#include "JSString.h"
#include "JSGlobalObject.h"
+#include "JSGlobalObjectFunctions.h"
#include "JSObject.h"
-#include "Operations.h"
+#include "JSCInlines.h"
#include "StringObject.h"
#include "StringPrototype.h"
+#include "StrongInlines.h"
namespace JSC {
-static const unsigned resolveRopeForSubstringCutoff = 4;
+const ClassInfo JSString::s_info = { "string", 0, 0, 0, CREATE_METHOD_TABLE(JSString) };
-// Overview: this methods converts a JSString from holding a string in rope form
-// down to a simple UString representation. It does so by building up the string
-// backwards, since we want to avoid recursion, we expect that the tree structure
-// representing the rope is likely imbalanced with more nodes down the left side
-// (since appending to the string is likely more common) - and as such resolving
-// in this fashion should minimize work queue size. (If we built the queue forwards
-// we would likely have to place all of the constituent UStringImpls into the
-// Vector before performing any concatenation, but by working backwards we likely
-// only fill the queue with the number of substrings at any given level in a
-// rope-of-ropes.)
-void JSString::resolveRope(ExecState* exec) const
+void JSRopeString::RopeBuilder::expand()
{
- ASSERT(isRope());
+ ASSERT(m_index == JSRopeString::s_maxInternalRopeLength);
+ JSString* jsString = m_jsString;
+ RELEASE_ASSERT(jsString);
+ m_jsString = jsStringBuilder(&m_vm);
+ m_index = 0;
+ append(jsString);
+}
- // Allocate the buffer to hold the final string, position initially points to the end.
- UChar* buffer;
- if (PassRefPtr<UStringImpl> newImpl = UStringImpl::tryCreateUninitialized(m_length, buffer))
- m_value = newImpl;
+void JSString::destroy(JSCell* cell)
+{
+ JSString* thisObject = static_cast<JSString*>(cell);
+ thisObject->JSString::~JSString();
+}
+
+void JSString::dumpToStream(const JSCell* cell, PrintStream& out)
+{
+ const JSString* thisObject = jsCast<const JSString*>(cell);
+ out.printf("<%p, %s, [%u], ", thisObject, thisObject->className(), thisObject->length());
+ if (thisObject->isRope())
+ out.printf("[rope]");
else {
- for (unsigned i = 0; i < m_fiberCount; ++i) {
- RopeImpl::deref(m_other.m_fibers[i]);
- m_other.m_fibers[i] = 0;
+ WTF::StringImpl* ourImpl = thisObject->m_value.impl();
+ if (ourImpl->is8Bit())
+ out.printf("[8 %p]", ourImpl->characters8());
+ else
+ out.printf("[16 %p]", ourImpl->characters16());
+ }
+ out.printf(">");
+}
+
+void JSString::visitChildren(JSCell* cell, SlotVisitor& visitor)
+{
+ JSString* thisObject = jsCast<JSString*>(cell);
+ Base::visitChildren(thisObject, visitor);
+
+ if (thisObject->isRope())
+ static_cast<JSRopeString*>(thisObject)->visitFibers(visitor);
+ else {
+ StringImpl* impl = thisObject->m_value.impl();
+ ASSERT(impl);
+ visitor.reportExtraMemoryUsage(thisObject, impl->costDuringGC());
+ }
+}
+
+void JSRopeString::visitFibers(SlotVisitor& visitor)
+{
+ for (size_t i = 0; i < s_maxInternalRopeLength && m_fibers[i]; ++i)
+ visitor.append(&m_fibers[i]);
+}
+
+static const unsigned maxLengthForOnStackResolve = 2048;
+
+void JSRopeString::resolveRopeInternal8(LChar* buffer) const
+{
+ for (size_t i = 0; i < s_maxInternalRopeLength && m_fibers[i]; ++i) {
+ if (m_fibers[i]->isRope()) {
+ resolveRopeSlowCase8(buffer);
+ return;
}
- m_fiberCount = 0;
- ASSERT(!isRope());
- ASSERT(m_value == UString());
- if (exec)
- throwOutOfMemoryError(exec);
- return;
}
- UChar* position = buffer + m_length;
-
- // Start with the current RopeImpl.
