Fix an entity bug for Naoki: Japanese entities weren't being correctly

pasted.  This also fixes the symptoms, but not the underlying cause,
of the html mail nbsp bug.  r=nhotta
This commit is contained in:
akkana%netscape.com 2000-03-28 00:10:44 +00:00
Родитель 9a6dbab6c9
Коммит 9e4f4dc539
8 изменённых файлов: 243 добавлений и 229 удалений

Просмотреть файл

@ -92,7 +92,11 @@ public:
OutputFormatFlowed = 64,
// Convert links, image src, and script src to absolute URLs when possible
OutputAbsoluteLinks = 128
OutputAbsoluteLinks = 128,
// Encode entities when outputting to a string.
// E.g. If set, we'll output   if clear, we'll output 0xa0.
OutputEncodeEntities = 256
};
static const nsIID& GetIID() { static nsIID iid = NS_IDOCUMENT_ENCODER_IID; return iid; }

Просмотреть файл

@ -37,7 +37,7 @@ function EditorGetHTML()
{
if (editorShell) {
dump("Getting HTML\n");
var outputHTML = editorShell.GetContentsAs("text/html", 0);
var outputHTML = editorShell.GetContentsAs("text/html", 256);
dump(outputHTML + "\n");
}
}

Просмотреть файл

@ -31,7 +31,6 @@
* model.
*/
#include "nsHTMLContentSinkStream.h"
#include "nsIParserNode.h"
#include <ctype.h>
@ -49,11 +48,8 @@
#include "nsNetUtil.h" // for NS_MakeAbsoluteURI
static NS_DEFINE_IID(kISupportsIID, NS_ISUPPORTS_IID);
static NS_DEFINE_IID(kIContentSinkIID, NS_ICONTENT_SINK_IID);
static NS_DEFINE_IID(kIHTMLContentSinkIID, NS_IHTML_CONTENT_SINK_IID);
static NS_DEFINE_IID(kIHTMLContentSinkStreamIID, NS_IHTMLCONTENTSINKSTREAM_IID);
static NS_DEFINE_CID(kSaveAsCharsetCID, NS_SAVEASCHARSET_CID);
static NS_DEFINE_CID(kEntityConverterCID, NS_ENTITYCONVERTER_CID);
static char* gHeaderComment = "<!-- This page was created by the Gecko output system. -->";
static char* gDocTypeHeader = "<!DOCTYPE HTML PUBLIC \"-//W3C//DTD HTML 3.2//EN\">";
@ -87,16 +83,16 @@ nsHTMLContentSinkStream::QueryInterface(const nsIID& aIID, void** aInstancePtr)
if (NULL == aInstancePtr) {
return NS_ERROR_NULL_POINTER;
}
if(aIID.Equals(kISupportsIID)) {
if (aIID.Equals(NS_GET_IID(nsISupports))) {
*aInstancePtr = (nsIContentSink*)(this);
}
else if(aIID.Equals(kIContentSinkIID)) {
else if (aIID.Equals(NS_GET_IID(nsIContentSink))) {
*aInstancePtr = (nsIContentSink*)(this);
}
else if(aIID.Equals(kIHTMLContentSinkIID)) {
else if (aIID.Equals(NS_GET_IID(nsIHTMLContentSink))) {
*aInstancePtr = (nsIHTMLContentSink*)(this);
}
else if(aIID.Equals(kIHTMLContentSinkStreamIID)) {
else if (aIID.Equals(NS_GET_IID(nsIHTMLContentSinkStream))) {
*aInstancePtr = (nsIHTMLContentSinkStream*)(this);
}
else {
@ -207,125 +203,134 @@ nsHTMLContentSinkStream::EndContext(PRInt32 aPosition)
* Initialize the Unicode encoder with our current mCharsetOverride.
*/
NS_IMETHODIMP
nsHTMLContentSinkStream::InitEncoder()
nsHTMLContentSinkStream::InitEncoders()
{
nsAutoString charsetName = mCharsetOverride;
nsresult res;
nsICharsetAlias* calias = nsnull;
res = nsServiceManager::GetService(kCharsetAliasCID,
kICharsetAliasIID,
(nsISupports**)&calias);
NS_ASSERTION(nsnull != calias, "cannot find charset alias");
if(NS_SUCCEEDED(res) && (nsnull != calias))
// Initialize an entity encoder if we're using the string interface:
if (mString && (mFlags & nsIDocumentEncoder::OutputEncodeEntities))
res = nsComponentManager::CreateInstance(kEntityConverterCID, NULL,
NS_GET_IID(nsIEntityConverter),
