michael@0: /* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
michael@0: /* This Source Code Form is subject to the terms of the Mozilla Public
michael@0:  * License, v. 2.0. If a copy of the MPL was not distributed with this
michael@0:  * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
michael@0: 
michael@0: /*
michael@0:  * A class which represents a fragment of text (eg inside a text
michael@0:  * node); if only codepoints below 256 are used, the text is stored as
michael@0:  * a char*; otherwise the text is stored as a char16_t*
michael@0:  */
michael@0: 
michael@0: #include "nsTextFragment.h"
michael@0: #include "nsCRT.h"
michael@0: #include "nsReadableUtils.h"
michael@0: #include "nsMemory.h"
michael@0: #include "nsBidiUtils.h"
michael@0: #include "nsUnicharUtils.h"
michael@0: #include "nsUTF8Utils.h"
michael@0: #include "mozilla/MemoryReporting.h"
michael@0: #include "mozilla/SSE.h"
michael@0: #include "nsTextFragmentImpl.h"
michael@0: #include <algorithm>
michael@0: 
michael@0: #define TEXTFRAG_WHITE_AFTER_NEWLINE 50
michael@0: #define TEXTFRAG_MAX_NEWLINES 7
michael@0: 
michael@0: // Static buffer used for common fragments
michael@0: static char* sSpaceSharedString[TEXTFRAG_MAX_NEWLINES + 1];
michael@0: static char* sTabSharedString[TEXTFRAG_MAX_NEWLINES + 1];
michael@0: static char sSingleCharSharedString[256];
michael@0: 
michael@0: // static
michael@0: nsresult
michael@0: nsTextFragment::Init()
michael@0: {
michael@0:   // Create whitespace strings
michael@0:   uint32_t i;
michael@0:   for (i = 0; i <= TEXTFRAG_MAX_NEWLINES; ++i) {
michael@0:     sSpaceSharedString[i] = new char[1 + i + TEXTFRAG_WHITE_AFTER_NEWLINE];
michael@0:     sTabSharedString[i] = new char[1 + i + TEXTFRAG_WHITE_AFTER_NEWLINE];
michael@0:     NS_ENSURE_TRUE(sSpaceSharedString[i] && sTabSharedString[i],
michael@0:                    NS_ERROR_OUT_OF_MEMORY);
michael@0:     sSpaceSharedString[i][0] = ' ';
michael@0:     sTabSharedString[i][0] = ' ';
michael@0:     uint32_t j;
michael@0:     for (j = 1; j < 1 + i; ++j) {
michael@0:       sSpaceSharedString[i][j] = '\n';
michael@0:       sTabSharedString[i][j] = '\n';
michael@0:     }
michael@0:     for (; j < (1 + i + TEXTFRAG_WHITE_AFTER_NEWLINE); ++j) {
michael@0:       sSpaceSharedString[i][j] = ' ';
michael@0:       sTabSharedString[i][j] = '\t';
michael@0:     }
michael@0:   }
michael@0: 
michael@0:   // Create single-char strings
michael@0:   for (i = 0; i < 256; ++i) {
michael@0:     sSingleCharSharedString[i] = i;
michael@0:   }
michael@0: 
michael@0:   return NS_OK;
michael@0: }
michael@0: 
michael@0: // static
michael@0: void
michael@0: nsTextFragment::Shutdown()
michael@0: {
michael@0:   uint32_t  i;
michael@0:   for (i = 0; i <= TEXTFRAG_MAX_NEWLINES; ++i) {
michael@0:     delete [] sSpaceSharedString[i];
michael@0:     delete [] sTabSharedString[i];
michael@0:     sSpaceSharedString[i] = nullptr;
michael@0:     sTabSharedString[i] = nullptr;
michael@0:   }
michael@0: }
michael@0: 
michael@0: nsTextFragment::~nsTextFragment()
michael@0: {
michael@0:   ReleaseText();
michael@0:   MOZ_COUNT_DTOR(nsTextFragment);
michael@0: }
michael@0: 
michael@0: void
michael@0: nsTextFragment::ReleaseText()
michael@0: {
michael@0:   if (mState.mLength && m1b && mState.mInHeap) {
