mozilla
/
pjs
зеркало из https://github.com/mozilla/pjs.git
1
0
Форкнуть 0
Код Задачи Пакеты Проекты Релизы Вики Активность
pjs/xpcom/base/nsCycleCollector.cpp

2642 строки
72 KiB
C++
Исходник Ответственный История

/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
/* vim: set cindent tabstop=4 expandtab shiftwidth=4: */
/* ***** BEGIN LICENSE BLOCK *****
* Version: MPL 1.1/GPL 2.0/LGPL 2.1
*
* The contents of this file are subject to the Mozilla Public License Version
* 1.1 (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
* http://www.mozilla.org/MPL/
*
* Software distributed under the License is distributed on an "AS IS" basis,
* WITHOUT WARRANTY OF ANY KIND, either express or implied. See the License
* for the specific language governing rights and limitations under the
* License.
*
* The Original Code is mozilla.org code.
*
* The Initial Developer of the Original Code is
* The Mozilla Foundation.
* Portions created by the Initial Developer are Copyright (C) 2006
* the Initial Developer. All Rights Reserved.
*
* Contributor(s):
* L. David Baron <dbaron@dbaron.org>, Mozilla Corporation
*
* Alternatively, the contents of this file may be used under the terms of
* either of the GNU General Public License Version 2 or later (the "GPL"),
* or the GNU Lesser General Public License Version 2.1 or later (the "LGPL"),
* in which case the provisions of the GPL or the LGPL are applicable instead
* of those above. If you wish to allow use of your version of this file only
* under the terms of either the GPL or the LGPL, and not to allow others to
* use your version of this file under the terms of the MPL, indicate your
* decision by deleting the provisions above and replace them with the notice
* and other provisions required by the GPL or the LGPL. If you do not delete
* the provisions above, a recipient may use your version of this file under
* the terms of any one of the MPL, the GPL or the LGPL.
*
* ***** END LICENSE BLOCK ***** */
//
// This file implements a garbage-cycle collector based on the paper
//
// Concurrent Cycle Collection in Reference Counted Systems
// Bacon & Rajan (2001), ECOOP 2001 / Springer LNCS vol 2072
//
// We are not using the concurrent or acyclic cases of that paper; so
// the green, red and orange colors are not used.
//
// The collector is based on tracking pointers of four colors:
//
// Black nodes are definitely live. If we ever determine a node is
// black, it's ok to forget about, drop from our records.
//
// White nodes are definitely garbage cycles. Once we finish with our
// scanning, we unlink all the white nodes and expect that by
// unlinking them they will self-destruct (since a garbage cycle is
// only keeping itself alive with internal links, by definition).
//
// Grey nodes are being scanned. Nodes that turn grey will turn
// either black if we determine that they're live, or white if we
// determine that they're a garbage cycle. After the main collection
// algorithm there should be no grey nodes.
//
// Purple nodes are *candidates* for being scanned. They are nodes we
// haven't begun scanning yet because they're not old enough, or we're
// still partway through the algorithm.
//
// XPCOM objects participating in garbage-cycle collection are obliged
// to inform us when they ought to turn purple; that is, when their
// refcount transitions from N+1 -> N, for nonzero N. Furthermore we
// require that *after* an XPCOM object has informed us of turning
// purple, they will tell us when they either transition back to being
// black (incremented refcount) or are ultimately deleted.
// Safety:
//
// An XPCOM object is either scan-safe or scan-unsafe, purple-safe or
// purple-unsafe.
//
// An object is scan-safe if:
//
// - It can be QI'ed to |nsXPCOMCycleCollectionParticipant|, though this
// operation loses ISupports identity (like nsIClassInfo).
// - The operation |traverse| on the resulting
// nsXPCOMCycleCollectionParticipant does not cause *any* refcount
// adjustment to occur (no AddRef / Release calls).
//
// An object is purple-safe if it satisfies the following properties:
//
// - The object is scan-safe.
// - If the object calls |nsCycleCollector::suspect(this)|,
// it will eventually call |nsCycleCollector::forget(this)|,
// exactly once per call to |suspect|, before being destroyed.
//
// When we receive a pointer |ptr| via
// |nsCycleCollector::suspect(ptr)|, we assume it is purple-safe. We
// can check the scan-safety, but have no way to ensure the
// purple-safety; objects must obey, or else the entire system falls
// apart. Don't involve an object in this scheme if you can't
// guarantee its purple-safety.
//
// When we have a scannable set of purple nodes ready, we begin
// our walks. During the walks, the nodes we |traverse| should only
// feed us more scan-safe nodes, and should not adjust the refcounts
// of those nodes.
//
// We do not |AddRef| or |Release| any objects during scanning. We
// rely on purple-safety of the roots that call |suspect| and
// |forget| to hold, such that we will forget about a purple pointer
// before it is destroyed. The pointers that are merely scan-safe,
// we hold only for the duration of scanning, and there should be no
// objects released from the scan-safe set during the scan (there
// should be no threads involved).
//
// We *do* call |AddRef| and |Release| on every white object, on
// either side of the calls to |Unlink|. This keeps the set of white
// objects alive during the unlinking.
//
#ifndef __MINGW32__
#ifdef WIN32
#include <crtdbg.h>
#include <errno.h>
#endif
#endif
#include "nsCycleCollectionParticipant.h"
#include "nsIProgrammingLanguage.h"
#include "nsBaseHashtable.h"
#include "nsHashKeys.h"
#include "nsDeque.h"
#include "nsCycleCollector.h"
#include "nsThreadUtils.h"
#include "prenv.h"
#include "prprf.h"
#include "plstr.h"
#include "prtime.h"
#include "nsPrintfCString.h"
#include "nsTArray.h"
#include "nsIObserverService.h"
#include "nsServiceManagerUtils.h"
#include <stdio.h>
#ifdef WIN32
#include <io.h>
#include <process.h>
#endif
#define DEFAULT_SHUTDOWN_COLLECTIONS 5
#ifdef DEBUG_CC
#define SHUTDOWN_COLLECTIONS(params) params.mShutdownCollections
#else
#define SHUTDOWN_COLLECTIONS(params) DEFAULT_SHUTDOWN_COLLECTIONS
#endif
// Various parameters of this collector can be tuned using environment
// variables.
struct nsCycleCollectorParams
{
PRBool mDoNothing;
#ifdef DEBUG_CC
PRBool mReportStats;
PRBool mHookMalloc;
PRBool mDrawGraphs;
PRBool mFaultIsFatal;
PRBool mLogPointers;
PRUint32 mShutdownCollections;
#endif
PRUint32 mScanDelay;
nsCycleCollectorParams() :
#ifdef DEBUG_CC
mDoNothing (PR_GetEnv("XPCOM_CC_DO_NOTHING") != NULL),
mReportStats (PR_GetEnv("XPCOM_CC_REPORT_STATS") != NULL),
mHookMalloc (PR_GetEnv("XPCOM_CC_HOOK_MALLOC") != NULL),
mDrawGraphs (PR_GetEnv("XPCOM_CC_DRAW_GRAPHS") != NULL),
mFaultIsFatal (PR_GetEnv("XPCOM_CC_FAULT_IS_FATAL") != NULL),
mLogPointers (PR_GetEnv("XPCOM_CC_LOG_POINTERS") != NULL),
mShutdownCollections(DEFAULT_SHUTDOWN_COLLECTIONS),
#else
mDoNothing (PR_FALSE),
#endif
// The default number of collections to "age" candidate
// pointers in the purple buffer before we decide that any
// garbage cycle they're in has stabilized and we want to
// consider scanning it.
//
// Making this number smaller causes:
// - More time to be spent in the collector (bad)
// - Less delay between forming garbage and collecting it (good)
mScanDelay(0)
{
#ifdef DEBUG_CC
char *s = PR_GetEnv("XPCOM_CC_SCAN_DELAY");
if (s)
PR_sscanf(s, "%d", &mScanDelay);
s = PR_GetEnv("XPCOM_CC_SHUTDOWN_COLLECTIONS");
if (s)
PR_sscanf(s, "%d", &mShutdownCollections);
#endif
}
};
#ifdef DEBUG_CC
// Various operations involving the collector are recorded in a
// statistics table. These are for diagnostics.
struct nsCycleCollectorStats
{
PRUint32 mFailedQI;
PRUint32 mSuccessfulQI;
PRUint32 mVisitedNode;
PRUint32 mWalkedGraph;
PRUint32 mCollectedBytes;
PRUint32 mFreeCalls;
PRUint32 mFreedBytes;
PRUint32 mSetColorGrey;
PRUint32 mSetColorBlack;
PRUint32 mSetColorWhite;
PRUint32 mFailedUnlink;
PRUint32 mCollectedNode;
PRUint32 mBumpGeneration;
PRUint32 mZeroGeneration;
PRUint32 mSuspectNode;
PRUint32 mSpills;
PRUint32 mForgetNode;
PRUint32 mFreedWhilePurple;
PRUint32 mCollection;
nsCycleCollectorStats()
{
memset(this, 0, sizeof(nsCycleCollectorStats));
}
void Dump()
{
fprintf(stderr, "\f\n");
#define DUMP(entry) fprintf(stderr, "%30.30s: %-20.20d\n", #entry, entry)
DUMP(mFailedQI);
DUMP(mSuccessfulQI);
DUMP(mVisitedNode);
DUMP(mWalkedGraph);
DUMP(mCollectedBytes);
DUMP(mFreeCalls);
DUMP(mFreedBytes);
DUMP(mSetColorGrey);
DUMP(mSetColorBlack);
DUMP(mSetColorWhite);
DUMP(mFailedUnlink);
DUMP(mCollectedNode);
DUMP(mBumpGeneration);
DUMP(mZeroGeneration);
DUMP(mSuspectNode);
DUMP(mSpills);
DUMP(mForgetNode);
DUMP(mFreedWhilePurple);
DUMP(mCollection);
#undef DUMP
}
};
#endif
#ifdef DEBUG_CC
static PRBool
nsCycleCollector_shouldSuppress(nsISupports *s);
#endif
////////////////////////////////////////////////////////////////////////
// Base types
////////////////////////////////////////////////////////////////////////
struct PtrInfo;
class EdgePool
{
public:
// EdgePool allocates arrays of void*, primarily to hold PtrInfo*.
