotations "predicate" literalId "=" predicateRef "/" int ";"
      |   qldoc? annotations "class" classname "=" type ";"
      |   qldoc? annotations "module" modulename "=" moduleExpr ";"

An alias introduces a binding from the new name to the binding referred to by the right-hand side in the current module’s visible predicate, type, or module environment respectively.

An alias is called a strong alias if and only if it has the final annotation. Otherwise, it is called a weak alias.

Two bindings A, B are called equal modulo weak aliasing if and only if one of the following conditions are satisfied:

  • A and B are the same binding or
  • A` is introduced by a weak alias for C, where B and C are equal modulo weak aliasing (or vice versa) or
  • A and B are introduced by the same strong alias and they are aliases for bindings that are equal modulo weak aliasing.

Note that the third condition is only relevant in Parameterized modules, where the binding introduced by the alias can depend on instantiation parameters.

Built-ins

A QL database includes a number of built-in predicates . This section defines a number of built-in predicates that all databases include. Each database also includes a number of additional non-member predicates that are not specified in this document.

This section gives several tables of built-in predicates. For each predicate, the table gives the result type of each predicate that has one, and the sequence of argument types.

Each table also specifies which ordered tuples are in the database content of each predicate. It specifies this with a description that holds true for exactly the tuples that are included. In each description, the “result” is the last element of each tuple, if the predicate has a result type. The “receiver” is the first element of each tuple. The “arguments” are all elements of each tuple other than the result and the receiver.

Non-member built-ins

The following built-in predicates are non-member predicates:

Name Result type Argument types Content
any     The empty tuple.
none     No tuples.
toUrl   string, int, int, int, int, string Let the arguments be file, startLine, startCol, endLine, endCol, and url. The predicate holds if url is equal to the string file://file:startLine:startCol:endLine:endCol.

Built-ins for boolean

The following built-in predicates are members of type boolean:

Name Result type Argument types Content
booleanAnd boolean boolean The result is the boolean and of the receiver and the argument.
booleanNot boolean   The result is the boolean not of the receiver.
booleanOr boolean boolean The result is the boolean or of the receiver and the argument.
booleanXor boolean boolean The result is the boolean exclusive or of the receiver and the argument.
toString string   The result is “true” if the receiver is true, otherwise “false.”

Built-ins for date

The following built-in predicates are members of type date:

Name Result type Argument types Content
daysTo int date The result is the number of days between but not including the receiver and the argument.
getDay int   The result is the day component of the receiver.
getHours int   The result is the hours component of the receiver.
getMinutes int   The result is the minutes component of the receiver.
getMonth string   The result is a string that is determined by the month component of the receiver. The string is one of January, February, March, April, May, June, July, August, September, October, November, or December.
getSeconds int   The result is the seconds component of the receiver.
getYear int   The result is the year component of the receiver.
toISO string   The result is a string representation of the date. The representation is left unspecified.
toString string   The result is a string representation of the date. The representation is left unspecified.

Built-ins for float

The following built-in predicates are members of type float:

Name Result type Argument types Content
abs float   The result is the absolute value of the receiver.
acos float   The result is the inverse cosine of the receiver.
asin float   The result is the inverse sine of the receiver.
atan float   The result is the inverse tangent of the receiver.
ceil int   The result is the smallest integer greater than or equal to the receiver.
copySign float float The result is the floating point number with the magnitude of the receiver and the sign of the argument.
cos float   The result is the cosine of the receiver.
cosh float   The result is the hyperbolic cosine of the receiver.
exp float   The result is the value of e, the base of the natural logarithm, raised to the power of the receiver.
floor int   The result is the largest integer that is not greater than the receiver.
log float   The result is the natural logarithm of the receiver.
log float float The result is the logarithm of the receiver with the base of the argument.
log float int The result is the logarithm of the receiver with the base of the argument.
log10 float   The result is the base-10 logarithm of the receiver.
log2 float   The result is the base-2 logarithm of the receiver.
maximum float float The result is the larger of the receiver and the argument.
maximum float int The result is the larger of the receiver and the argument.
minimum float float The result is the smaller of the receiver and the argument.
minimum float int The result is the smaller of the receiver and the argument.
nextAfter float float The result is the number adjacent to the receiver in the direction of the argument.
nextDown float   The result is the number adjacent to the receiver in the direction of negative infinity.
nextUp float   The result is the number adjacent to the receiver in the direction of positive infinity.
pow float float The result is the receiver raised to the power of the argument.
pow float int The result is the receiver raised to the power of the argument.
signum float   The result is the sign of the receiver: zero if it is zero, 1.0 if it is greater than zero, -1.0 if it is less than zero.
sin float   The result is the sine of the receiver.
sinh float   The result is the hyperbolic sine of the receiver.
sqrt float   The result is the square root of the receiver.
tan float   The result is the tangent of the receiver.
tanh float   The result is the hyperbolic tangent of the receiver.
toString string   The decimal representation of the number as a string.
ulp float   The result is the ULP (unit in last place) of the receiver.

