gecko-dev/layout/generic/nsBulletFrame.cpp

1890 строки
60 KiB
C++

/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
/* ***** BEGIN LICENSE BLOCK *****
* Version: NPL 1.1/GPL 2.0/LGPL 2.1
*
* The contents of this file are subject to the Netscape 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/NPL/
*
* 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 Communicator client code.
*
* The Initial Developer of the Original Code is
* Netscape Communications Corporation.
* Portions created by the Initial Developer are Copyright (C) 1998
* the Initial Developer. All Rights Reserved.
*
* Contributor(s):
*
* Alternatively, the contents of this file may be used under the terms of
* either 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 NPL, 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 NPL, the GPL or the LGPL.
*
* ***** END LICENSE BLOCK ***** */
#include "nsCOMPtr.h"
#include "nsBulletFrame.h"
#include "nsHTMLAtoms.h"
#include "nsHTMLParts.h"
#include "nsHTMLValue.h"
#include "nsHTMLContainerFrame.h"
#include "nsIFontMetrics.h"
#include "nsIHTMLContent.h"
#include "nsIPresContext.h"
#include "nsIPresShell.h"
#include "nsIDocument.h"
#include "nsReflowPath.h"
#include "nsIRenderingContext.h"
#include "nsILoadGroup.h"
#include "nsIURL.h"
#include "nsNetUtil.h"
#include "nsLayoutAtoms.h"
#include "prprf.h"
#ifdef IBMBIDI
#include "nsBidiPresUtils.h"
#endif // IBMBIDI
#include "imgILoader.h"
#include "imgIContainer.h"
#include "imgIDecoderObserver.h"
#include "nsIServiceManager.h"
#include "nsIComponentManager.h"
class nsBulletListener : public imgIDecoderObserver
{
public:
nsBulletListener();
virtual ~nsBulletListener();
NS_DECL_ISUPPORTS
NS_DECL_IMGIDECODEROBSERVER
NS_DECL_IMGICONTAINEROBSERVER
void SetFrame(nsBulletFrame *frame) { mFrame = frame; }
private:
nsBulletFrame *mFrame;
};
nsBulletFrame::nsBulletFrame() :
mPresContext(nsnull)
{
}
nsBulletFrame::~nsBulletFrame()
{
NS_PRECONDITION(!mPresContext, "Never got destroyed properly!");
}
NS_IMETHODIMP
nsBulletFrame::Destroy(nsIPresContext* aPresContext)
{
// Stop image loading first
if (mImageRequest) {
mImageRequest->Cancel(NS_ERROR_FAILURE);
mImageRequest = nsnull;
}
if (mListener)
NS_REINTERPRET_CAST(nsBulletListener*, mListener.get())->SetFrame(nsnull);
mPresContext = nsnull; // clear weak pointer
// Let base class do the rest
return nsFrame::Destroy(aPresContext);
}
NS_IMETHODIMP
nsBulletFrame::Init(nsIPresContext* aPresContext,
nsIContent* aContent,
nsIFrame* aParent,
nsStyleContext* aContext,
nsIFrame* aPrevInFlow)
{
mPresContext = aPresContext;
nsresult rv = nsFrame::Init(aPresContext, aContent, aParent, aContext, aPrevInFlow);
nsIURI *imgURI = GetStyleList()->mListStyleImage;
if (imgURI) {
nsCOMPtr<imgILoader> il(do_GetService("@mozilla.org/image/loader;1", &rv));
if (NS_FAILED(rv))
return rv;
nsCOMPtr<nsILoadGroup> loadGroup;
GetLoadGroup(aPresContext, getter_AddRefs(loadGroup));
// Get the document URI for the referrer...
nsIURI *documentURI = nsnull;
nsCOMPtr<nsIDocument> doc;
if (mContent) {
doc = mContent->GetDocument();
if (doc) {
documentURI = doc->GetDocumentURL();
}
}
if (!mListener) {
nsBulletListener *listener;
NS_NEWXPCOM(listener, nsBulletListener);
NS_ADDREF(listener);
listener->SetFrame(this);
listener->QueryInterface(NS_GET_IID(imgIDecoderObserver), getter_AddRefs(mListener));
NS_ASSERTION(mListener, "queryinterface for the listener failed");
NS_RELEASE(listener);
}
// XXX: initialDocumentURI is NULL !