// However, at the end of a block, the last two pointers are a null
// and then a void** pointing to the next block. This allows
// EdgePool::Iterators to be a single word but still capable of crossing
// block boundaries.
EdgePool()
{
mSentinelAndBlocks[0].block = nsnull;
mSentinelAndBlocks[1].block = nsnull;
}
~EdgePool()
{
Block *b = Blocks();
while (b) {
Block *next = b->Next();
delete b;
b = next;
}
}
private:
struct Block;
union PtrInfoOrBlock {
// Use a union to avoid reinterpret_cast and the ensuing
// potential aliasing bugs.
PtrInfo *ptrInfo;
Block *block;
};
struct Block {
enum { BlockSize = 64 * 1024 };
PtrInfoOrBlock mPointers[BlockSize];
Block() {
mPointers[BlockSize - 2].block = nsnull; // sentinel
mPointers[BlockSize - 1].block = nsnull; // next block pointer
}
Block*& Next()
{ return mPointers[BlockSize - 1].block; }
PtrInfoOrBlock* Start()
{ return &mPointers[0]; }
PtrInfoOrBlock* End()
{ return &mPointers[BlockSize - 2]; }
};
// Store the null sentinel so that we can have valid iterators
// before adding any edges and without adding any blocks.
PtrInfoOrBlock mSentinelAndBlocks[2];
Block*& Blocks() { return mSentinelAndBlocks[1].block; }
public:
class Iterator
{
public:
Iterator() : mPointer(nsnull) {}
Iterator(PtrInfoOrBlock *aPointer) : mPointer(aPointer) {}
Iterator(const Iterator& aOther) : mPointer(aOther.mPointer) {}
Iterator& operator++()
{
if (mPointer->ptrInfo == nsnull) {
// Null pointer is a sentinel for link to the next block.
mPointer = (mPointer + 1)->block->mPointers;
}
++mPointer;
return *this;
}
PtrInfo* operator*() const
{
if (mPointer->ptrInfo == nsnull) {
// Null pointer is a sentinel for link to the next block.
return (mPointer + 1)->block->mPointers->ptrInfo;
}
return mPointer->ptrInfo;
}
PRBool operator==(const Iterator& aOther) const
{ return mPointer == aOther.mPointer; }
PRBool operator!=(const Iterator& aOther) const
{ return mPointer != aOther.mPointer; }
private:
PtrInfoOrBlock *mPointer;
};
class Builder;
friend class Builder;
class Builder {
public:
Builder(EdgePool &aPool)
: mCurrent(&aPool.mSentinelAndBlocks[0]),
mBlockEnd(&aPool.mSentinelAndBlocks[0]),
mNextBlockPtr(&aPool.Blocks())
{
}
Iterator Mark() { return Iterator(mCurrent); }
void Add(PtrInfo* aEdge) {
if (mCurrent == mBlockEnd) {
Block *b = new Block();
if (!b) {
// This means we just won't collect (some) cycles.
NS_NOTREACHED("out of memory, ignoring edges");
return;
}
*mNextBlockPtr = b;
mCurrent = b->Start();
mBlockEnd = b->End();
mNextBlockPtr = &b->Next();
}
(mCurrent++)->ptrInfo = aEdge;
}
private:
// mBlockEnd points to space for null sentinel
PtrInfoOrBlock *mCurrent, *mBlockEnd;
Block **mNextBlockPtr;
};
};
#ifdef DEBUG_CC
struct ReversedEdge {
PtrInfo *mTarget;
ReversedEdge *mNext;
};
#endif
enum NodeColor { black, white, grey };
// This structure should be kept as small as possible; we may expect
// a million of them to be allocated and touched repeatedly during
// each cycle collection.
struct PtrInfo
{
void *mPointer;
nsCycleCollectionParticipant *mParticipant;
PRUint32 mColor : 2;
PRUint32 mInternalRefs : 30;
PRUint32 mRefCount : 31;
PRUint32 mWasPurple : 1;
EdgePool::Iterator mFirstChild; // first
EdgePool::Iterator mLastChild; // one after last
#ifdef DEBUG_CC
size_t mBytes;
char *mName;
PRUint32 mLangID;
// For finding roots in ExplainLiveExpectedGarbage (when there are
// missing calls to suspect or failures to unlink).
PRUint32 mSCCIndex; // strongly connected component
// For finding roots in ExplainLiveExpectedGarbage (when nodes
// expected to be garbage are black).
ReversedEdge* mReversedEdges; // linked list
PtrInfo* mShortestPathToExpectedGarbage;
#endif
PtrInfo(void *aPointer, nsCycleCollectionParticipant *aParticipant
IF_DEBUG_CC_PARAM(PRUint32 aLangID)
)
: mPointer(aPointer),
mParticipant(aParticipant),
mColor(grey),
mInternalRefs(0),
mRefCount(0),
mWasPurple(PR_FALSE),
mFirstChild(),
mLastChild()
#ifdef DEBUG_CC
, mBytes(0),
mName(nsnull),
mLangID(aLangID),
mSCCIndex(0),
mReversedEdges(nsnull),
mShortestPathToExpectedGarbage(nsnull)
#endif
{
}
// Allow uninitialized values in large arrays.
PtrInfo() {}
};
/**
* A structure designed to be used like a linked list of PtrInfo, except
* that allocates the PtrInfo 32K-at-a-time.
*/
class NodePool
{
private:
enum { BlockSize = 32 * 1024 }; // could be int template parameter
struct Block {
Block* mNext;
PtrInfo mEntries[BlockSize];
Block() : mNext(nsnull) {}
};
public:
NodePool()
: mBlocks(nsnull),
mLast(nsnull)
{
}
~NodePool()
{
#ifdef DEBUG_CC
{
Enumerator queue(*this);
while (!queue.IsDone()) {
PL_strfree(queue.GetNext()->mName);
}
}
#endif
Block *b = mBlocks;
while (b) {
Block *n = b->mNext;
delete b;
b = n;
}
}
class Builder;
friend class Builder;
class Builder {
public:
Builder(NodePool& aPool)
: mNextBlock(&aPool.mBlocks),
mNext(aPool.mLast),
mBlockEnd(nsnull)
{
NS_ASSERTION(aPool.mBlocks == nsnull && aPool.mLast == nsnull,
"pool not empty");
}
PtrInfo *Add(void *aPointer, nsCycleCollectionParticipant *aParticipant
IF_DEBUG_CC_PARAM(PRUint32 aLangID)
)
{
if (mNext == mBlockEnd) {
Block *block;
if (!(*mNextBlock = block = new Block()))
return nsnull;
mNext = block->mEntries;
mBlockEnd = block->mEntries + BlockSize;
mNextBlock = &block->mNext;
}
return new (mNext++) PtrInfo(aPointer, aParticipant
IF_DEBUG_CC_PARAM(aLangID)
);
}
private:
Block **mNextBlock;
PtrInfo *&mNext;
PtrInfo *mBlockEnd;
};
class Enumerator;
friend class Enumerator;
class Enumerator {
public:
Enumerator(NodePool& aPool)
: mFirstBlock(aPool.mBlocks),
mCurBlock(nsnull),
mNext(nsnull),
mBlockEnd(nsnull),
mLast(aPool.mLast)
{
}
PRBool IsDone() const
{
return mNext == mLast;
}
PtrInfo* GetNext()
{
NS_ASSERTION(!IsDone(), "calling GetNext when done");
if (mNext == mBlockEnd) {
Block *nextBlock = mCurBlock ? mCurBlock->mNext : mFirstBlock;
mNext = nextBlock->mEntries;
mBlockEnd = mNext + BlockSize;
mCurBlock = nextBlock;
}
return mNext++;
}
private:
Block *mFirstBlock, *mCurBlock;
// mNext is the next value we want to return, unless mNext == mBlockEnd
// NB: mLast is a reference to allow enumerating while building!
