| AllocationRecording.h |
This struct holds the information needed to create a profiler marker payload
that can represent a JS allocation. It translates JS engine specific classes,
into something that can be used in the profiler.
|
2654 |
| AllocPolicy.h |
JS allocation policies.
The allocators here are for system memory with lifetimes which are not
managed by the GC. See the comment at the top of vm/MallocProvider.h.
|
5973 |
| ArrayBuffer.h |
ArrayBuffer functionality. |
9648 |
| BuildId.h |
Embedding-provided build ID information, used by SpiderMonkey to tag cached
compilation data so that cached data can be reused when possible, or
discarded and regenerated if necessary.
|
2137 |
| CallArgs.h |
[SMDOC] JS::CallArgs API
Helper classes encapsulating access to the callee, |this| value, arguments,
and argument count for a call/construct operation.
JS::CallArgs encapsulates access to a JSNative's un-abstracted
|unsigned argc, Value* vp| arguments. The principal way to create a
JS::CallArgs is using JS::CallArgsFromVp:
// If provided no arguments or a non-numeric first argument, return zero.
// Otherwise return |this| exactly as given, without boxing.
static bool
Func(JSContext* cx, unsigned argc, JS::Value* vp)
{
JS::CallArgs args = JS::CallArgsFromVp(argc, vp);
// Guard against no arguments or a non-numeric arg0.
if (args.length() == 0 || !args[0].isNumber()) {
args.rval().setInt32(0);
return true;
}
// Access to the callee must occur before accessing/setting
// the return value.
JSObject& callee = args.callee();
args.rval().setObject(callee);
// callee() and calleev() will now assert.
// It's always fine to access thisv().
HandleValue thisv = args.thisv();
args.rval().set(thisv);
// As the return value was last set to |this|, returns |this|.
return true;
}
CallArgs is exposed publicly and used internally. Not all parts of its
public interface are meant to be used by embedders! See inline comments to
for details.
It's possible (albeit deprecated) to manually index into |vp| to access the
callee, |this|, and arguments of a function, and to set its return value.
This does not have the error-handling or moving-GC correctness of CallArgs.
New code should use CallArgs instead whenever possible.
The eventual plan is to change JSNative to take |const CallArgs&| directly,
for automatic assertion of correct use and to make calling functions more
efficient. Embedders should start internally switching away from using
|argc| and |vp| directly, except to create a |CallArgs|. Then, when an
eventual release making that change occurs, porting efforts will require
changing methods' signatures but won't require invasive changes to the
methods' implementations, potentially under time pressure.
|
11651 |
| CallNonGenericMethod.h |
js_CallNonGenericMethod_h |
4839 |
| CharacterEncoding.h |
By default, all C/C++ 1-byte-per-character strings passed into the JSAPI
are treated as ISO/IEC 8859-1, also known as Latin-1. That is, each
byte is treated as a 2-byte character, and there is no way to pass in a
string containing characters beyond U+00FF.
|
13993 |
| Class.h |
JSClass definition and its component types, plus related interfaces. |
34108 |
| CompilationAndEvaluation.h |
Functions for compiling and evaluating scripts. |
15543 |
| CompileOptions.h |
Options for JavaScript compilation.
In the most common use case, a CompileOptions instance is allocated on the
stack, and holds non-owning references to non-POD option values: strings,
principals, objects, and so on. The code declaring the instance guarantees
that such option values will outlive the CompileOptions itself: objects are
otherwise rooted, principals have had their reference counts bumped, and
strings won't be freed until the CompileOptions goes out of scope. In this
situation, CompileOptions only refers to things others own, so it can be
lightweight.
In some cases, however, we need to hold compilation options with a
non-stack-like lifetime. For example, JS::CompileOffThread needs to save
compilation options where a worker thread can find them, then return
immediately. The worker thread will come along at some later point, and use
the options.
The compiler itself just needs to be able to access a collection of options;
it doesn't care who owns them, or what's keeping them alive. It does its
own addrefs/copies/tracing/etc.
Furthermore, in some cases compile options are propagated from one entity to
another (e.g. from a script to a function defined in that script). This
involves copying over some, but not all, of the options.
