<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN" "http://www.w3.org/TR/html4/strict.dtd">
<html>
<head>
<title>FFI Semantics</title>
<meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1">
<meta name="Author" content="Mike Pall">
<meta name="Copyright" content="Copyright (C) 2005-2012, Mike Pall">
<meta name="Language" content="en">
<link rel="stylesheet" type="text/css" href="bluequad.css" media="screen">
<link rel="stylesheet" type="text/css" href="bluequad-print.css" media="print">
<style type="text/css">
table.convtable { line-height: 1.2; }
tr.convhead td { font-weight: bold; }
td.convin { width: 11em; }
td.convop { font-style: italic; width: 16em; }
</style>
</head>
<body>
<div id="site">
<a href="http://luajit.org"><span>Lua<span id="logo">JIT</span></span></a>
</div>
<div id="head">
<h1>FFI Semantics</h1>
</div>
<div id="nav">
<ul><li>
<a href="luajit.html">LuaJIT</a>
<ul><li>
<a href="install.html">Installation</a>
</li><li>
<a href="running.html">Running</a>
</li></ul>
</li><li>
<a href="extensions.html">Extensions</a>
<ul><li>
<a href="ext_ffi.html">FFI Library</a>
<ul><li>
<a href="ext_ffi_tutorial.html">FFI Tutorial</a>
</li><li>
<a href="ext_ffi_api.html">ffi.* API</a>
</li><li>
<a class="current" href="ext_ffi_semantics.html">FFI Semantics</a>
</li></ul>
</li><li>
<a href="ext_jit.html">jit.* Library</a>
</li><li>
<a href="ext_c_api.html">Lua/C API</a>
</li></ul>
</li><li>
<a href="status.html">Status</a>
<ul><li>
<a href="changes.html">Changes</a>
</li></ul>
</li><li>
<a href="faq.html">FAQ</a>
</li><li>
<a href="http://luajit.org/performance.html">Performance <span class="ext">»</span></a>
</li><li>
<a href="http://luajit.org/download.html">Download <span class="ext">»</span></a>
</li><li>
<a href="http://wiki.luajit.org/">Wiki <span class="ext">»</span></a>
</li><li>
<a href="http://luajit.org/list.html">Mailing List <span class="ext">»</span></a>
</li></ul>
</div>
<div id="main">
<p>
This page describes the detailed semantics underlying the FFI library
and its interaction with both Lua and C code.
</p>
<p>
Given that the FFI library is designed to interface with C code
and that declarations can be written in plain C syntax, <b>it
closely follows the C language semantics</b>, wherever possible.
Some minor concessions are needed for smoother interoperation with Lua
language semantics.
</p>
<p>
Please don't be overwhelmed by the contents of this page — this
is a reference and you may need to consult it, if in doubt. It doesn't
hurt to skim this page, but most of the semantics "just work" as you'd
expect them to work. It should be straightforward to write
applications using the LuaJIT FFI for developers with a C or C++
background.
</p>
<p class="indent" style="color: #c00000;">
Please note: this doesn't comprise the final specification for the FFI
semantics, yet. Some semantics may need to be changed, based on your
feedback. Please <a href="contact.html">report</a> any problems you may
encounter or any improvements you'd like to see — thank you!
</p>
<h2 id="clang">C Language Support</h2>
<p>
The FFI library has a built-in C parser with a minimal memory
footprint. It's used by the <a href="ext_ffi_api.html">ffi.* library
functions</a> to declare C types or external symbols.
</p>
<p>
It's only purpose is to parse C declarations, as found e.g. in
C header files. Although it does evaluate constant expressions,
it's <em>not</em> a C compiler. The body of <tt>inline</tt>
C function definitions is simply ignored.
</p>
<p>
Also, this is <em>not</em> a validating C parser. It expects and
accepts correctly formed C declarations, but it may choose to
ignore bad declarations or show rather generic error messages. If in
doubt, please check the input against your favorite C compiler.
</p>
<p>
The C parser complies to the <b>C99 language standard</b> plus
the following extensions:
</p>
<ul>
<li>The <tt>'\e'</tt> escape in character and string literals.</li>
<li>The C99/C++ boolean type, declared with the keywords <tt>bool</tt>
or <tt>_Bool</tt>.</li>
<li>Complex numbers, declared with the keywords <tt>complex</tt> or
<tt>_Complex</tt>.</li>
<li>Two complex number types: <tt>complex</tt> (aka
<tt>complex double</tt>) and <tt>complex float</tt>.</li>
<li>Vector types, declared with the GCC <tt>mode</tt> or
<tt>vector_size</tt> attribute.</li>
<li>Unnamed ('transparent') <tt>struct</tt>/<tt>union</tt> fields
inside a <tt>struct</tt>/<tt>union</tt>.</li>
<li>Incomplete <tt>enum</tt> declarations, handled like incomplete
<tt>struct</tt> declarations.</li>
<li>Unnamed <tt>enum</tt> fields inside a
<tt>struct</tt>/<tt>union</tt>. This is similar to a scoped C++
<tt>enum</tt>, except that declared constants are visible in the
global namespace, too.</li>
<li>Scoped <tt>static const</tt> declarations inside a
<tt>struct</tt>/<tt>union</tt> (from C++).</li>
<li>Zero-length arrays (<tt>[0]</tt>), empty
<tt>struct</tt>/<tt>union</tt>, variable-length arrays (VLA,
<tt>[?]</tt>) and variable-length structs (VLS, with a trailing
VLA).</li>
<li>C++ reference types (<tt>int &x</tt>).</li>
<li>Alternate GCC keywords with '<tt>__</tt>', e.g.
