glib/docs/reference/glib/gvariant-text.xml
Matthias Clasen 06bb6c75a2 More consistent doc formatting
Move some things around, make capitalization of short descriptions
more consistent.
2011-11-12 21:54:42 -05:00

620 lines
23 KiB
XML

<?xml version='1.0' encoding='utf-8'?>
<refentry id='gvariant-text'>
<refmeta>
<refentrytitle>GVariant Text Format</refentrytitle>
</refmeta>
<refnamediv>
<refname>GVariant Text Format</refname>
<refpurpose>textual representation of GVariants</refpurpose>
</refnamediv>
<refsect1>
<title>GVariant Text Format</title>
<para>
This page attempts to document the GVariant text format as produced by
<link linkend='g-variant-print'><function>g_variant_print()</function></link> and parsed by the
<link linkend='g-variant-parse'><function>g_variant_parse()</function></link> family of functions. In most
cases the style closely resembles the formatting of literals in Python but there are some additions and
exceptions.
</para>
<para>
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.
</para>
<para>
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:
<informalexample><programlisting>[[1, 2, 3], [4, 5, 6]]</programlisting></informalexample>
is parsed as an array of arrays of integers (type '<literal>aai</literal>'), but
<informalexample><programlisting>[[1, 2, 3], [4, 5, 6.0]]</programlisting></informalexample>
is parsed as a array of arrays of doubles (type '<literal>aad</literal>').
</para>
<para>
As another example, GVariant is able to determine that
<informalexample><programlisting>["hello", nothing]</programlisting></informalexample>
is an array of maybe strings (type '<literal>ams</literal>').
</para>
<para>
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.
</para>
</refsect1>
<refsect1>
<title>Syntax Summary</title>
<para>
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.
</para>
<informaltable>
<tgroup cols='2'>
<colspec colname='col_0'/>
<colspec colname='col_1'/>
<tbody>
<row rowsep='1'>
<entry colsep='1' rowsep='1'>
<para>
<emphasis role='strong'>Symbol</emphasis>
</para>
</entry>
<entry colsep='1' rowsep='1'>
<para>
<emphasis role='strong'>Meaning</emphasis>
</para>
</entry>
</row>
<row rowsep='1'>
<entry colsep='1' rowsep='1'>
<para>
<emphasis role='strong'><literal>true</literal></emphasis>,
<emphasis role='strong'><literal>false</literal></emphasis>
</para>
</entry>
<entry colsep='1' rowsep='1'>
<para>
<link linkend='gvariant-text-booleans'>Booleans</link>.
</para>
</entry>
</row>
<row rowsep='1'>
<entry colsep='1' rowsep='1'>
<para>
<emphasis role='strong'><literal>""</literal></emphasis>,
<emphasis role='strong'><literal>''</literal></emphasis>
</para>
</entry>
<entry colsep='1' rowsep='1'>
<para>
String literal. See <link linkend='gvariant-text-strings'>Strings</link> below.
</para>
</entry>
</row>
<row rowsep='1'>
<entry colsep='1' rowsep='1'>
<para>
numbers
</para>
</entry>
<entry colsep='1' rowsep='1'>
<para>
See <link linkend='gvariant-text-numbers'>Numbers</link> below.
</para>
</entry>
</row>
<row rowsep='1'>
<entry colsep='1' rowsep='1'>
<para>
<emphasis role='strong'><literal>()</literal></emphasis>
</para>
</entry>
<entry colsep='1' rowsep='1'>
<para>
<link linkend='gvariant-text-tuples'>Tuples</link>.
</para>
</entry>
</row>
<row rowsep='1'>
<entry colsep='1' rowsep='1'>
<para>
<emphasis role='strong'><literal>[]</literal></emphasis>
</para>
</entry>
<entry colsep='1' rowsep='1'>
<para>
<link linkend='gvariant-text-arrays'>Arrays</link>.
</para>
</entry>
</row>
<row rowsep='1'>
<entry colsep='1' rowsep='1'>
<para>
<emphasis role='strong'><literal>{}</literal></emphasis>
</para>
</entry>
<entry colsep='1' rowsep='1'>
<para>
<link linkend='gvariant-text-dictionaries'>Dictionaries and Dictionary Entries</link>.
</para>
</entry>
</row>
<row rowsep='1'>
<entry colsep='1' rowsep='1'>
<para>
<emphasis role='strong'><literal>&lt;&gt;</literal></emphasis>
</para>
</entry>
<entry colsep='1' rowsep='1'>
<para>
<link linkend='gvariant-text-variants'>Variants</link>.