So we have a class hierarchy that reflects these four use cases:
- TransitiveCompileOptions is the common base class, representing options
that should get propagated from a script to functions defined in that
script. This class is abstract and is only ever used as a subclass.
- ReadOnlyCompileOptions is the only subclass of TransitiveCompileOptions,
representing a full set of compile options. It can be used by code that
simply needs to access options set elsewhere, like the compiler. This
class too is abstract and is only ever used as a subclass.
- The usual CompileOptions class must be stack-allocated, and holds
non-owning references to the filename, element, and so on. It's derived
from ReadOnlyCompileOptions, so the compiler can use it.
- OwningCompileOptions roots / copies / reference counts of all its values,
and unroots / frees / releases them when it is destructed. It too is
derived from ReadOnlyCompileOptions, so the compiler accepts it.
|
15653 |
| ContextOptions.h |
JavaScript API. |
5069 |
| Conversions.h |
ECMAScript conversion operations. |
20709 |
| Date.h |
JavaScript date/time computation and creation functions. |
8784 |
| Debug.h |
|
14069 |
| Equality.h |
Equality operations. |
2341 |
| ErrorReport.h |
Error-reporting types and structures.
Despite these types and structures existing in js/public, significant parts
of their heritage date back to distant SpiderMonkey past, and they are not
all universally well-thought-out as ideal, intended-to-be-permanent API.
We may eventually replace this with something more consistent with
ECMAScript the language and less consistent with '90s-era JSAPI inventions,
but it's doubtful this will happen any time soon.
|
8189 |
| experimental |
|
1 |
| ForOfIterator.h |
A convenience class that makes it easy to perform the operations of a for-of
loop.
|
3342 |
| GCAnnotations.h |
js_GCAnnotations_h |
3647 |
| GCAPI.h |
High-level interface to the JS garbage collector.
|
36386 |
| GCHashTable.h |
|
23758 |
| GCPolicyAPI.h |
|
7182 |
| GCVariant.h |
|
5575 |
| GCVector.h |
|
10788 |
| HashTable.h |
js_HashTable_h |
1135 |
| HeapAPI.h |
These values are private to the JS engine. |
21265 |
| Id.h |
Only JSStrings that have been interned via the JSAPI can be turned into
jsids by API clients.
N.B. if a jsid is backed by a string which has not been interned, that
string must be appropriately rooted to avoid being collected by the GC.
|
7688 |
| Initialization.h |
SpiderMonkey initialization and shutdown APIs. |
4557 |
| JSON.h |
JSON serialization and deserialization operations.
|
3621 |
| LocaleSensitive.h |
Functions and structures related to locale-sensitive behavior, including
exposure of the default locale (used by operations like toLocaleString).
|
3810 |
| MemoryFunctions.h |
Low-level memory-allocation functions. |
3221 |
| MemoryMetrics.h |
These are the measurements used by Servo. |
29357 |
| Modules.h |
JavaScript module (as in, the syntactic construct) operations. |
6324 |
| OffThreadScriptCompilation.h |
Types and functions related to the compilation of JavaScript off the
direct JSAPI-using thread.
|
5918 |
| Principals.h |
JSPrincipals and related interfaces. |
5012 |
| Printf.h |
Wrappers for mozilla::Smprintf and friends that are used throughout
JS. |
1212 |
| ProfilingCategory.h |
|
6391 |
| ProfilingFrameIterator.h |
After each sample run, this method should be called with the current buffer
position at which the buffer contents start. This will update the
corresponding field on the JSRuntime.
See the field |profilerSampleBufferRangeStart| on JSRuntime for documentation
about what this value is used for.
|
7496 |
| ProfilingStack.h |
|
21839 |
| Promise.h |
Abstract base class for an ECMAScript Job Queue:
https://www.ecma-international.org/ecma-262/9.0/index.html#sec-jobs-and-job-queues
SpiderMonkey doesn't schedule Promise resolution jobs itself; instead, the
embedding can provide an instance of this class SpiderMonkey can use to do
that scheduling.
