duktape/website/guide/compiling.html

<h1 id="compiling">Compiling</h1>

<h2>Automatic defaults</h2>

<p>If you compile Duktape with no compiler options, Duktape will detect the
compiler and the platform automatically and select defaults appropriate in
most cases.</p>

<p>The default features are, at a high level:</p>
<ul>
<li>Full Ecmascript compliance
    (including the optional 
    <a href="http://www.ecma-international.org/ecma-262/5.1/#sec-B">Annex B</a>
    features)</li>
<li>Packed value representation (8 bytes per value) when available,
    unpacked value representation (12-16 bytes per value) when not</li>
<li>Reference counting and mark-and-sweep garbage collection</li>
<li>Full error messages and tracebacks</li>
<li>No debug printing, no asserts, etc</li>
</ul>

<h2>Feature options (DUK_OPT_xxx)</h2>

<p>If you wish to modify the defaults, you can provide feature options in the
form of <code>DUK_OPT_xxx</code> compiler defines.  These will be taken into
account by the internal <code>duk_features.h</code> file, which resolves the
final internal features based on feature requests, compiler features, and
platform features.</p>

<p>The available feature options can be found in <code>duk_features.h</code>.
The table below summarizes the available options, in no particular order:</p>

<table>
<thead>
<tr>
<th>Define</th>
<th>Description</th>
</tr>
</thead>
<tbody>
<tr>
<td class="definename">DUK_OPT_NO_PACKED_TVAL</td>
<td>Don't use the packed 8-byte internal value representation even if otherwise
    possible.  The packed representation has more platform/compiler portability
    issues than the unpacked one.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_NO_REFERENCE_COUNTING</td>
<td>Disable reference counting and use only mark-and-sweep for garbage collection.
    Although this reduces memory footprint of heap objects, the downside is much
    more fluctuation in memory usage.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_NO_MARK_AND_SWEEP</td>
<td>Disable mark-and-sweep and use only reference counting for garbage collection.
    This reduces code footprint and eliminates garbage collection pauses, but
    objects participating in unreachable reference cycles won't be collected until
    the Duktape heap is destroyed.  In particular, function instances won't be
    collected because they're always in a reference cycle with their default
    prototype object.  Unreachable objects are collected if you break reference
    cycles manually (and are always freed when a heap is destroyed).</td>
</tr>
<tr>
<td class="definename">DUK_OPT_NO_MS_STRINGTABLE_RESIZE</td>
<td>Disable forced string intern table resize during mark-and-sweep garbage
    collection.  This may be useful when reference counting is disabled, as
    mark-and-sweep collections will be more frequent and thus more expensive.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_GC_TORTURE</td>
<td>Development time option: force full mark-and-sweep on every allocation to
    stress test memory management.</td>
</tr>
<td class="definename">DUK_OPT_NO_AUGMENT_ERRORS</td>
<td>Don't augment Ecmascript error objects with custom fields like
    <code>fileName</code>, <code>lineNumber</code>, and traceback data.
    Implies <code>DUK_OPT_NO_TRACEBACKS</code>.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_NO_TRACEBACKS</td>
<td>Don't record traceback data into Ecmascript error objects (but still record
    <code>fileName</code> and <code>lineNumber</code>).  Reduces footprint and
    makes error handling a bit faster, at the cost of less nformative Ecmascript
    errors.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_NO_VERBOSE_ERRORS</td>
<td>Don't provide error message strings or filename/line information for
    errors generated by Duktape.  Reduces footprint, at the cost of much
    less informative Ecmascript errors.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_TRACEBACK_DEPTH</td>
<td>Override default traceback collection depth.  The default is currently 10.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_NO_PC2LINE</td>
<td>Don't record a "pc2line" map into function instances.  Without this map,
    exceptions won't have meaningful line numbers (virtual machine program
    counter values cannot be translated to line numbers) but function instances
    will have a smaller footprint.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_NO_REGEXP_SUPPORT</td>
<td>Disable support for regular expressions.  Regexp literals are treated as a
    <code>SyntaxError</code>, RegExp constructor and prototype functions throw
    an error, <code>String.prototype.replace()</code> throws an error if given
    a regexp search value, <code>String.prototype.split()</code> throws an error
    if given a regexp separator value, <code>String.prototype.search()</code>
    and <code>String.prototype.match()</code> throw an error unconditionally.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_STRICT_UTF8_SOURCE</td>
<td>Enable strict UTF-8 parsing of source code.  When enabled, non-shortest
    encodings (normally invalid UTF-8) and surrogate pair codepoints are accepted
    as valid source code characters.  This option breaks compatibility with
    some test262 tests.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_NO_OCTAL_SUPPORT</td>
<td>Disable optional octal number support (Ecmascript E5/E5.1
    <a href="http://www.ecma-international.org/ecma-262/5.1/#sec-B">Annex B</a>).</td>
</tr>
<tr>
<td class="definename">DUK_OPT_NO_SOURCE_NONBMP</td>
<td>Disable accurate Unicode support for non-BMP characters in source code.
    Non-BMP characters are then always accepted as identifier characters.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_NO_BROWSER_LIKE</td>
<td>Disable browser-like functions.  Makes <code>print()</code> and
    <code>alert()</code> throw an error.  This option is confusing when
    used with the Duktape command line tool, as the command like tool
    will immediately panic.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_NO_SECTION_B</td>
<td>Disable optional features in Ecmascript specification
    <a href="http://www.ecma-international.org/ecma-262/5.1/#sec-B">Annex B</a>.
    Causes <code>escape()</code>, <code>unescape()</code>, and
    <code>String.prototype.substr()</code> to throw an error.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_NO_JSONX</td>
<td>Disable support for the JSONX format.  Reduces code footprint.  Causes
    JSONX calls to throw an error.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_NO_JSONC</td>
<td>Disable support for the JSONC format.  Reduces code footprint.  Causes
    JSONC calls to throw an error.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_NO_FILE_IO</td>
<td>Disable use of ANSI C file I/O which might be a portability issue on some
    platforms.  Causes <code>duk_eval_file()</code> to throw an error,
    makes built-in <code>print()</code> and <code>alert()</code> no-ops,
    and suppresses writing of a panic message to <code>stderr</code> on panic.
    This option does not suppress debug printing so don't enable debug printing
    if you wish to avoid I/O.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_NO_INTERRUPT_COUNTER</td>
<td>Disable the internal bytecode executor periodic interrupt counter.
    The mechanism is used to implement e.g. execution step limit, custom
    profiling, and debugger interaction.  Disabling the interrupt counter
    improves bytecode execution performance very slightly but disables all
    features depending on it.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_SEGFAULT_ON_PANIC</td>
<td>Cause the default panic handler to cause a segfault instead of using
    <code>abort()</code> or <code>exit()</code>.  This is useful when debugging
    with valgrind, as a segfault provides a nice C traceback in valgrind.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_SELF_TESTS</td>
<td>Perform run-time self tests when a Duktape heap is created.  Catches
    platform/compiler problems which cannot be reliably detected during
    compile time.  Not enabled by default because of the extra footprint.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_ASSERTIONS</td>
<td>Enable internal assert checks.  These slow down execution considerably
    so only use when debugging.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_DEBUG</td>
<td>Enable debug printouts.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_DDEBUG</td>
<td>Enable more debug printouts.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_DDDEBUG</td>
<td>Enable even more debug printouts.  Not recommended unless you have
    grep handy.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_DPRINT_COLORS</td>
<td>Enable coloring of debug prints with 
    <a href="http://en.wikipedia.org/wiki/ANSI_escape_code">ANSI escape codes</a>.
    The behavior is not sensitive to terminal settings.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_DPRINT_RDTSC</td>
<td>Print RDTSC cycle count in debug prints if available.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_DEBUG_BUFSIZE</td>
<td>Debug code uses a static buffer as a formatting temporary to avoid side
    effects in debug prints.  The static buffer is large by default, which may
    be an issue in constrained environments.  You can set the buffer size
    manually with this option.  Example:
    <code>-DDUK_OPT_DEBUG_BUFSIZE=2048</code>.</td>
</tr>
<tr>
<td class="definename">DUK_OPT_HAVE_CUSTOM_H</td>
<td>Enable user-provided <code>duk_custom.h</code> customization header
    (see below for details).  Not recommended unless really necessary.</td>
</tr>
</tbody>
</table>