Built-ins for int

The following built-in predicates are members of type int:

Name Result type Argument types Content
abs int   The result is the absolute value of the receiver.
acos float   The result is the inverse cosine of the receiver.
asin float   The result is the inverse sine of the receiver.
atan float   The result is the inverse tangent of the receiver.
cos float   The result is the cosine of the receiver.
cosh float   The result is the hyperbolic cosine of the receiver.
exp float   The result is the value of value of e, the base of the natural logarithm, raised to the power of the receiver.
gcd int int The result is the greatest common divisor of the receiver and the argument.
log float   The result is the natural logarithm of the receiver.
log float float The result is the logarithm of the receiver with the base of the argument.
log float int The result is the logarithm of the receiver with the base of the argument.
log10 float   The result is the base-10 logarithm of the receiver.
log2 float   The result is the base-2 logarithm of the receiver.
maximum float float The result is the larger of the receiver and the argument.
maximum int int The result is the larger of the receiver and the argument.
minimum float float The result is the smaller of the receiver and the argument.
minimum int int The result is the smaller of the receiver and the argument.
pow float float The result is the receiver raised to the power of the argument.
pow float int The result is the receiver raised to the power of the argument.
sin float   The result is the sine of the receiver.
sinh float   The result is the hyperbolic sine of the receiver.
sqrt float   The result is the square root of the receiver.
tan float   The result is the tangent of the receiver.
tanh float   The result is the hyperbolic tangent of the receiver.
bitAnd int int The result is the bitwise and of the receiver and the argument.
bitOr int int The result is the bitwise or of the receiver and the argument.
bitXor int int The result is the bitwise xor of the receiver and the argument.
bitNot int   The result is the bitwise complement of the receiver.
bitShiftLeft int int The result is the bitwise left shift of the receiver by the argument, modulo 32.
bitShiftRight int int The result is the bitwise right shift of the receiver by the argument, modulo 32.
bitShiftRightSigned int int The result is the signed bitwise right shift of the receiver by the argument, modulo 32.
toString string   The result is the decimal representation of the number as a string.
toUnicode string   The result is the unicode character for the receiver seen as a unicode code point.

The leftmost bit after bitShiftRightSigned depends on sign extension, whereas after bitShiftRight it is zero.

Built-ins for string

The following built-in predicates are members of type string:

Name Result type Argument types Content
charAt string int The result is a 1-character string containing the character in the receiver at the index given by the argument. The first element of the string is at index 0.
indexOf int string The result is an index into the receiver where the argument occurs.
indexOf int string, int, int Let the arguments be s, n, and start. The result is the index of occurrence n of substring s in the receiver that is no earlier in the string than start.
isLowercase     The receiver contains no upper-case letters.
isUppercase     The receiver contains no lower-case letters.
length int   The result is the number of characters in the receiver.
matches   string The argument is a pattern that matches the receiver, in the same way as the LIKE operator in SQL. Patterns may include _ to match a single character and % to match any sequence of characters. A backslash can be used to escape an underscore, a percent, or a backslash. Otherwise, all characters in the pattern other than _ and % and \\ must match exactly.
prefix string int The result is the prefix of the receiver that has a length exactly equal to the argument. If the argument is negative or greater than the receiver’s length, then there is no result.
regexpCapture string string, int The receiver exactly matches the regex in the first argument, and the result is the group of the match numbered by the second argument.
regexpFind string string, int, int The receiver contains one or more occurrences of the regex in the first argument. The result is the substring which matches the regex, the second argument is the occurrence number, and the third argument is the index within the receiver at which the occurrence begins.
regexpMatch   string The receiver matches the argument as a regex.
regexpReplaceAll string string, string The result is obtained by replacing all occurrences in the receiver of the first argument as a regex by the second argument.
replaceAll string string, string The result is obtained by replacing all occurrences in the receiver of the first argument by the second.
splitAt string string The result is one of the strings obtained by splitting the receiver at every occurrence of the argument.
splitAt string string, int Let the arguments be delim and i. The result is field number i of the fields obtained by splitting the receiver at every occurrence of delim.
substring string int, int The result is the substring of the receiver starting at the index of the first argument and ending just before the index of the second argument.
suffix string int The result is the suffix of the receiver that has a length exactly equal to the receiver’s length minus the argument. If the argument is negative or greater than the receiver’s length, then there is no result. As a result, the identity s.prefix(i)+s.suffix(i)=s holds for i in [0, s.length()].
toDate date   The result is a date value determined by the receiver. The format of the receiver is unspecified, except that if (d, s) is in date.toString, (s, d) is in string.toDate.
toFloat float   The result is the float whose value is represented by the receiver. If the receiver cannot be parsed as a float then there is no result.
toInt int   The result is the integer whose value is represented by the receiver. If the receiver cannot be parsed as an integer or cannot be represented as a QL int, then there is no result. The parser accepts an optional leading - or + character, followed by one or more decimal digits.
toLowerCase string   The result is the receiver with all letters converted to lower case.
toString string   The result is the receiver.
toUpperCase string   The result is the receiver with all letters converted to upper case.
trim string   The result is the receiver with all whitespace removed from the beginning and end of the string.
codePointAt int int The result is the unicode code point at the index given by the argument.
codePointCount int int, int The result is the number of unicode code points in the receiver between the given indices.

Regular expressions are as defined by java.util.regex.Pattern in Java. For more information, see the Java API Documentation.

Evaluation

This section specifies the evaluation of a QL program. Evaluation happens in three phases. First, the program is stratified into a number of layers. Second, the layers are evaluated one by one. Finally, the queries in the QL file are evaluated to produce sets of ordered tuples.

Stratification

A QL program can be stratified to a sequence of layers. A layer is a set of predicates and types.

A valid stratification must include each predicate and type in the QL program that is fully instantiated. It must not include any other predicates or types.

A valid stratification must not include the same predicate in multiple layers.

Each non-abstract predicate has an associated body. For predicates inside declared modules, this is the predicate declaration. The body of an instantiation-nested predicate is the body of the underlying nested predicate where all references and calls have been substituted with the instantiation-relative entity or alias.

Formulas, variable declarations and expressions within a predicate body have a negation polarity that is positive, negative, or zero. Positive and negative are opposites of each other, while zero is the opposite of itself. The negation polarity of a formula or expression is then determined as follows:

  • The body of a predicate is positive.
  • The formula within a negation formula has the opposite polarity to that of the negation formula.
  • The condition of a conditional formula has zero polarity.
  • The formula on the left of an implication formula has the opposite polarity to that of the implication.
  • The formula and variable declarations of an aggregate have zero polarity.
  • If the monotonicAggregates language pragma is not enabled, or the original formula and variable declarations are both omitted, then the expressions and order by expressions of the aggregate have zero polarity.
  • If the monotonicAggregates language pragma is enabled, and the original formula and variable declarations were not both omitted, then the expressions and order by expressions of the aggregate have the polarity of the aggregate.
  • If a forall has two quantified formulas, then the first quantified formula has the opposite polarity to that of the forall.
  • The variable declarations of a forall have the opposite polarity to that of the forall.
  • If a forex has two quantified formulas, then the first quantified formula has zero polarity.
  • The variable declarations of a forex have zero polarity.
  • In all other cases, a formula or expression has the same polarity as its immediately enclosing formula or expression.

For a member predicate p we define the strict dispatch dependencies. The strict dispatch dependencies are defined as:

  • The strict dispatch dependencies of any predicates that override p.
  • If p is not abstract, C.class for any class C with a predicate that overrides p.

For a member predicate p we define the dispatch dependencies. The dispatch dependencies are defined as:

  • The dispatch dependencies of predicates that override p.
  • The predicate p itself.
  • C.class where C is the class that defines p.