The JavaScript shell includes a simple implementation adequate for running
tests. Browsers need to augment job handling to meet their own additional
requirements, so they can provide their own implementation.
|
24303 |
| PropertyDescriptor.h |
Property descriptors and flags. |
13333 |
| PropertySpec.h |
Property descriptors and flags. |
14657 |
| ProtoKey.h |
A higher-order macro for enumerating all JSProtoKey values. |
8463 |
| Proxy.h |
|
29289 |
| Realm.h |
Ways to get various per-Realm objects. All the getters declared below operate
on the JSContext's current Realm.
|
4660 |
| RealmOptions.h |
Options specified when creating a realm to determine its behavior, immutable
options determining the behavior of an existing realm, and mutable options on
an existing realm that may be changed when desired.
|
9587 |
| RefCounted.h |
js_RefCounted_h |
2444 |
| RegExp.h |
Regular expression-related operations. |
3158 |
| RegExpFlags.h |
Regular expression flags. |
3764 |
| RequiredDefines.h |
Various #defines required to build SpiderMonkey. Embedders should add this
file to the start of the command line via -include or a similar mechanism,
or SpiderMonkey public headers may not work correctly.
|
1210 |
| Result.h |
[SMDOC] JS::Result
`Result` is used as the return type of many SpiderMonkey functions that
can either succeed or fail. See "/mfbt/Result.h".
## Which return type to use
`Result` is for return values. Obviously, if you're writing a function that
can't fail, don't use Result. Otherwise:
JS::Result<> - function can fail, doesn't return anything on success
(defaults to `JS::Result<JS::Ok, JS::Error&>`)
JS::Result<JS::OOM&> - like JS::Result<>, but fails only on OOM
JS::Result<Data> - function can fail, returns Data on success
JS::Result<Data, JS::OOM&> - returns Data, fails only on OOM
mozilla::GenericErrorResult<JS::Error&> - always fails
That last type is like a Result with no success type. It's used for
functions like `js::ReportNotFunction` that always return an error
result. `GenericErrorResult<E>` implicitly converts to `Result<V, E>`,
regardless of V.
## Checking Results when your return type is Result
When you call a function that returns a `Result`, use the `MOZ_TRY` macro to
check for errors:
MOZ_TRY(DefenestrateObject(cx, obj));
If `DefenestrateObject` returns a success result, `MOZ_TRY` is done, and
control flows to the next statement. If `DefenestrateObject` returns an
error result, `MOZ_TRY` will immediately return it, propagating the error to
your caller. It's kind of like exceptions, but more explicit -- you can see
in the code exactly where errors can happen.
You can do a tail call instead of using `MOZ_TRY`:
return DefenestrateObject(cx, obj);
Indicate success with `return Ok();`.
If the function returns a value on success, use `MOZ_TRY_VAR` to get it:
RootedValue thrug(cx);
MOZ_TRY_VAR(thrug, GetObjectThrug(cx, obj));
This behaves the same as `MOZ_TRY` on error. On success, the success
value of `GetObjectThrug(cx, obj)` is assigned to the variable `thrug`.
## Checking Results when your return type is not Result
This header defines alternatives to MOZ_TRY and MOZ_TRY_VAR for when you
need to call a `Result` function from a function that uses false or nullptr
to indicate errors:
JS_TRY_OR_RETURN_FALSE(cx, DefenestrateObject(cx, obj));
JS_TRY_VAR_OR_RETURN_FALSE(cx, v, GetObjectThrug(cx, obj));
JS_TRY_OR_RETURN_NULL(cx, DefenestrateObject(cx, obj));
JS_TRY_VAR_OR_RETURN_NULL(cx, v, GetObjectThrug(cx, obj));
When TRY is not what you want, because you need to do some cleanup or
recovery on error, use this idiom:
if (!cx->resultToBool(expr_that_is_a_Result)) {
... your recovery code here ...
}
In place of a tail call, you can use one of these methods:
return cx->resultToBool(expr); // false on error
return cx->resultToPtr(expr); // null on error
Once we are using `Result` everywhere, including in public APIs, all of
these will go away.