getter_AddRefs(mEntityConverter));
// Initialize a charset encoder if we're using the stream interface
if (mStream)
{
res = calias->GetPreferred(mCharsetOverride, charsetName);
nsServiceManager::ReleaseService(kCharsetAliasCID, calias);
}
if (NS_FAILED(res))
{
// failed - unknown alias , fallback to ISO-8859-1
charsetName = "ISO-8859-1";
nsAutoString charsetName = mCharsetOverride;
NS_WITH_SERVICE(nsICharsetAlias, calias, kCharsetAliasCID, &res);
if (NS_SUCCEEDED(res) && calias)
res = calias->GetPreferred(mCharsetOverride, charsetName);
if (NS_FAILED(res))
{
// failed - unknown alias , fallback to ISO-8859-1
charsetName = "ISO-8859-1";
}
res = nsComponentManager::CreateInstance(kSaveAsCharsetCID, NULL,
NS_GET_IID(nsISaveAsCharset),
getter_AddRefs(mCharsetEncoder));
if (NS_FAILED(res))
return res;
// SaveAsCharset requires a const char* in its first argument:
nsCAutoString charsetCString (charsetName);
// For ISO-8859-1 only, convert to entity first (always generate entites like &nbsp;).
res = mCharsetEncoder->Init(charsetCString,
charsetName.EqualsIgnoreCase("ISO-8859-1") ?
nsISaveAsCharset::attr_htmlTextDefault :
nsISaveAsCharset::attr_EntityAfterCharsetConv
+ nsISaveAsCharset::attr_FallbackDecimalNCR,
nsIEntityConverter::html40);
}
res = nsComponentManager::CreateInstance(kSaveAsCharsetCID, NULL,
NS_GET_IID(nsISaveAsCharset),
getter_AddRefs(mUnicodeEncoder));
if (NS_FAILED(res))
return res;
// SaveAsCharset requires a const char* in its first argument:
nsCAutoString charsetCString (charsetName);
// For ISO-8859-1 only, convert to entity first (always generate entites like &nbsp;).
res = mUnicodeEncoder->Init(charsetCString,
charsetName.EqualsIgnoreCase("ISO-8859-1") ?
nsISaveAsCharset::attr_htmlTextDefault :
nsISaveAsCharset::attr_EntityAfterCharsetConv + nsISaveAsCharset::attr_FallbackDecimalNCR,
nsIEntityConverter::html40);
return res;
}
/**
* Use mUnicodeEncoder to encode to the buffer;
* this also encodes entities, so it's useful even for the default charset.
*
* @param aSrc - the string to be encoded.
*/
void
nsHTMLContentSinkStream::EncodeToBuffer(const nsString& aSrc)
{
char *encodedBuffer = nsnull;
nsresult res;
// Initialize the encoder if we haven't already
if (!mUnicodeEncoder)
InitEncoder();
if (mUnicodeEncoder)
{
// Call the converter to convert to the target charset.
// Convert() takes a char* output param even though it's writing unicode.
res = mUnicodeEncoder->Convert(aSrc.GetUnicode(), &encodedBuffer);
if (!NS_SUCCEEDED(res))
{
#ifdef DEBUG_akkana
printf("Unicode convert didn't work!\n");
#endif
encodedBuffer = aSrc.ToNewUTF8String();
}
}
else
{
#ifdef DEBUG_akkana
printf("Unicode convert didn't work!\n");
#endif
encodedBuffer = aSrc.ToNewCString();
}
mBufferLength = 0;
if (encodedBuffer)
{
mBufferLength = nsCRT::strlen(encodedBuffer);
EnsureBufferSize(mBufferLength+1);
nsCRT::memcpy(mBuffer, encodedBuffer, mBufferLength+1);
}
}
void nsHTMLContentSinkStream::Write(const nsString& aString)
//
// Write(nsString) returns the number of characters written.
// If we do both string and stream output, stream chars will override string.
//
PRInt32 nsHTMLContentSinkStream::Write(const nsString& aString)
{
if (mBodyOnly && !mInBody)
return;
return 0;
// If an encoder is being used then convert first convert the input string
// Ideally, we should do this only for the stream case,