</para>
</entry>
</row>
<row rowsep='1'>
<entry colsep='1' rowsep='1'>
<para>
<emphasis role='strong'><literal>just</literal></emphasis>,
<emphasis role='strong'><literal>nothing</literal></emphasis>
</para>
</entry>
<entry colsep='1' rowsep='1'>
<para>
<link linkend='gvariant-text-maybe-types'>Maybe Types</link>.
</para>
</entry>
</row>
<row rowsep='1'>
<entry colsep='1' rowsep='1'>
<para>
<emphasis role='strong'><literal>@</literal></emphasis>
</para>
</entry>
<entry colsep='1' rowsep='1'>
<para>
<link linkend='gvariant-text-type-annotations'>Type Annotations</link>.
</para>
</entry>
</row>
<row rowsep='1'>
<entry colsep='1' rowsep='1'>
<para>
type keywords
</para>
</entry>
<entry colsep='1' rowsep='1'>
<para>
<literal>boolean</literal>,
<literal>byte</literal>,
<literal>int16</literal>,
<literal>uint16</literal>,
<literal>int32</literal>,
<literal>uint32</literal>,
<literal>handle</literal>,
<literal>int64</literal>,
<literal>uint64</literal>,
<literal>double</literal>,
<literal>string</literal>,
<literal>objectpath</literal>,
<literal>signature</literal>
</para>
<para>
See <link linkend='gvariant-text-type-annotations'>Type Annotations</link> below.
</para>
</entry>
</row>
<row rowsep='1'>
<entry colsep='1' rowsep='1'>
<para>
<emphasis role='strong'><literal>b""</literal></emphasis>,
<emphasis role='strong'><literal>b''</literal></emphasis>
</para>
</entry>
<entry colsep='1' rowsep='1'>
<para>
<link linkend='gvariant-text-bytestrings'>Bytestrings</link>.
</para>
</entry>
</row>
<row rowsep='1'>
<entry colsep='1' rowsep='1'>
<para>
<emphasis role='strong'><literal>%</literal></emphasis>
</para>
</entry>
<entry colsep='1' rowsep='1'>
<para>
<link linkend='gvariant-text-positional'>Positional Parameters</link>.
</para>
</entry>
</row>
</tbody>
</tgroup>
</informaltable>
<refsect2 id='gvariant-text-booleans'>
<title>Booleans</title>
<para>
The strings <literal>true</literal> and <literal>false</literal> are parsed as booleans. This is the only
way to specify a boolean value.
</para>
</refsect2>
<refsect2 id='gvariant-text-strings'>
<title>Strings</title>
<para>
Strings literals must be quoted using <literal>""</literal> or <literal>''</literal>. The two are
completely equivalent (except for the fact that each one is unable to contain itself unescaped).
</para>
<para>
Strings are Unicode strings with no particular encoding. For example, to specify the character
<literal>é</literal>, you just write <literal>'é'</literal>. You could also give the Unicode codepoint of
that character (U+E9) as the escape sequence <literal>'\u00e9'</literal>. 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.
</para>
<para>
Unicode escapes of the form <literal>\uxxxx</literal> and <literal>\Uxxxxxxxx</literal> are supported, in
hexidecimal. The usual control sequence escapes <literal>\a</literal>, <literal>\b</literal>,
<literal>\f</literal>, <literal>\n</literal>, <literal>\r</literal>, <literal>\t</literal> and
<literal>\v</literal> are supported. Additionally, a <literal>\</literal> before a newline character causes
the newline to be ignored. Finally, any other character following <literal>\</literal> is copied literally
(for example, <literal>\"</literal> or <literal>\\</literal>) but for forwards compatibility with future
additions you should only use this feature when necessary for escaping backslashes or quotes.
</para>
<para>
The usual octal and hexidecimal escapes <literal>\0nnn</literal> and <literal>\xnn</literal> are not
supported here. Those escapes are used to encode byte values and GVariant strings are Unicode.
</para>
<para>
Single-character strings are not interpreted as bytes. Bytes must be specified by their numerical value.
</para>
</refsect2>
<refsect2 id='gvariant-text-numbers'>
<title>Numbers</title>
<para>
Numbers are given by default as decimal values. Octal and hex values can be given in the usual way (by
prefixing with <literal>0</literal> or <literal>0x</literal>). Note that GVariant considers bytes to be
unsigned integers and will print them as a two digit hexidecimal number by default.
</para>
<para>
Floating point numbers can also be given in the usual ways, including scientific and hexidecimal notations.
</para>
<para>
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.