<tt>__const__</tt>.</li>
<li>GCC <tt>__attribute__</tt> with the following attributes:
<tt>aligned</tt>, <tt>packed</tt>, <tt>mode</tt>,
<tt>vector_size</tt>, <tt>cdecl</tt>, <tt>fastcall</tt>,
<tt>stdcall</tt>.</li>
<li>The GCC <tt>__extension__</tt> keyword and the GCC
<tt>__alignof__</tt> operator.</li>
<li>GCC <tt>__asm__("symname")</tt> symbol name redirection for
function declarations.</li>
<li>MSVC keywords for fixed-length types: <tt>__int8</tt>,
<tt>__int16</tt>, <tt>__int32</tt> and <tt>__int64</tt>.</li>
<li>MSVC <tt>__cdecl</tt>, <tt>__fastcall</tt>, <tt>__stdcall</tt>,
<tt>__ptr32</tt>, <tt>__ptr64</tt>, <tt>__declspec(align(n))</tt>
and <tt>#pragma pack</tt>.</li>
<li>All other GCC/MSVC-specific attributes are ignored.</li>
</ul>
<p>
The following C types are pre-defined by the C parser (like
a <tt>typedef</tt>, except re-declarations will be ignored):
</p>
<ul>
<li>Vararg handling: <tt>va_list</tt>, <tt>__builtin_va_list</tt>,
<tt>__gnuc_va_list</tt>.</li>
<li>From <tt><stddef.h></tt>: <tt>ptrdiff_t</tt>,
<tt>size_t</tt>, <tt>wchar_t</tt>.</li>
<li>From <tt><stdint.h></tt>: <tt>int8_t</tt>, <tt>int16_t</tt>,
<tt>int32_t</tt>, <tt>int64_t</tt>, <tt>uint8_t</tt>,
<tt>uint16_t</tt>, <tt>uint32_t</tt>, <tt>uint64_t</tt>,
<tt>intptr_t</tt>, <tt>uintptr_t</tt>.</li>
</ul>
<p>
You're encouraged to use these types in preference to
compiler-specific extensions or target-dependent standard types.
E.g. <tt>char</tt> differs in signedness and <tt>long</tt> differs in
size, depending on the target architecture and platform ABI.
</p>
<p>
The following C features are <b>not</b> supported:
</p>
<ul>
<li>A declaration must always have a type specifier; it doesn't
default to an <tt>int</tt> type.</li>
<li>Old-style empty function declarations (K&R) are not allowed.
All C functions must have a proper prototype declaration. A
function declared without parameters (<tt>int foo();</tt>) is
treated as a function taking zero arguments, like in C++.</li>
<li>The <tt>long double</tt> C type is parsed correctly, but
there's no support for the related conversions, accesses or arithmetic
operations.</li>
<li>Wide character strings and character literals are not
supported.</li>
<li><a href="#status">See below</a> for features that are currently
not implemented.</li>
</ul>
<h2 id="convert">C Type Conversion Rules</h2>
<h3 id="convert_tolua">Conversions from C types to Lua objects</h3>
<p>
These conversion rules apply for <em>read accesses</em> to
C types: indexing pointers, arrays or
<tt>struct</tt>/<tt>union</tt> types; reading external variables or
constant values; retrieving return values from C calls:
</p>
<table class="convtable">
<tr class="convhead">
<td class="convin">Input</td>
<td class="convop">Conversion</td>
<td class="convout">Output</td>
</tr>
<tr class="odd separate">
<td class="convin"><tt>int8_t</tt>, <tt>int16_t</tt></td><td class="convop">→<sup>sign-ext</sup> <tt>int32_t</tt> → <tt>double</tt></td><td class="convout">number</td></tr>
<tr class="even">
<td class="convin"><tt>uint8_t</tt>, <tt>uint16_t</tt></td><td class="convop">→<sup>zero-ext</sup> <tt>int32_t</tt> → <tt>double</tt></td><td class="convout">number</td></tr>
<tr class="odd">
<td class="convin"><tt>int32_t</tt>, <tt>uint32_t</tt></td><td class="convop">→ <tt>double</tt></td><td class="convout">number</td></tr>
<tr class="even">
<td class="convin"><tt>int64_t</tt>, <tt>uint64_t</tt></td><td class="convop">boxed value</td><td class="convout">64 bit int cdata</td></tr>
<tr class="odd separate">
<td class="convin"><tt>double</tt>, <tt>float</tt></td><td class="convop">→ <tt>double</tt></td><td class="convout">number</td></tr>
<tr class="even separate">
<td class="convin"><tt>bool</tt></td><td class="convop">0 → <tt>false</tt>, otherwise <tt>true</tt></td><td class="convout">boolean</td></tr>
<tr class="odd separate">
<td class="convin"><tt>enum</tt></td><td class="convop">boxed value</td><td class="convout">enum cdata</td></tr>
<tr class="even">
<td class="convin">Complex number</td><td class="convop">boxed value</td><td class="convout">complex cdata</td></tr>
<tr class="odd">
<td class="convin">Vector</td><td class="convop">boxed value</td><td class="convout">vector cdata</td></tr>
<tr class="even">
<td class="convin">Pointer</td><td class="convop">boxed value</td><td class="convout">pointer cdata</td></tr>
<tr class="odd separate">
<td class="convin">Array</td><td class="convop">boxed reference</td><td class="convout">reference cdata</td></tr>
<tr class="even">
<td class="convin"><tt>struct</tt>/<tt>union</tt></td><td class="convop">boxed reference</td><td class="convout">reference cdata</td></tr>
</table>
<p>
Bitfields are treated like their underlying type.
</p>
<p>
Reference types are dereferenced <em>before</em> a conversion can take
place — the conversion is applied to the C type pointed to
by the reference.
</p>
<h3 id="convert_fromlua">Conversions from Lua objects to C types</h3>
<p>
These conversion rules apply for <em>write accesses</em> to
C types: indexing pointers, arrays or
<tt>struct</tt>/<tt>union</tt> types; initializing cdata objects;
casts to C types; writing to external variables; passing
arguments to C calls:
</p>
<table class="convtable">
<tr class="convhead">
<td class="convin">Input</td>
<td class="convop">Conversion</td>
<td class="convout">Output</td>
</tr>
<tr class="odd separate">
<td class="convin">number</td><td class="convop">→</td><td class="convout"><tt>double</tt></td></tr>
<tr class="even">
<td class="convin">boolean</td><td class="convop"><tt>false</tt> → 0, <tt>true</tt> → 1</td><td class="convout"><tt>bool</tt></td></tr>
<tr class="odd separate">
<td class="convin">nil</td><td class="convop"><tt>NULL</tt> →</td><td class="convout"><tt>(void *)</tt></td></tr>
<tr class="even">
<td class="convin">lightuserdata</td><td class="convop">lightuserdata address →</td><td class="convout"><tt>(void *)</tt></td></tr>
<tr class="odd">
<td class="convin">userdata</td><td class="convop">userdata payload →</td><td class="convout"><tt>(void *)</tt></td></tr>
<tr class="even">
<td class="convin">io.* file</td><td class="convop">get FILE * handle →</td><td class="convout"><tt>(void *)</tt></td></tr>
<tr class="odd separate">
<td class="convin">string</td><td class="convop">match against <tt>enum</tt> constant</td><td class="convout"><tt>enum</tt></td></tr>
<tr class="even">
<td class="convin">string</td><td class="convop">copy string data + zero-byte</td><td class="convout"><tt>int8_t[]</tt>, <tt>uint8_t[]</tt></td></tr>
<tr class="odd">
<td class="convin">string</td><td class="convop">string data →</td><td class="convout"><tt>const char[]</tt></td></tr>
<tr class="even separate">
<td class="convin">function</td><td class="convop"><a href="#callback">create callback</a> →</td><td class="convout">C function type</td></tr>
<tr class="odd separate">
<td class="convin">table</td><td class="convop"><a href="#init_table">table initializer</a></td><td class="convout">Array</td></tr>
<tr class="even">
<td class="convin">table</td><td class="convop"><a href="#init_table">table initializer</a></td><td class="convout"><tt>struct</tt>/<tt>union</tt></td></tr>
<tr class="odd separate">
<td class="convin">cdata</td><td class="convop">cdata payload →</td><td class="convout">C type</td></tr>
</table>
<p>
If the result type of this conversion doesn't match the
C type of the destination, the
<a href="#convert_between">conversion rules between C types</a>
are applied.