<h2>DUK_OPT_HAVE_CUSTOM_H and duk_custom.h</h2>

<p>Normally you define <code>DUK_OPT_xxx</code> feature options and the
internal <code>duk_features.h</code> header resolves these with platform/compiler
constraints to determine effective compilation options for Duktape internals.
The effective options are provided as <code>DUK_USE_xxx</code> defines which
you normally never see.</p>

<p>If you define <code>DUK_OPT_HAVE_CUSTOM_H</code>, Duktape will include
<code>duk_custom.h</code> after determining the appropriate <code>DUK_USE_xxx</code>
defines but before compiling any code.  The <code>duk_custom.h</code> header,
which you provide, can then tweak the active <code>DUK_USE_xxx</code> defines
freely.  See <code>duk_features.h</code> for the available defines.</p>

<p>This approach is useful when the <code>DUK_OPT_xxx</code> feature options
don't provide enough flexibility to tweak the build.  The downside is that
you can easily create inconsistent <code>DUK_USE_xxx</code> flags, the
customization header will be version specific, and you need to peek into
Duktape internals to know what defines to tweak.</p>

<h2>DUK_PANIC_HANDLER</h2>

<p>The default panic handler will print an error message to stdout
unless I/O is disabled by <code>DUK_OPT_NO_FILE_IO</code>.  It will then call
<code>abort()</code> or cause a segfault if
<code>DUK_OPT_SEGFAULT_ON_PANIC</code> is defined.</p>