PtrInfo *mNext, *mBlockEnd, *&mLast;
};
private:
Block *mBlocks;
PtrInfo *mLast;
};
struct GCGraph;
static GCGraph *sCurrGraph = nsnull;
struct GCGraph
{
NodePool mNodes;
EdgePool mEdges;
PRUint32 mRootCount;
#ifdef DEBUG_CC
ReversedEdge *mReversedEdges;
#endif
GCGraph() : mRootCount(0) {
sCurrGraph = this;
}
~GCGraph() {
sCurrGraph = nsnull;
}
};
// XXX Would be nice to have an nsHashSet<KeyType> API that has
// Add/Remove/Has rather than PutEntry/RemoveEntry/GetEntry.
typedef nsTHashtable<nsVoidPtrHashKey> PointerSet;
typedef nsBaseHashtable<nsVoidPtrHashKey, PRUint32, PRUint32>
PointerSetWithGeneration;
static void
WriteGraph(FILE *stream, GCGraph &graph, const void *redPtr);
struct nsPurpleBuffer
{
#define ASSOCIATIVITY 2
#define INDEX_LOW_BIT 6
#define N_INDEX_BITS 13
#define N_ENTRIES (1 << N_INDEX_BITS)
#define N_POINTERS (N_ENTRIES * ASSOCIATIVITY)
#define TOTAL_BYTES (N_POINTERS * PR_BYTES_PER_WORD)
#define INDEX_MASK PR_BITMASK(N_INDEX_BITS)
#define POINTER_INDEX(P) ((((PRUword)P) >> INDEX_LOW_BIT) & (INDEX_MASK))
#if (INDEX_LOW_BIT + N_INDEX_BITS > (8 * PR_BYTES_PER_WORD))
#error "index bit overflow"
#endif
// This class serves as a generational wrapper around a pldhash
// table: a subset of generation zero lives in mCache, the
// remainder spill into the mBackingStore hashtable. The idea is
// to get a higher hit rate and greater locality of reference for
// generation zero, in which the vast majority of suspect/forget
// calls annihilate one another.
nsCycleCollectorParams &mParams;
#ifdef DEBUG_CC
nsCycleCollectorStats &mStats;
#endif
void* mCache[N_POINTERS];
PRUint32 mCurrGen;
PointerSetWithGeneration mBackingStore;
nsDeque *mTransferBuffer;
#ifdef DEBUG_CC
nsPurpleBuffer(nsCycleCollectorParams &params,
nsCycleCollectorStats &stats)
: mParams(params),
mStats(stats),
mCurrGen(0),
mTransferBuffer(nsnull)
{
Init();
}
#else
nsPurpleBuffer(nsCycleCollectorParams &params)
: mParams(params),
mCurrGen(0),
mTransferBuffer(nsnull)
{
Init();
}
#endif
~nsPurpleBuffer()
{
memset(mCache, 0, sizeof(mCache));
mBackingStore.Clear();
}
void Init()
{
memset(mCache, 0, sizeof(mCache));
mBackingStore.Init();
}
void BumpGeneration();
void SelectAgedPointers(nsDeque *transferBuffer);
PRBool Exists(void *p)
{
PRUint32 idx = POINTER_INDEX(p);
for (PRUint32 i = 0; i < ASSOCIATIVITY; ++i) {
if (mCache[idx+i] == p)
return PR_TRUE;
}
PRUint32 gen;
return mBackingStore.Get(p, &gen);
}
void Put(void *p)
{
PRUint32 idx = POINTER_INDEX(p);
for (PRUint32 i = 0; i < ASSOCIATIVITY; ++i) {
if (!mCache[idx+i]) {
mCache[idx+i] = p;
return;
}
}
#ifdef DEBUG_CC
mStats.mSpills++;
#endif
SpillOne(p);
}
void Remove(void *p)
{
PRUint32 idx = POINTER_INDEX(p);
for (PRUint32 i = 0; i < ASSOCIATIVITY; ++i) {
if (mCache[idx+i] == p) {
mCache[idx+i] = (void*)0;
return;
}
}
mBackingStore.Remove(p);
}
void SpillOne(void* &p)
{
mBackingStore.Put(p, mCurrGen);
p = (void*)0;
}
void SpillAll()
{
for (PRUint32 i = 0; i < N_POINTERS; ++i) {
if (mCache[i]) {
SpillOne(mCache[i]);
}
}
}
};
static PR_CALLBACK PLDHashOperator
zeroGenerationCallback(const void* ptr,
PRUint32& generation,
void* userArg)
{
#ifdef DEBUG_CC
nsPurpleBuffer *purp = static_cast<nsPurpleBuffer*>(userArg);
purp->mStats.mZeroGeneration++;
#endif
generation = 0;
return PL_DHASH_NEXT;
}
void nsPurpleBuffer::BumpGeneration()
{
SpillAll();
if (mCurrGen == 0xffffffff) {
mBackingStore.Enumerate(zeroGenerationCallback, this);
mCurrGen = 0;
} else {
++mCurrGen;
}
#ifdef DEBUG_CC
mStats.mBumpGeneration++;
#endif
}
static inline PRBool
SufficientlyAged(PRUint32 generation, nsPurpleBuffer *p)
{
return generation + p->mParams.mScanDelay < p->mCurrGen;
}
static PR_CALLBACK PLDHashOperator
ageSelectionCallback(const void* ptr,
PRUint32& generation,
void* userArg)
{
nsPurpleBuffer *purp = static_cast<nsPurpleBuffer*>(userArg);
if (SufficientlyAged(generation, purp)) {
nsISupports *root = static_cast<nsISupports *>(const_cast<void*>(ptr));
purp->mTransferBuffer->Push(root);
}
return PL_DHASH_NEXT;
}
void
nsPurpleBuffer::SelectAgedPointers(nsDeque *transferBuffer)
{
// Rely on our caller having done a BumpGeneration first, which in
// turn calls SpillAll.
mTransferBuffer = transferBuffer;
mBackingStore.Enumerate(ageSelectionCallback, this);
mTransferBuffer = nsnull;
}
////////////////////////////////////////////////////////////////////////
// Implement the LanguageRuntime interface for C++/XPCOM
////////////////////////////////////////////////////////////////////////
struct nsCycleCollectionXPCOMRuntime :
public nsCycleCollectionLanguageRuntime
{
nsresult BeginCycleCollection()
{
return NS_OK;
}
nsresult FinishCycleCollection()
{
return NS_OK;
}
inline nsCycleCollectionParticipant *ToParticipant(void *p);
#ifdef DEBUG_CC
virtual void PrintAllReferencesTo(void *p) {}
virtual void SuspectExtraPointers() {}
#endif
};
struct nsCycleCollector
{
PRBool mCollectionInProgress;
PRBool mScanInProgress;
nsCycleCollectionLanguageRuntime *mRuntimes[nsIProgrammingLanguage::MAX+1];
nsCycleCollectionXPCOMRuntime mXPCOMRuntime;
// The buffer |mBuf| serves a variety of purposes; mostly involving the
// transfer of pointers from a hashtable iterator routine to some outer
// logic that might also need to mutate the hashtable.
nsDeque mBuf;
nsCycleCollectorParams mParams;
nsPurpleBuffer mPurpleBuf;
void RegisterRuntime(PRUint32 langID,
nsCycleCollectionLanguageRuntime *rt);
void ForgetRuntime(PRUint32 langID);
void SelectPurple();
void MarkRoots(GCGraph &graph);
void ScanRoots(GCGraph &graph);
PRBool CollectWhite(GCGraph &graph); // returns whether anything collected
nsCycleCollector();
~nsCycleCollector();
PRBool Suspect(nsISupports *n, PRBool current = PR_FALSE);
PRBool Forget(nsISupports *n);
void Allocated(void *n, size_t sz);
void Freed(void *n);
void Collect(PRUint32 aTryCollections = 1);
void Shutdown();
#ifdef DEBUG_CC
nsCycleCollectorStats mStats;
FILE *mPtrLog;
void MaybeDrawGraphs(GCGraph &graph);
void ExplainLiveExpectedGarbage();
PRBool CreateReversedEdges(GCGraph &graph);
void DestroyReversedEdges(GCGraph &graph);
void ShouldBeFreed(nsISupports *n);
void WasFreed(nsISupports *n);
PointerSet mExpectedGarbage;
#endif
};
class GraphWalker
{
private:
void DoWalk(nsDeque &aQueue);
public:
void Walk(PtrInfo *s0);
void WalkFromRoots(GCGraph &aGraph);
// Provided by concrete walker subtypes.
virtual PRBool ShouldVisitNode(PtrInfo const *pi) = 0;
virtual void VisitNode(PtrInfo *pi) = 0;
};
////////////////////////////////////////////////////////////////////////
// The static collector object
////////////////////////////////////////////////////////////////////////
static nsCycleCollector *sCollector = nsnull;
////////////////////////////////////////////////////////////////////////
// Utility functions
////////////////////////////////////////////////////////////////////////
static void
Fault(const char *msg, const void *ptr=nsnull)
{
#ifdef DEBUG_CC
// This should be nearly impossible, but just in case.
if (!sCollector)
return;
if (sCollector->mParams.mFaultIsFatal) {
if (ptr)
printf("Fatal fault in cycle collector: %s (ptr: %p)\n", msg, ptr);
else
printf("Fatal fault in cycle collector: %s\n", msg);
if (sCurrGraph) {
FILE *stream;
#ifdef WIN32
const char fname[] = "c:\\fault-graph.dot";
#else
const char fname[] = "/tmp/fault-graph.dot";
#endif
printf("depositing faulting cycle-collection graph in %s\n", fname);
stream = fopen(fname, "w+");
WriteGraph(stream, *sCurrGraph, ptr);
fclose(stream);
}
exit(1);
}
#endif
NS_NOTREACHED(nsPrintfCString(256,
"Fault in cycle collector: %s (ptr: %p)\n",
msg, ptr).get());
// When faults are not fatal, we assume we're running in a
// production environment and we therefore want to disable the
// collector on a fault. This will unfortunately cause the browser
// to leak pretty fast wherever creates cyclical garbage, but it's
// probably a better user experience than crashing. Besides, we
// *should* never hit a fault.
sCollector->mParams.mDoNothing = PR_TRUE;
}
#ifdef DEBUG_CC
static void
Fault(const char *msg, PtrInfo *pi)
{
printf("Fault in cycle collector: %s\n"
" while operating on pointer %p %s\n",
msg, pi->mPointer, pi->mName);
if (pi->mInternalRefs) {
printf(" which has internal references from:\n");
NodePool::Enumerator queue(sCurrGraph->mNodes);
while (!queue.IsDone()) {
PtrInfo *ppi = queue.GetNext();
for (EdgePool::Iterator e = ppi->mFirstChild, e_end = ppi->mLastChild;
e != e_end; ++e) {
if (*e == pi) {
printf(" %p %s\n", ppi->mPointer, ppi->mName);
}
}
}
}
Fault(msg, pi->mPointer);
}
#else
inline void
Fault(const char *msg, PtrInfo *pi)
{
Fault(msg, pi->mPointer);
}
#endif
static nsISupports *
canonicalize(nsISupports *in)
{
nsCOMPtr<nsISupports> child;
in->QueryInterface(NS_GET_IID(nsCycleCollectionISupports),
getter_AddRefs(child));
return child.get();
}
static inline void
ToParticipant(nsISupports *s, nsXPCOMCycleCollectionParticipant **cp)
{
// We use QI to move from an nsISupports to an
// nsXPCOMCycleCollectionParticipant, which is a per-class singleton helper
// object that implements traversal and unlinking logic for the nsISupports
// in question.