</p>
<p>
Reference types are immutable after initialization ("no re-seating of
references"). For initialization purposes or when passing values to
reference parameters, they are treated like pointers. Note that unlike
in C++, there's no way to implement automatic reference generation of
variables under the Lua language semantics. If you want to call a
function with a reference parameter, you need to explicitly pass a
one-element array.
</p>
<h3 id="convert_between">Conversions between C types</h3>
<p>
These conversion rules are more or less the same as the standard
C conversion rules. Some rules only apply to casts, or require
pointer or type compatibility:
</p>
<table class="convtable">
<tr class="convhead">
<td class="convin">Input</td>
<td class="convop">Conversion</td>
<td class="convout">Output</td>
</tr>
<tr class="odd separate">
<td class="convin">Signed integer</td><td class="convop">→<sup>narrow or sign-extend</sup></td><td class="convout">Integer</td></tr>
<tr class="even">
<td class="convin">Unsigned integer</td><td class="convop">→<sup>narrow or zero-extend</sup></td><td class="convout">Integer</td></tr>
<tr class="odd">
<td class="convin">Integer</td><td class="convop">→<sup>round</sup></td><td class="convout"><tt>double</tt>, <tt>float</tt></td></tr>
<tr class="even">
<td class="convin"><tt>double</tt>, <tt>float</tt></td><td class="convop">→<sup>trunc</sup> <tt>int32_t</tt> →<sup>narrow</sup></td><td class="convout"><tt>(u)int8_t</tt>, <tt>(u)int16_t</tt></td></tr>
<tr class="odd">
<td class="convin"><tt>double</tt>, <tt>float</tt></td><td class="convop">→<sup>trunc</sup></td><td class="convout"><tt>(u)int32_t</tt>, <tt>(u)int64_t</tt></td></tr>
<tr class="even">
<td class="convin"><tt>double</tt>, <tt>float</tt></td><td class="convop">→<sup>round</sup></td><td class="convout"><tt>float</tt>, <tt>double</tt></td></tr>
<tr class="odd separate">
<td class="convin">Number</td><td class="convop">n == 0 → 0, otherwise 1</td><td class="convout"><tt>bool</tt></td></tr>
<tr class="even">
<td class="convin"><tt>bool</tt></td><td class="convop"><tt>false</tt> → 0, <tt>true</tt> → 1</td><td class="convout">Number</td></tr>
<tr class="odd separate">
<td class="convin">Complex number</td><td class="convop">convert real part</td><td class="convout">Number</td></tr>
<tr class="even">
<td class="convin">Number</td><td class="convop">convert real part, imag = 0</td><td class="convout">Complex number</td></tr>
<tr class="odd">
<td class="convin">Complex number</td><td class="convop">convert real and imag part</td><td class="convout">Complex number</td></tr>
<tr class="even separate">
<td class="convin">Number</td><td class="convop">convert scalar and replicate</td><td class="convout">Vector</td></tr>
<tr class="odd">
<td class="convin">Vector</td><td class="convop">copy (same size)</td><td class="convout">Vector</td></tr>
<tr class="even separate">
<td class="convin"><tt>struct</tt>/<tt>union</tt></td><td class="convop">take base address (compat)</td><td class="convout">Pointer</td></tr>
<tr class="odd">
<td class="convin">Array</td><td class="convop">take base address (compat)</td><td class="convout">Pointer</td></tr>
<tr class="even">
<td class="convin">Function</td><td class="convop">take function address</td><td class="convout">Function pointer</td></tr>
<tr class="odd separate">
<td class="convin">Number</td><td class="convop">convert via <tt>uintptr_t</tt> (cast)</td><td class="convout">Pointer</td></tr>
<tr class="even">
<td class="convin">Pointer</td><td class="convop">convert address (compat/cast)</td><td class="convout">Pointer</td></tr>
<tr class="odd">
<td class="convin">Pointer</td><td class="convop">convert address (cast)</td><td class="convout">Integer</td></tr>
<tr class="even">
<td class="convin">Array</td><td class="convop">convert base address (cast)</td><td class="convout">Integer</td></tr>
<tr class="odd separate">
<td class="convin">Array</td><td class="convop">copy (compat)</td><td class="convout">Array</td></tr>
<tr class="even">
<td class="convin"><tt>struct</tt>/<tt>union</tt></td><td class="convop">copy (identical type)</td><td class="convout"><tt>struct</tt>/<tt>union</tt></td></tr>
</table>
<p>
Bitfields or <tt>enum</tt> types are treated like their underlying
type.
</p>
<p>
Conversions not listed above will raise an error. E.g. it's not
possible to convert a pointer to a complex number or vice versa.