<p>You can override the entire panic handler by defining
<code>DUK_PANIC_HANDLER</code>.  For example, you could add the
following to your compiler options:</p>
<pre>
'-DDUK_PANIC_HANDLER(code,msg)={printf("*** %d:%s\n",(code),(msg));abort();}'
</pre>

<p>Or perhaps:</p>
<pre>
'-DDUK_PANIC_HANDLER(code,msg)={my_panic_handler((code),(msg))}'
</pre>

<p>which calls your custom handler:</p>
<pre class="c-code">
void my_panic_handler(int code, const char *msg) {
    /* Your panic handling.  Must not return. */
}
</pre>

<h2>Adding new feature options</h2>

<p>This section only applies if you customize Duktape internals and wish
to submit a patch to be included in the mainline distribution:</p>

<ul>
<li>Add a descriptive <code>DUK_OPT_xxx</code> for the custom feature.
    The custom feature should only be enabled if the feature option is
    explicitly given.</li>
<li>Modify <code>duk_features.h</code> to detect your custom feature
    option and define appropriate internal <code>DUK_USE_xxx</code>
    define(s).  Conflicts with other features should be detected.
    Code outside <code>duk_features.h</code> should only listen
    to <code>DUK_USE_xxx</code> defines so that the resolution process
    is fully contained in <code>duk_features.h</code>.</li>
</ul>

<h2>Memory management alternatives</h2>

<p>There are three supported memory management alternatives:</p>
<ul>
<li><b>Reference counting and mark-and-sweep (default)</b>: heap objects are
    freed immediately when they become unreachable except for objects
    participating in unreachable reference cycles.  Such objects are freed by
    a periodic voluntary, stop the world mark-and-sweep collection.
    Mark-and-sweep is also used as the emergency garbage collector if
    memory allocation fails.</li>
<li><b>Reference counting only</b>: reduces code footprint and eliminates
    garbage collection pauses, but objects in unreachable reference cycles
    are not collected until the Duktape heap is destroyed.  See note below
    on function instances and reference cycles.</li>
<li><b>Mark-and-sweep only</b>: reduces code footprint and memory footprint
    (heap headers don't need to store a reference count), but there is more
    memory usage variance than in the default case.  The frequency of voluntary,
    stop the world mark-and-sweep collections is also higher than in the default
    case where reference counting is expected to handle almost all memory
    management.</li>
</ul>

<p>When using only reference counting it is important to avoid creating
unreachable reference cycles.  Reference cycles are usually easy to avoid in
application code e.g. by using only forward pointers in data structures.  Even
if reference cycles are necessary, garbage collection can be allowed to work
simply by breaking the cycles before deleting the final references to such objects.
For example, if you have a tree structure where nodes maintain references to
both children and parents (creating reference cycles for each node) you could
walk the tree and set the parent reference to <code>null</code> before deleting
the final reference to the tree.</p>

<p>Unfortunately every Ecmascript function instance is, by default, in a
reference loop with an automatic prototype object created for the object.
The function instance's <code>prototype</code> property points to the prototype
object, and the prototype's <code>constructor</code> property points back to the
function instance.  Only mark-and-sweep is able to collect these reference
loops at the moment.  If you build with reference counting only, function
instances may appear to leak memory; the memory will be released when the
relevant heap is destroyed.  You can also break the reference loops manually
(although this is a bit cumbersome):</p>
<pre class="ecmascript-code">
var f = function() { };
var g = function() { };
var h = function() { };
Duktape.fin(f, function() { print('finalizer for f'); });
Duktape.fin(g, function() { print('finalizer for g'); });
Duktape.fin(h, function() { print('finalizer for h'); });

// not collected until heap destruction in a reference counting only build
f = null;            // not collected immediately

// break cycle by deleting 'prototype' reference (alternative 1)
g.prototype = null;
g = null;            // collected immediately, finalizer runs

// break cycle by deleting 'constructor' reference (alternative 2)
h.prototype.constructor = null;
h = null;            // collected immediately, finalizer runs

// no-op with refcount only, with mark-and-sweep finalizer for 'f' runs
Duktape.gc();
</pre>

<h2>Compiler warnings</h2>

<p>Current goal is for the Duktape compile to be clean when:</p>
<ul>
<li>using a major compiler (e.g. gcc, clang, MSVC, mingw);</li>
<li>the compiler is in C99 mode; and</li>
<li>warnings are enabled (e.g. <code>-Wall</code> in gcc/clang).</li>
</ul>

<p>There are still some warnings present when you compile with
<code>-Wextra</code> or equivalent option.</p>

<p>When your compiler is not C99 compliant, Duktape uses an awkward
replacement for variadic macros.  This may cause, as a side effect, a
lot of harmless warnings if you set the compiler warning level too high.
This is difficult to fix, so C99 compilation may not be clean at the
moment.</p>