Source code

Revision control

Copy as Markdown

Other Tools

/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*-
* vim: set ts=8 sts=2 et sw=2 tw=80:
* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
#ifndef js_UbiNodeCensus_h
#define js_UbiNodeCensus_h
#include "js/GCVector.h"
#include "js/UbiNode.h"
#include "js/UbiNodeBreadthFirst.h"
// A census is a ubi::Node traversal that assigns each node to one or more
// buckets, and returns a report with the size of each bucket.
//
// We summarize the results of a census with counts broken down according to
// criteria selected by the API consumer code that is requesting the census. For
// example, the following breakdown might give an interesting overview of the
// heap:
//
// - all nodes
// - objects
// - objects with a specific [[Class]] *
// - strings
// - scripts
// - DOM nodes
// - nsINodes with a specific name (found in nsINode::NodeName()) *
// - all other Node types
// - nodes with a specific ubi::Node::typeName *
//
// Obviously, the parts of this tree marked with * represent many separate
// counts, depending on how many distinct [[Class]] values and ubi::Node type
// names we encounter.
//
// The supported types of breakdowns are documented in
// js/src/doc/Debugger/Debugger.Memory.md.
//
// When we parse the 'breakdown' argument to takeCensus, we build a tree of
// CountType nodes. For example, for the breakdown shown in the
// Debugger.Memory.prototype.takeCensus, documentation:
//
// {
// by: "coarseType",
// objects: { by: "objectClass" },
// other: { by: "internalType" },
// domNode: { by: "descriptiveType" }
// }
//
// we would build the following tree of CountType subclasses:
//
// ByCoarseType
// objects: ByObjectClass
// each class: SimpleCount
// scripts: SimpleCount
// strings: SimpleCount
// other: ByUbinodeType
// each type: SimpleCount
// domNode: ByDomObjectClass
// each type: SimpleCount
//
// The interior nodes are all breakdown types that categorize nodes according to
// one characteristic or another; and the leaf nodes are all SimpleType.
//
// Each CountType has its own concrete C++ type that holds the counts it
// produces. SimpleCount::Count just holds totals. ByObjectClass::Count has a
// hash table whose keys are object class names and whose values are counts of
// some other type (in the example above, SimpleCount).
//
// To keep actual count nodes small, they have no vtable. Instead, each count
// points to its CountType, which knows how to carry out all the operations we
// need on a Count. A CountType can produce new count nodes; process nodes as we
// visit them; build a JS object reporting the results; and destruct count
// nodes.
namespace JS {
namespace ubi {
struct Census;
class CountBase;
struct CountDeleter {
JS_PUBLIC_API void operator()(CountBase*);
};
using CountBasePtr = js::UniquePtr<CountBase, CountDeleter>;
// Abstract base class for CountType nodes.
struct CountType {
explicit CountType() = default;
virtual ~CountType() = default;
// Destruct a count tree node that this type instance constructed.
virtual void destructCount(CountBase& count) = 0;
// Return a fresh node for the count tree that categorizes nodes according
// to this type. Return a nullptr on OOM.
virtual CountBasePtr makeCount() = 0;
// Trace |count| and all its children, for garbage collection.
virtual void traceCount(CountBase& count, JSTracer* trc) = 0;
// Implement the 'count' method for counts returned by this CountType
// instance's 'newCount' method.
[[nodiscard]] virtual bool count(CountBase& count,
mozilla::MallocSizeOf mallocSizeOf,
const Node& node) = 0;
// Implement the 'report' method for counts returned by this CountType
// instance's 'newCount' method.
[[nodiscard]] virtual bool report(JSContext* cx, CountBase& count,
MutableHandleValue report) = 0;
};
using CountTypePtr = js::UniquePtr<CountType>;
// An abstract base class for count tree nodes.
class CountBase {
// In lieu of a vtable, each CountBase points to its type, which
// carries not only the implementations of the CountBase methods, but also
// additional parameters for the type's behavior, as specified in the
// breakdown argument passed to takeCensus.
CountType& type;
protected:
~CountBase() = default;
public:
explicit CountBase(CountType& type)
: type(type), total_(0), smallestNodeIdCounted_(SIZE_MAX) {}
// Categorize and count |node| as appropriate for this count's type.
[[nodiscard]] bool count(mozilla::MallocSizeOf mallocSizeOf,
const Node& node) {
total_++;
auto id = node.identifier();
if (id < smallestNodeIdCounted_) {
smallestNodeIdCounted_ = id;
}
#ifdef DEBUG
size_t oldTotal = total_;
#endif
bool ret = type.count(*this, mallocSizeOf, node);
MOZ_ASSERT(total_ == oldTotal,
"CountType::count should not increment total_, CountBase::count "
"handles that");
return ret;
}
// Construct a JavaScript object reporting the counts recorded in this
// count, and store it in |report|. Return true on success, or false on
// failure.
[[nodiscard]] bool report(JSContext* cx, MutableHandleValue report) {
return type.report(cx, *this, report);
}
// Down-cast this CountBase to its true type, based on its 'type' member,
// and run its destructor.
void destruct() { return type.destructCount(*this); }
// Trace this count for garbage collection.
void trace(JSTracer* trc) { type.traceCount(*this, trc); }
size_t total_;
// The smallest JS::ubi::Node::identifier() passed to this instance's
// count() method. This provides a stable way to sort sets.
Node::Id smallestNodeIdCounted_;
};
using RootedCount = JS::Rooted<CountBasePtr>;
// Common data for a census traversal, shared across all CountType nodes.
struct Census {
JSContext* const cx;
// If the targetZones set is non-empty, then only consider nodes whose zone
// is an element of the set. If the targetZones set is empty, then nodes in
// all zones are considered.
JS::ZoneSet targetZones;
explicit Census(JSContext* cx) : cx(cx) {}
};
// A BreadthFirst handler type that conducts a census, using a CountBase to
// categorize and count each node.
class CensusHandler {
Census& census;
JS::Handle<CountBasePtr> rootCount;
mozilla::MallocSizeOf mallocSizeOf;
public:
CensusHandler(Census& census, JS::Handle<CountBasePtr> rootCount,
mozilla::MallocSizeOf mallocSizeOf)
: census(census), rootCount(rootCount), mallocSizeOf(mallocSizeOf) {}
[[nodiscard]] bool report(JSContext* cx, MutableHandleValue report) {
return rootCount->report(cx, report);
}
// This class needs to retain no per-node data.
class NodeData {};
[[nodiscard]] JS_PUBLIC_API bool operator()(
BreadthFirst<CensusHandler>& traversal, Node origin, const Edge& edge,
NodeData* referentData, bool first);
};
using CensusTraversal = BreadthFirst<CensusHandler>;
// Examine the census options supplied by the API consumer, and (among other
// things) use that to build a CountType tree.
[[nodiscard]] JS_PUBLIC_API bool ParseCensusOptions(JSContext* cx,
Census& census,
HandleObject options,
CountTypePtr& outResult);
// Parse the breakdown language (as described in
// js/src/doc/Debugger/Debugger.Memory.md) into a CountTypePtr. A null pointer
// is returned on error and is reported to the cx.
JS_PUBLIC_API CountTypePtr
ParseBreakdown(JSContext* cx, HandleValue breakdownValue,
MutableHandle<JS::GCVector<JSLinearString*>> seen);
} // namespace ubi
} // namespace JS
#endif // js_UbiNodeCensus_h