</p>
<h3 id="convert_vararg">Conversions for vararg C function arguments</h3>
<p>
The following default conversion rules apply when passing Lua objects
to the variable argument part of vararg C functions:
</p>
<table class="convtable">
<tr class="convhead">
<td class="convin">Input</td>
<td class="convop">Conversion</td>
<td class="convout">Output</td>
</tr>
<tr class="odd separate">
<td class="convin">number</td><td class="convop">→</td><td class="convout"><tt>double</tt></td></tr>
<tr class="even">
<td class="convin">boolean</td><td class="convop"><tt>false</tt> → 0, <tt>true</tt> → 1</td><td class="convout"><tt>bool</tt></td></tr>
<tr class="odd separate">
<td class="convin">nil</td><td class="convop"><tt>NULL</tt> →</td><td class="convout"><tt>(void *)</tt></td></tr>
<tr class="even">
<td class="convin">userdata</td><td class="convop">userdata payload →</td><td class="convout"><tt>(void *)</tt></td></tr>
<tr class="odd">
<td class="convin">lightuserdata</td><td class="convop">lightuserdata address →</td><td class="convout"><tt>(void *)</tt></td></tr>
<tr class="even separate">
<td class="convin">string</td><td class="convop">string data →</td><td class="convout"><tt>const char *</tt></td></tr>
<tr class="odd separate">
<td class="convin"><tt>float</tt> cdata</td><td class="convop">→</td><td class="convout"><tt>double</tt></td></tr>
<tr class="even">
<td class="convin">Array cdata</td><td class="convop">take base address</td><td class="convout">Element pointer</td></tr>
<tr class="odd">
<td class="convin"><tt>struct</tt>/<tt>union</tt> cdata</td><td class="convop">take base address</td><td class="convout"><tt>struct</tt>/<tt>union</tt> pointer</td></tr>
<tr class="even">
<td class="convin">Function cdata</td><td class="convop">take function address</td><td class="convout">Function pointer</td></tr>
<tr class="odd">
<td class="convin">Any other cdata</td><td class="convop">no conversion</td><td class="convout">C type</td></tr>
</table>
<p>
To pass a Lua object, other than a cdata object, as a specific type,
you need to override the conversion rules: create a temporary cdata
object with a constructor or a cast and initialize it with the value
to pass:
</p>
<p>
Assuming <tt>x</tt> is a Lua number, here's how to pass it as an
integer to a vararg function:
</p>
<pre class="code">
ffi.cdef[[
int printf(const char *fmt, ...);
]]
ffi.C.printf("integer value: %d\n", ffi.new("int", x))
</pre>
<p>
If you don't do this, the default Lua number → <tt>double</tt>
conversion rule applies. A vararg C function expecting an integer
will see a garbled or uninitialized value.
</p>
<h2 id="init">Initializers</h2>
<p>
Creating a cdata object with
<a href="ext_ffi_api.html#ffi_new"><tt>ffi.new()</tt></a> or the
equivalent constructor syntax always initializes its contents, too.
Different rules apply, depending on the number of optional
initializers and the C types involved:
</p>
<ul>
<li>If no initializers are given, the object is filled with zero bytes.</li>
<li>Scalar types (numbers and pointers) accept a single initializer.
The Lua object is <a href="#convert_fromlua">converted to the scalar
C type</a>.</li>
<li>Valarrays (complex numbers and vectors) are treated like scalars
when a single initializer is given. Otherwise they are treated like
regular arrays.</li>
<li>Aggregate types (arrays and structs) accept either a single
<a href="#init_table">table initializer</a> or a flat list of
initializers.</li>
<li>The elements of an array are initialized, starting at index zero.
If a single initializer is given for an array, it's repeated for all
remaining elements. This doesn't happen if two or more initializers
are given: all remaining uninitialized elements are filled with zero
bytes.</li>
<li>Byte arrays may also be initialized with a Lua string. This copies
the whole string plus a terminating zero-byte. The copy stops early only
if the array has a known, fixed size.</li>
<li>The fields of a <tt>struct</tt> are initialized in the order of
their declaration. Uninitialized fields are filled with zero
bytes.</li>
<li>Only the first field of a <tt>union</tt> can be initialized with a
flat initializer.</li>
<li>Elements or fields which are aggregates themselves are initialized
with a <em>single</em> initializer, but this may be a table
initializer or a compatible aggregate.</li>
<li>Excess initializers cause an error.</li>
</ul>
<h2 id="init_table">Table Initializers</h2>
<p>
The following rules apply if a Lua table is used to initialize an
Array or a <tt>struct</tt>/<tt>union</tt>:
</p>
<ul>
<li>If the table index <tt>[0]</tt> is non-<tt>nil</tt>, then the
table is assumed to be zero-based. Otherwise it's assumed to be
one-based.</li>
<li>Array elements, starting at index zero, are initialized one-by-one
with the consecutive table elements, starting at either index
<tt>[0]</tt> or <tt>[1]</tt>. This process stops at the first
<tt>nil</tt> table element.</li>
<li>If exactly one array element was initialized, it's repeated for
all the remaining elements. Otherwise all remaining uninitialized
elements are filled with zero bytes.</li>
<li>The above logic only applies to arrays with a known fixed size.
A VLA is only initialized with the element(s) given in the table.
Depending on the use case, you may need to explicitly add a
<tt>NULL</tt> or <tt>0</tt> terminator to a VLA.</li>
<li>If the table has a non-empty hash part, a
<tt>struct</tt>/<tt>union</tt> is initialized by looking up each field
name (as a string key) in the table. Each non-<tt>nil</tt> value is
used to initialize the corresponding field.</li>
<li>Otherwise a <tt>struct</tt>/<tt>union</tt> is initialized in the
order of the declaration of its fields. Each field is initialized with
the consecutive table elements, starting at either index <tt>[0]</tt>
or <tt>[1]</tt>. This process stops at the first <tt>nil</tt> table
element.</li>
<li>Uninitialized fields of a <tt>struct</tt> are filled with zero
bytes, except for the trailing VLA of a VLS.</li>
<li>Initialization of a <tt>union</tt> stops after one field has been
initialized. If no field has been initialized, the <tt>union</tt> is
filled with zero bytes.</li>
<li>Elements or fields which are aggregates themselves are initialized
with a <em>single</em> initializer, but this may be a nested table
initializer (or a compatible aggregate).</li>
<li>Excess initializers for an array cause an error. Excess
initializers for a <tt>struct</tt>/<tt>union</tt> are ignored.