- Vector<RopeImpl::Fiber, 32> workQueue;
- RopeImpl::Fiber currentFiber;
- for (unsigned i = 0; i < (m_fiberCount - 1); ++i)
- workQueue.append(m_other.m_fibers[i]);
- currentFiber = m_other.m_fibers[m_fiberCount - 1];
- while (true) {
- if (RopeImpl::isRope(currentFiber)) {
- RopeImpl* rope = static_cast<RopeImpl*>(currentFiber);
- // Copy the contents of the current rope into the workQueue, with the last item in 'currentFiber'
- // (we will be working backwards over the rope).
- unsigned fiberCountMinusOne = rope->fiberCount() - 1;
- for (unsigned i = 0; i < fiberCountMinusOne; ++i)
- workQueue.append(rope->fibers()[i]);
- currentFiber = rope->fibers()[fiberCountMinusOne];
- } else {
- UStringImpl* string = static_cast<UStringImpl*>(currentFiber);
- unsigned length = string->length();
- position -= length;
- UStringImpl::copyChars(position, string->characters(), length);
-
- // Was this the last item in the work queue?
- if (workQueue.isEmpty()) {
- // Create a string from the UChar buffer, clear the rope RefPtr.
- ASSERT(buffer == position);
- for (unsigned i = 0; i < m_fiberCount; ++i) {
- RopeImpl::deref(m_other.m_fibers[i]);
- m_other.m_fibers[i] = 0;
- }
- m_fiberCount = 0;
-
- ASSERT(!isRope());
- return;
- }
-
- // No! - set the next item up to process.
- currentFiber = workQueue.last();
- workQueue.removeLast();
+
+ LChar* position = buffer;
+ for (size_t i = 0; i < s_maxInternalRopeLength && m_fibers[i]; ++i) {
+ const StringImpl& fiberString = *m_fibers[i]->m_value.impl();
+ unsigned length = fiberString.length();
+ StringImpl::copyChars(position, fiberString.characters8(), length);
+ position += length;
+ }
+ ASSERT((buffer + m_length) == position);
+}
+
+void JSRopeString::resolveRopeInternal16(UChar* buffer) const
+{
+ for (size_t i = 0; i < s_maxInternalRopeLength && m_fibers[i]; ++i) {
+ if (m_fibers[i]->isRope()) {
+ resolveRopeSlowCase(buffer);
+ return;
}
}
+
+ UChar* position = buffer;
+ for (size_t i = 0; i < s_maxInternalRopeLength && m_fibers[i]; ++i) {
+ const StringImpl& fiberString = *m_fibers[i]->m_value.impl();
+ unsigned length = fiberString.length();
+ if (fiberString.is8Bit())
+ StringImpl::copyChars(position, fiberString.characters8(), length);
+ else
+ StringImpl::copyChars(position, fiberString.characters16(), length);
+ position += length;
+ }
+ ASSERT((buffer + m_length) == position);
}
-
-// This function construsts a substring out of a rope without flattening by reusing the existing fibers.
-// This can reduce memory usage substantially. Since traversing ropes is slow the function will revert
-// back to flattening if the rope turns out to be long.