Predicates, and types can depend and strictly depend on each other. Such dependencies exist in the following circumstances:

  • If A strictly depends on B, then A depends on B.
  • If A depends on B, then A also depends on anything on which B depends.
  • If A strictly depends on B, then A and anything depending on A strictly depend on anything on which B depends (including B itself).
  • If a predicate has a parameter whose declared type is a class type C, it depends on C.class.
  • If a predicate declares a result type which is a class type C, it depends on C.class.
  • A member predicate of class C depends on C.class.
  • If a predicate contains a variable declaration of a variable whose declared type is a class type C, then the predicate depends on C.class. If the declaration has negative or zero polarity then the dependency is strict.
  • If a predicate contains a variable declaration with negative or zero polarity of a variable whose declared type is a class type C, then the predicate strictly depends on C.class.
  • If a predicate contains an expression whose type is a class type C which is not a variable reference, then the predicate depends on C.class. If the expression has negative or zero polarity then the dependency is strict.
  • A predicate containing a predicate call depends on the predicate to which the call resolves. If the call has negative or zero polarity then the dependency is strict.
  • A predicate containing a predicate call, which resolves to a member predicate, where the call is not direct, depends on the dispatch dependencies of the root definitions of the target of the call. If the call has negative or zero polarity then the dependencies are strict. The predicate strictly depends on the strict dispatch dependencies of the root definitions.
  • For each class C in the program, for each base class B of C, C.extends depends on B.B.
  • For each class C in the program, for each instanceof type B of C, C.extends depends on B.
  • For each class C in the program, for each base type B of C that is not a class type, C.extends depends on B.
  • For each class C in the program, C.class depends on C.C.
  • For each class C in the program, C.C depends on C.extends.
  • For each class C in the program that declares a field of class type B, C.C depends on B.class.
  • For each class C with a characteristic predicate, C.C depends on the characteristic predicate.
  • For each abstract class A in the program, for each type C that inherits from A and has A as a base type, A.class depends on C.class.
  • A predicate with a higher-order body may strictly depend or depend on each predicate reference within the body. The exact dependencies are left unspecified.

A valid stratification must have no predicate that depends on a predicate in a later layer. Additionally, it must have no predicate that strictly depends on a predicate in the same layer.

If a QL program has no valid stratification, then the program itself is not valid. If it does have a stratification, a QL implementation must choose exactly one stratification. The precise stratification chosen is left unspecified.

Layer evaluation

The store is first initialized with the database content of all built-in predicates and external predicates. The database content of a predicate is a set of ordered tuples that are included in the database.

Each layer of the stratification is populated in order. To populate a layer, each predicate in the layer is repeatedly populated until the store stops changing. The way that a predicate is populated is as follows:

  • To populate a predicate that has a formula as a body, find each named tuple t that has the following properties:

    • The tuple matches the body formula.
    • The variables should be the predicate’s arguments.
    • If the predicate has a result, then the tuples should additionally have a value for result.
    • If the predicate is a member predicate or characteristic predicate of a class C then the tuples should additionally have a value for this and each visible field on the class.
    • The values corresponding to the arguments should all be a member of the declared types of the arguments.
    • The values corresponding to result should all be a member of the result type.
    • The values corresponding to the fields should all be a member of the declared types of the fields.
    • If the predicate is a member predicate of a class C and not a characteristic predicate, then the tuples should additionally extend some tuple in C.class.
    • If the predicate is a characteristic predicate of a class C, then there should be a tuple t' in C.extends such that for each visible field in C, any field that is equal to or overrides a field in t' should have the same value in t. this should also map to the same value in t and t'.

    For each such tuple remove any components that correspond to fields and add it to the predicate in the store.

  • To populate an abstract predicate, do nothing.

  • The population of predicates with a higher-order body is left only partially specified. A number of tuples are added to the given predicate in the store. The tuples that are added must be fully determined by the QL program and by the state of the store.

  • To populate the type C.extends for a class C, identify each named tuple that has the following properties:

    • The value of this is in all non-class base types of C.
    • The value of this is in all instanceof types of C.
    • The keys of the tuple are this and the union of the public fields from each base type.
    • For each class base type B of C there is a named tuple with variables from the public fields of B and this that the given tuple and some tuple in B.B both extend.

    For each such tuple add it to C.extends.

  • To populate the type C.C for a class C, if C has a characteristic predicate, then add all tuples from that predicate to the store. Otherwise add all tuples t such that:

    • The variables of t should be this and the visible fields of C.
    • The values corresponding to the fields should all be a member of the declared types of the fields.
    • If the predicate is a characteristic predicate of a class C, then there should be a tuple t' in C.extends such that for each visible field in C, any field that is equal to or overrides a field in t' should have the same value in t. this should also map to the same value in t and t'.
  • To populate the type C.class for a non-abstract class type C, add each tuple in C.C to C.class.