</para>
<para>
Some examples:
</para>
<para>
<literal>5</literal> parses as the int32 value five.
</para>
<para>
<literal>37.5</literal> parses as a floating point value.
</para>
<para>
<literal>3.75e1</literal> parses the same as the value above.
</para>
<para>
<literal>uint64 7</literal> parses seven as a uint64.
See <link linkend='gvariant-text-type-annotations'>Type Annotations</link>.
</para>
</refsect2>
<refsect2 id='gvariant-text-tuples'>
<title>Tuples</title>
<para>
Tuples are formed using the same syntax as Python. Here are some examples:
</para>
<para>
<literal>()</literal> parses as the empty tuple.
</para>
<para>
<literal>(5,)</literal> is a tuple containing a single value.
</para>
<para>
<literal>("hello", 42)</literal> is a pair. Note that values of different types are permitted.
</para>
</refsect2>
<refsect2 id='gvariant-text-arrays'>
<title>Arrays</title>
<para>
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.
</para>
<para>
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.
</para>
<para>
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:
</para>
<para>
<literal>[1]</literal> parses (without additional type information) as a one-item array of signed integers.
</para>
<para>
<literal>[1, 2, 3]</literal> parses (similarly) as a three-item array.
</para>
<para>
<literal>[1, 2, 3.0]</literal> parses as an array of doubles. This is the most simple case of the type
inferencing in action.
</para>
<para>
<literal>[(1, 2), (3, 4.0)]</literal> causes the 2 to also be parsed as a double (but the 1 and 4 are still
integers).
</para>
<para>
<literal>["", nothing]</literal> parses as an array of maybe strings. The presence of
"<literal>nothing</literal>" clearly implies that the array elements are nullable.
</para>
<para>
<literal>[[], [""]]</literal> 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).
</para>
<para>
<literal>[b'hello', []]</literal> looks odd but will parse properly.
See <link linkend='gvariant-text-bytestrings'>Bytestrings</link>
</para>
<para>
And some examples of errors:
</para>
<para>
<literal>["hello", 42]</literal> fails to parse due to conflicting types.
</para>
<para>
<literal>[]</literal> will fail to parse without additional type information.
</para>
</refsect2>
<refsect2 id='gvariant-text-dictionaries'>
<title>Dictionaries and Dictionary Entries</title>
<para>
Dictionaries and dictionary entries are both specified using the <literal>{}</literal> characters.
</para>
<para>
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).
</para>
<para>
The dictionary syntax is the same as Python's syntax for dictionaries. Some examples:
</para>
<para>
<literal>@a{sv} {}</literal> parses as the empty dictionary of everyone's favourite type.
</para>
<para>
<literal>@a{sv} []</literal> is the same as above (owing to the fact that dictionaries are really arrays).
</para>
<para>
<literal>{1: "one", 2: "two", 3: "three"}</literal> parses as a dictionary mapping integers to strings.
</para>
<para>
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:
</para>
<para>
<literal>{1, "one"}</literal> is a free-standing dictionary entry that can be parsed on its own or as part
of another container value.
</para>
<para>
<literal>[{1, "one"}, {2, "two"}, {3, "three"}]</literal> is exactly equivalent to the dictionary example
given above.
</para>
</refsect2>
<refsect2 id='gvariant-text-variants'>
<title>Variants</title>
<para>
Variants are denoted using angle brackets (aka "XML brackets"), <literal>&lt;&gt;</literal>. They may not
be omitted.
</para>
<para>
Using <literal>&lt;&gt;</literal> effectively disrupts the type inferencing that occurs between array
elements. This can have positive and negative effects.
</para>
<para>
<literal>[&lt;"hello"&gt;, &lt;42&gt;]</literal> will parse whereas <literal>["hello", 42]</literal> would
not.
</para>
<para>
<literal>[&lt;['']&gt;, &lt;[]&gt;]</literal> will fail to parse even though <literal>[[''], []]</literal>
parses successfully. You would need to specify <literal>[&lt;['']&gt;, &lt;@as []&gt;]</literal>.
</para>
<para>
<literal>{"title": &lt;"frobit"&gt;, "enabled": &lt;true&gt;, width: &lt;800&gt;}</literal> is an example of
perhaps the most pervasive use of both dictionaries and variants.
</para>
</refsect2>
<refsect2 id='gvariant-text-maybe-types'>
<title>Maybe Types</title>
<para>
The syntax for specifying maybe types is inspired by Haskell.