-JSString* JSString::substringFromRope(ExecState* exec, unsigned substringStart, unsigned substringLength)
+
+void JSRopeString::resolveRopeToAtomicString(ExecState* exec) const
{
- ASSERT(isRope());
+ if (m_length > maxLengthForOnStackResolve) {
+ resolveRope(exec);
+ m_value = AtomicString(m_value);
+ return;
+ }
- JSGlobalData* globalData = &exec->globalData();
+ if (is8Bit()) {
+ LChar buffer[maxLengthForOnStackResolve];
+ resolveRopeInternal8(buffer);
+ m_value = AtomicString(buffer, m_length);
+ } else {
+ UChar buffer[maxLengthForOnStackResolve];
+ resolveRopeInternal16(buffer);
+ m_value = AtomicString(buffer, m_length);
+ }
- UString substringFibers[3];
-
- unsigned fiberCount = 0;
- unsigned substringFiberCount = 0;
- unsigned substringEnd = substringStart + substringLength;
- unsigned fiberEnd = 0;
-
- RopeIterator end;
- for (RopeIterator it(m_other.m_fibers, m_fiberCount); it != end; ++it) {
- ++fiberCount;
- UStringImpl* fiberString = *it;
- unsigned fiberStart = fiberEnd;
- fiberEnd = fiberStart + fiberString->length();
- if (fiberEnd <= substringStart)
- continue;
- unsigned copyStart = std::max(substringStart, fiberStart);
- unsigned copyEnd = std::min(substringEnd, fiberEnd);
- if (copyStart == fiberStart && copyEnd == fiberEnd)
- substringFibers[substringFiberCount++] = UString(fiberString);
- else
- substringFibers[substringFiberCount++] = UString(UStringImpl::create(fiberString, copyStart - fiberStart, copyEnd - copyStart));
- if (fiberEnd >= substringEnd)
- break;
- if (fiberCount > resolveRopeForSubstringCutoff || substringFiberCount >= 3) {
- // This turned out to be a really inefficient rope. Just flatten it.
- resolveRope(exec);
- return jsSubstring(&exec->globalData(), m_value, substringStart, substringLength);
+ clearFibers();
+
+ // If we resolved a string that didn't previously exist, notify the heap that we've grown.
+ if (m_value.impl()->hasOneRef())
+ Heap::heap(this)->reportExtraMemoryCost(m_value.impl()->cost());
+}
+
+void JSRopeString::clearFibers() const
+{
+ for (size_t i = 0; i < s_maxInternalRopeLength && m_fibers[i]; ++i)
+ m_fibers[i].clear();
+}
+
+AtomicStringImpl* JSRopeString::resolveRopeToExistingAtomicString(ExecState* exec) const
+{
+ if (m_length > maxLengthForOnStackResolve) {
+ resolveRope(exec);
+ if (AtomicStringImpl* existingAtomicString = AtomicString::find(m_value.impl())) {
+ m_value = *existingAtomicString;
+ clearFibers();
+ return existingAtomicString;
}
+ return nullptr;
}
- ASSERT(substringFiberCount && substringFiberCount <= 3);
- if (substringLength == 1) {
- ASSERT(substringFiberCount == 1);
- UChar c = substringFibers[0].data()[0];
- if (c <= 0xFF)
- return globalData->smallStrings.singleCharacterString(globalData, c);
+ if (is8Bit()) {
+ LChar buffer[maxLengthForOnStackResolve];
+ resolveRopeInternal8(buffer);
+ if (AtomicStringImpl* existingAtomicString = AtomicString::find(buffer, m_length)) {
+ m_value = *existingAtomicString;
+ clearFibers();
+ return existingAtomicString;
+ }
+ } else {
+ UChar buffer[maxLengthForOnStackResolve];
+ resolveRopeInternal16(buffer);
+ if (AtomicStringImpl* existingAtomicString = AtomicString::find(buffer, m_length)) {
+ m_value = *existingAtomicString;
+ clearFibers();
+ return existingAtomicString;
+ }
}
- if (substringFiberCount == 1)
- return new (globalData) JSString(globalData, substringFibers[0]);
- if (substringFiberCount == 2)
- return new (globalData) JSString(globalData, substringFibers[0], substringFibers[1]);
- return new (globalData) JSString(globalData, substringFibers[0], substringFibers[1], substringFibers[2]);
+
+ return nullptr;