  • To populate the type C.class for an abstract class type C, identify each named tuple that has the following properties:
    • It is a member of C.C.
    • For each class D that inherits from C and has C as a base type then there is a named tuple with variables from the public fields of C and this that the given tuple and a tuple in D.class both extend.

Query evaluation

A query is evaluated as follows:

  1. Identify all facts about query predicates.
  2. If there is a select clause then find all named tuples with the variables declared in the from clause that match the formula given in the where clause, if there is one. For each named tuple, convert it to a set of ordered tuples where each element of the ordered tuple is, in the context of the named tuple, a value of one of the corresponding select expressions. Then sequence the ordered tuples lexicographically. The first elements of the lexicographic order are the tuple elements specified by the ordering directives of the predicate targeted by the query, if it has any. Each such element is ordered either ascending (asc) or descending (desc) as specified by the ordering directive, or ascending if the ordering directive does not specify. This lexicographic order is only a partial order, if there are fewer ordering directives than elements of the tuples. An implementation may produce any sequence of the ordered tuples that satisfies this partial order.
  3. The result is the facts from the query predicates plus the list of ordered tuples from the select clause if it exists.

Summary of syntax

The complete grammar for QL is as follows:

ql ::= qldoc? moduleBody

module ::= annotation* "module" modulename parameters? implements? "{" moduleBody "}"

parameters ::= "<" signatureExpr parameterName ("," signatureExpr parameterName)* ">"

implements ::= "implements" moduleSignatureExpr ("," moduleSignatureExpr)*

moduleBody ::= (import | predicate | class | module | alias | select)*

import ::= annotations "import" importModuleExpr ("as" modulename)?

qualId ::= simpleId | qualId "." simpleId

importModuleExpr ::= qualId | importModuleExpr "::" modulename arguments?

arguments ::= "<" argument ("," argument)* ">"

argument ::= moduleExpr | type | predicateRef "/" int

signature ::= predicateSignature | typeSignature | moduleSignature

predicateSignature ::= qldoc? annotations "signature" head ";"

typeSignature ::= qldoc? annotations "signature" "class" classname ("extends" type ("," type)*)? (";" | "{" signaturePredicate* "}")

moduleSignature ::= qldoc? annotation* "signature" "module" moduleSignatureName parameters? "{" moduleSignatureBody "}"

moduleSignatureBody ::= (signaturePredicate | defaultPredicate | signatureType)*

signaturePredicate ::= qldoc? annotations head ";"

defaultPredicate ::= qldoc? annotations "default" head "{" formula "}"

signatureType ::= qldoc? annotations "class" classname ("extends" type ("," type)*)? "{" signaturePredicate* "}"

select ::= ("from" var_decls)? ("where" formula)? "select" as_exprs ("order" "by" orderbys)?

as_exprs ::= as_expr ("," as_expr)*

as_expr ::= expr ("as" lowerId)?

orderbys ::= orderby ("," orderby)*

orderby ::= lowerId ("asc" | "desc")?

predicate ::= qldoc? annotations head optbody

annotations ::= annotation*

annotation ::= simpleAnnotation | argsAnnotation

simpleAnnotation ::= "abstract"
                 |   "cached"
                 |   "external"
                 |   "extensible"
                 |   "final"
                 |   "transient"
                 |   "library"
                 |   "private"
                 |   "deprecated"
                 |   "override"
                 |   "additional"
                 |   "query"

argsAnnotation ::= "pragma" "[" ("inline" | "inline_late" | "noinline" | "nomagic" | "noopt" | "assume_small_delta") "]"
               |   "language" "[" "monotonicAggregates" "]"
               |   "bindingset" "[" (variable ( "," variable)*)? "]"

head ::= ("predicate" | type) predicateName "(" var_decls ")"

optbody ::= ";"
        |  "{" formula "}"
        |  "=" literalId "(" (predicateRef "/" int ("," predicateRef "/" int)*)? ")" "(" (exprs)? ")"

class ::= qldoc? annotations "class" classname ("extends" type ("," type)*)? ("instanceof" type ("," type)*)?  "{" member* "}"

member ::= character | predicate | field

character ::= qldoc? annotations classname "(" ")" "{" formula "}"

field ::= qldoc? annotations var_decl ";"

moduleExpr ::= modulename arguments? | moduleExpr "::" modulename arguments?