CallQueryInterface(s, cp);
#ifdef DEBUG_CC
if (cp)
++sCollector->mStats.mSuccessfulQI;
else
++sCollector->mStats.mFailedQI;
#endif
}
nsCycleCollectionParticipant *
nsCycleCollectionXPCOMRuntime::ToParticipant(void *p)
{
nsXPCOMCycleCollectionParticipant *cp;
::ToParticipant(static_cast<nsISupports*>(p), &cp);
return cp;
}
void
GraphWalker::Walk(PtrInfo *s0)
{
nsDeque queue;
queue.Push(s0);
DoWalk(queue);
}
void
GraphWalker::WalkFromRoots(GCGraph& aGraph)
{
nsDeque queue;
NodePool::Enumerator etor(aGraph.mNodes);
for (PRUint32 i = 0; i < aGraph.mRootCount; ++i) {
queue.Push(etor.GetNext());
}
DoWalk(queue);
}
void
GraphWalker::DoWalk(nsDeque &aQueue)
{
// Use a aQueue to match the breadth-first traversal used when we
// built the graph, for hopefully-better locality.
while (aQueue.GetSize() > 0) {
PtrInfo *pi = static_cast<PtrInfo*>(aQueue.PopFront());
if (this->ShouldVisitNode(pi)) {
this->VisitNode(pi);
for (EdgePool::Iterator child = pi->mFirstChild,
child_end = pi->mLastChild;
child != child_end; ++child) {
aQueue.Push(*child);
}
}
};
#ifdef DEBUG_CC
sCollector->mStats.mWalkedGraph++;
#endif
}
////////////////////////////////////////////////////////////////////////
// Bacon & Rajan's |MarkRoots| routine.
////////////////////////////////////////////////////////////////////////
struct PtrToNodeEntry : public PLDHashEntryHdr
{
// The key is mNode->mPointer
PtrInfo *mNode;
};
PR_STATIC_CALLBACK(PRBool)
PtrToNodeMatchEntry(PLDHashTable *table,
const PLDHashEntryHdr *entry,
const void *key)
{
const PtrToNodeEntry *n = static_cast<const PtrToNodeEntry*>(entry);
return n->mNode->mPointer == key;
}
static PLDHashTableOps PtrNodeOps = {
PL_DHashAllocTable,
PL_DHashFreeTable,
PL_DHashVoidPtrKeyStub,
PtrToNodeMatchEntry,
PL_DHashMoveEntryStub,
PL_DHashClearEntryStub,
PL_DHashFinalizeStub,
nsnull
};
class GCGraphBuilder : private nsCycleCollectionTraversalCallback
{
private:
NodePool::Builder mNodeBuilder;
EdgePool::Builder mEdgeBuilder;
PLDHashTable mPtrToNodeMap;
PtrInfo *mCurrPi;
nsCycleCollectionLanguageRuntime **mRuntimes; // weak, from nsCycleCollector
public:
GCGraphBuilder(GCGraph &aGraph,
nsCycleCollectionLanguageRuntime **aRuntimes);
~GCGraphBuilder();
PRUint32 Count() const { return mPtrToNodeMap.entryCount; }
#ifdef DEBUG_CC
PtrInfo* AddNode(void *s, nsCycleCollectionParticipant *aParticipant,
PRUint32 aLangID);
#else
PtrInfo* AddNode(void *s, nsCycleCollectionParticipant *aParticipant);
PtrInfo* AddNode(void *s, nsCycleCollectionParticipant *aParticipant,
PRUint32 aLangID)
{
return AddNode(s, aParticipant);
}
#endif
void Traverse(PtrInfo* aPtrInfo);
private:
// nsCycleCollectionTraversalCallback methods.
#ifdef DEBUG_CC
NS_IMETHOD_(void) DescribeNode(nsrefcnt refCount, size_t objSz, const char *objName);
#else
NS_IMETHOD_(void) DescribeNode(nsrefcnt refCount);
#endif
NS_IMETHOD_(void) NoteXPCOMChild(nsISupports *child);
NS_IMETHOD_(void) NoteNativeChild(void *child,
nsCycleCollectionParticipant *participant);
NS_IMETHOD_(void) NoteScriptChild(PRUint32 langID, void *child);
};
GCGraphBuilder::GCGraphBuilder(GCGraph &aGraph,
nsCycleCollectionLanguageRuntime **aRuntimes)
: mNodeBuilder(aGraph.mNodes),
mEdgeBuilder(aGraph.mEdges),
mRuntimes(aRuntimes)
{
if (!PL_DHashTableInit(&mPtrToNodeMap, &PtrNodeOps, nsnull,
sizeof(PtrToNodeEntry), 32768))
mPtrToNodeMap.ops = nsnull;
}
GCGraphBuilder::~GCGraphBuilder()
{
if (mPtrToNodeMap.ops)
PL_DHashTableFinish(&mPtrToNodeMap);
}
PtrInfo*
GCGraphBuilder::AddNode(void *s, nsCycleCollectionParticipant *aParticipant
IF_DEBUG_CC_PARAM(PRUint32 aLangID)
)
{
PtrToNodeEntry *e = static_cast<PtrToNodeEntry*>(PL_DHashTableOperate(&mPtrToNodeMap, s, PL_DHASH_ADD));
PtrInfo *result;
if (!e->mNode) {
// New entry.
result = mNodeBuilder.Add(s, aParticipant
IF_DEBUG_CC_PARAM(aLangID)
);
if (!result) {
PL_DHashTableRawRemove(&mPtrToNodeMap, e);
return nsnull;
}
e->mNode = result;
} else {
result = e->mNode;
NS_ASSERTION(result->mParticipant == aParticipant,
"nsCycleCollectionParticipant shouldn't change!");
}
return result;
}
void
GCGraphBuilder::Traverse(PtrInfo* aPtrInfo)
{
mCurrPi = aPtrInfo;
#ifdef DEBUG_CC
if (!mCurrPi->mParticipant) {
Fault("unknown pointer during walk", aPtrInfo);
return;
}
#endif
mCurrPi->mFirstChild = mEdgeBuilder.Mark();
nsresult rv = aPtrInfo->mParticipant->Traverse(aPtrInfo->mPointer, *this);
if (NS_FAILED(rv)) {
Fault("script pointer traversal failed", aPtrInfo);
}
mCurrPi->mLastChild = mEdgeBuilder.Mark();
}
NS_IMETHODIMP_(void)
#ifdef DEBUG_CC
GCGraphBuilder::DescribeNode(nsrefcnt refCount, size_t objSz, const char *objName)
#else
GCGraphBuilder::DescribeNode(nsrefcnt refCount)
#endif
{
#ifdef DEBUG_CC
mCurrPi->mBytes = objSz;
mCurrPi->mName = PL_strdup(objName);
#endif
if (refCount == 0)
Fault("zero refcount", mCurrPi);
mCurrPi->mRefCount = refCount;
#ifdef DEBUG_CC
sCollector->mStats.mVisitedNode++;
#endif
}
NS_IMETHODIMP_(void)
GCGraphBuilder::NoteXPCOMChild(nsISupports *child)
{
if (!child || !(child = canonicalize(child)))
return;
#ifdef DEBUG_CC
if (nsCycleCollector_shouldSuppress(child))
return;
#endif
nsXPCOMCycleCollectionParticipant *cp;
ToParticipant(child, &cp);
if (cp) {
PtrInfo *childPi = AddNode(child, cp, nsIProgrammingLanguage::CPLUSPLUS);
if (!childPi)
return;
mEdgeBuilder.Add(childPi);
++childPi->mInternalRefs;
}
}
NS_IMETHODIMP_(void)
GCGraphBuilder::NoteNativeChild(void *child,
nsCycleCollectionParticipant *participant)
{
if (!child)
return;
NS_ASSERTION(participant, "Need a nsCycleCollectionParticipant!");
PtrInfo *childPi = AddNode(child, participant, nsIProgrammingLanguage::CPLUSPLUS);
if (!childPi)
return;
mEdgeBuilder.Add(childPi);
++childPi->mInternalRefs;
}
NS_IMETHODIMP_(void)
GCGraphBuilder::NoteScriptChild(PRUint32 langID, void *child)
{
if (!child)
return;
if (langID > nsIProgrammingLanguage::MAX || !mRuntimes[langID]) {
Fault("traversing pointer for unregistered language", child);
return;
}
nsCycleCollectionParticipant *cp = mRuntimes[langID]->ToParticipant(child);
if (!cp)
return;
PtrInfo *childPi = AddNode(child, cp, langID);
if (!childPi)
return;
mEdgeBuilder.Add(childPi);
++childPi->mInternalRefs;
}
void
nsCycleCollector::SelectPurple()
{
mPurpleBuf.BumpGeneration();
mPurpleBuf.SelectAgedPointers(&mBuf);
}
void
nsCycleCollector::MarkRoots(GCGraph &graph)
{
if (mBuf.GetSize() == 0)
return;
GCGraphBuilder builder(graph, mRuntimes);
int i;
for (i = 0; i < mBuf.GetSize(); ++i) {
nsISupports *s = static_cast<nsISupports *>(mBuf.ObjectAt(i));
nsXPCOMCycleCollectionParticipant *cp;
ToParticipant(s, &cp);
if (cp) {
PtrInfo *pinfo = builder.AddNode(canonicalize(s), cp,
nsIProgrammingLanguage::CPLUSPLUS);
if (pinfo)
pinfo->mWasPurple = PR_TRUE;
}
}
graph.mRootCount = builder.Count();
// read the PtrInfo out of the graph that we are building
NodePool::Enumerator queue(graph.mNodes);
while (!queue.IsDone()) {
PtrInfo *pi = queue.GetNext();
builder.Traverse(pi);
}
}
////////////////////////////////////////////////////////////////////////
// Bacon & Rajan's |ScanRoots| routine.