michael@0:     moz_free(m2b); // m1b == m2b as far as moz_free is concerned
michael@0:   }
michael@0: 
michael@0:   m1b = nullptr;
michael@0:   mState.mIsBidi = false;
michael@0: 
michael@0:   // Set mState.mIs2b, mState.mInHeap, and mState.mLength = 0 with mAllBits;
michael@0:   mAllBits = 0;
michael@0: }
michael@0: 
michael@0: nsTextFragment&
michael@0: nsTextFragment::operator=(const nsTextFragment& aOther)
michael@0: {
michael@0:   ReleaseText();
michael@0: 
michael@0:   if (aOther.mState.mLength) {
michael@0:     if (!aOther.mState.mInHeap) {
michael@0:       m1b = aOther.m1b; // This will work even if aOther is using m2b
michael@0:     }
michael@0:     else {
michael@0:       size_t m2bSize = aOther.mState.mLength *
michael@0:         (aOther.mState.mIs2b ? sizeof(char16_t) : sizeof(char));
michael@0: 
michael@0:       m2b = static_cast<char16_t*>(moz_malloc(m2bSize));
michael@0:       if (m2b) {
michael@0:         memcpy(m2b, aOther.m2b, m2bSize);
michael@0:       } else {
michael@0:         // allocate a buffer for a single REPLACEMENT CHARACTER
michael@0:         m2b = static_cast<char16_t*>(moz_xmalloc(sizeof(char16_t)));
michael@0:         m2b[0] = 0xFFFD; // REPLACEMENT CHARACTER
michael@0:         mState.mIs2b = true;
michael@0:         mState.mInHeap = true;
michael@0:         mState.mLength = 1;
michael@0:       }
michael@0:     }
michael@0: 
michael@0:     if (m1b) {
michael@0:       mAllBits = aOther.mAllBits;
michael@0:     }
michael@0:   }
michael@0: 
michael@0:   return *this;
michael@0: }
michael@0: 
michael@0: static inline int32_t
michael@0: FirstNon8BitUnvectorized(const char16_t *str, const char16_t *end)
michael@0: {
michael@0:   typedef Non8BitParameters<sizeof(size_t)> p;
michael@0:   const size_t mask = p::mask();
michael@0:   const uint32_t alignMask = p::alignMask();
michael@0:   const uint32_t numUnicharsPerWord = p::numUnicharsPerWord();
michael@0:   const int32_t len = end - str;
michael@0:   int32_t i = 0;
michael@0: 
michael@0:   // Align ourselves to a word boundary.
michael@0:   int32_t alignLen =
michael@0:     std::min(len, int32_t(((-NS_PTR_TO_INT32(str)) & alignMask) / sizeof(char16_t)));
michael@0:   for (; i < alignLen; i++) {
michael@0:     if (str[i] > 255)
michael@0:       return i;
michael@0:   }
michael@0: 
michael@0:   // Check one word at a time.
michael@0:   const int32_t wordWalkEnd = ((len - i) / numUnicharsPerWord) * numUnicharsPerWord;
michael@0:   for (; i < wordWalkEnd; i += numUnicharsPerWord) {
michael@0:     const size_t word = *reinterpret_cast<const size_t*>(str + i);
michael@0:     if (word & mask)
michael@0:       return i;
michael@0:   }
michael@0: 
michael@0:   // Take care of the remainder one character at a time.
michael@0:   for (; i < len; i++) {
michael@0:     if (str[i] > 255)
michael@0:       return i;
michael@0:   }
michael@0: 
michael@0:   return -1;
michael@0: }
michael@0: 
michael@0: #ifdef MOZILLA_MAY_SUPPORT_SSE2
michael@0: namespace mozilla {
michael@0:   namespace SSE2 {
michael@0:     int32_t FirstNon8Bit(const char16_t *str, const char16_t *end);
michael@0:   }
michael@0: }
michael@0: #endif
michael@0: 
michael@0: /*
michael@0:  * This function returns -1 if all characters in str are 8 bit characters.