// and encode entities (using nsIEntityEncoder) but not charsets
// in the string case.
if (mUnicodeEncoder)
EncodeToBuffer(aString);
int charsWritten = 0;
// For the string case, we don't want to do charset conversion,
// but we still want to encode entities.
if (mString)
{
if (mUnicodeEncoder)
mString->Append(mBuffer, mBufferLength);
else
if (!mEntityConverter && (mFlags & nsIDocumentEncoder::OutputEncodeEntities))
InitEncoders();
if (mEntityConverter && (mFlags & nsIDocumentEncoder::OutputEncodeEntities))
{
nsresult res;
PRUnichar *encodedBuffer = nsnull;
res = mEntityConverter->ConvertToEntities(aString.GetUnicode(),
nsIEntityConverter::html40Latin1,
&encodedBuffer);
if (NS_SUCCEEDED(res) && encodedBuffer)
{
PRInt32 len = nsCRT::strlen(encodedBuffer);
mString->Append(encodedBuffer, len);
nsCRT::free(encodedBuffer);
charsWritten = len;
}
else {
charsWritten = aString.Length();
mString->Append(aString);
}
}
else {
charsWritten = aString.Length();
mString->Append(aString);
}
}
if (!mStream)
return;
return charsWritten;
// Now handle the stream case:
nsOutputStream out(mStream);
// Test again in case there was a problem initializing the unicode encoder:
if (mUnicodeEncoder)
// If an encoder is being used then convert first convert the input string
char *encodedBuffer = nsnull;
nsresult res;
// Initialize the encoder if we haven't already
if (!mCharsetEncoder)
InitEncoders();
if (mCharsetEncoder)
{
out.write(mBuffer, mBufferLength);
// Call the converter to convert to the target charset.
// Convert() takes a char* output param even though it's writing unicode.
res = mCharsetEncoder->Convert(aString.GetUnicode(), &encodedBuffer);
if (NS_SUCCEEDED(res) && encodedBuffer)
{
charsWritten = nsCRT::strlen(encodedBuffer);
out.write(encodedBuffer, charsWritten);
nsCRT::free(encodedBuffer);
}
// If it didn't work, just write the unicode
else
{
const PRUnichar* unicode = aString.GetUnicode();
charsWritten = aString.Length();
out.write(unicode, charsWritten);
}
}
// else just write the unicode
// If we couldn't get an encoder, just write the unicode
else
{
const PRUnichar* unicode = aString.GetUnicode();
out.write(unicode, aString.Length());
charsWritten = aString.Length();
out.write(unicode, charsWritten);
}
return charsWritten;
}
void nsHTMLContentSinkStream::Write(const char* aData)
@ -949,13 +954,11 @@ nsHTMLContentSinkStream::AddLeaf(const nsIParserNode& aNode){
}
else if (type == eHTMLTag_entity)
{
const nsString& entity = aNode.GetText();
EncodeToBuffer(entity);
Write('&');
Write(mBuffer);
const nsString& entity = aNode.GetText();
mColPos += Write(entity) + 1;
// Don't write the semicolon;
// rely on the DTD to include it if one is wanted.
mColPos += entity.Length() + 2;
}
else if (type == eHTMLTag_text)
{
@ -1016,7 +1019,7 @@ PRBool nsHTMLContentSinkStream::HasLongLines(const nsString& text)
{
PRInt32 eol = text.FindChar('\n', PR_FALSE, start);
if (eol < 0) eol = text.Length();
if (eol - start > longLineLen)
if ((PRInt32)(eol - start) > longLineLen)
return PR_TRUE;
start = eol+1;
}
@ -1165,7 +1168,7 @@ nsHTMLContentSinkStream::OpenContainer(const nsIParserNode& aNode)
const nsString& value=aNode.GetValueAt(i);
if (mCharsetOverride.IsEmpty())
mCharsetOverride.Assign(value);
InitEncoder();
InitEncoders();
}
else if (key.Equals("uri"))
{

