/* -*- Mode: C++; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ /* 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/. */ #include "txNodeSet.h" #include "txLog.h" #include "txXPathTreeWalker.h" #include /** * Implementation of an XPath nodeset */ #ifdef NS_BUILD_REFCNT_LOGGING # define LOG_CHUNK_MOVE(_start, _new_start, _count) \ { \ txXPathNode* start = const_cast(_start); \ while (start < _start + _count) { \ NS_LogDtor(start, "txXPathNode", sizeof(*start)); \ ++start; \ } \ start = const_cast(_new_start); \ while (start < _new_start + _count) { \ NS_LogCtor(start, "txXPathNode", sizeof(*start)); \ ++start; \ } \ } #else # define LOG_CHUNK_MOVE(_start, _new_start, _count) #endif static const int32_t kTxNodeSetMinSize = 4; static const int32_t kTxNodeSetGrowFactor = 2; #define kForward 1 #define kReversed -1 txNodeSet::txNodeSet(txResultRecycler* aRecycler) : txAExprResult(aRecycler), mStart(nullptr), mEnd(nullptr), mStartBuffer(nullptr), mEndBuffer(nullptr), mDirection(kForward), mMarks(nullptr) {} txNodeSet::txNodeSet(const txXPathNode& aNode, txResultRecycler* aRecycler) : txAExprResult(aRecycler), mStart(nullptr), mEnd(nullptr), mStartBuffer(nullptr), mEndBuffer(nullptr), mDirection(kForward), mMarks(nullptr) { if (!ensureGrowSize(1)) { return; } new (mStart) txXPathNode(aNode); ++mEnd; } txNodeSet::txNodeSet(const txNodeSet& aSource, txResultRecycler* aRecycler) : txAExprResult(aRecycler), mStart(nullptr), mEnd(nullptr), mStartBuffer(nullptr), mEndBuffer(nullptr), mDirection(kForward), mMarks(nullptr) { append(aSource); } txNodeSet::~txNodeSet() { delete[] mMarks; if (mStartBuffer) { destroyElements(mStart, mEnd); free(mStartBuffer); } } nsresult txNodeSet::add(const txXPathNode& aNode) { NS_ASSERTION(mDirection == kForward, "only append(aNode) is supported on reversed nodesets"); if (isEmpty()) { return append(aNode); } bool dupe; txXPathNode* pos = findPosition(aNode, mStart, mEnd, dupe); if (dupe) { return NS_OK; } // save pos, ensureGrowSize messes with the pointers int32_t moveSize = mEnd - pos; int32_t offset = pos - mStart; if (!ensureGrowSize(1)) { return NS_ERROR_OUT_OF_MEMORY; } // set pos to where it was pos = mStart + offset; if (moveSize > 0) { LOG_CHUNK_MOVE(pos, pos + 1, moveSize); memmove(pos + 1, pos, moveSize * sizeof(txXPathNode)); } new (pos) txXPathNode(aNode); ++mEnd; return NS_OK; } nsresult txNodeSet::add(const txNodeSet& aNodes) { return add(aNodes, copyElements, nullptr); } nsresult txNodeSet::addAndTransfer(txNodeSet* aNodes) { // failure is out-of-memory, transfer didn't happen nsresult rv = add(*aNodes, transferElements, destroyElements); NS_ENSURE_SUCCESS(rv, rv); #ifdef TX_DONT_RECYCLE_BUFFER if (aNodes->mStartBuffer) { free(aNodes->mStartBuffer); aNodes->mStartBuffer = aNodes->mEndBuffer = nullptr; } #endif aNodes->mStart = aNodes->mEnd = aNodes->mStartBuffer; return NS_OK; } /** * add(aNodeSet, aTransferOp) * * The code is optimized to make a minimum number of calls to * Node::compareDocumentPosition. The idea is this: * We have the two nodesets (number indicate "document position") * * 1 3 7 <- source 1 * 2 3 6 8 9 <- source 2 * _ _ _ _ _ _ _ _ <- result * * * When merging these nodesets into the result, the nodes are transfered * in chunks to the end of the buffer so that each chunk does not contain * a node from the other nodeset, in document order. * * We select the last non-transfered node in the first nodeset and find * where in the other nodeset it would be inserted. In this case we would * take the 7 from the first nodeset and find the position between the * 6 and 8 in the second. We then take the nodes after the insert-position * and transfer them to the end of the resulting nodeset. Which in this case * means that we first transfered the 8 and 9 nodes, giving us the following: * * 1 3 7 <- source 1 * 2 3 6 <- source 2 * _ _ _ _ _ _ 8 9 <- result * * The corresponding procedure is done for the second nodeset, that is * the insertion position of the 6 in the first nodeset is found, which * is between the 3 and the 7. The 7 is memmoved (as it stays within * the same nodeset) to the result buffer. * * As the result buffer is filled from the end, it is safe to share the * buffer between this nodeset and the