</para>
<para>
The null case is specified using the keyword <literal>nothing</literal> and the non-null case is explicitly
specified using the keyword <literal>just</literal>. GVariant allows <literal>just</literal> 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:
</para>
<itemizedlist>
<listitem>
<para>when using nested maybes, in order to specify the <literal>just nothing</literal> case</para>
</listitem>
<listitem>
<para>
to establish the nullability of the type of a value without explicitly specifying its full type
</para>
</listitem>
</itemizedlist>
<para>
Some examples:
</para>
<para>
<literal>just 'hello'</literal> parses as a non-null nullable string.
</para>
<para>
<literal>@ms 'hello'</literal> is the same (demonstrating how <literal>just</literal> can be dropped if the type is already
known).
</para>
<para>
<literal>nothing</literal> will not parse wtihout extra type information.
</para>
<para>
<literal>@ms nothing</literal> parses as a null nullable string.
</para>
<para>
<literal>[just 3, nothing]</literal> is an array of nullable integers
</para>
<para>
<literal>[3, nothing]</literal> is the same as the above (demonstrating another place were
<literal>just</literal> can be dropped).
</para>
<para>
<literal>[3, just nothing]</literal> parses as an array of maybe maybe integers (type
<literal>'ammi'</literal>).
</para>
</refsect2>
<refsect2 id='gvariant-text-type-annotations'>
<title>Type Annotations</title>
<para>
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.
</para>
<para>
Type annotations come in two forms: type codes and type keywords.
</para>
<para>
Type keywords can be seen as more verbose (and more legible) versions of a common subset of the type codes.
The type keywords <literal>boolean</literal>, <literal>byte</literal>, <literal>int16</literal>,
<literal>uint16</literal>, <literal>int32</literal>, <literal>uint32</literal>, <literal>handle</literal>,
<literal>int64</literal>, <literal>uint64</literal>, <literal>double</literal>, <literal>string</literal>,
<literal>objectpath</literal> and literal <literal>signature</literal> are each exactly equivalent to their
corresponding type code.
</para>
<para>
Type codes are an <literal>@</literal> ("at" sign) followed by a definite GVariant type string. Some
examples:
</para>
<para>
<literal>uint32 5</literal> causes the number to be parsed unsigned instead of signed (the default).
</para>
<para>
<literal>@u 5</literal> is the same
</para>
<para>
<literal>objectpath "/org/gnome/xyz"</literal> creates an object path instead of a normal string
</para>
<para>
<literal>@au []</literal> specifies the type of the empty array (which would not parse otherwise)
</para>
<para>
<literal>@ms ""</literal> indicates that a string value is meant to have a maybe type
</para>
</refsect2>
<refsect2 id='gvariant-text-bytestrings'>
<title>Bytestrings</title>
<para>
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 '<literal>ay</literal>') with a nul terminator at the end.
</para>
<para>
Bytestrings are specified with either <literal>b""</literal> or <literal>b''</literal>. As with strings,
there is no fundamental difference between the two different types of quotes.
</para>
<para>
Bytestrings support the full range of escapes that you would expect (ie: those supported by
<link linkend='g-strcompress'><function>g_strcompress()</function></link>. This includes the normal control
sequence escapes (as mentioned in the section on strings) as well as octal and hexidecimal escapes of the
forms <literal>\0nnn</literal> and <literal>\xnn</literal>.
</para>
<para>
<literal>b'abc'</literal> is equivalent to <literal>[byte 0x97, 0x98, 0x99, 0]</literal>.
</para>
<para>
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.
</para>
</refsect2>
<refsect2 id='gvariant-text-positional'>
<title>Positional Parameters</title>
<para>
Positional parameters are not a part of the normal GVariant text format, but they are mentioned here because
they can be used with <link linkend='g-variant-new-parsed'><function>g_variant_new_parsed()</function></link>.
</para>
<para>
A positional parameter is indicated with a <literal>%</literal> followed by any valid
<link linkend='gvariant-format-strings'>GVariant Format String</link>. Variable arguments are collected as
specified by the format string and the resulting value is inserted at the current position.
</para>
<para>
This feature is best explained by example:
</para>
<informalexample><programlisting><![CDATA[char *t = "xyz";
gboolean en = false;
GVariant *value;
value = g_variant_new_parsed ("{'title': <%s>, 'enabled': <%b>}", t, en);]]></programlisting></informalexample>
<para>
This constructs a dictionary mapping strings to variants (type '<literal>a{sv}</literal>') 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.
</para>
<para>
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
<literal>%</literal> signs that appear. You can also give format strings that collect no arguments, but
there's no good reason to do so.
</para>
</refsect2>
</refsect1>
</refentry>