Просмотреть файл

@ -48,7 +48,8 @@
#include "nsParserCIID.h"
#include "nsCOMPtr.h"
#include "nsHTMLTokens.h" // for eHTMLTags
#include "nsISaveAsCharset.h" // Some compilers can't do nsCOMPtr on a forward type
#include "nsISaveAsCharset.h"
#include "nsIEntityConverter.h"
#include "nsIURI.h"
#define NS_IHTMLCONTENTSINKSTREAM_IID \
@ -154,10 +155,9 @@ protected:
void AddIndent();
void EnsureBufferSize(PRInt32 aNewSize);
void EncodeToBuffer(const nsString& aString);
NS_IMETHOD InitEncoder();
NS_IMETHOD InitEncoders();
void Write(const nsString& aString);
PRInt32 Write(const nsString& aString); // returns # chars written
void Write(const char* aCharBuffer);
void Write(char aChar);
@ -170,7 +170,6 @@ protected:
protected:
nsIOutputStream* mStream;
nsString mStreamBuffer;
nsString* mString;
int mTabLevel;
@ -195,7 +194,8 @@ protected:
PRInt32 mMaxColumn;
nsCOMPtr<nsISaveAsCharset> mUnicodeEncoder;
nsCOMPtr<nsISaveAsCharset> mCharsetEncoder;
nsCOMPtr<nsIEntityConverter> mEntityConverter;
nsCAutoString mCharsetOverride;
};

Просмотреть файл

@ -92,7 +92,11 @@ public:
OutputFormatFlowed = 64,
// Convert links, image src, and script src to absolute URLs when possible
OutputAbsoluteLinks = 128
OutputAbsoluteLinks = 128,
// Encode entities when outputting to a string.
// E.g. If set, we'll output &nbsp; if clear, we'll output 0xa0.
OutputEncodeEntities = 256
};
static const nsIID& GetIID() { static nsIID iid = NS_IDOCUMENT_ENCODER_IID; return iid; }

Просмотреть файл

@ -516,8 +516,8 @@ void nsXIFConverter::WriteDebugFile() {
#endif
ofstream out(filename);
char* s = mBuffer.ToNewCString();
if(s) {
char* s = mBuffer.ToNewUTF8String();
if (s) {
out << s;
Recycle(s);
}