il->LoadImage(imgURI, nsnull, documentURI, loadGroup, mListener, aPresContext, nsIRequest::LOAD_NORMAL, nsnull, nsnull, getter_AddRefs(mImageRequest));
}
return NS_OK;
}
#ifdef NS_DEBUG
NS_IMETHODIMP
nsBulletFrame::GetFrameName(nsAString& aResult) const
{
return MakeFrameName(NS_LITERAL_STRING("Bullet"), aResult);
}
#endif
nsIAtom*
nsBulletFrame::GetType() const
{
return nsLayoutAtoms::bulletFrame;
}
#include "nsIDOMNode.h"
NS_IMETHODIMP
nsBulletFrame::Paint(nsIPresContext* aPresContext,
nsIRenderingContext& aRenderingContext,
const nsRect& aDirtyRect,
nsFramePaintLayer aWhichLayer,
PRUint32 aFlags)
{
if (NS_FRAME_PAINT_LAYER_FOREGROUND != aWhichLayer) {
return NS_OK;
}
PRBool isVisible;
if (NS_SUCCEEDED(IsVisibleForPainting(aPresContext, aRenderingContext, PR_TRUE, &isVisible)) && isVisible) {
const nsStyleList* myList = GetStyleList();
PRUint8 listStyleType = myList->mListStyleType;
if (myList->mListStyleImage && mImageRequest) {
PRUint32 status;
mImageRequest->GetImageStatus(&status);
if (status & imgIRequest::STATUS_LOAD_COMPLETE &&
!(status & imgIRequest::STATUS_ERROR)) {
nsCOMPtr<imgIContainer> imageCon;
mImageRequest->GetImage(getter_AddRefs(imageCon));
if (imageCon) {
nsRect innerArea(mPadding.left, mPadding.top,
mRect.width - (mPadding.left + mPadding.right),
mRect.height - (mPadding.top + mPadding.bottom));
nsPoint p(innerArea.x, innerArea.y);
innerArea.x = innerArea.y = 0;
aRenderingContext.DrawImage(imageCon, &innerArea, &p);
return NS_OK;
}
}
}
const nsStyleFont* myFont = GetStyleFont();
const nsStyleColor* myColor = GetStyleColor();
nsCOMPtr<nsIFontMetrics> fm;
aRenderingContext.SetColor(myColor->mColor);
#ifdef IBMBIDI
nsCharType charType = eCharType_LeftToRight;
PRUint8 level = 0;
PRBool isBidiSystem = PR_FALSE;
const nsStyleVisibility* vis = GetStyleVisibility();
PRUint32 hints = 0;
#endif // IBMBIDI
nsAutoString text;
switch (listStyleType) {
case NS_STYLE_LIST_STYLE_NONE:
break;
default:
case NS_STYLE_LIST_STYLE_BASIC:
case NS_STYLE_LIST_STYLE_DISC:
aRenderingContext.FillEllipse(mPadding.left, mPadding.top,
mRect.width - (mPadding.left + mPadding.right),
mRect.height - (mPadding.top + mPadding.bottom));
break;
case NS_STYLE_LIST_STYLE_CIRCLE:
aRenderingContext.DrawEllipse(mPadding.left, mPadding.top,
mRect.width - (mPadding.left + mPadding.right),
mRect.height - (mPadding.top + mPadding.bottom));
break;
case NS_STYLE_LIST_STYLE_SQUARE:
aRenderingContext.FillRect(mPadding.left, mPadding.top,
mRect.width - (mPadding.left + mPadding.right),
mRect.height - (mPadding.top + mPadding.bottom));
break;
case NS_STYLE_LIST_STYLE_DECIMAL:
case NS_STYLE_LIST_STYLE_OLD_DECIMAL:
case NS_STYLE_LIST_STYLE_DECIMAL_LEADING_ZERO:
#ifdef IBMBIDI
GetListItemText(aPresContext, *myList, text);
charType = eCharType_EuropeanNumber;
break;
case NS_STYLE_LIST_STYLE_MOZ_ARABIC_INDIC:
if (GetListItemText(aPresContext, *myList, text))
charType = eCharType_ArabicNumber;
else
charType = eCharType_EuropeanNumber;
break;
case NS_STYLE_LIST_STYLE_HEBREW:
aRenderingContext.GetHints(hints);
isBidiSystem = (hints & NS_RENDERING_HINT_BIDI_REORDERING);
if (!isBidiSystem) {
if (GetListItemText(aPresContext, *myList, text)) {
charType = eCharType_RightToLeft;
level = 1;
} else {
charType = eCharType_EuropeanNumber;
}
if (NS_STYLE_DIRECTION_RTL == vis->mDirection) {
text.Cut(0, 1);
text.Append(NS_LITERAL_STRING("."));
}
break;
}
// else fall through
#endif // IBMBIDI
case NS_STYLE_LIST_STYLE_LOWER_ROMAN:
case NS_STYLE_LIST_STYLE_UPPER_ROMAN:
case NS_STYLE_LIST_STYLE_LOWER_ALPHA:
case NS_STYLE_LIST_STYLE_UPPER_ALPHA:
case NS_STYLE_LIST_STYLE_OLD_LOWER_ROMAN:
case NS_STYLE_LIST_STYLE_OLD_UPPER_ROMAN:
case NS_STYLE_LIST_STYLE_OLD_LOWER_ALPHA:
case NS_STYLE_LIST_STYLE_OLD_UPPER_ALPHA:
case NS_STYLE_LIST_STYLE_LOWER_GREEK:
#ifndef IBMBIDI
case NS_STYLE_LIST_STYLE_HEBREW:
#endif
case NS_STYLE_LIST_STYLE_ARMENIAN:
case NS_STYLE_LIST_STYLE_GEORGIAN:
case NS_STYLE_LIST_STYLE_CJK_IDEOGRAPHIC:
case NS_STYLE_LIST_STYLE_HIRAGANA:
case NS_STYLE_LIST_STYLE_KATAKANA:
case NS_STYLE_LIST_STYLE_HIRAGANA_IROHA:
case NS_STYLE_LIST_STYLE_KATAKANA_IROHA:
case NS_STYLE_LIST_STYLE_MOZ_SIMP_CHINESE_INFORMAL:
case NS_STYLE_LIST_STYLE_MOZ_SIMP_CHINESE_FORMAL:
case NS_STYLE_LIST_STYLE_MOZ_TRAD_CHINESE_INFORMAL:
case NS_STYLE_LIST_STYLE_MOZ_TRAD_CHINESE_FORMAL:
case NS_STYLE_LIST_STYLE_MOZ_JAPANESE_INFORMAL:
case NS_STYLE_LIST_STYLE_MOZ_JAPANESE_FORMAL:
case NS_STYLE_LIST_STYLE_MOZ_CJK_HEAVENLY_STEM:
case NS_STYLE_LIST_STYLE_MOZ_CJK_EARTHLY_BRANCH:
#ifndef IBMBIDI
case NS_STYLE_LIST_STYLE_MOZ_ARABIC_INDIC:
#endif
case NS_STYLE_LIST_STYLE_MOZ_PERSIAN:
case NS_STYLE_LIST_STYLE_MOZ_URDU:
case NS_STYLE_LIST_STYLE_MOZ_DEVANAGARI:
case NS_STYLE_LIST_STYLE_MOZ_GURMUKHI:
case NS_STYLE_LIST_STYLE_MOZ_GUJARATI:
case NS_STYLE_LIST_STYLE_MOZ_ORIYA:
case NS_STYLE_LIST_STYLE_MOZ_KANNADA:
case NS_STYLE_LIST_STYLE_MOZ_MALAYALAM:
case NS_STYLE_LIST_STYLE_MOZ_BENGALI:
case NS_STYLE_LIST_STYLE_MOZ_TAMIL:
case NS_STYLE_LIST_STYLE_MOZ_TELUGU:
case NS_STYLE_LIST_STYLE_MOZ_THAI:
case NS_STYLE_LIST_STYLE_MOZ_LAO:
case NS_STYLE_LIST_STYLE_MOZ_MYANMAR:
case NS_STYLE_LIST_STYLE_MOZ_KHMER:
case NS_STYLE_LIST_STYLE_MOZ_HANGUL:
case NS_STYLE_LIST_STYLE_MOZ_HANGUL_CONSONANT:
case NS_STYLE_LIST_STYLE_MOZ_ETHIOPIC_HALEHAME:
case NS_STYLE_LIST_STYLE_MOZ_ETHIOPIC_NUMERIC:
case NS_STYLE_LIST_STYLE_MOZ_ETHIOPIC_HALEHAME_AM:
case NS_STYLE_LIST_STYLE_MOZ_ETHIOPIC_HALEHAME_TI_ER:
case NS_STYLE_LIST_STYLE_MOZ_ETHIOPIC_HALEHAME_TI_ET:
aPresContext->GetMetricsFor(myFont->mFont, getter_AddRefs(fm));
#ifdef IBMBIDI
// If we can't render our numeral using the chars in the numbering
// system, we'll be using "decimal"...