moduleSignatureExpr ::= (moduleExpr "::")? moduleSignatureName arguments?

signatureExpr : (moduleExpr "::")? simpleId ("/" Integer | arguments)?;

type ::= (moduleExpr "::")? classname | dbasetype | "boolean" | "date" | "float" | "int" | "string"

exprs ::= expr ("," expr)*

alias ::= qldoc? annotations "predicate" literalId "=" predicateRef "/" int ";"
      |  qldoc? annotations "class" classname "=" type ";"
      |  qldoc? annotations "module" modulename "=" moduleExpr ";"

var_decls ::= (var_decl ("," var_decl)*)?

var_decl ::= type lowerId

formula ::= fparen
        |   disjunction
        |   conjunction
        |   implies
        |   ifthen
        |   negated
        |   quantified
        |   comparison
        |   instanceof
        |   inrange
        |   call

fparen ::= "(" formula ")"

disjunction ::= formula "or" formula

conjunction ::= formula "and" formula

implies ::= formula "implies" formula

ifthen ::= "if" formula "then" formula "else" formula

negated ::= "not" formula

quantified ::= "exists" "(" expr ")"
           |   "exists" "(" var_decls ("|" formula)? ("|" formula)? ")"
           |   "forall" "(" var_decls ("|" formula)? "|" formula ")"
           |   "forex"  "(" var_decls ("|" formula)? "|" formula ")"

comparison ::= expr compop expr

compop ::= "=" | "!=" | "<" | ">" | "<=" | ">="

instanceof ::= expr "instanceof" type

inrange ::= expr "in" (range | setliteral)

call ::= predicateRef (closure)? "(" (exprs)? ")"
     |   primary "." predicateName (closure)? "(" (exprs)? ")"

closure ::= "*" | "+"

expr ::= dontcare
     |   unop
     |   binop
     |   cast
     |   primary


primary ::= eparen
        |   literal
        |   variable
        |   super_expr
        |   postfix_cast
        |   callwithresults
        |   aggregation
        |   expression_pragma
        |   any
        |   range
        |   setliteral

eparen ::= "(" expr ")"

dontcare ::= "_"

literal ::= "false" | "true" | int | float | string

unop ::= "+" expr
     |   "-" expr

binop ::= expr "+" expr
      |   expr "-" expr
      |   expr "*" expr
      |   expr "/" expr
      |   expr "%" expr

variable ::= varname | "this" | "result"

super_expr ::= "super" | type "." "super"

cast ::= "(" type ")" expr

postfix_cast ::= primary "." "(" type ")"

aggregation ::= aggid ("[" expr "]")? "(" var_decls ("|" (formula)? ("|" as_exprs ("order" "by" aggorderbys)?)?)? ")"
            |   aggid ("[" expr "]")? "(" as_exprs ("order" "by" aggorderbys)? ")"
            |   "unique" "(" var_decls "|" (formula)? ("|" as_exprs)? ")"

expression_pragma ::= "pragma" "[" expression_pragma_type "]" "(" expr ")"

expression_pragma_type ::= "only_bind_out" | "only_bind_into"

aggid ::= "avg" | "concat" | "count" | "max" | "min" | "rank" | "strictconcat" | "strictcount" | "strictsum" | "sum"

aggorderbys ::= aggorderby ("," aggorderby)*

aggorderby ::= expr ("asc" | "desc")?

any ::= "any" "(" var_decls ("|" (formula)? ("|" expr)?)? ")"

callwithresults ::= predicateRef (closure)? "(" (exprs)? ")"
                |   primary "." predicateName (closure)? "(" (exprs)? ")"

range ::= "[" expr ".." expr "]"

setliteral ::= "[" expr ("," expr)* ","? "]"

simpleId ::= lowerId | upperId

modulename ::= simpleId

moduleSignatureName ::= upperId

classname ::= upperId

dbasetype ::= atLowerId

predicateRef ::= (moduleExpr "::")? literalId

predicateName ::= lowerId

parameterName ::= simpleId

varname ::= lowerId

literalId ::= lowerId | atLowerId | "any" | "none"
  • © GitHub, Inc.
  • Terms
  • Privacy