Просмотреть файл

@ -31,7 +31,6 @@
* model.
*/
#include "nsHTMLContentSinkStream.h"
#include "nsIParserNode.h"
#include <ctype.h>
@ -49,11 +48,8 @@
#include "nsNetUtil.h" // for NS_MakeAbsoluteURI
static NS_DEFINE_IID(kISupportsIID, NS_ISUPPORTS_IID);
static NS_DEFINE_IID(kIContentSinkIID, NS_ICONTENT_SINK_IID);
static NS_DEFINE_IID(kIHTMLContentSinkIID, NS_IHTML_CONTENT_SINK_IID);
static NS_DEFINE_IID(kIHTMLContentSinkStreamIID, NS_IHTMLCONTENTSINKSTREAM_IID);
static NS_DEFINE_CID(kSaveAsCharsetCID, NS_SAVEASCHARSET_CID);
static NS_DEFINE_CID(kEntityConverterCID, NS_ENTITYCONVERTER_CID);
static char* gHeaderComment = "<!-- This page was created by the Gecko output system. -->";
static char* gDocTypeHeader = "<!DOCTYPE HTML PUBLIC \"-//W3C//DTD HTML 3.2//EN\">";
@ -87,16 +83,16 @@ nsHTMLContentSinkStream::QueryInterface(const nsIID& aIID, void** aInstancePtr)
if (NULL == aInstancePtr) {
return NS_ERROR_NULL_POINTER;
}
if(aIID.Equals(kISupportsIID)) {
if (aIID.Equals(NS_GET_IID(nsISupports))) {
*aInstancePtr = (nsIContentSink*)(this);
}
else if(aIID.Equals(kIContentSinkIID)) {
else if (aIID.Equals(NS_GET_IID(nsIContentSink))) {
*aInstancePtr = (nsIContentSink*)(this);
}
else if(aIID.Equals(kIHTMLContentSinkIID)) {
else if (aIID.Equals(NS_GET_IID(nsIHTMLContentSink))) {
*aInstancePtr = (nsIHTMLContentSink*)(this);
}
else if(aIID.Equals(kIHTMLContentSinkStreamIID)) {
else if (aIID.Equals(NS_GET_IID(nsIHTMLContentSinkStream))) {
*aInstancePtr = (nsIHTMLContentSinkStream*)(this);
}
else {
@ -207,125 +203,134 @@ nsHTMLContentSinkStream::EndContext(PRInt32 aPosition)
* Initialize the Unicode encoder with our current mCharsetOverride.
*/
NS_IMETHODIMP
nsHTMLContentSinkStream::InitEncoder()
nsHTMLContentSinkStream::InitEncoders()
{
nsAutoString charsetName = mCharsetOverride;
nsresult res;
nsICharsetAlias* calias = nsnull;
res = nsServiceManager::GetService(kCharsetAliasCID,
kICharsetAliasIID,
(nsISupports**)&calias);
NS_ASSERTION(nsnull != calias, "cannot find charset alias");
if(NS_SUCCEEDED(res) && (nsnull != calias))
// Initialize an entity encoder if we're using the string interface:
if (mString && (mFlags & nsIDocumentEncoder::OutputEncodeEntities))
res = nsComponentManager::CreateInstance(kEntityConverterCID, NULL,
NS_GET_IID(nsIEntityConverter),
getter_AddRefs(mEntityConverter));
// Initialize a charset encoder if we're using the stream interface
if (mStream)
{
res = calias->GetPreferred(mCharsetOverride, charsetName);
nsServiceManager::ReleaseService(kCharsetAliasCID, calias);
}
if (NS_FAILED(res))
{
// failed - unknown alias , fallback to ISO-8859-1
charsetName = "ISO-8859-1";
nsAutoString charsetName = mCharsetOverride;
NS_WITH_SERVICE(nsICharsetAlias, calias, kCharsetAliasCID, &res);
if (NS_SUCCEEDED(res) && calias)
res = calias->GetPreferred(mCharsetOverride, charsetName);
if (NS_FAILED(res))
{
// failed - unknown alias , fallback to ISO-8859-1
charsetName = "ISO-8859-1";
}
res = nsComponentManager::CreateInstance(kSaveAsCharsetCID, NULL,
NS_GET_IID(nsISaveAsCharset),
getter_AddRefs(mCharsetEncoder));
if (NS_FAILED(res))
return res;
// SaveAsCharset requires a const char* in its first argument:
nsCAutoString charsetCString (charsetName);
// For ISO-8859-1 only, convert to entity first (always generate entites like &nbsp;).
res = mCharsetEncoder->Init(charsetCString,
charsetName.EqualsIgnoreCase("ISO-8859-1") ?
nsISaveAsCharset::attr_htmlTextDefault :
nsISaveAsCharset::attr_EntityAfterCharsetConv
+ nsISaveAsCharset::attr_FallbackDecimalNCR,
nsIEntityConverter::html40);
}
res = nsComponentManager::CreateInstance(kSaveAsCharsetCID, NULL,
NS_GET_IID(nsISaveAsCharset),
getter_AddRefs(mUnicodeEncoder));
if (NS_FAILED(res))
return res;
// SaveAsCharset requires a const char* in its first argument:
nsCAutoString charsetCString (charsetName);
// For ISO-8859-1 only, convert to entity first (always generate entites like &nbsp;).
res = mUnicodeEncoder->Init(charsetCString,
charsetName.EqualsIgnoreCase("ISO-8859-1") ?
nsISaveAsCharset::attr_htmlTextDefault :
nsISaveAsCharset::attr_EntityAfterCharsetConv + nsISaveAsCharset::attr_FallbackDecimalNCR,
nsIEntityConverter::html40);
return res;
}
/**
* Use mUnicodeEncoder to encode to the buffer;
* this also encodes entities, so it's useful even for the default charset.
*
* @param aSrc - the string to be encoded.
*/
void
nsHTMLContentSinkStream::EncodeToBuffer(const nsString& aSrc)
{
char *encodedBuffer = nsnull;
nsresult res;