////////////////////////////////////////////////////////////////////////
struct ScanBlackWalker : public GraphWalker
{
PRBool ShouldVisitNode(PtrInfo const *pi)
{
return pi->mColor != black;
}
void VisitNode(PtrInfo *pi)
{
pi->mColor = black;
#ifdef DEBUG_CC
sCollector->mStats.mSetColorBlack++;
#endif
}
};
struct scanWalker : public GraphWalker
{
PRBool ShouldVisitNode(PtrInfo const *pi)
{
return pi->mColor == grey;
}
void VisitNode(PtrInfo *pi)
{
if (pi->mColor != grey)
Fault("scanning non-grey node", pi);
if (pi->mInternalRefs > pi->mRefCount)
Fault("traversed refs exceed refcount", pi);
if (pi->mInternalRefs == pi->mRefCount) {
pi->mColor = white;
#ifdef DEBUG_CC
sCollector->mStats.mSetColorWhite++;
#endif
} else {
ScanBlackWalker().Walk(pi);
NS_ASSERTION(pi->mColor == black,
"Why didn't ScanBlackWalker make pi black?");
}
}
};
void
nsCycleCollector::ScanRoots(GCGraph &graph)
{
// On the assumption that most nodes will be black, it's
// probably faster to use a GraphWalker than a
// NodePool::Enumerator.
scanWalker().WalkFromRoots(graph);
#ifdef DEBUG_CC
// Sanity check: scan should have colored all grey nodes black or
// white. So we ensure we have no grey nodes at this point.
NodePool::Enumerator etor(graph.mNodes);
while (!etor.IsDone())
{
PtrInfo *pinfo = etor.GetNext();
if (pinfo->mColor == grey) {
Fault("valid grey node after scanning", pinfo);
}
}
#endif
}
////////////////////////////////////////////////////////////////////////
// Bacon & Rajan's |CollectWhite| routine, somewhat modified.
////////////////////////////////////////////////////////////////////////
PRBool
nsCycleCollector::CollectWhite(GCGraph &graph)
{
// Explanation of "somewhat modified": we have no way to collect the
// set of whites "all at once", we have to ask each of them to drop
// their outgoing links and assume this will cause the garbage cycle
// to *mostly* self-destruct (except for the reference we continue
// to hold).
//
// To do this "safely" we must make sure that the white nodes we're
// operating on are stable for the duration of our operation. So we
// make 3 sets of calls to language runtimes:
//
// - Root(whites), which should pin the whites in memory.
// - Unlink(whites), which drops outgoing links on each white.
// - Unroot(whites), which returns the whites to normal GC.
nsresult rv;
mBuf.Empty();
#if defined(DEBUG_CC) && !defined(__MINGW32__) && defined(WIN32)
struct _CrtMemState ms1, ms2;
_CrtMemCheckpoint(&ms1);
#endif
NodePool::Enumerator etor(graph.mNodes);
while (!etor.IsDone())
{
PtrInfo *pinfo = etor.GetNext();
void *p = pinfo->mPointer;
if (pinfo->mColor == white) {
mBuf.Push(pinfo);
if (pinfo->mWasPurple) {
nsISupports* s = static_cast<nsISupports*>(p);
PRBool forgetResult = Forget(s);
NS_ASSERTION(forgetResult, "Forget failed");
}
}
else if (pinfo->mWasPurple) {
nsISupports* s = static_cast<nsISupports*>(p);
nsXPCOMCycleCollectionParticipant* cp =
static_cast<nsXPCOMCycleCollectionParticipant*>
(pinfo->mParticipant);
#ifdef DEBUG
nsXPCOMCycleCollectionParticipant* checkcp;
CallQueryInterface(s, &checkcp);
NS_ASSERTION(checkcp == cp,
"QI should return the same participant!");
#endif
cp->UnmarkPurple(s);
PRBool forgetResult = Forget(s);
NS_ASSERTION(forgetResult, "Forget failed");
}
}
PRUint32 i, count = mBuf.GetSize();
for (i = 0; i < count; ++i) {
PtrInfo *pinfo = static_cast<PtrInfo*>(mBuf.ObjectAt(i));
rv = pinfo->mParticipant->Root(pinfo->mPointer);
if (NS_FAILED(rv))
Fault("Failed root call while unlinking", pinfo);
}
for (i = 0; i < count; ++i) {
PtrInfo *pinfo = static_cast<PtrInfo*>(mBuf.ObjectAt(i));
rv = pinfo->mParticipant->Unlink(pinfo->mPointer);
if (NS_FAILED(rv)) {
Fault("Failed unlink call while unlinking", pinfo);
#ifdef DEBUG_CC
mStats.mFailedUnlink++;
#endif
}
else {
#ifdef DEBUG_CC
++mStats.mCollectedNode;
#endif
}
}
for (i = 0; i < count; ++i) {
PtrInfo *pinfo = static_cast<PtrInfo*>(mBuf.ObjectAt(i));
rv = pinfo->mParticipant->Unroot(pinfo->mPointer);
if (NS_FAILED(rv))
Fault("Failed unroot call while unlinking", pinfo);
}
mBuf.Empty();
#if defined(DEBUG_CC) && !defined(__MINGW32__) && defined(WIN32)
_CrtMemCheckpoint(&ms2);
if (ms2.lTotalCount < ms1.lTotalCount)
mStats.mFreedBytes += (ms1.lTotalCount - ms2.lTotalCount);
#endif
return count > 0;
}
#ifdef DEBUG_CC
////////////////////////////////////////////////////////////////////////
// Memory-hooking stuff
// When debugging wild pointers, it sometimes helps to hook malloc and
// free. This stuff is disabled unless you set an environment variable.
////////////////////////////////////////////////////////////////////////
static PRBool hookedMalloc = PR_FALSE;
#ifdef __GLIBC__
#include <malloc.h>
static void* (*old_memalign_hook)(size_t, size_t, const void *);
static void* (*old_realloc_hook)(void *, size_t, const void *);
static void* (*old_malloc_hook)(size_t, const void *);
static void (*old_free_hook)(void *, const void *);
static void* my_memalign_hook(size_t, size_t, const void *);
static void* my_realloc_hook(void *, size_t, const void *);
static void* my_malloc_hook(size_t, const void *);
static void my_free_hook(void *, const void *);
static inline void
install_old_hooks()
{
__memalign_hook = old_memalign_hook;
__realloc_hook = old_realloc_hook;
__malloc_hook = old_malloc_hook;
__free_hook = old_free_hook;
}
static inline void
save_old_hooks()
{
// Glibc docs recommend re-saving old hooks on
// return from recursive calls. Strangely when
// we do this, we find ourselves in infinite
// recursion.