michael@0:  * Otherwise, it returns a value less than or equal to the index of the first
michael@0:  * non-8bit character in str. For example, if first non-8bit character is at
michael@0:  * position 25, it may return 25, or for example 24, or 16. But it guarantees
michael@0:  * there is no non-8bit character before returned value.
michael@0:  */
michael@0: static inline int32_t
michael@0: FirstNon8Bit(const char16_t *str, const char16_t *end)
michael@0: {
michael@0: #ifdef MOZILLA_MAY_SUPPORT_SSE2
michael@0:   if (mozilla::supports_sse2()) {
michael@0:     return mozilla::SSE2::FirstNon8Bit(str, end);
michael@0:   }
michael@0: #endif
michael@0: 
michael@0:   return FirstNon8BitUnvectorized(str, end);
michael@0: }
michael@0: 
michael@0: bool
michael@0: nsTextFragment::SetTo(const char16_t* aBuffer, int32_t aLength, bool aUpdateBidi)
michael@0: {
michael@0:   ReleaseText();
michael@0: 
michael@0:   if (aLength == 0) {
michael@0:     return true;
michael@0:   }
michael@0:   
michael@0:   char16_t firstChar = *aBuffer;
michael@0:   if (aLength == 1 && firstChar < 256) {
michael@0:     m1b = sSingleCharSharedString + firstChar;
michael@0:     mState.mInHeap = false;
michael@0:     mState.mIs2b = false;
michael@0:     mState.mLength = 1;
michael@0: 
michael@0:     return true;
michael@0:   }
michael@0: 
michael@0:   const char16_t *ucp = aBuffer;
michael@0:   const char16_t *uend = aBuffer + aLength;
michael@0: 
michael@0:   // Check if we can use a shared string
michael@0:   if (aLength <= 1 + TEXTFRAG_WHITE_AFTER_NEWLINE + TEXTFRAG_MAX_NEWLINES &&
michael@0:      (firstChar == ' ' || firstChar == '\n' || firstChar == '\t')) {
michael@0:     if (firstChar == ' ') {
michael@0:       ++ucp;
michael@0:     }
michael@0: 
michael@0:     const char16_t* start = ucp;
michael@0:     while (ucp < uend && *ucp == '\n') {
michael@0:       ++ucp;
michael@0:     }
michael@0:     const char16_t* endNewLine = ucp;
michael@0: 
michael@0:     char16_t space = ucp < uend && *ucp == '\t' ? '\t' : ' ';
michael@0:     while (ucp < uend && *ucp == space) {
michael@0:       ++ucp;
michael@0:     }
michael@0: 
michael@0:     if (ucp == uend &&
michael@0:         endNewLine - start <= TEXTFRAG_MAX_NEWLINES &&
michael@0:         ucp - endNewLine <= TEXTFRAG_WHITE_AFTER_NEWLINE) {
michael@0:       char** strings = space == ' ' ? sSpaceSharedString : sTabSharedString;
michael@0:       m1b = strings[endNewLine - start];
michael@0: 
michael@0:       // If we didn't find a space in the beginning, skip it now.