Unrelated table entries are ignored, too.</li>
</ul>
<p>
Example:
</p>
<pre class="code">
local ffi = require("ffi")
ffi.cdef[[
struct foo { int a, b; };
union bar { int i; double d; };
struct nested { int x; struct foo y; };
]]
ffi.new("int[3]", {}) --> 0, 0, 0
ffi.new("int[3]", {1}) --> 1, 1, 1
ffi.new("int[3]", {1,2}) --> 1, 2, 0
ffi.new("int[3]", {1,2,3}) --> 1, 2, 3
ffi.new("int[3]", {[0]=1}) --> 1, 1, 1
ffi.new("int[3]", {[0]=1,2}) --> 1, 2, 0
ffi.new("int[3]", {[0]=1,2,3}) --> 1, 2, 3
ffi.new("int[3]", {[0]=1,2,3,4}) --> error: too many initializers
ffi.new("struct foo", {}) --> a = 0, b = 0
ffi.new("struct foo", {1}) --> a = 1, b = 0
ffi.new("struct foo", {1,2}) --> a = 1, b = 2
ffi.new("struct foo", {[0]=1,2}) --> a = 1, b = 2
ffi.new("struct foo", {b=2}) --> a = 0, b = 2
ffi.new("struct foo", {a=1,b=2,c=3}) --> a = 1, b = 2 'c' is ignored
ffi.new("union bar", {}) --> i = 0, d = 0.0
ffi.new("union bar", {1}) --> i = 1, d = ?
ffi.new("union bar", {[0]=1,2}) --> i = 1, d = ? '2' is ignored
ffi.new("union bar", {d=2}) --> i = ?, d = 2.0
ffi.new("struct nested", {1,{2,3}}) --> x = 1, y.a = 2, y.b = 3
ffi.new("struct nested", {x=1,y={2,3}}) --> x = 1, y.a = 2, y.b = 3
</pre>
<h2 id="cdata_ops">Operations on cdata Objects</h2>
<p>
All of the standard Lua operators can be applied to cdata objects or a
mix of a cdata object and another Lua object. The following list shows
the pre-defined operations.
</p>
<p>
Reference types are dereferenced <em>before</em> performing each of
the operations below — the operation is applied to the
C type pointed to by the reference.
</p>
<p>
The pre-defined operations are always tried first before deferring to a
metamethod or index table (if any) for the corresponding ctype (except
for <tt>__new</tt>). An error is raised if the metamethod lookup or
index table lookup fails.
</p>
<h3 id="cdata_array">Indexing a cdata object</h3>
<ul>
<li><b>Indexing a pointer/array</b>: a cdata pointer/array can be
indexed by a cdata number or a Lua number. The element address is
computed as the base address plus the number value multiplied by the
element size in bytes. A read access loads the element value and
<a href="#convert_tolua">converts it to a Lua object</a>. A write
access <a href="#convert_fromlua">converts a Lua object to the element
type</a> and stores the converted value to the element. An error is
raised if the element size is undefined or a write access to a
constant element is attempted.</li>
<li><b>Dereferencing a <tt>struct</tt>/<tt>union</tt> field</b>: a
cdata <tt>struct</tt>/<tt>union</tt> or a pointer to a
<tt>struct</tt>/<tt>union</tt> can be dereferenced by a string key,
giving the field name. The field address is computed as the base
address plus the relative offset of the field. A read access loads the
field value and <a href="#convert_tolua">converts it to a Lua
object</a>. A write access <a href="#convert_fromlua">converts a Lua
object to the field type</a> and stores the converted value to the
field. An error is raised if a write access to a constant
<tt>struct</tt>/<tt>union</tt> or a constant field is attempted.
Scoped enum constants or static constants are treated like a constant
field.</li>
<li><b>Indexing a complex number</b>: a complex number can be indexed
either by a cdata number or a Lua number with the values 0 or 1, or by
the strings <tt>"re"</tt> or <tt>"im"</tt>. A read access loads the
real part (<tt>[0]</tt>, <tt>.re</tt>) or the imaginary part
(<tt>[1]</tt>, <tt>.im</tt>) part of a complex number and
<a href="#convert_tolua">converts it to a Lua number</a>. The
sub-parts of a complex number are immutable — assigning to an
index of a complex number raises an error. Accessing out-of-bound
indexes returns unspecified results, but is guaranteed not to trigger
memory access violations.</li>
<li><b>Indexing a vector</b>: a vector is treated like an array for
indexing purposes, except the vector elements are immutable —
assigning to an index of a vector raises an error.</li>
</ul>
<p>
A ctype object can be indexed with a string key, too. The only
pre-defined operation is reading scoped constants of
<tt>struct</tt>/<tt>union</tt> types. All other accesses defer
to the corresponding metamethods or index tables (if any).
</p>
<p>
Note: since there's (deliberately) no address-of operator, a cdata
object holding a value type is effectively immutable after
initialization. The JIT compiler benefits from this fact when applying
certain optimizations.
</p>
<p>
As a consequence, the <em>elements</em> of complex numbers and
vectors are immutable. But the elements of an aggregate holding these
types <em>may</em> be modified of course. I.e. you cannot assign to
<tt>foo.c.im</tt>, but you can assign a (newly created) complex number
to <tt>foo.c</tt>.
</p>
<p>
The JIT compiler implements strict aliasing rules: accesses to different
types do <b>not</b> alias, except for differences in signedness (this
applies even to <tt>char</tt> pointers, unlike C99). Type punning
through unions is explicitly detected and allowed.