// old_memalign_hook = __memalign_hook;
// old_realloc_hook = __realloc_hook;
// old_malloc_hook = __malloc_hook;
// old_free_hook = __free_hook;
}
static inline void
install_new_hooks()
{
__memalign_hook = my_memalign_hook;
__realloc_hook = my_realloc_hook;
__malloc_hook = my_malloc_hook;
__free_hook = my_free_hook;
}
static void*
my_realloc_hook(void *ptr, size_t size, const void *caller)
{
void *result;
install_old_hooks();
result = realloc(ptr, size);
save_old_hooks();
if (sCollector) {
sCollector->Freed(ptr);
sCollector->Allocated(result, size);
}
install_new_hooks();
return result;
}
static void*
my_memalign_hook(size_t size, size_t alignment, const void *caller)
{
void *result;
install_old_hooks();
result = memalign(size, alignment);
save_old_hooks();
if (sCollector)
sCollector->Allocated(result, size);
install_new_hooks();
return result;
}
static void
my_free_hook (void *ptr, const void *caller)
{
install_old_hooks();
free(ptr);
save_old_hooks();
if (sCollector)
sCollector->Freed(ptr);
install_new_hooks();
}
static void*
my_malloc_hook (size_t size, const void *caller)
{
void *result;
install_old_hooks();
result = malloc (size);
save_old_hooks();
if (sCollector)
sCollector->Allocated(result, size);
install_new_hooks();
return result;
}
static void
InitMemHook(void)
{
if (!hookedMalloc) {
save_old_hooks();
install_new_hooks();
hookedMalloc = PR_TRUE;
}
}
#elif defined(WIN32)
#ifndef __MINGW32__
static int
AllocHook(int allocType, void *userData, size_t size, int
blockType, long requestNumber, const unsigned char *filename, int
lineNumber)
{
if (allocType == _HOOK_FREE)
sCollector->Freed(userData);
return 1;
}
static void InitMemHook(void)
{
if (!hookedMalloc) {
_CrtSetAllocHook (AllocHook);
hookedMalloc = PR_TRUE;
}
}
#endif // __MINGW32__
#elif 0 // defined(XP_MACOSX)
#include <malloc/malloc.h>
static void (*old_free)(struct _malloc_zone_t *zone, void *ptr);
static void
freehook(struct _malloc_zone_t *zone, void *ptr)
{
if (sCollector)
sCollector->Freed(ptr);
old_free(zone, ptr);
}
static void
InitMemHook(void)
{
if (!hookedMalloc) {
malloc_zone_t *default_zone = malloc_default_zone();
old_free = default_zone->free;
default_zone->free = freehook;
hookedMalloc = PR_TRUE;
}
}
#else
static void
InitMemHook(void)
{
}
#endif // GLIBC / WIN32 / OSX
#endif // DEBUG_CC
////////////////////////////////////////////////////////////////////////
// Collector implementation
////////////////////////////////////////////////////////////////////////
nsCycleCollector::nsCycleCollector() :
mCollectionInProgress(PR_FALSE),
mScanInProgress(PR_FALSE),
#ifdef DEBUG_CC
mPurpleBuf(mParams, mStats),
mPtrLog(nsnull)
#else
mPurpleBuf(mParams)
#endif
{
#ifdef DEBUG_CC
mExpectedGarbage.Init();
#endif
memset(mRuntimes, 0, sizeof(mRuntimes));
mRuntimes[nsIProgrammingLanguage::CPLUSPLUS] = &mXPCOMRuntime;
}
nsCycleCollector::~nsCycleCollector()
{
}
void
nsCycleCollector::RegisterRuntime(PRUint32 langID,
nsCycleCollectionLanguageRuntime *rt)
{
if (mParams.mDoNothing)
return;
if (langID > nsIProgrammingLanguage::MAX)
Fault("unknown language runtime in registration");
if (mRuntimes[langID])
Fault("multiple registrations of language runtime", rt);
mRuntimes[langID] = rt;
}
void
nsCycleCollector::ForgetRuntime(PRUint32 langID)
{
if (mParams.mDoNothing)
return;
if (langID > nsIProgrammingLanguage::MAX)
Fault("unknown language runtime in deregistration");
if (! mRuntimes[langID])
Fault("forgetting non-registered language runtime");
mRuntimes[langID] = nsnull;
}
#ifdef DEBUG_CC
static void
WriteGraph(FILE *stream, GCGraph &graph, const void *redPtr)
{
fprintf(stream,
"digraph collection {\n"
"rankdir=LR\n"
"node [fontname=fixed, fontsize=10, style=filled, shape=box]\n"
);
NodePool::Enumerator etor(graph.mNodes);
while (!etor.IsDone()) {
PtrInfo *pi = etor.GetNext();
const void *p = pi->mPointer;
fprintf(stream,
"n%p [label=\"%s\\n%p\\n%u/%u refs found\", "
"fillcolor=%s, fontcolor=%s]\n",
p,
pi->mName,
p,
pi->mInternalRefs, pi->mRefCount,
(redPtr && redPtr == p ? "red" : (pi->mColor == black ? "black" : "white")),
(pi->mColor == black ? "white" : "black"));
for (EdgePool::Iterator child = pi->mFirstChild,
child_end = pi->mLastChild;
child != child_end; ++child) {
fprintf(stream, "n%p -> n%p\n", p, (*child)->mPointer);
}
}
fprintf(stream, "\n}\n");
}
void
nsCycleCollector::MaybeDrawGraphs(GCGraph &graph)
{
if (mParams.mDrawGraphs) {
// We draw graphs only if there were any white nodes.
PRBool anyWhites = PR_FALSE;
NodePool::Enumerator fwetor(graph.mNodes);
while (!fwetor.IsDone())
{
PtrInfo *pinfo = fwetor.GetNext();
if (pinfo->mColor == white) {
anyWhites = PR_TRUE;
break;
}
}
if (anyWhites) {
// We can't just use _popen here because graphviz-for-windows
// doesn't set up its stdin stream properly, sigh.
FILE *stream;
#ifdef WIN32
stream = fopen("c:\\cycle-graph.dot", "w+");
#else
stream = popen("dotty -", "w");
#endif
WriteGraph(stream, graph, nsnull);
#ifdef WIN32
fclose(stream);
// Even dotty doesn't work terribly well on windows, since
// they execute lefty asynchronously. So we'll just run
// lefty ourselves.
_spawnlp(_P_WAIT,
"lefty",
"lefty",
"-e",
"\"load('dotty.lefty');"
"dotty.simple('c:\\cycle-graph.dot');\"",
NULL);
unlink("c:\\cycle-graph.dot");
#else
pclose(stream);
#endif
}
}
}
class Suppressor :
public nsCycleCollectionTraversalCallback
{
protected:
static char *sSuppressionList;
static PRBool sInitialized;
PRBool mSuppressThisNode;
public:
Suppressor()
{
}
PRBool shouldSuppress(nsISupports *s)
{
if (!sInitialized) {
sSuppressionList = PR_GetEnv("XPCOM_CC_SUPPRESS");
sInitialized = PR_TRUE;
}
if (sSuppressionList == nsnull) {
mSuppressThisNode = PR_FALSE;
} else {
nsresult rv;
nsXPCOMCycleCollectionParticipant *cp;
rv = CallQueryInterface(s, &cp);
if (NS_FAILED(rv)) {
Fault("checking suppression on wrong type of pointer", s);
return PR_TRUE;
}
cp->Traverse(s, *this);
}
return mSuppressThisNode;
}
NS_IMETHOD_(void) DescribeNode(nsrefcnt refCount, size_t objSz, const char *objName)
{
mSuppressThisNode = (PL_strstr(sSuppressionList, objName) != nsnull);
}
NS_IMETHOD_(void) NoteXPCOMChild(nsISupports *child) {}
NS_IMETHOD_(void) NoteScriptChild(PRUint32 langID, void *child) {}
NS_IMETHOD_(void) NoteNativeChild(void *child,
nsCycleCollectionParticipant *participant) {}
};
char *Suppressor::sSuppressionList = nsnull;
PRBool Suppressor::sInitialized = PR_FALSE;
static PRBool
nsCycleCollector_shouldSuppress(nsISupports *s)
{
Suppressor supp;
return supp.shouldSuppress(s);
}
#endif
#ifdef DEBUG
static PRBool
nsCycleCollector_isScanSafe(nsISupports *s)
{
if (!s)
return PR_FALSE;
nsXPCOMCycleCollectionParticipant *cp;
ToParticipant(s, &cp);
return cp != nsnull;
}
#endif
PRBool
nsCycleCollector::Suspect(nsISupports *n, PRBool current)
{
// Re-entering ::Suspect during collection used to be a fault, but
// we are canonicalizing nsISupports pointers using QI, so we will
// see some spurious refcount traffic here.
if (mScanInProgress)
return PR_FALSE;
NS_ASSERTION(nsCycleCollector_isScanSafe(n),
"suspected a non-scansafe pointer");
NS_ASSERTION(NS_IsMainThread(), "trying to suspect from non-main thread");
if (mParams.mDoNothing)
return PR_FALSE;
#ifdef DEBUG_CC
mStats.mSuspectNode++;
if (nsCycleCollector_shouldSuppress(n))
return PR_FALSE;
#ifndef __MINGW32__
if (mParams.mHookMalloc)
InitMemHook();
#endif
if (mParams.mLogPointers) {
if (!mPtrLog)
mPtrLog = fopen("pointer_log", "w");
fprintf(mPtrLog, "S %p\n", static_cast<void*>(n));
}
#endif
if (current)
mBuf.Push(n);
else
mPurpleBuf.Put(n);
return PR_TRUE;
}
PRBool
nsCycleCollector::Forget(nsISupports *n)
{
// Re-entering ::Forget during collection used to be a fault, but
// we are canonicalizing nsISupports pointers using QI, so we will
// see some spurious refcount traffic here.