michael@0:       if (firstChar != ' ') {
michael@0:         ++m1b;
michael@0:       }
michael@0: 
michael@0:       mState.mInHeap = false;
michael@0:       mState.mIs2b = false;
michael@0:       mState.mLength = aLength;
michael@0: 
michael@0:       return true;        
michael@0:     }
michael@0:   }
michael@0: 
michael@0:   // See if we need to store the data in ucs2 or not
michael@0:   int32_t first16bit = FirstNon8Bit(ucp, uend);
michael@0: 
michael@0:   if (first16bit != -1) { // aBuffer contains no non-8bit character
michael@0:     // Use ucs2 storage because we have to
michael@0:     size_t m2bSize = aLength * sizeof(char16_t);
michael@0:     m2b = (char16_t *)moz_malloc(m2bSize);
michael@0:     if (!m2b) {
michael@0:       return false;
michael@0:     }
michael@0:     memcpy(m2b, aBuffer, m2bSize);
michael@0: 
michael@0:     mState.mIs2b = true;
michael@0:     if (aUpdateBidi) {
michael@0:       UpdateBidiFlag(aBuffer + first16bit, aLength - first16bit);
michael@0:     }
michael@0: 
michael@0:   } else {
michael@0:     // Use 1 byte storage because we can
michael@0:     char* buff = (char *)moz_malloc(aLength * sizeof(char));
michael@0:     if (!buff) {
michael@0:       return false;
michael@0:     }
michael@0: 
michael@0:     // Copy data
michael@0:     LossyConvertEncoding16to8 converter(buff);
michael@0:     copy_string(aBuffer, aBuffer+aLength, converter);
michael@0:     m1b = buff;
michael@0:     mState.mIs2b = false;
michael@0:   }
michael@0: 
michael@0:   // Setup our fields
michael@0:   mState.mInHeap = true;
michael@0:   mState.mLength = aLength;
michael@0: 
michael@0:   return true;
michael@0: }
michael@0: 
michael@0: void
michael@0: nsTextFragment::CopyTo(char16_t *aDest, int32_t aOffset, int32_t aCount)
michael@0: {
michael@0:   NS_ASSERTION(aOffset >= 0, "Bad offset passed to nsTextFragment::CopyTo()!");
michael@0:   NS_ASSERTION(aCount >= 0, "Bad count passed to nsTextFragment::CopyTo()!");
michael@0: 
michael@0:   if (aOffset < 0) {
michael@0:     aOffset = 0;
michael@0:   }
michael@0: 
michael@0:   if (uint32_t(aOffset + aCount) > GetLength()) {
michael@0:     aCount = mState.mLength - aOffset;
michael@0:   }
michael@0: 
michael@0:   if (aCount != 0) {
michael@0:     if (mState.mIs2b) {
michael@0:       memcpy(aDest, m2b + aOffset, sizeof(char16_t) * aCount);
michael@0:     } else {
michael@0:       const char *cp = m1b + aOffset;
michael@0:       const char *end = cp + aCount;
michael@0:       LossyConvertEncoding8to16 converter(aDest);
michael@0:       copy_string(cp, end, converter);
michael@0:     }
michael@0:   }
michael@0: }
michael@0: 
michael@0: bool
michael@0: nsTextFragment::Append(const char16_t* aBuffer, uint32_t aLength, bool aUpdateBidi)
michael@0: {
michael@0:   // This is a common case because some callsites create a textnode
michael@0:   // with a value by creating the node and then calling AppendData.
michael@0:   if (mState.mLength == 0) {
michael@0:     return SetTo(aBuffer, aLength, aUpdateBidi);
michael@0:   }
michael@0: 
michael@0:   // Should we optimize for aData.Length() == 0?
michael@0: 
michael@0:   if (mState.mIs2b) {
michael@0:     // Already a 2-byte string so the result will be too
michael@0:     char16_t* buff = (char16_t*)moz_realloc(m2b, (mState.mLength + aLength) * sizeof(char16_t));
michael@0:     if (!buff) {
michael@0:       return false;
michael@0:     }
michael@0: 
michael@0:     memcpy(buff + mState.mLength, aBuffer, aLength * sizeof(char16_t));
michael@0:     mState.mLength += aLength;
michael@0:     m2b = buff;
michael@0: 
michael@0:     if (aUpdateBidi) {
michael@0:       UpdateBidiFlag(aBuffer, aLength);
michael@0:     }
michael@0: 
michael@0:     return true;
michael@0:   }
michael@0: 
michael@0:   // Current string is a 1-byte string, check if the new data fits in one byte too.