</p>
<h3 id="cdata_call">Calling a cdata object</h3>
<ul>
<li><b>Constructor</b>: a ctype object can be called and used as a
<a href="ext_ffi_api.html#ffi_new">constructor</a>. This is equivalent
to <tt>ffi.new(ct, ...)</tt>, unless a <tt>__new</tt> metamethod is
defined. The <tt>__new</tt> metamethod is called with the ctype object
plus any other arguments passed to the contructor. Note that you have to
use <tt>ffi.new</tt> inside of it, since calling <tt>ct(...)</tt> would
cause infinite recursion.</li>
<li><b>C function call</b>: a cdata function or cdata function
pointer can be called. The passed arguments are
<a href="#convert_fromlua">converted to the C types</a> of the
parameters given by the function declaration. Arguments passed to the
variable argument part of vararg C function use
<a href="#convert_vararg">special conversion rules</a>. This
C function is called and the return value (if any) is
<a href="#convert_tolua">converted to a Lua object</a>.<br>
On Windows/x86 systems, <tt>__stdcall</tt> functions are automatically
detected and a function declared as <tt>__cdecl</tt> (the default) is
silently fixed up after the first call.</li>
</ul>
<h3 id="cdata_arith">Arithmetic on cdata objects</h3>
<ul>
<li><b>Pointer arithmetic</b>: a cdata pointer/array and a cdata
number or a Lua number can be added or subtracted. The number must be
on the right hand side for a subtraction. The result is a pointer of
the same type with an address plus or minus the number value
multiplied by the element size in bytes. An error is raised if the
element size is undefined.</li>
<li><b>Pointer difference</b>: two compatible cdata pointers/arrays
can be subtracted. The result is the difference between their
addresses, divided by the element size in bytes. An error is raised if
the element size is undefined or zero.</li>
<li><b>64 bit integer arithmetic</b>: the standard arithmetic
operators (<tt>+ - * / % ^</tt> and unary
minus) can be applied to two cdata numbers, or a cdata number and a
Lua number. If one of them is an <tt>uint64_t</tt>, the other side is
converted to an <tt>uint64_t</tt> and an unsigned arithmetic operation
is performed. Otherwise both sides are converted to an
<tt>int64_t</tt> and a signed arithmetic operation is performed. The
result is a boxed 64 bit cdata object.<br>
If one of the operands is an <tt>enum</tt> and the other operand is a
string, the string is converted to the value of a matching <tt>enum</tt>
constant before the above conversion.<br>
These rules ensure that 64 bit integers are "sticky". Any
expression involving at least one 64 bit integer operand results
in another one. The undefined cases for the division, modulo and power
operators return <tt>2LL ^ 63</tt> or
<tt>2ULL ^ 63</tt>.<br>
You'll have to explicitly convert a 64 bit integer to a Lua
number (e.g. for regular floating-point calculations) with
<tt>tonumber()</tt>. But note this may incur a precision loss.</li>
</ul>
<h3 id="cdata_comp">Comparisons of cdata objects</h3>
<ul>
<li><b>Pointer comparison</b>: two compatible cdata pointers/arrays
can be compared. The result is the same as an unsigned comparison of
their addresses. <tt>nil</tt> is treated like a <tt>NULL</tt> pointer,
which is compatible with any other pointer type.</li>
<li><b>64 bit integer comparison</b>: two cdata numbers, or a
cdata number and a Lua number can be compared with each other. If one
of them is an <tt>uint64_t</tt>, the other side is converted to an
<tt>uint64_t</tt> and an unsigned comparison is performed. Otherwise
both sides are converted to an <tt>int64_t</tt> and a signed
comparison is performed.<br>
If one of the operands is an <tt>enum</tt> and the other operand is a
string, the string is converted to the value of a matching <tt>enum</tt>
constant before the above conversion.<br>
<li><b>Comparisons for equality/inequality</b> never raise an error.
Even incompatible pointers can be compared for equality by address. Any
other incompatible comparison (also with non-cdata objects) treats the
two sides as unequal.</li>
</ul>
<h3 id="cdata_key">cdata objects as table keys</h3>
<p>
Lua tables may be indexed by cdata objects, but this doesn't provide
any useful semantics — <b>cdata objects are unsuitable as table
keys!</b>
</p>
<p>
A cdata object is treated like any other garbage-collected object and
is hashed and compared by its address for table indexing. Since
there's no interning for cdata value types, the same value may be
boxed in different cdata objects with different addresses. Thus
<tt>t[1LL+1LL]</tt> and <tt>t[2LL]</tt> usually <b>do not</b> point to
the same hash slot and they certainly <b>do not</b> point to the same
hash slot as <tt>t[2]</tt>.
</p>
<p>
It would seriously drive up implementation complexity and slow down
the common case, if one were to add extra handling for by-value
hashing and comparisons to Lua tables. Given the ubiquity of their use
inside the VM, this is not acceptable.
</p>
<p>
There are three viable alternatives, if you really need to use cdata
objects as keys:
</p>
<ul>
<li>If you can get by with the precision of Lua numbers
(52 bits), then use <tt>tonumber()</tt> on a cdata number or
combine multiple fields of a cdata aggregate to a Lua number. Then use
the resulting Lua number as a key when indexing tables.<br>
One obvious benefit: <tt>t[tonumber(2LL)]</tt> <b>does</b> point to
the same slot as <tt>t[2]</tt>.</li>
<li>Otherwise use either <tt>tostring()</tt> on 64 bit integers
or complex numbers or combine multiple fields of a cdata aggregate to
a Lua string (e.g. with
<a href="ext_ffi_api.html#ffi_string"><tt>ffi.string()</tt></a>). Then
use the resulting Lua string as a key when indexing tables.</li>
<li>Create your own specialized hash table implementation using the
C types provided by the FFI library, just like you would in
C code. Ultimately this may give much better performance than the
other alternatives or what a generic by-value hash table could
possibly provide.</li>
</ul>
<h2 id="gc">Garbage Collection of cdata Objects</h2>
<p>
All explicitly (<tt>ffi.new()</tt>, <tt>ffi.cast()</tt> etc.) or
implicitly (accessors) created cdata objects are garbage collected.
You need to ensure to retain valid references to cdata objects
somewhere on a Lua stack, an upvalue or in a Lua table while they are
still in use. Once the last reference to a cdata object is gone, the
garbage collector will automatically free the memory used by it (at
the end of the next GC cycle).
</p>
<p>
Please note that pointers themselves are cdata objects, however they
are <b>not</b> followed by the garbage collector. So e.g. if you
assign a cdata array to a pointer, you must keep the cdata object
holding the array alive as long as the pointer is still in use:
</p>
<pre class="code">
ffi.cdef[[
typedef struct { int *a; } foo_t;
]]
local s = ffi.new("foo_t", ffi.new("int[10]")) -- <span style="color:#c00000;">WRONG!</span>
local a = ffi.new("int[10]") -- <span style="color:#00a000;">OK</span>
local s = ffi.new("foo_t", a)
-- Now do something with 's', but keep 'a' alive until you're done.
</pre>
<p>
Similar rules apply for Lua strings which are implicitly converted to
<tt>"const char *"</tt>: the string object itself must be
referenced somewhere or it'll be garbage collected eventually. The
pointer will then point to stale data, which may have already been
overwritten. Note that <em>string literals</em> are automatically kept
alive as long as the function containing it (actually its prototype)
is not garbage collected.