## GC safety
When a function returns a `JS::Result<JSObject*>`, it is the program's
responsibility to check for errors and root the object before continuing:
RootedObject wrapper(cx);
MOZ_TRY_VAR(wrapper, Enwrapify(cx, thing));
This is ideal. On error, there is no object to root; on success, the
assignment to wrapper roots it. GC safety is ensured.
`Result` has methods .isOk(), .isErr(), .unwrap(), and .unwrapErr(), but if
you're actually using them, it's possible to create a GC hazard. The static
analysis will catch it if so, but that's hardly convenient. So try to stick
to the idioms shown above.
## Future directions
At present, JS::Error and JS::OOM are empty structs. The plan is to make them
GC things that contain the actual error information (including the exception
value and a saved stack).
The long-term plan is to remove JS_IsExceptionPending and
JS_GetPendingException in favor of JS::Error. Exception state will no longer
exist.
|
8038 |
| RootingAPI.h |
[SMDOC] Stack Rooting
Moving GC Stack Rooting
A moving GC may change the physical location of GC allocated things, even
when they are rooted, updating all pointers to the thing to refer to its new
location. The GC must therefore know about all live pointers to a thing,
not just one of them, in order to behave correctly.
The |Rooted| and |Handle| classes below are used to root stack locations
whose value may be held live across a call that can trigger GC. For a
code fragment such as:
JSObject* obj = NewObject(cx);
DoSomething(cx);
... = obj->lastProperty();
If |DoSomething()| can trigger a GC, the stack location of |obj| must be
rooted to ensure that the GC does not move the JSObject referred to by
|obj| without updating |obj|'s location itself. This rooting must happen
regardless of whether there are other roots which ensure that the object
itself will not be collected.
If |DoSomething()| cannot trigger a GC, and the same holds for all other
calls made between |obj|'s definitions and its last uses, then no rooting
is required.
SpiderMonkey can trigger a GC at almost any time and in ways that are not
always clear. For example, the following innocuous-looking actions can
cause a GC: allocation of any new GC thing; JSObject::hasProperty;
JS_ReportError and friends; and ToNumber, among many others. The following
dangerous-looking actions cannot trigger a GC: js_malloc, cx->malloc_,
rt->malloc_, and friends and JS_ReportOutOfMemory.
The following family of three classes will exactly root a stack location.
Incorrect usage of these classes will result in a compile error in almost
all cases. Therefore, it is very hard to be incorrectly rooted if you use
these classes exclusively. These classes are all templated on the type T of
the value being rooted.
- Rooted<T> declares a variable of type T, whose value is always rooted.
Rooted<T> may be automatically coerced to a Handle<T>, below. Rooted<T>
should be used whenever a local variable's value may be held live across a
call which can trigger a GC.
- Handle<T> is a const reference to a Rooted<T>. Functions which take GC
things or values as arguments and need to root those arguments should
generally use handles for those arguments and avoid any explicit rooting.
This has two benefits. First, when several such functions call each other
then redundant rooting of multiple copies of the GC thing can be avoided.
Second, if the caller does not pass a rooted value a compile error will be
generated, which is quicker and easier to fix than when relying on a
separate rooting analysis.
- MutableHandle<T> is a non-const reference to Rooted<T>. It is used in the
same way as Handle<T> and includes a |set(const T& v)| method to allow
updating the value of the referenced Rooted<T>. A MutableHandle<T> can be
created with an implicit cast from a Rooted<T>*.
In some cases the small performance overhead of exact rooting (measured to
be a few nanoseconds on desktop) is too much. In these cases, try the
following:
- Move all Rooted<T> above inner loops: this allows you to re-use the root
on each iteration of the loop.
- Pass Handle<T> through your hot call stack to avoid re-rooting costs at
every invocation.