result. * * This is repeated until both of the nodesets are empty. * * If we find a duplicate node when searching for where insertposition we * check for sequences of duplicate nodes, which can be optimized. * */ nsresult txNodeSet::add(const txNodeSet& aNodes, transferOp aTransfer, destroyOp aDestroy) { NS_ASSERTION(mDirection == kForward, "only append(aNode) is supported on reversed nodesets"); if (aNodes.isEmpty()) { return NS_OK; } if (!ensureGrowSize(aNodes.size())) { return NS_ERROR_OUT_OF_MEMORY; } // This is probably a rather common case, so lets try to shortcut. if (mStart == mEnd || txXPathNodeUtils::comparePosition(mEnd[-1], *aNodes.mStart) < 0) { aTransfer(mEnd, aNodes.mStart, aNodes.mEnd); mEnd += aNodes.size(); return NS_OK; } // Last element in this nodeset txXPathNode* thisPos = mEnd; // Last element of the other nodeset txXPathNode* otherPos = aNodes.mEnd; // Pointer to the insertion point in this nodeset txXPathNode* insertPos = mEndBuffer; bool dupe; txXPathNode* pos; int32_t count; while (thisPos > mStart || otherPos > aNodes.mStart) { // Find where the last remaining node of this nodeset would // be inserted in the other nodeset. if (thisPos > mStart) { pos = findPosition(thisPos[-1], aNodes.mStart, otherPos, dupe); if (dupe) { const txXPathNode* deletePos = thisPos; --thisPos; // this is already added // check dupe sequence while (thisPos > mStart && pos > aNodes.mStart && thisPos[-1] == pos[-1]) { --thisPos; --pos; } if (aDestroy) { aDestroy(thisPos, deletePos); } } } else { pos = aNodes.mStart; } // Transfer the otherNodes after the insertion point to the result count = otherPos - pos; if (count > 0) { insertPos -= count; aTransfer(insertPos, pos, otherPos); otherPos -= count; } // Find where the last remaining node of the otherNodeset would // be inserted in this nodeset. if (otherPos > aNodes.mStart) { pos = findPosition(otherPos[-1], mStart, thisPos, dupe); if (dupe) { const txXPathNode* deletePos = otherPos; --otherPos; // this is already added // check dupe sequence while (otherPos > aNodes.mStart && pos > mStart && otherPos[-1] == pos[-1]) { --otherPos; --pos; } if (aDestroy) { aDestroy(otherPos, deletePos); } } } else { pos = mStart; } // Move the nodes from this nodeset after the insertion point // to the result count = thisPos - pos; if (count > 0) { insertPos -= count; LOG_CHUNK_MOVE(pos, insertPos, count); memmove(insertPos, pos, count * sizeof(txXPathNode)); thisPos -= count; } } mStart = insertPos; mEnd = mEndBuffer; return NS_OK; } /** * Append API * These functions should be used with care. * They are intended to be used when the caller assures that the resulting * nodeset remains in document order. * Abuse will break document order, and cause errors in the result. * These functions are significantly faster than the add API, as no * order info operations will be performed. */ nsresult txNodeSet::append(const txXPathNode& aNode) { if (!ensureGrowSize(1)) { return NS_ERROR_OUT_OF_MEMORY; } if (mDirection == kForward) { new (mEnd) txXPathNode(aNode); ++mEnd; return NS_OK; } new (--mStart) txXPathNode(aNode); return NS_OK; } nsresult txNodeSet::append(const txNodeSet& aNodes) { NS_ASSERTION(mDirection == kForward, "only append(aNode) is supported on reversed nodesets"); if (aNodes.isEmpty()) { return NS_OK; } int32_t appended = aNodes.size(); if (!ensureGrowSize(appended)) { return NS_ERROR_OUT_OF_MEMORY; } copyElements(mEnd, aNodes.mStart, aNodes.mEnd); mEnd += appended; return NS_OK; } nsresult txNodeSet::mark(int32_t aIndex) { NS_ASSERTION(aIndex >= 0 && mStart && mEnd - mStart > aIndex, "index out of bounds"); if (!mMarks) { int32_t length = size(); mMarks = new bool[length]; memset(mMarks, 0, length * sizeof(bool)); } if (mDirection == kForward) { mMarks[aIndex] = true; } else { mMarks[size() - aIndex - 1] = true; } return NS_OK; } nsresult txNodeSet::sweep() { if (!mMarks) { // sweep everything clear(); } int32_t chunk, pos = 0; int32_t length = size(); txXPathNode* insertion = mStartBuffer; while (pos < length) { while (pos < length && !mMarks[pos]) { // delete unmarked mStart[pos].