if (mScanInProgress)
return PR_FALSE;
NS_ASSERTION(NS_IsMainThread(), "trying to forget from non-main thread");
if (mParams.mDoNothing)
return PR_TRUE; // it's as good as forgotten
#ifdef DEBUG_CC
mStats.mForgetNode++;
#ifndef __MINGW32__
if (mParams.mHookMalloc)
InitMemHook();
#endif
if (mParams.mLogPointers) {
if (!mPtrLog)
mPtrLog = fopen("pointer_log", "w");
fprintf(mPtrLog, "F %p\n", static_cast<void*>(n));
}
#endif
mPurpleBuf.Remove(n);
return PR_TRUE;
}
#ifdef DEBUG_CC
void
nsCycleCollector::Allocated(void *n, size_t sz)
{
}
void
nsCycleCollector::Freed(void *n)
{
mStats.mFreeCalls++;
if (!n) {
// Ignore null pointers coming through
return;
}
if (mPurpleBuf.Exists(n)) {
mStats.mForgetNode++;
mStats.mFreedWhilePurple++;
Fault("freed while purple", n);
mPurpleBuf.Remove(n);
if (mParams.mLogPointers) {
if (!mPtrLog)
mPtrLog = fopen("pointer_log", "w");
fprintf(mPtrLog, "R %p\n", n);
}
}
}
#endif
void
nsCycleCollector::Collect(PRUint32 aTryCollections)
{
#if defined(DEBUG_CC) && !defined(__MINGW32__)
if (!mParams.mDoNothing && mParams.mHookMalloc)
InitMemHook();
#endif
// This can legitimately happen in a few cases. See bug 383651.
if (mCollectionInProgress)
return;
#ifdef COLLECT_TIME_DEBUG
printf("cc: Starting nsCycleCollector::Collect(%d)\n", aTryCollections);
PRTime start = PR_Now(), now;
#endif
mCollectionInProgress = PR_TRUE;
nsCOMPtr<nsIObserverService> obs =
do_GetService("@mozilla.org/observer-service;1");
if (obs) {
obs->NotifyObservers(nsnull, "cycle-collector-begin", nsnull);
}
#ifdef DEBUG_CC
PRUint32 origTryCollections = aTryCollections;
#endif
while (aTryCollections > 0) {
// This triggers a JS GC. Our caller assumes we always trigger at
// least one JS GC -- they rely on this fact to avoid redundant JS
// GC calls -- so it's essential that we actually execute this
// step!
//
// It is also essential to empty mBuf here because starting up
// collection in language runtimes may force some "current" suspects
// into mBuf.
mBuf.Empty();
#ifdef COLLECT_TIME_DEBUG
now = PR_Now();
#endif
for (PRUint32 i = 0; i <= nsIProgrammingLanguage::MAX; ++i) {
if (mRuntimes[i])
mRuntimes[i]->BeginCycleCollection();
}
#ifdef COLLECT_TIME_DEBUG
printf("cc: mRuntimes[*]->BeginCycleCollection() took %lldms\n",
(PR_Now() - now) / PR_USEC_PER_MSEC);
#endif
if (mParams.mDoNothing) {
aTryCollections = 0;
} else {
#ifdef COLLECT_TIME_DEBUG
now = PR_Now();
#endif
#ifdef DEBUG_CC
PRUint32 purpleStart = mBuf.GetSize();
#endif
SelectPurple();
#ifdef DEBUG_CC
PRUint32 purpleEnd = mBuf.GetSize();
#endif
#ifdef COLLECT_TIME_DEBUG
printf("cc: SelectPurple() took %lldms\n",
(PR_Now() - now) / PR_USEC_PER_MSEC);
#endif
if (mBuf.GetSize() == 0) {
aTryCollections = 0;
} else {
mScanInProgress = PR_TRUE;
GCGraph graph;
// The main Bacon & Rajan collection algorithm.
#ifdef COLLECT_TIME_DEBUG
now = PR_Now();
#endif
MarkRoots(graph);
#ifdef COLLECT_TIME_DEBUG
{
PRTime then = PR_Now();
printf("cc: MarkRoots() took %lldms\n",
(then - now) / PR_USEC_PER_MSEC);
now = then;
}
#endif
ScanRoots(graph);
#ifdef COLLECT_TIME_DEBUG
printf("cc: ScanRoots() took %lldms\n",
(PR_Now() - now) / PR_USEC_PER_MSEC);
#endif
#ifdef DEBUG_CC
MaybeDrawGraphs(graph);
#endif
mScanInProgress = PR_FALSE;
#ifdef DEBUG_CC
if (aTryCollections != origTryCollections && purpleStart != purpleEnd) {
PRUint32 i = 0;
NodePool::Enumerator queue(graph.mNodes);
while (i++ < purpleStart) {
queue.GetNext();
}
while (i++ < purpleEnd) {
PtrInfo *pi = queue.GetNext();
if (pi->mColor == white) {
printf("nsCycleCollector: a later shutdown collection collected the additional\n"
" suspect %p %s\n"
" (which could be fixed by improving traversal)\n",
pi->mPointer, pi->mName);
}
}
}
#endif
#ifdef COLLECT_TIME_DEBUG
now = PR_Now();
#endif
PRBool collected = CollectWhite(graph);
#ifdef COLLECT_TIME_DEBUG
printf("cc: CollectWhite() took %lldms\n",
(PR_Now() - now) / PR_USEC_PER_MSEC);
#endif
// Some additional book-keeping.
--aTryCollections;
// Since runtimes may add wrappers to the purple buffer
// (which will mean we won't stop repeating due to the
// mBuf.GetSize() == 0 check above), we should stop
// repeating collections if we didn't collect anything
// this time.
if (!collected)
aTryCollections = 0;
}
#ifdef DEBUG_CC
mStats.mCollection++;
if (mParams.mReportStats)
mStats.Dump();
#endif
}
for (PRUint32 i = 0; i <= nsIProgrammingLanguage::MAX; ++i) {
if (mRuntimes[i])
mRuntimes[i]->FinishCycleCollection();
}
}
mCollectionInProgress = PR_FALSE;
#ifdef COLLECT_TIME_DEBUG
printf("cc: Collect() took %lldms\n",
(PR_Now() - start) / PR_USEC_PER_MSEC);
#endif
#ifdef DEBUG_CC
ExplainLiveExpectedGarbage();
#endif
}
void
nsCycleCollector::Shutdown()
{
// Here we want to run a final collection on everything we've seen
// buffered, irrespective of age; then permanently disable
// the collector because the program is shutting down.
mParams.mScanDelay = 0;
Collect(SHUTDOWN_COLLECTIONS(mParams));
#ifdef DEBUG_CC
SelectPurple();
if (mBuf.GetSize() != 0) {
printf("Might have been able to release more cycles if the cycle collector would "
"run once more at shutdown.\n");
}
#endif
mParams.mDoNothing = PR_TRUE;
}
#ifdef DEBUG_CC
PR_STATIC_CALLBACK(PLDHashOperator)
AddExpectedGarbage(nsVoidPtrHashKey *p, void *arg)
{
nsCycleCollector *c = static_cast<nsCycleCollector*>(arg);
c->mBuf.Push(const_cast<void*>(p->GetKey()));
return PL_DHASH_NEXT;
}
struct SetSCCWalker : public GraphWalker
{
SetSCCWalker(PRUint32 aIndex) : mIndex(aIndex) {}
PRBool ShouldVisitNode(PtrInfo const *pi) { return pi->mSCCIndex == 0; }
void VisitNode(PtrInfo *pi) { pi->mSCCIndex = mIndex; }
private:
PRUint32 mIndex;
};
struct SetNonRootGreyWalker : public GraphWalker
{
PRBool ShouldVisitNode(PtrInfo const *pi) { return pi->mColor == white; }
void VisitNode(PtrInfo *pi) { pi->mColor = grey; }
};
void
nsCycleCollector::ExplainLiveExpectedGarbage()
{
if (mScanInProgress || mCollectionInProgress)
Fault("can't explain expected garbage during collection itself");
if (mParams.mDoNothing) {
printf("nsCycleCollector: not explaining expected garbage since\n"
" cycle collection disabled\n");
return;
}
mBuf.Empty();
for (PRUint32 i = 0; i <= nsIProgrammingLanguage::MAX; ++i) {
if (mRuntimes[i]) {
mRuntimes[i]->BeginCycleCollection();
mRuntimes[i]->SuspectExtraPointers();
}
}
mCollectionInProgress = PR_TRUE;
mScanInProgress = PR_TRUE;
{
GCGraph graph;
// Instead of filling mBuf from the purple buffer, we fill it
// from the list of nodes we were expected to collect.
PRUint32 suspectCurrentCount = mBuf.GetSize();
mExpectedGarbage.EnumerateEntries(&AddExpectedGarbage, this);
MarkRoots(graph);
ScanRoots(graph);
mScanInProgress = PR_FALSE;
PRBool describeExtraRefcounts = PR_FALSE;
PRBool findCycleRoots = PR_FALSE;
{
NodePool::Enumerator queue(graph.mNodes);
PRUint32 i = 0;
while (!queue.IsDone()) {
PtrInfo *pi = queue.GetNext();
if (pi->mColor == white) {
findCycleRoots = PR_TRUE;
}
if (pi->mInternalRefs != pi->mRefCount && i >= suspectCurrentCount) {
describeExtraRefcounts = PR_TRUE;
}
++i;
}
}
// The describeExtraRefcounts check isn't much use now that
// we're traversing from suspectCurrent roots too. But it's
// just extra work, not extra output.
if (describeExtraRefcounts && CreateReversedEdges(graph)) {
// Note that the external references may have been external
// to a different node in the cycle collection that just
// happened, if that different node was purple and then
// black.