The following diagram explains the list of supported, implicit type
conversions between classes of this family:
Rooted<T> ----> Handle<T>
| ^
| |
| |
+---> MutableHandle<T>
(via &)
All of these types have an implicit conversion to raw pointers.
|
55031 |
| SavedFrameAPI.h |
Functions and types related to SavedFrame objects created by the Debugger
API.
|
6180 |
| SharedArrayBuffer.h |
ArrayBuffer functionality. |
2016 |
| SliceBudget.h |
This class records how much work has been done in a given collection slice,
so that we can return before pausing for too long. Some slices are allowed
to run for unlimited time, and others are bounded. To reduce the number of
gettimeofday calls, we only check the time every 1000 operations.
|
2551 |
| SourceText.h |
SourceText encapsulates a count of char16_t (UTF-16) or Utf8Unit (UTF-8)
code units (note: code *units*, not bytes or code points) and those code
units ("source units"). (Latin-1 is not supported: all places where Latin-1
must be compiled first convert to a supported encoding.)
A SourceText either observes without owning, or takes ownership of, source
units passed to |SourceText::init|. Thus SourceText can be used to
efficiently avoid copying.
Rules for use:
1) The passed-in source units must be allocated with js_malloc(),
js_calloc(), or js_realloc() if |SourceText::init| is instructed to take
ownership of the source units.
2) If |SourceText::init| merely borrows the source units, the user must
keep them alive until associated JS compilation is complete.
3) Code that calls |SourceText::take{Chars,Units}()| must keep the source
units alive until JS compilation completes. Normally only the JS engine
should call |SourceText::take{Chars,Units}()|.
4) Use the appropriate SourceText parameterization depending on the source
units encoding.
Example use:
size_t length = 512;
char16_t* chars = js_pod_malloc<char16_t>(length);
if (!chars) {
JS_ReportOutOfMemory(cx);
return false;
}
JS::SourceText<char16_t> srcBuf;
if (!srcBuf.init(cx, chars, length, JS::SourceOwnership::TakeOwnership)) {
return false;
}
JS::Rooted<JSScript*> script(cx);
if (!JS::Compile(cx, options, srcBuf, &script)) {
return false;
}
|
10854 |
| StableStringChars.h |
Safely access the contents of a string even as GC can cause the string's
contents to move around in memory.
|
4167 |
| Stream.h |
JSAPI functions and callbacks related to WHATWG Stream objects.
Much of the API here mirrors the standard algorithms and standard JS methods
of the objects defined in the Streams standard. One difference is that the
functionality of the JS controller object is exposed to C++ as functions
taking ReadableStream instances instead, for convenience.
|
19803 |
| StructuredClone.h |
API for safe passing of structured data, HTML 2018 Feb 21 section 2.7.
<https://html.spec.whatwg.org/multipage/structured-data.html>
This is a serialization scheme for JS values, somewhat like JSON. It
preserves some aspects of JS objects (strings, numbers, own data properties
with string keys, array elements) but not others (methods, getters and
setters, prototype chains). Unlike JSON, structured data:
- can contain cyclic references.
- handles Maps, Sets, and some other object types.
- supports *transferring* objects of certain types from one realm to
another, rather than cloning them.
- is specified by a living standard, and continues to evolve.
- is encoded in a nonstandard binary format, and is never exposed to Web
content in its serialized form. It's used internally by the browser to
send data from one thread/realm/domain to another, not across the
network.
|
27639 |
| SweepingAPI.h |
|
2590 |
| Symbol.h |
Symbols. |
3738 |
| TraceKind.h |
name type canBeGray inCCGraph |
9900 |
| TraceLoggerAPI.h |
SpiderMonkey TraceLogger APIs. |
7936 |
| TracingAPI.h |
Returns a static string equivalent of |kind|. |
16971 |
| TrackedOptimizationInfo.h |
|
10611 |
| Transcoding.h |
Structures and functions for transcoding compiled scripts and functions to
and from memory.
|
3786 |
| TypeDecls.h |
js_TypeDecls_h |
4018 |
| UbiNode.h |
|
46091 |
| UbiNodeBreadthFirst.h |
|
10134 |
| UbiNodeCensus.h |
|
8169 |
| UbiNodeDominatorTree.h |
In a directed graph with a root node `R`, a node `A` is said to "dominate" a
node `B` iff every path from `R` to `B` contains `A`. A node `A` is said to
be the "immediate dominator" of a node `B` iff it dominates `B`, is not `B`
itself, and does not dominate any other nodes which also dominate `B` in
turn.