}
-JSValue JSString::replaceCharacter(ExecState* exec, UChar character, const UString& replacement)
+void JSRopeString::resolveRope(ExecState* exec) const
{
- if (!isRope()) {
- unsigned matchPosition = m_value.find(character);
- if (matchPosition == UString::NotFound)
- return JSValue(this);
- return jsString(exec, m_value.substr(0, matchPosition), replacement, m_value.substr(matchPosition + 1));
+ ASSERT(isRope());
+
+ if (is8Bit()) {
+ LChar* buffer;
+ if (RefPtr<StringImpl> newImpl = StringImpl::tryCreateUninitialized(m_length, buffer)) {
+ Heap::heap(this)->reportExtraMemoryCost(newImpl->cost());
+ m_value = newImpl.release();
+ } else {
+ outOfMemory(exec);
+ return;
+ }
+ resolveRopeInternal8(buffer);
+ clearFibers();
+ ASSERT(!isRope());
+ return;
}
- RopeIterator end;
+ UChar* buffer;
+ if (RefPtr<StringImpl> newImpl = StringImpl::tryCreateUninitialized(m_length, buffer)) {
+ Heap::heap(this)->reportExtraMemoryCost(newImpl->cost());
+ m_value = newImpl.release();
+ } else {
+ outOfMemory(exec);
+ return;
+ }
+
+ resolveRopeInternal16(buffer);
+ clearFibers();
+ ASSERT(!isRope());
+}
+
+// Overview: These functions convert a JSString from holding a string in rope form
+// down to a simple String representation. It does so by building up the string
+// backwards, since we want to avoid recursion, we expect that the tree structure
+// representing the rope is likely imbalanced with more nodes down the left side
+// (since appending to the string is likely more common) - and as such resolving
+// in this fashion should minimize work queue size. (If we built the queue forwards
+// we would likely have to place all of the constituent StringImpls into the
+// Vector before performing any concatenation, but by working backwards we likely
+// only fill the queue with the number of substrings at any given level in a
+// rope-of-ropes.)
+void JSRopeString::resolveRopeSlowCase8(LChar* buffer) const
+{
+ LChar* position = buffer + m_length; // We will be working backwards over the rope.
+ Vector<JSString*, 32, UnsafeVectorOverflow> workQueue; // Putting strings into a Vector is only OK because there are no GC points in this method.
- // Count total fibers and find matching string.
- size_t fiberCount = 0;
- UStringImpl* matchString = 0;
- int matchPosition = -1;
- for (RopeIterator it(m_other.m_fibers, m_fiberCount); it != end; ++it) {
- ++fiberCount;
- if (matchString)
- continue;
+ for (size_t i = 0; i < s_maxInternalRopeLength && m_fibers[i]; ++i)
+ workQueue.append(m_fibers[i].get());
+
+ while (!workQueue.isEmpty()) {
+ JSString* currentFiber = workQueue.last();
+ workQueue.removeLast();
- UStringImpl* string = *it;
- matchPosition = string->find(character);
- if (matchPosition == -1)
+ if (currentFiber->isRope()) {
+ JSRopeString* currentFiberAsRope = static_cast<JSRopeString*>(currentFiber);
+ for (size_t i = 0; i < s_maxInternalRopeLength && currentFiberAsRope->m_fibers[i]; ++i)
+ workQueue.append(currentFiberAsRope->m_fibers[i].get());
continue;
- matchString = string;
+ }
+
+ StringImpl* string = static_cast<StringImpl*>(currentFiber->m_value.impl());
+ unsigned length = string->length();
+ position -= length;
+ StringImpl::copyChars(position, string->characters8(), length);
}
- if (!matchString)
- return this;
+ ASSERT(buffer == position);
+}
- RopeBuilder builder(replacement.size() ? fiberCount + 2 : fiberCount + 1);
- if (UNLIKELY(builder.isOutOfMemory()))
- return throwOutOfMemoryError(exec);
+void JSRopeString::resolveRopeSlowCase(UChar* buffer) const
+{
+ UChar* position = buffer + m_length; // We will be working backwards over the rope.
