GVariant Text Format

This page attempts to document the GVariant text format as produced by g_variant_print() and parsed by the g_variant_parse() family of functions. In most cases the style closely resembles the formatting of literals in Python but there are some additions and exceptions.

The functions that deal with GVariant text format absolutely always deal in utf-8. Conceptually, GVariant text format is a string of Unicode characters -- not bytes. Non-ASCII but otherwise printable Unicode characters are not treated any differently from normal ASCII characters.

The parser makes two passes. The purpose of the first pass is to determine the type of the value being parsed. The second pass does the actual parsing. Based on the fact that all elements in an array have to have the same type, GVariant is able to make some deductions that would not otherwise be possible. As an example:

1
[[1, 2, 3], [4, 5, 6]]

is parsed as an array of arrays of integers (type 'aai'), but

1
[[1, 2, 3], [4, 5, 6.0]]

is parsed as a array of arrays of doubles (type 'aad').

As another example, GVariant is able to determine that

1
["hello", nothing]

is an array of maybe strings (type 'ams').

What the parser accepts as valid input is dependent on context. The API permits for out-of-band type information to be supplied to the parser (which will change its behaviour). This can be seen in the GSettings and GDBus command line utilities where the type information is available from the schema or the remote introspection information. The additional information can cause parses to succeed when they would not otherwise have been able to (by resolving ambiguous type information) or can cause them to fail (due to conflicting type information). Unless stated otherwise, the examples given in this section assume that no out-of-band type data has been given to the parser.

Syntax Summary

The following table describes the rough meaning of symbols that may appear inside GVariant text format. Each symbol is described in detail in its own section, including usage examples.

Symbol

Meaning

true, false

Booleans.

"", ''

String literal. See Strings below.

numbers

See Numbers below.

()

Tuples.

[]

Arrays.

{}

Dictionaries and Dictionary Entries.

<>

Variants.

just, nothing

Maybe Types.

@

Type Annotations.

type keywords

boolean, byte, int16, uint16, int32, uint32, handle, int64, uint64, double, string, objectpath, signature

See Type Annotations below.

b"", b''

Bytestrings.

%

Positional Parameters.

Booleans

The strings true and false are parsed as booleans. This is the only way to specify a boolean value.


Strings

Strings literals must be quoted using "" or ''. The two are completely equivalent (except for the fact that each one is unable to contain itself unescaped).

Strings are Unicode strings with no particular encoding. For example, to specify the character é, you just write 'é'. You could also give the Unicode codepoint of that character (U+E9) as the escape sequence '\u00e9'. Since the strings are pure Unicode, you should not attempt to encode the utf-8 byte sequence corresponding to the string using escapes; it won't work and you'll end up with the individual characters corresponding to each byte.

Unicode escapes of the form \uxxxx and \Uxxxxxxxx are supported, in hexidecimal. The usual control sequence escapes \a, \b, \f, \n, \r, \t and \v are supported. Additionally, a \ before a newline character causes the newline to be ignored. Finally, any other character following \ is copied literally (for example, \" or \\) but for forwards compatibility with future additions you should only use this feature when necessary for escaping backslashes or quotes.

The usual octal and hexidecimal escapes \0nnn and \xnn are not supported here. Those escapes are used to encode byte values and GVariant strings are Unicode.

Single-character strings are not interpreted as bytes. Bytes must be specified by their numerical value.


Numbers

Numbers are given by default as decimal values. Octal and hex values can be given in the usual way (by prefixing with 0 or 0x). Note that GVariant considers bytes to be unsigned integers and will print them as a two digit hexidecimal number by default.

Floating point numbers can also be given in the usual ways, including scientific and hexidecimal notations.

For lack of additional information, integers will be parsed as int32 values by default. If the number has a point or an 'e' in it, then it will be parsed as a double precision floating point number by default. If type information is available (either explicitly or inferred) then that type will be used instead.

Some examples:

5 parses as the int32 value five.

37.5 parses as a floating point value.

3.75e1 parses the same as the value above.

uint64 7 parses seven as a uint64. See Type Annotations.


Tuples

Tuples are formed using the same syntax as Python. Here are some examples:

() parses as the empty tuple.

(5,) is a tuple containing a single value.

("hello", 42) is a pair. Note that values of different types are permitted.


Arrays

Arrays are formed using the same syntax as Python uses for lists (which is arguably the term that GVariant should have used). Note that, unlike Python lists, GVariant arrays are statically typed. This has two implications.

First, all items in the array must have the same type. Second, the type of the array must be known, even in the case that it is empty. This means that (unless there is some other way to infer it) type information will need to be given explicitly for empty arrays.

The parser is able to infer some types based on the fact that all items in an array must have the same type. See the examples below:

[1] parses (without additional type information) as a one-item array of signed integers.

[1, 2, 3] parses (similarly) as a three-item array.

[1, 2, 3.0] parses as an array of doubles. This is the most simple case of the type inferencing in action.

[(1, 2), (3, 4.0)] causes the 2 to also be parsed as a double (but the 1 and 4 are still integers).

["", nothing] parses as an array of maybe strings. The presence of "nothing" clearly implies that the array elements are nullable.

[[], [""]] will parse properly because the type of the first (empty) array can be inferred to be equal to the type of the second array (both are arrays of strings).