PRBool usedChars =
#endif // IBMBIDI
GetListItemText(aPresContext, *myList, text);
#ifdef IBMBIDI
if (!usedChars)
charType = eCharType_EuropeanNumber;
#endif
aRenderingContext.SetFont(fm);
nscoord ascent;
fm->GetMaxAscent(ascent);
aRenderingContext.DrawString(text, mPadding.left, mPadding.top + ascent);
break;
}
#ifdef IBMBIDI
if (charType != eCharType_LeftToRight) {
aPresContext->GetMetricsFor(myFont->mFont, getter_AddRefs(fm));
aRenderingContext.SetFont(fm);
nscoord ascent;
fm->GetMaxAscent(ascent);
nsBidiPresUtils* bidiUtils;
aPresContext->GetBidiUtils(&bidiUtils);
if (bidiUtils) {
const PRUnichar* buffer = text.get();
PRInt32 textLength = text.Length();
if (eCharType_RightToLeft == charType) {
bidiUtils->FormatUnicodeText(aPresContext, (PRUnichar*)buffer, textLength,
charType, level, PR_FALSE);
}
else {
//Mohamed
aRenderingContext.GetHints(hints);
isBidiSystem = (hints & NS_RENDERING_HINT_ARABIC_SHAPING);
bidiUtils->FormatUnicodeText(aPresContext, (PRUnichar*)buffer, textLength,
charType, level, isBidiSystem);//Mohamed
}
}
aRenderingContext.DrawString(text, mPadding.left, mPadding.top + ascent);
}
#endif // IBMBIDI
}
DO_GLOBAL_REFLOW_COUNT_DSP("nsBulletFrame", &aRenderingContext);
return NS_OK;
}
PRInt32
nsBulletFrame::SetListItemOrdinal(PRInt32 aNextOrdinal,
PRBool* aChanged)
{
// Assume that the ordinal comes from the caller
PRInt32 oldOrdinal = mOrdinal;
mOrdinal = aNextOrdinal;
// Try to get value directly from the list-item, if it specifies a
// value attribute. Note: we do this with our parent's content
// because our parent is the list-item.
nsHTMLValue value;
nsIContent* parentContent = mParent->GetContent();
if (parentContent) {
nsCOMPtr<nsIHTMLContent> hc = do_QueryInterface(parentContent);
if (hc) {
if (NS_CONTENT_ATTR_HAS_VALUE ==
hc->GetHTMLAttribute(nsHTMLAtoms::value, value)) {
if (eHTMLUnit_Integer == value.GetUnit()) {
// Use ordinal specified by the value attribute
mOrdinal = value.GetIntValue();
}
}
}
}
*aChanged = oldOrdinal != mOrdinal;
return mOrdinal + 1;
}
// XXX change roman/alpha to use unsigned math so that maxint and
// maxnegint will work
/**
* For all functions below, a return value of PR_TRUE means that we
* could represent mOrder in the desired numbering system. PR_FALSE
* means we had to fall back to decimal
*/
static PRBool DecimalToText(PRInt32 ordinal, nsString& result)
{
char cbuf[40];
PR_snprintf(cbuf, sizeof(cbuf), "%ld", ordinal);
result.AppendWithConversion(cbuf);
return PR_TRUE;
}
static PRBool DecimalLeadingZeroToText(PRInt32 ordinal, nsString& result)
{
char cbuf[40];
PR_snprintf(cbuf, sizeof(cbuf), "%02ld", ordinal);
result.AppendWithConversion(cbuf);
return PR_TRUE;
}
static PRBool OtherDecimalToText(PRInt32 ordinal, PRUnichar zeroChar, nsString& result)
{
PRUnichar diff = zeroChar - PRUnichar('0');
DecimalToText(ordinal, result);
PRUnichar* p = NS_CONST_CAST(PRUnichar*, result.get());
if (ordinal < 0) {
// skip the leading '-'
++p;
}
for(; nsnull != *p ; p++)
*p += diff;
return PR_TRUE;
}
static PRBool TamilToText(PRInt32 ordinal, nsString& result)
{
PRUnichar diff = 0x0BE6 - PRUnichar('0');
DecimalToText(ordinal, result);
if (ordinal < 1 || ordinal > 9999) {
// Can't do those in this system.
