Skip to main content

Syntax

Schemas are written in Glean's Angle language.

Named schemas​

We usually create a new named schema for a family of related types and predicates. For example cxx is the schema for facts about C/C++ and Objective C code, java is the schema for facts about Java code, python is the schema for facts about Python code and so on.

schema java.1 {
# your types and predicates go here
}
note

What does the .1 suffix mean? This is the version of the schema. Versions are a somewhat legacy feature and you should normally leave everything at version 1.

The name of the schema will be also used as the namespace for the generated Thrift types. You'll notice that each schema gets a separate generated Thrift file for working with the data in your code.

Predicates​

A predicate definition looks like this:

predicate P : KeyType

where

  • P is the name of the predicate. e.g. src.File
  • KeyType is the key type

For example, the example schema that we saw in the Introduction contains the Class predicate:

schema example.1 {
predicate Class :
{
name : string,
line : nat,
}
}

This defines:

  • A predicate called example.Class.1 (you can usually refer to it by its unversioned name, example.Class).
  • With a key type that is a record, consisting of
    • a field name, that has type string
    • a field line, of type nat

A predicate can also have a value:

predicate P : KeyType -> ValueType

We sometimes call these key-value predicates. Most predicates you encounter will only have a key.

Referring to other predicates​

A key feature of Glean is the ability to refer directly to other types to build up nested structures.

To refer to a predicate defined in the same schema, we just use its unqualified name:

schema example.1 {
predicate Class :
{
name : string,
line : nat,
}

predicate Parent :
{
child : Class,
parent : Class,
}
}

We can also refer to predicates defined in other schemas, but we have to bring them into scope using an import declaration. For example, the src schema contains a File predicate:

schema src.1 {
predicate File : string
}

If we want to extend our Class predicate with the file that contains it, we can do it like this:

schema example.1 {
import src.1

predicate Class :
{
name : string,
line : nat,
file : src.File
}
}

Note that to refer to a predicate from another schema, we have to qualify it with the name of its schema, as in src.File above. We don't have to add the version though - there can only be one version of any given predicate in scope at a time.

Named types​

It's useful to be able to name a type so that it can be used in multiple places. The form of a type definition is similar to a predicate:

type <name> = <type>

For example, the src schema has a type for source locations:

schema src.1 {
predicate File : string

# Common source code location type
type Loc = {
file : File,
line : nat,
column : nat,
}
}

What is the difference between a predicate and a type?​

Predicates describe facts that are stored in the DB. A type has no facts, it's just a name for a type. Types are expanded everywhere they're mentioned.

Predicates are useful for increasing sharing. For example, if we have

predicate P: { a : string, b : [nat] }

and we notice that many P facts have the same array of nat for the b field, we could store the data more compactly by making it a predicate:

predicate List: [nat]

predicate P: { a : string, b : List }

Now there will only be one fact for each unique value of List stored in the DB. Using a type would not have the same effect:

type List: [nat]

predicate P: { a : string, b : List }

This is exactly equivalent to the original version; every P fact will contain the full [nat].

Finally, type declarations cannot be recursive or mutually recursive. If you want a recursive type, the cycle must go through at least one predicate. For more details, see Recursion.

Importing and inheriting​

There are two ways for schemas to refer to each other:

Importing​

schema example.1 {
import src.1
...
}

The import declaration brings into scope all the predicates and types defined by the named schema. Those predicates and types are referred to by qualified names, e.g. src.File for the File predicate defined by the schema src. The version number is dropped; each schema exports a single version of a predicate or type, so the unversioned name is unambiguous.

Inheritance and revising schemas​

danger

Schema inheritance is a legacy feature and may be removed in the future. The process for safely changing schemas is described in Changing the Schema.

schema example.2 : example.1 {
predicate Class :
{
# new definition of Class
}
}

Inheritance is useful for making changes to a schema by creating a new schema version:

  • Inheriting from a schema brings into scope all the types and predicates of that schema, both qualified and unqualified.
  • The new schema also exports all the types and predicates defined in the schemas it inherits from, except those that are re-defined.

Specifically, in the above example:

  • We can import example.2 anywhere and get all the predicates defined in example.1, except that we'll get the new Class defined in example.2.
  • We can still import example.1 and get the old version of the schema.

Note that if you have predicates that depend on a predicate that was revised in this way, you must also copy those predicates to the new schema, because the existing predicates will refer to the old version of the predicate you revised. (In due course Glean will probably provide a convenient way to do this; in the meantime you have to copy & paste. Not a big deal because you'll usually delete the old one at some point, and you can't modify it anyway.)

Named schemas can not form cycles through their import or inheritance declarations.

Naming rules and conventions​

Names take the form of a dot-separated sequence of alphanumeric words. For example, sys.Blob, clang.File, or cxx.objc.Name. The words up to the last dot are the namespace, the final word is the name.

See Names for full details.

Briefly:

  • Namespaces (schema names) are dot-separated sequences of identifiers each beginning with a lower-case letter
  • Names and namespaces can contain only alphanumeric characters, '_', or '.' (namespaces only)
  • There is a set of reserved words that can't be used for names, e.g. class. Syncing the schema will fail with an error if you use a reserved word.