~txXPathNode(); ++pos; } // find chunk to move chunk = 0; while (pos < length && mMarks[pos]) { ++pos; ++chunk; } // move chunk if (chunk > 0) { LOG_CHUNK_MOVE(mStart + pos - chunk, insertion, chunk); memmove(insertion, mStart + pos - chunk, chunk * sizeof(txXPathNode)); insertion += chunk; } } mStart = mStartBuffer; mEnd = insertion; delete[] mMarks; mMarks = nullptr; return NS_OK; } void txNodeSet::clear() { destroyElements(mStart, mEnd); #ifdef TX_DONT_RECYCLE_BUFFER if (mStartBuffer) { free(mStartBuffer); mStartBuffer = mEndBuffer = nullptr; } #endif mStart = mEnd = mStartBuffer; delete[] mMarks; mMarks = nullptr; mDirection = kForward; } int32_t txNodeSet::indexOf(const txXPathNode& aNode, uint32_t aStart) const { NS_ASSERTION(mDirection == kForward, "only append(aNode) is supported on reversed nodesets"); if (!mStart || mStart == mEnd) { return -1; } txXPathNode* pos = mStart + aStart; for (; pos < mEnd; ++pos) { if (*pos == aNode) { return pos - mStart; } } return -1; } const txXPathNode& txNodeSet::get(int32_t aIndex) const { if (mDirection == kForward) { return mStart[aIndex]; } return mEnd[-aIndex - 1]; } short txNodeSet::getResultType() { return txAExprResult::NODESET; } bool txNodeSet::booleanValue() { return !isEmpty(); } double txNodeSet::numberValue() { nsAutoString str; stringValue(str); return txDouble::toDouble(str); } void txNodeSet::stringValue(nsString& aStr) { NS_ASSERTION(mDirection == kForward, "only append(aNode) is supported on reversed nodesets"); if (isEmpty()) { return; } txXPathNodeUtils::appendNodeValue(get(0), aStr); } const nsString* txNodeSet::stringValuePointer() { return nullptr; } bool txNodeSet::ensureGrowSize(int32_t aSize) { // check if there is enough place in the buffer as is if (mDirection == kForward && aSize <= mEndBuffer - mEnd) { return true; } if (mDirection == kReversed && aSize <= mStart - mStartBuffer) { return true; } // check if we just have to align mStart to have enough space int32_t oldSize = mEnd - mStart; int32_t oldLength = mEndBuffer - mStartBuffer; int32_t ensureSize = oldSize + aSize; if (ensureSize <= oldLength) { // just move the buffer txXPathNode* dest = mStartBuffer; if (mDirection == kReversed) { dest = mEndBuffer - oldSize; } LOG_CHUNK_MOVE(mStart, dest, oldSize); memmove(dest, mStart, oldSize * sizeof(txXPathNode)); mStart = dest; mEnd = dest + oldSize; return true; } // This isn't 100% safe. But until someone manages to make a 1gig nodeset // it should be ok. int32_t newLength = std::max(oldLength, kTxNodeSetMinSize); while (newLength < ensureSize) { newLength *= kTxNodeSetGrowFactor; } txXPathNode* newArr = static_cast(moz_xmalloc(newLength * sizeof(txXPathNode))); txXPathNode* dest = newArr; if (mDirection == kReversed) { dest += newLength - oldSize; } if (oldSize > 0) { LOG_CHUNK_MOVE(mStart, dest, oldSize); memcpy(dest, mStart, oldSize * sizeof(txXPathNode)); } if (mStartBuffer) { #ifdef DEBUG memset(mStartBuffer, 0, (mEndBuffer - mStartBuffer) * sizeof(txXPathNode)); #endif free(mStartBuffer); } mStartBuffer = newArr; mEndBuffer = mStartBuffer + newLength; mStart = dest; mEnd = dest + oldSize; return true; } txXPathNode* txNodeSet::findPosition(const txXPathNode& aNode, txXPathNode* aFirst, txXPathNode* aLast, bool& aDupe) const { aDupe = false; if (aLast - aFirst <= 2) { // If we search 2 nodes or less there is no point in further divides txXPathNode* pos = aFirst; for (; pos < aLast; ++pos) { int cmp = txXPathNodeUtils::comparePosition(aNode, *pos); if (cmp < 0) { return pos; } if (cmp == 0) { aDupe = true; return pos; } } return pos; } // (cannot add two pointers) txXPathNode* midpos = aFirst + (aLast - aFirst) / 2; int cmp = txXPathNodeUtils::comparePosition(aNode, *midpos); if (cmp == 0) { aDupe = true; return midpos; } if (cmp > 0) { return findPosition(aNode, midpos + 1, aLast, aDupe); } // midpos excluded as end of range return findPosition(aNode, aFirst, midpos, aDupe); } /* static */ void txNodeSet::copyElements(txXPathNode* aDest, const txXPathNode* aStart, const txXPathNode* aEnd) { const txXPathNode* pos = aStart; while (pos < aEnd) { new (aDest) txXPathNode(*pos); ++aDest; ++pos; } } /* static */ void txNodeSet::transferElements(txXPathNode* aDest, const txXPathNode* aStart, const txXPathNode* aEnd) { LOG_CHUNK_MOVE(aStart, aDest, (aEnd - aStart)); memcpy(aDest, aStart, (aEnd - aStart) * sizeof(txXPathNode)); }