[b'hello', []] looks odd but will parse properly. See Bytestrings

And some examples of errors:

["hello", 42] fails to parse due to conflicting types.

[] will fail to parse without additional type information.


Dictionaries and Dictionary Entries

Dictionaries and dictionary entries are both specified using the {} characters.

The dictionary syntax is more commonly used. This is what the printer elects to use in the normal case of dictionary entries appearing in an array (aka "a dictionary"). The separate syntax for dictionary entries is typically only used for when the entries appear on their own, outside of an array (which is valid but unusual). Of course, you are free to use the dictionary entry syntax within arrays but there is no good reason to do so (and the printer itself will never do so). Note that, as with arrays, the type of empty dictionaries must be established (either explicitly or through inference).

The dictionary syntax is the same as Python's syntax for dictionaries. Some examples:

@a{sv} {} parses as the empty dictionary of everyone's favourite type.

@a{sv} [] is the same as above (owing to the fact that dictionaries are really arrays).

{1: "one", 2: "two", 3: "three"} parses as a dictionary mapping integers to strings.

The dictionary entry syntax looks just like a pair (2-tuple) that uses braces instead of parens. The presence of a comma immediately following the key differentiates it from the dictionary syntax (which features a colon after the first key). Some examples:

{1, "one"} is a free-standing dictionary entry that can be parsed on its own or as part of another container value.

[{1, "one"}, {2, "two"}, {3, "three"}] is exactly equivalent to the dictionary example given above.


Variants

Variants are denoted using angle brackets (aka "XML brackets"), <>. They may not be omitted.

Using <> effectively disrupts the type inferencing that occurs between array elements. This can have positive and negative effects.

[<"hello">, <42>] will parse whereas ["hello", 42] would not.

[<['']>, <[]>] will fail to parse even though [[''], []] parses successfully. You would need to specify [<['']>, <@as []>].

{"title": <"frobit">, "enabled": <true>, width: <800>} is an example of perhaps the most pervasive use of both dictionaries and variants.


Maybe Types

The syntax for specifying maybe types is inspired by Haskell.

The null case is specified using the keyword nothing and the non-null case is explicitly specified using the keyword just. GVariant allows just to be omitted in every case that it is able to unambiguously determine the intention of the writer. There are two cases where it must be specified:

  • when using nested maybes, in order to specify the just nothing case

  • to establish the nullability of the type of a value without explicitly specifying its full type

Some examples:

just 'hello' parses as a non-null nullable string.

@ms 'hello' is the same (demonstrating how just can be dropped if the type is already known).

nothing will not parse wtihout extra type information.

@ms nothing parses as a null nullable string.

[just 3, nothing] is an array of nullable integers

[3, nothing] is the same as the above (demonstrating another place were just can be dropped).

[3, just nothing] parses as an array of maybe maybe integers (type 'ammi').


Type Annotations

Type annotations allow additional type information to be given to the parser. Depending on the context, this type information can change the output of the parser, cause an error when parsing would otherwise have succeeded or resolve an error when parsing would have otherwise failed.

Type annotations come in two forms: type codes and type keywords.

Type keywords can be seen as more verbose (and more legible) versions of a common subset of the type codes. The type keywords boolean, byte, int16, uint16, int32, uint32, handle, int64, uint64, double, string, objectpath and literal signature are each exactly equivalent to their corresponding type code.

Type codes are an @ ("at" sign) followed by a definite GVariant type string. Some examples:

uint32 5 causes the number to be parsed unsigned instead of signed (the default).

@u 5 is the same

objectpath "/org/gnome/xyz" creates an object path instead of a normal string

@au [] specifies the type of the empty array (which would not parse otherwise)

@ms "" indicates that a string value is meant to have a maybe type


Bytestrings

The bytestring syntax is a piece of syntactic sugar meant to complement the bytestring APIs in GVariant. It constructs arrays of non-nul bytes (type 'ay') with a nul terminator at the end.

Bytestrings are specified with either b"" or b''. As with strings, there is no fundamental difference between the two different types of quotes.

Bytestrings support the full range of escapes that you would expect (ie: those supported by g_strcompress(). This includes the normal control sequence escapes (as mentioned in the section on strings) as well as octal and hexidecimal escapes of the forms \0nnn and \xnn.

b'abc' is equivalent to [byte 0x97, 0x98, 0x99, 0].

When formatting arrays of bytes, the printer will choose to display the array as a bytestring if it contains a nul character at the end and no other nul bytes within. Otherwise, it is formatted as a normal array.


Positional Parameters

Positional parameters are not a part of the normal GVariant text format, but they are mentioned here because they can be used with g_variant_new_parsed().

A positional parameter is indicated with a % followed by any valid GVariant Format String. Variable arguments are collected as specified by the format string and the resulting value is inserted at the current position.

This feature is best explained by example:

1
2
3
4
5
char *t = "xyz";
gboolean en = false;
GVariant *value;

value = g_variant_new_parsed ("{'title': <%s>, 'enabled': <%b>}", t, en);

This constructs a dictionary mapping strings to variants (type 'a{sv}') with two items in it. The key names are parsed from the string and the values for those keys are taken as variable arguments parameters.

The arguments are always collected in the order that they appear in the string to be parsed. Format strings that collect multiple arguments are permitted, so you may require more varargs parameters than the number of % signs that appear. You can also give format strings that collect no arguments, but there's no good reason to do so.