// Initialize the encoder if we haven't already
if (!mUnicodeEncoder)
InitEncoder();
if (mUnicodeEncoder)
{
// Call the converter to convert to the target charset.
// Convert() takes a char* output param even though it's writing unicode.
res = mUnicodeEncoder->Convert(aSrc.GetUnicode(), &encodedBuffer);
if (!NS_SUCCEEDED(res))
{
#ifdef DEBUG_akkana
printf("Unicode convert didn't work!\n");
#endif
encodedBuffer = aSrc.ToNewUTF8String();
}
}
else
{
#ifdef DEBUG_akkana
printf("Unicode convert didn't work!\n");
#endif
encodedBuffer = aSrc.ToNewCString();
}
mBufferLength = 0;
if (encodedBuffer)
{
mBufferLength = nsCRT::strlen(encodedBuffer);
EnsureBufferSize(mBufferLength+1);
nsCRT::memcpy(mBuffer, encodedBuffer, mBufferLength+1);
}
}
void nsHTMLContentSinkStream::Write(const nsString& aString)
//
// Write(nsString) returns the number of characters written.
// If we do both string and stream output, stream chars will override string.
//
PRInt32 nsHTMLContentSinkStream::Write(const nsString& aString)
{
if (mBodyOnly && !mInBody)
return;
return 0;
// If an encoder is being used then convert first convert the input string
// Ideally, we should do this only for the stream case,
// and encode entities (using nsIEntityEncoder) but not charsets
// in the string case.
if (mUnicodeEncoder)
EncodeToBuffer(aString);
int charsWritten = 0;
// For the string case, we don't want to do charset conversion,
// but we still want to encode entities.
if (mString)
{
if (mUnicodeEncoder)
mString->Append(mBuffer, mBufferLength);
else
if (!mEntityConverter && (mFlags & nsIDocumentEncoder::OutputEncodeEntities))
InitEncoders();
if (mEntityConverter && (mFlags & nsIDocumentEncoder::OutputEncodeEntities))
{
nsresult res;
PRUnichar *encodedBuffer = nsnull;
res = mEntityConverter->ConvertToEntities(aString.GetUnicode(),
nsIEntityConverter::html40Latin1,
&encodedBuffer);
if (NS_SUCCEEDED(res) && encodedBuffer)
{
PRInt32 len = nsCRT::strlen(encodedBuffer);
mString->Append(encodedBuffer, len);
nsCRT::free(encodedBuffer);
charsWritten = len;
}
else {
charsWritten = aString.Length();
mString->Append(aString);
}
}
else {
charsWritten = aString.Length();
mString->Append(aString);
}
}
if (!mStream)
return;
return charsWritten;
// Now handle the stream case:
nsOutputStream out(mStream);
// Test again in case there was a problem initializing the unicode encoder:
if (mUnicodeEncoder)
// If an encoder is being used then convert first convert the input string
char *encodedBuffer = nsnull;
nsresult res;
// Initialize the encoder if we haven't already
if (!mCharsetEncoder)
InitEncoders();
if (mCharsetEncoder)
{
out.write(mBuffer, mBufferLength);
// Call the converter to convert to the target charset.
// Convert() takes a char* output param even though it's writing unicode.
res = mCharsetEncoder->Convert(aString.GetUnicode(), &encodedBuffer);
if (NS_SUCCEEDED(res) && encodedBuffer)
{
charsWritten = nsCRT::strlen(encodedBuffer);
out.write(encodedBuffer, charsWritten);
nsCRT::free(encodedBuffer);
}
// If it didn't work, just write the unicode
else
{
const PRUnichar* unicode = aString.GetUnicode();
charsWritten = aString.Length();
out.write(unicode, charsWritten);
}
}
// else just write the unicode
// If we couldn't get an encoder, just write the unicode
else
{
const PRUnichar* unicode = aString.GetUnicode();
out.write(unicode, aString.Length());
charsWritten = aString.Length();
out.write(unicode, charsWritten);
}
return charsWritten;
}
void nsHTMLContentSinkStream::Write(const char* aData)
@ -949,13 +954,11 @@ nsHTMLContentSinkStream::AddLeaf(const nsIParserNode& aNode){
}
else if (type == eHTMLTag_entity)
{
const nsString& entity = aNode.GetText();
EncodeToBuffer(entity);
Write('&');
Write(mBuffer);
const nsString& entity = aNode.GetText();
mColPos += Write(entity) + 1;
// Don't write the semicolon;
// rely on the DTD to include it if one is wanted.
mColPos += entity.Length() + 2;
}
else if (type == eHTMLTag_text)
{
@ -1016,7 +1019,7 @@ PRBool nsHTMLContentSinkStream::HasLongLines(const nsString& text)
{
PRInt32 eol = text.FindChar('\n', PR_FALSE, start);
if (eol < 0) eol = text.Length();
if (eol - start > longLineLen)
if ((PRInt32)(eol - start) > longLineLen)
return PR_TRUE;
start = eol+1;
}
@ -1165,7 +1168,7 @@ nsHTMLContentSinkStream::OpenContainer(const nsIParserNode& aNode)
const nsString& value=aNode.GetValueAt(i);
if (mCharsetOverride.IsEmpty())
mCharsetOverride.Assign(value);
InitEncoder();
InitEncoders();
}
else if (key.Equals("uri"))
{