+ Vector<JSString*, 32, UnsafeVectorOverflow> workQueue; // These strings are kept alive by the parent rope, so using a Vector is OK.
- for (RopeIterator it(m_other.m_fibers, m_fiberCount); it != end; ++it) {
- UStringImpl* string = *it;
- if (string != matchString) {
- builder.append(UString(string));
+ for (size_t i = 0; i < s_maxInternalRopeLength && m_fibers[i]; ++i)
+ workQueue.append(m_fibers[i].get());
+
+ while (!workQueue.isEmpty()) {
+ JSString* currentFiber = workQueue.last();
+ workQueue.removeLast();
+
+ if (currentFiber->isRope()) {
+ JSRopeString* currentFiberAsRope = static_cast<JSRopeString*>(currentFiber);
+ for (size_t i = 0; i < s_maxInternalRopeLength && currentFiberAsRope->m_fibers[i]; ++i)
+ workQueue.append(currentFiberAsRope->m_fibers[i].get());
continue;
}
- builder.append(UString(string).substr(0, matchPosition));
- if (replacement.size())
- builder.append(replacement);
- builder.append(UString(string).substr(matchPosition + 1));
- matchString = 0;
+ StringImpl* string = static_cast<StringImpl*>(currentFiber->m_value.impl());
+ unsigned length = string->length();
+ position -= length;
+ if (string->is8Bit())
+ StringImpl::copyChars(position, string->characters8(), length);
+ else
+ StringImpl::copyChars(position, string->characters16(), length);
}
- JSGlobalData* globalData = &exec->globalData();
- return JSValue(new (globalData) JSString(globalData, builder.release()));
+ ASSERT(buffer == position);
}
-JSString* JSString::getIndexSlowCase(ExecState* exec, unsigned i)
+void JSRopeString::outOfMemory(ExecState* exec) const
+{
+ clearFibers();
+ ASSERT(isRope());
+ ASSERT(m_value.isNull());
+ if (exec)
+ throwOutOfMemoryError(exec);
+}
+
+JSString* JSRopeString::getIndexSlowCase(ExecState* exec, unsigned i)
{
ASSERT(isRope());
resolveRope(exec);
// Return a safe no-value result, this should never be used, since the excetion will be thrown.
if (exec->exception())
- return jsString(exec, "");
+ return jsEmptyString(exec);
ASSERT(!isRope());
- ASSERT(i < m_value.size());
+ RELEASE_ASSERT(i < m_value.length());
return jsSingleCharacterSubstring(exec, m_value, i);
}
return const_cast<JSString*>(this);
}
-bool JSString::getPrimitiveNumber(ExecState* exec, double& number, JSValue& result)
+bool JSString::getPrimitiveNumber(ExecState* exec, double& number, JSValue& result) const
{
result = this;
- number = value(exec).toDouble();
+ number = jsToNumber(value(exec));
return false;
}
-bool JSString::toBoolean(ExecState*) const
+bool JSString::toBoolean() const
{
return m_length;
}
double JSString::toNumber(ExecState* exec) const
{
- return value(exec).toDouble();
-}
-
-UString JSString::toString(ExecState* exec) const
-{
- return value(exec);
+ return jsToNumber(value(exec));
}
-inline StringObject* StringObject::create(ExecState* exec, JSString* string)
+inline StringObject* StringObject::create(VM& vm, JSGlobalObject* globalObject, JSString* string)
{
- return new (exec) StringObject(exec->lexicalGlobalObject()->stringObjectStructure(), string);
+ StringObject* object = new (NotNull, allocateCell<StringObject>(vm.heap)) StringObject(vm, globalObject->stringObjectStructure());
+ object->finishCreation(vm, string);
+ return object;
}
-JSObject* JSString::toObject(ExecState* exec) const
+JSObject* JSString::toObject(ExecState* exec, JSGlobalObject* globalObject) const
{
- return StringObject::create(exec, const_cast<JSString*>(this));
+ return StringObject::create(exec->vm(), globalObject, const_cast<JSString*>(this));
}
-JSObject* JSString::toThisObject(ExecState* exec) const
+JSValue JSString::toThis(JSCell* cell, ExecState* exec, ECMAMode ecmaMode)
{
- return StringObject::create(exec, const_cast<JSString*>(this));
+ if (ecmaMode == StrictMode)
+ return cell;
+ return StringObject::create(exec->vm(), exec->lexicalGlobalObject(), jsCast<JSString*>(cell));
}
-bool JSString::getOwnPropertySlot(ExecState* exec, const Identifier& propertyName, PropertySlot& slot)
-{
- // The semantics here are really getPropertySlot, not getOwnPropertySlot.