</p>
<p>
Objects which are passed as an argument to an external C function
are kept alive until the call returns. So it's generally safe to
create temporary cdata objects in argument lists. This is a common
idiom for <a href="#convert_vararg">passing specific C types to
vararg functions</a>.
</p>
<p>
Memory areas returned by C functions (e.g. from <tt>malloc()</tt>)
must be manually managed, of course (or use
<a href="ext_ffi_api.html#ffi_gc"><tt>ffi.gc()</tt></a>). Pointers to
cdata objects are indistinguishable from pointers returned by C
functions (which is one of the reasons why the GC cannot follow them).
</p>
<h2 id="callback">Callbacks</h2>
<p>
The LuaJIT FFI automatically generates special callback functions
whenever a Lua function is converted to a C function pointer. This
associates the generated callback function pointer with the C type
of the function pointer and the Lua function object (closure).
</p>
<p>
This can happen implicitly due to the usual conversions, e.g. when
passing a Lua function to a function pointer argument. Or you can use
<tt>ffi.cast()</tt> to explicitly cast a Lua function to a
C function pointer.
</p>
<p>
Currently only certain C function types can be used as callback
functions. Neither C vararg functions nor functions with
pass-by-value aggregate argument or result types are supported. There
are no restrictions for the kind of Lua functions that can be called
from the callback — no checks for the proper number of arguments
are made. The return value of the Lua function will be converted to the
result type and an error will be thrown for invalid conversions.
</p>
<p>
It's allowed to throw errors across a callback invocation, but it's not
advisable in general. Do this only if you know the C function, that
called the callback, copes with the forced stack unwinding and doesn't
leak resources.
</p>
<h3 id="callback_resources">Callback resource handling</h3>
<p>
Callbacks take up resources — you can only have a limited number
of them at the same time (500 - 1000, depending on the
architecture). The associated Lua functions are anchored to prevent
garbage collection, too.
</p>
<p>
<b>Callbacks due to implicit conversions are permanent!</b> There is no
way to guess their lifetime, since the C side might store the
function pointer for later use (typical for GUI toolkits). The associated
resources cannot be reclaimed until termination:
</p>
<pre class="code">
ffi.cdef[[
typedef int (__stdcall *WNDENUMPROC)(void *hwnd, intptr_t l);
int EnumWindows(WNDENUMPROC func, intptr_t l);
]]
-- Implicit conversion to a callback via function pointer argument.
local count = 0
ffi.C.EnumWindows(function(hwnd, l)
count = count + 1
return true
end, 0)
-- The callback is permanent and its resources cannot be reclaimed!
-- Ok, so this may not be a problem, if you do this only once.
</pre>
<p>
Note: this example shows that you <em>must</em> properly declare
<tt>__stdcall</tt> callbacks on Windows/x86 systems. The calling
convention cannot be automatically detected, unlike for
<tt>__stdcall</tt> calls <em>to</em> Windows functions.
</p>
<p>
For some use cases it's necessary to free up the resources or to
dynamically redirect callbacks. Use an explicit cast to a
C function pointer and keep the resulting cdata object. Then use
the <a href="ext_ffi_api.html#callback_free"><tt>cb:free()</tt></a>
or <a href="ext_ffi_api.html#callback_set"><tt>cb:set()</tt></a> methods
on the cdata object:
</p>
<pre class="code">
-- Explicitly convert to a callback via cast.
local count = 0
local cb = ffi.cast("WNDENUMPROC", function(hwnd, l)
count = count + 1
return true
end)
-- Pass it to a C function.
ffi.C.EnumWindows(cb, 0)
-- EnumWindows doesn't need the callback after it returns, so free it.
cb:free()
-- The callback function pointer is no longer valid and its resources
-- will be reclaimed. The created Lua closure will be garbage collected.
</pre>
<h3 id="callback_performance">Callback performance</h3>
<p>
<b>Callbacks are slow!</b> First, the C to Lua transition itself
has an unavoidable cost, similar to a <tt>lua_call()</tt> or
<tt>lua_pcall()</tt>. Argument and result marshalling add to that cost.
And finally, neither the C compiler nor LuaJIT can inline or
optimize across the language barrier and hoist repeated computations out
of a callback function.
</p>
<p>
Do not use callbacks for performance-sensitive work: e.g. consider a
numerical integration routine which takes a user-defined function to
integrate over. It's a bad idea to call a user-defined Lua function from
C code millions of times. The callback overhead will be absolutely
detrimental for performance.
</p>
<p>
It's considerably faster to write the numerical integration routine
itself in Lua — the JIT compiler will be able to inline the
user-defined function and optimize it together with its calling context,
with very competitive performance.
</p>
<p>
As a general guideline: <b>use callbacks only when you must</b>, because
of existing C APIs. E.g. callback performance is irrelevant for a
GUI application, which waits for user input most of the time, anyway.
</p>
<p>
For new designs <b>avoid push-style APIs</b> (C function repeatedly
calling a callback for each result). Instead <b>use pull-style APIs</b>
(call a C function repeatedly to get a new result). Calls from Lua
to C via the FFI are much faster than the other way round. Most well-designed
libraries already use pull-style APIs (read/write, get/put).
</p>
<h2 id="clib">C Library Namespaces</h2>
<p>
A C library namespace is a special kind of object which allows
access to the symbols contained in shared libraries or the default
symbol namespace. The default
<a href="ext_ffi_api.html#ffi_C"><tt>ffi.C</tt></a> namespace is
automatically created when the FFI library is loaded. C library
namespaces for specific shared libraries may be created with the
<a href="ext_ffi_api.html#ffi_load"><tt>ffi.load()</tt></a> API
function.
</p>
<p>
Indexing a C library namespace object with a symbol name (a Lua
string) automatically binds it to the library. First the symbol type
is resolved — it must have been declared with
<a href="ext_ffi_api.html#ffi_cdef"><tt>ffi.cdef</tt></a>. Then the
symbol address is resolved by searching for the symbol name in the
associated shared libraries or the default symbol namespace. Finally,
the resulting binding between the symbol name, the symbol type and its
address is cached. Missing symbol declarations or nonexistent symbol
names cause an error.