Просмотреть файл

@ -48,7 +48,8 @@
#include "nsParserCIID.h"
#include "nsCOMPtr.h"
#include "nsHTMLTokens.h" // for eHTMLTags
#include "nsISaveAsCharset.h" // Some compilers can't do nsCOMPtr on a forward type
#include "nsISaveAsCharset.h"
#include "nsIEntityConverter.h"
#include "nsIURI.h"
#define NS_IHTMLCONTENTSINKSTREAM_IID \
@ -154,10 +155,9 @@ protected:
void AddIndent();
void EnsureBufferSize(PRInt32 aNewSize);
void EncodeToBuffer(const nsString& aString);
NS_IMETHOD InitEncoder();
NS_IMETHOD InitEncoders();
void Write(const nsString& aString);
PRInt32 Write(const nsString& aString); // returns # chars written
void Write(const char* aCharBuffer);
void Write(char aChar);
@ -170,7 +170,6 @@ protected:
protected:
nsIOutputStream* mStream;
nsString mStreamBuffer;
nsString* mString;
int mTabLevel;
@ -195,7 +194,8 @@ protected:
PRInt32 mMaxColumn;
nsCOMPtr<nsISaveAsCharset> mUnicodeEncoder;
nsCOMPtr<nsISaveAsCharset> mCharsetEncoder;
nsCOMPtr<nsIEntityConverter> mEntityConverter;
nsCAutoString mCharsetOverride;
};