- // This function should only be called by JSValue::get.
- if (getStringPropertySlot(exec, propertyName, slot))
- return true;
- if (propertyName == exec->propertyNames().underscoreProto) {
- slot.setValue(exec->lexicalGlobalObject()->stringPrototype());
- return true;
- }
- slot.setBase(this);
- JSObject* object;
- for (JSValue prototype = exec->lexicalGlobalObject()->stringPrototype(); !prototype.isNull(); prototype = object->prototype()) {
- object = asObject(prototype);
- if (object->getOwnPropertySlot(exec, propertyName, slot))
- return true;
- }
- slot.setUndefined();
- return true;
-}
-
-bool JSString::getStringPropertyDescriptor(ExecState* exec, const Identifier& propertyName, PropertyDescriptor& descriptor)
+bool JSString::getStringPropertyDescriptor(ExecState* exec, PropertyName propertyName, PropertyDescriptor& descriptor)
{
if (propertyName == exec->propertyNames().length) {
- descriptor.setDescriptor(jsNumber(exec, m_length), DontEnum | DontDelete | ReadOnly);
+ descriptor.setDescriptor(jsNumber(m_length), DontEnum | DontDelete | ReadOnly);
return true;
}
- bool isStrictUInt32;
- unsigned i = propertyName.toStrictUInt32(&isStrictUInt32);
- if (isStrictUInt32 && i < m_length) {
+ unsigned i = propertyName.asIndex();
+ if (i < m_length) {
+ ASSERT(i != PropertyName::NotAnIndex); // No need for an explicit check, the above test would always fail!
descriptor.setDescriptor(getIndex(exec, i), DontDelete | ReadOnly);
return true;
}
return false;
}
-bool JSString::getOwnPropertyDescriptor(ExecState* exec, const Identifier& propertyName, PropertyDescriptor& descriptor)
-{
- if (getStringPropertyDescriptor(exec, propertyName, descriptor))
- return true;
- if (propertyName != exec->propertyNames().underscoreProto)
- return false;
- descriptor.setDescriptor(exec->lexicalGlobalObject()->stringPrototype(), DontEnum);
- return true;
-}
-
-bool JSString::getOwnPropertySlot(ExecState* exec, unsigned propertyName, PropertySlot& slot)
+JSString* jsStringWithCacheSlowCase(VM& vm, StringImpl& stringImpl)
{
- // The semantics here are really getPropertySlot, not getOwnPropertySlot.
- // This function should only be called by JSValue::get.
- if (getStringPropertySlot(exec, propertyName, slot))
- return true;
- return JSString::getOwnPropertySlot(exec, Identifier::from(exec, propertyName), slot);
+ auto addResult = vm.stringCache.add(&stringImpl, nullptr);
+ if (addResult.isNewEntry)
+ addResult.iterator->value = jsString(&vm, String(stringImpl));
+ vm.lastCachedString.set(vm, addResult.iterator->value.get());
+ return addResult.iterator->value.get();
}
} // namespace JSC
projects / apple / javascriptcore.git / blobdiff