If we take every node from a graph `G` and create a new graph `T` with edges
to each node from its immediate dominator, then `T` is a tree (each node has
only one immediate dominator, or none if it is the root). This tree is called
a "dominator tree".
This class represents a dominator tree constructed from a `JS::ubi::Node`
heap graph. The domination relationship and dominator trees are useful tools
for analyzing heap graphs because they tell you:
- Exactly what could be reclaimed by the GC if some node `A` became
unreachable: those nodes which are dominated by `A`,
- The "retained size" of a node in the heap graph, in contrast to its
"shallow size". The "shallow size" is the space taken by a node itself,
not counting anything it references. The "retained size" of a node is its
shallow size plus the size of all the things that would be collected if
the original node wasn't (directly or indirectly) referencing them. In
other words, the retained size is the shallow size of a node plus the
shallow sizes of every other node it dominates. For example, the root
node in a binary tree might have a small shallow size that does not take
up much space itself, but it dominates the rest of the binary tree and
its retained size is therefore significant (assuming no external
references into the tree).
The simple, engineered algorithm presented in "A Simple, Fast Dominance
Algorithm" by Cooper el al[0] is used to find dominators and construct the
dominator tree. This algorithm runs in O(n^2) time, but is faster in practice
than alternative algorithms with better theoretical running times, such as
Lengauer-Tarjan which runs in O(e * log(n)). The big caveat to that statement
is that Cooper et al found it is faster in practice *on control flow graphs*
and I'm not convinced that this property also holds on *heap* graphs. That
said, the implementation of this algorithm is *much* simpler than
Lengauer-Tarjan and has been found to be fast enough at least for the time
being.
[0]: http://www.cs.rice.edu/~keith/EMBED/dom.pdf
|
24250 |
| UbiNodePostOrder.h |
A post-order depth-first traversal of `ubi::Node` graphs.
No GC may occur while an instance of `PostOrder` is live.
The `NodeVisitor` type provided to `PostOrder::traverse` must have the
following member:
bool operator()(Node& node)
The node visitor method. This method is called once for each `node`
reachable from the start set in post-order.
This visitor function should return true on success, or false if an error
occurs. A false return value terminates the traversal immediately, and
causes `PostOrder::traverse` to return false.
The `EdgeVisitor` type provided to `PostOrder::traverse` must have the
following member:
bool operator()(Node& origin, Edge& edge)
The edge visitor method. This method is called once for each outgoing
`edge` from `origin` that is reachable from the start set.
NB: UNLIKE NODES, THERE IS NO GUARANTEED ORDER IN WHICH EDGES AND THEIR
ORIGINS ARE VISITED!
This visitor function should return true on success, or false if an error
occurs. A false return value terminates the traversal immediately, and
causes `PostOrder::traverse` to return false.
|
5488 |
| UbiNodeShortestPaths.h |
A back edge along a path in the heap graph.
|
10411 |
| UbiNodeUtils.h |
|
1239 |
| UniquePtr.h |
js_UniquePtr_h |
1547 |
| Utility.h |
The public JS engine namespace. |
23244 |
| Value.h |
JS::Value implementation. |
41929 |
| Vector.h |
js_Vector_h |
1035 |
| Warnings.h |
Functionality for issuing and handling warnings.
Warnings are situations that aren't inherently full-blown errors (and perhaps
for spec compliance *can't* be), but that may represent dubious programming
practice that embeddings may wish to know about. (That said, SpiderMonkey
exposes various options that, when active, automatically upgrade warnings to
errors. See |JS::TransitiveCompileOptions::werrorOptions| and
|JS::ContextOptions::werror()|.)
SpiderMonkey recognizes an unspecified set of syntactic patterns and runtime
behaviors as triggering a warning. Embeddings may also recognize and report
additional warnings.
|
3424 |
| WeakMapPtr.h |
namespace JS |
1415 |
| Wrapper.h |
Helper for Wrapper::New default options.
Callers of Wrapper::New() who wish to specify a prototype for the created
Wrapper, *MUST* construct a WrapperOptions with a JSContext.
|
23698 |