michael@0:   int32_t first16bit = FirstNon8Bit(aBuffer, aBuffer + aLength);
michael@0: 
michael@0:   if (first16bit != -1) { // aBuffer contains no non-8bit character
michael@0:     // The old data was 1-byte, but the new is not so we have to expand it
michael@0:     // all to 2-byte
michael@0:     char16_t* buff = (char16_t*)moz_malloc((mState.mLength + aLength) *
michael@0:                                                   sizeof(char16_t));
michael@0:     if (!buff) {
michael@0:       return false;
michael@0:     }
michael@0: 
michael@0:     // Copy data into buff
michael@0:     LossyConvertEncoding8to16 converter(buff);
michael@0:     copy_string(m1b, m1b+mState.mLength, converter);
michael@0: 
michael@0:     memcpy(buff + mState.mLength, aBuffer, aLength * sizeof(char16_t));
michael@0:     mState.mLength += aLength;
michael@0:     mState.mIs2b = true;
michael@0: 
michael@0:     if (mState.mInHeap) {
michael@0:       moz_free(m2b);
michael@0:     }
michael@0:     m2b = buff;
michael@0: 
michael@0:     mState.mInHeap = true;
michael@0: 
michael@0:     if (aUpdateBidi) {
michael@0:       UpdateBidiFlag(aBuffer + first16bit, aLength - first16bit);
michael@0:     }
michael@0: 
michael@0:     return true;
michael@0:   }
michael@0: 
michael@0:   // The new and the old data is all 1-byte
michael@0:   char* buff;
michael@0:   if (mState.mInHeap) {
michael@0:     buff = (char*)moz_realloc(const_cast<char*>(m1b),
michael@0:                                     (mState.mLength + aLength) * sizeof(char));
michael@0:     if (!buff) {
michael@0:       return false;
michael@0:     }
michael@0:   }
michael@0:   else {
michael@0:     buff = (char*)moz_malloc((mState.mLength + aLength) * sizeof(char));
michael@0:     if (!buff) {
michael@0:       return false;
michael@0:     }
michael@0: 
michael@0:     memcpy(buff, m1b, mState.mLength);
michael@0:     mState.mInHeap = true;
michael@0:   }
michael@0: 
michael@0:   // Copy aBuffer into buff.
michael@0:   LossyConvertEncoding16to8 converter(buff + mState.mLength);
michael@0:   copy_string(aBuffer, aBuffer + aLength, converter);
michael@0: 
michael@0:   m1b = buff;
michael@0:   mState.mLength += aLength;
michael@0: 
michael@0:   return true;
michael@0: }
michael@0: 
michael@0: /* virtual */ size_t
michael@0: nsTextFragment::SizeOfExcludingThis(mozilla::MallocSizeOf aMallocSizeOf) const
michael@0: {
michael@0:   if (Is2b()) {
michael@0:     return aMallocSizeOf(m2b);
michael@0:   }
michael@0: 
michael@0:   if (mState.mInHeap) {
michael@0:     return aMallocSizeOf(m1b);
michael@0:   }
michael@0: 
michael@0:   return 0;
michael@0: }
michael@0: 
michael@0: // To save time we only do this when we really want to know, not during
michael@0: // every allocation
michael@0: void
michael@0: nsTextFragment::UpdateBidiFlag(const char16_t* aBuffer, uint32_t aLength)
michael@0: {
michael@0:   if (mState.mIs2b && !mState.mIsBidi) {
michael@0:     const char16_t* cp = aBuffer;
michael@0:     const char16_t* end = cp + aLength;
michael@0:     while (cp < end) {
michael@0:       char16_t ch1 = *cp++;
michael@0:       uint32_t utf32Char = ch1;
michael@0:       if (NS_IS_HIGH_SURROGATE(ch1) &&
michael@0:           cp < end &&
michael@0:           NS_IS_LOW_SURROGATE(*cp)) {
michael@0:         char16_t ch2 = *cp++;
michael@0:         utf32Char = SURROGATE_TO_UCS4(ch1, ch2);
michael@0:       }
michael@0:       if (UTF32_CHAR_IS_BIDI(utf32Char) || IsBidiControl(utf32Char)) {
michael@0:         mState.mIsBidi = true;
michael@0:         break;
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: }