</p>
<p>
This is what happens on a <b>read access</b> for the different kinds of
symbols:
</p>
<ul>
<li>External functions: a cdata object with the type of the function
and its address is returned.</li>
<li>External variables: the symbol address is dereferenced and the
loaded value is <a href="#convert_tolua">converted to a Lua object</a>
and returned.</li>
<li>Constant values (<tt>static const</tt> or <tt>enum</tt>
constants): the constant is <a href="#convert_tolua">converted to a
Lua object</a> and returned.</li>
</ul>
<p>
This is what happens on a <b>write access</b>:
</p>
<ul>
<li>External variables: the value to be written is
<a href="#convert_fromlua">converted to the C type</a> of the
variable and then stored at the symbol address.</li>
<li>Writing to constant variables or to any other symbol type causes
an error, like any other attempted write to a constant location.</li>
</ul>
<p>
C library namespaces themselves are garbage collected objects. If
the last reference to the namespace object is gone, the garbage
collector will eventually release the shared library reference and
remove all memory associated with the namespace. Since this may
trigger the removal of the shared library from the memory of the
running process, it's generally <em>not safe</em> to use function
cdata objects obtained from a library if the namespace object may be
unreferenced.
</p>
<p>
Performance notice: the JIT compiler specializes to the identity of
namespace objects and to the strings used to index it. This
effectively turns function cdata objects into constants. It's not
useful and actually counter-productive to explicitly cache these
function objects, e.g. <tt>local strlen = ffi.C.strlen</tt>. OTOH it
<em>is</em> useful to cache the namespace itself, e.g. <tt>local C =
ffi.C</tt>.
</p>
<h2 id="policy">No Hand-holding!</h2>
<p>
The FFI library has been designed as <b>a low-level library</b>. The
goal is to interface with C code and C data types with a
minimum of overhead. This means <b>you can do anything you can do
from C</b>: access all memory, overwrite anything in memory, call
machine code at any memory address and so on.
</p>
<p>
The FFI library provides <b>no memory safety</b>, unlike regular Lua
code. It will happily allow you to dereference a <tt>NULL</tt>
pointer, to access arrays out of bounds or to misdeclare
C functions. If you make a mistake, your application might crash,
just like equivalent C code would.
</p>
<p>
This behavior is inevitable, since the goal is to provide full
interoperability with C code. Adding extra safety measures, like
bounds checks, would be futile. There's no way to detect
misdeclarations of C functions, since shared libraries only
provide symbol names, but no type information. Likewise there's no way
to infer the valid range of indexes for a returned pointer.
</p>
<p>
Again: the FFI library is a low-level library. This implies it needs
to be used with care, but it's flexibility and performance often
outweigh this concern. If you're a C or C++ developer, it'll be easy
to apply your existing knowledge. OTOH writing code for the FFI
library is not for the faint of heart and probably shouldn't be the
first exercise for someone with little experience in Lua, C or C++.
</p>
<p>
As a corollary of the above, the FFI library is <b>not safe for use by
untrusted Lua code</b>. If you're sandboxing untrusted Lua code, you
definitely don't want to give this code access to the FFI library or
to <em>any</em> cdata object (except 64 bit integers or complex
numbers). Any properly engineered Lua sandbox needs to provide safety
wrappers for many of the standard Lua library functions —
similar wrappers need to be written for high-level operations on FFI
data types, too.
</p>
<h2 id="status">Current Status</h2>
<p>
The initial release of the FFI library has some limitations and is
missing some features. Most of these will be fixed in future releases.
</p>
<p>
<a href="#clang">C language support</a> is
currently incomplete:
</p>
<ul>
<li>C declarations are not passed through a C pre-processor,
yet.</li>
<li>The C parser is able to evaluate most constant expressions
commonly found in C header files. However it doesn't handle the
full range of C expression semantics and may fail for some
obscure constructs.</li>
<li><tt>static const</tt> declarations only work for integer types
up to 32 bits. Neither declaring string constants nor
floating-point constants is supported.</li>
<li>Packed <tt>struct</tt> bitfields that cross container boundaries
are not implemented.</li>
<li>Native vector types may be defined with the GCC <tt>mode</tt> or
<tt>vector_size</tt> attribute. But no operations other than loading,
storing and initializing them are supported, yet.</li>
<li>The <tt>volatile</tt> type qualifier is currently ignored by
compiled code.</li>
<li><a href="ext_ffi_api.html#ffi_cdef"><tt>ffi.cdef</tt></a> silently
ignores all re-declarations.</li>
</ul>
<p>
The JIT compiler already handles a large subset of all FFI operations.
It automatically falls back to the interpreter for unimplemented
operations (you can check for this with the
<a href="running.html#opt_j"><tt>-jv</tt></a> command line option).
The following operations are currently not compiled and may exhibit
suboptimal performance, especially when used in inner loops:
</p>
<ul>
<li>Array/<tt>struct</tt> copies and bulk initializations.</li>
<li>Bitfield accesses and initializations.</li>
<li>Vector operations.</li>
<li>Table initializers.</li>
<li>Initialization of nested <tt>struct</tt>/<tt>union</tt> types.</li>
<li>Allocations of variable-length arrays or structs.</li>
<li>Allocations of C types with a size > 128 bytes or an
alignment > 8 bytes.</li>
<li>Conversions from lightuserdata to <tt>void *</tt>.</li>
<li>Pointer differences for element sizes that are not a power of
two.</li>
<li>Calls to C functions with aggregates passed or returned by
value.</li>
<li>Calls to ctype metamethods which are not plain functions.</li>
<li>ctype <tt>__newindex</tt> tables and non-string lookups in ctype
<tt>__index</tt> tables.</li>
<li><tt>tostring()</tt> for cdata types.</li>
<li>Calls to the following <a href="ext_ffi_api.html">ffi.* API</a>
functions: <tt>cdef</tt>, <tt>load</tt>, <tt>typeof</tt>,
<tt>metatype</tt>, <tt>gc</tt>, <tt>sizeof</tt>, <tt>alignof</tt>,
<tt>offsetof</tt>.</li>
</ul>
<p>
Other missing features:
</p>
<ul>
<li>Bit operations for 64 bit types.</li>
<li>Arithmetic for <tt>complex</tt> numbers.</li>
<li>Passing structs by value to vararg C functions.</li>
<li><a href="extensions.html#exceptions">C++ exception interoperability</a>
does not extend to C functions called via the FFI, if the call is
compiled.</li>
</ul>
<br class="flush">
</div>
<div id="foot">
<hr class="hide">
Copyright © 2005-2012 Mike Pall
<span class="noprint">
·
<a href="contact.html">Contact</a>
</span>
</div>
</body>
</html>