// Use mSCCIndex temporarily to track whether we've reached
// nodes in the breadth-first search.
const PRUint32 INDEX_UNREACHED = 0;
const PRUint32 INDEX_REACHED = 1;
NodePool::Enumerator etor_clear(graph.mNodes);
while (!etor_clear.IsDone()) {
PtrInfo *pi = etor_clear.GetNext();
pi->mSCCIndex = INDEX_UNREACHED;
}
nsDeque queue; // for breadth-first search
NodePool::Enumerator etor_roots(graph.mNodes);
for (PRUint32 i = 0; i < graph.mRootCount; ++i) {
PtrInfo *root_pi = etor_roots.GetNext();
if (i >= suspectCurrentCount) {
root_pi->mSCCIndex = INDEX_REACHED;
root_pi->mShortestPathToExpectedGarbage = root_pi;
queue.Push(root_pi);
}
}
while (queue.GetSize() > 0) {
PtrInfo *pi = (PtrInfo*)queue.PopFront();
for (ReversedEdge *e = pi->mReversedEdges; e; e = e->mNext) {
if (e->mTarget->mSCCIndex == INDEX_UNREACHED) {
e->mTarget->mSCCIndex = INDEX_REACHED;
PtrInfo *target = e->mTarget;
if (!target->mShortestPathToExpectedGarbage)
target->mShortestPathToExpectedGarbage = pi;
queue.Push(target);
}
}
if (pi->mInternalRefs != pi->mRefCount) {
printf("nsCycleCollector: %s %p was not collected due "
"to %d\n"
" external references (%d total - %d known)\n",
pi->mName, pi->mPointer,
pi->mRefCount - pi->mInternalRefs,
pi->mRefCount, pi->mInternalRefs);
printf(" An object expected to be garbage could be "
"reached from it by the path:\n");
for (PtrInfo *path = pi, *prev = nsnull; prev != path;
prev = path,
path = path->mShortestPathToExpectedGarbage)
printf(" %s %p\n", path->mName, path->mPointer);
printf(" The %d known references to it were from:\n",
pi->mInternalRefs);
for (ReversedEdge *e = pi->mReversedEdges;
e; e = e->mNext) {
printf(" %s %p\n",
e->mTarget->mName, e->mTarget->mPointer);
}
mRuntimes[pi->mLangID]->PrintAllReferencesTo(pi->mPointer);
}
}
DestroyReversedEdges(graph);
}
if (findCycleRoots) {
// NOTE: This code changes the white nodes that are not
// roots to gray.
// Put the nodes in post-order traversal order from a
// depth-first search.
nsDeque DFSPostOrder;
{
// Use mSCCIndex temporarily to track the DFS numbering:
const PRUint32 INDEX_UNREACHED = 0;
const PRUint32 INDEX_TRAVERSING = 1;
const PRUint32 INDEX_NUMBERED = 2;
NodePool::Enumerator etor_clear(graph.mNodes);
while (!etor_clear.IsDone()) {
PtrInfo *pi = etor_clear.GetNext();
pi->mSCCIndex = INDEX_UNREACHED;
}
nsDeque stack;
NodePool::Enumerator etor_roots(graph.mNodes);
for (PRUint32 i = 0; i < graph.mRootCount; ++i) {
PtrInfo *root_pi = etor_roots.GetNext();
stack.Push(root_pi);
}
while (stack.GetSize() > 0) {
PtrInfo *pi = (PtrInfo*)stack.Peek();
if (pi->mSCCIndex == INDEX_UNREACHED) {
pi->mSCCIndex = INDEX_TRAVERSING;
for (EdgePool::Iterator child = pi->mFirstChild,
child_end = pi->mLastChild;
child != child_end; ++child) {
stack.Push(*child);
}
} else {
stack.Pop();
// Somebody else might have numbered it already
// (since this is depth-first, not breadth-first).
// This happens if a node is pushed on the stack
// a second time while it is on the stack in
// UNREACHED state.
if (pi->mSCCIndex == INDEX_TRAVERSING) {
pi->mSCCIndex = INDEX_NUMBERED;
DFSPostOrder.Push(pi);
}
}
}
}
// Put the nodes into strongly-connected components.
{
NodePool::Enumerator etor_clear(graph.mNodes);
while (!etor_clear.IsDone()) {
PtrInfo *pi = etor_clear.GetNext();
pi->mSCCIndex = 0;
}
PRUint32 currentSCC = 1;
while (DFSPostOrder.GetSize() > 0) {
SetSCCWalker(currentSCC).Walk((PtrInfo*)DFSPostOrder.PopFront());
++currentSCC;
}
}
// Mark any white nodes reachable from other components as
// grey.
{
NodePool::Enumerator queue(graph.mNodes);
while (!queue.IsDone()) {
PtrInfo *pi = queue.GetNext();
if (pi->mColor != white)
continue;
for (EdgePool::Iterator child = pi->mFirstChild,
child_end = pi->mLastChild;
child != child_end; ++child) {
if ((*child)->mSCCIndex != pi->mSCCIndex) {
SetNonRootGreyWalker().Walk(*child);
}
}
}
}
{
NodePool::Enumerator queue(graph.mNodes);
while (!queue.IsDone()) {
PtrInfo *pi = queue.GetNext();
if (pi->mColor == white) {
printf("nsCycleCollector: %s %p in component %d\n"
" was not collected due to missing call to "
"suspect, failure to unlink,\n"
" or deficiency in traverse that causes "
"cycles referenced only from other\n"
" cycles to require multiple rounds of cycle "
"collection\n",
pi->mName, pi->mPointer, pi->mSCCIndex);
}
}
}
}
}
mCollectionInProgress = PR_FALSE;
for (PRUint32 i = 0; i <= nsIProgrammingLanguage::MAX; ++i) {
if (mRuntimes[i])
mRuntimes[i]->FinishCycleCollection();
}
}
PRBool
nsCycleCollector::CreateReversedEdges(GCGraph &graph)
{
// Count the edges in the graph.
PRUint32 edgeCount = 0;
NodePool::Enumerator countQueue(graph.mNodes);
while (!countQueue.IsDone()) {
PtrInfo *pi = countQueue.GetNext();
for (EdgePool::Iterator e = pi->mFirstChild, e_end = pi->mLastChild;
e != e_end; ++e, ++edgeCount) {
}
}
// Allocate a pool to hold all of the edges.
graph.mReversedEdges = new ReversedEdge[edgeCount];
if (graph.mReversedEdges == nsnull) {
NS_NOTREACHED("allocation failure creating reversed edges");
return PR_FALSE;
}
// Fill in the reversed edges by scanning all forward edges.
ReversedEdge *current = graph.mReversedEdges;
NodePool::Enumerator buildQueue(graph.mNodes);
while (!buildQueue.IsDone()) {
PtrInfo *pi = buildQueue.GetNext();
for (EdgePool::Iterator e = pi->mFirstChild, e_end = pi->mLastChild;
e != e_end; ++e) {
current->mTarget = pi;
current->mNext = (*e)->mReversedEdges;
(*e)->mReversedEdges = current;
++current;
}
}
NS_ASSERTION(current - graph.mReversedEdges == edgeCount, "misallocation");
return PR_TRUE;
}
void
nsCycleCollector::DestroyReversedEdges(GCGraph &graph)
{
NodePool::Enumerator queue(graph.mNodes);
while (!queue.IsDone()) {
PtrInfo *pi = queue.GetNext();
pi->mReversedEdges = nsnull;
}
delete graph.mReversedEdges;
graph.mReversedEdges = nsnull;
}
void
nsCycleCollector::ShouldBeFreed(nsISupports *n)
{
mExpectedGarbage.PutEntry(n);
}
void
nsCycleCollector::WasFreed(nsISupports *n)
{
mExpectedGarbage.RemoveEntry(n);
}
#endif
////////////////////////////////////////////////////////////////////////
// Module public API (exported in nsCycleCollector.h)
// Just functions that redirect into the singleton, once it's built.
////////////////////////////////////////////////////////////////////////
void
nsCycleCollector_registerRuntime(PRUint32 langID,
nsCycleCollectionLanguageRuntime *rt)
{
if (sCollector)
sCollector->RegisterRuntime(langID, rt);
}
void
nsCycleCollector_forgetRuntime(PRUint32 langID)
{
if (sCollector)
sCollector->ForgetRuntime(langID);
}
PRBool
NS_CycleCollectorSuspect(nsISupports *n)
{
if (sCollector)
return sCollector->Suspect(n);
return PR_FALSE;
}
void
nsCycleCollector_suspectCurrent(nsISupports *n)
{
if (sCollector) {
PRBool res = sCollector->Suspect(n, PR_TRUE);
NS_ASSERTION(res || sCollector->mParams.mDoNothing,
"suspectCurrent should not fail");
}
}
PRBool
NS_CycleCollectorForget(nsISupports *n)
{
if (sCollector)
return sCollector->Forget(n);
return PR_FALSE;
}
void
nsCycleCollector_collect()
{
if (sCollector)
sCollector->Collect();
}
nsresult
nsCycleCollector_startup()
{
NS_ASSERTION(!sCollector, "Forgot to call nsCycleCollector_shutdown?");
sCollector = new nsCycleCollector();
return sCollector ? NS_OK : NS_ERROR_OUT_OF_MEMORY;
}
void
nsCycleCollector_shutdown()
{
if (sCollector) {
sCollector->Shutdown();
delete sCollector;
sCollector = nsnull;
}
}
#ifdef DEBUG
void
nsCycleCollector_DEBUG_shouldBeFreed(nsISupports *n)
{
#ifdef DEBUG_CC
if (sCollector)
sCollector->ShouldBeFreed(n);
#endif
}
void
nsCycleCollector_DEBUG_wasFreed(nsISupports *n)
{
#ifdef DEBUG_CC
if (sCollector)
sCollector->WasFreed(n);
#endif
}
#endif
Ссылка в новой задаче Показать git blame Копировать постоянную ссылку