return PR_FALSE;
}
PRUnichar* p = (PRUnichar*)result.get();
for(; nsnull != *p ; p++)
if(*p != PRUnichar('0'))
*p += diff;
return PR_TRUE;
}
static const char gLowerRomanCharsA[] = "ixcm";
static const char gUpperRomanCharsA[] = "IXCM";
static const char gLowerRomanCharsB[] = "vld";
static const char gUpperRomanCharsB[] = "VLD";
static PRBool RomanToText(PRInt32 ordinal, nsString& result, const char* achars, const char* bchars)
{
if (ordinal < 1 || ordinal > 3999) {
DecimalToText(ordinal, result);
return PR_FALSE;
}
nsAutoString addOn, decStr;
decStr.AppendInt(ordinal, 10);
PRIntn len = decStr.Length();
const PRUnichar* dp = decStr.get();
const PRUnichar* end = dp + len;
PRIntn romanPos = len;
PRIntn n;
for (; dp < end; dp++) {
romanPos--;
addOn.SetLength(0);
switch(*dp) {
case '3': addOn.Append(PRUnichar(achars[romanPos]));
case '2': addOn.Append(PRUnichar(achars[romanPos]));
case '1': addOn.Append(PRUnichar(achars[romanPos]));
break;
case '4':
addOn.Append(PRUnichar(achars[romanPos]));
// FALLTHROUGH
case '5': case '6':
case '7': case '8':
addOn.Append(PRUnichar(bchars[romanPos]));
for(n=0;n<(*dp-'5');n++) {
addOn.Append(PRUnichar(achars[romanPos]));
}
break;
case '9':
addOn.Append(PRUnichar(achars[romanPos]));
addOn.Append(PRUnichar(achars[romanPos+1]));
break;
default:
break;
}
result.Append(addOn);
}
return PR_TRUE;
}
#define ALPHA_SIZE 26
static const PRUnichar gLowerAlphaChars[ALPHA_SIZE] =
{
0x0061, 0x0062, 0x0063, 0x0064, 0x0065, // A B C D E
0x0066, 0x0067, 0x0068, 0x0069, 0x006A, // F G H I J
0x006B, 0x006C, 0x006D, 0x006E, 0x006F, // K L M N O
0x0070, 0x0071, 0x0072, 0x0073, 0x0074, // P Q R S T
0x0075, 0x0076, 0x0077, 0x0078, 0x0079, // U V W X Y
0x007A // Z
};
static const PRUnichar gUpperAlphaChars[ALPHA_SIZE] =
{
0x0041, 0x0042, 0x0043, 0x0044, 0x0045, // A B C D E
0x0046, 0x0047, 0x0048, 0x0049, 0x004A, // F G H I J
0x004B, 0x004C, 0x004D, 0x004E, 0x004F, // K L M N O
0x0050, 0x0051, 0x0052, 0x0053, 0x0054, // P Q R S T
0x0055, 0x0056, 0x0057, 0x0058, 0x0059, // U V W X Y
0x005A // Z
};
#define KATAKANA_CHARS_SIZE 48
// Page 94 Writing Systems of The World
// after modification by momoi
static const PRUnichar gKatakanaChars[KATAKANA_CHARS_SIZE] =
{
0x30A2, 0x30A4, 0x30A6, 0x30A8, 0x30AA, // a i u e o
0x30AB, 0x30AD, 0x30AF, 0x30B1, 0x30B3, // ka ki ku ke ko
0x30B5, 0x30B7, 0x30B9, 0x30BB, 0x30BD, // sa shi su se so
0x30BF, 0x30C1, 0x30C4, 0x30C6, 0x30C8, // ta chi tsu te to
0x30CA, 0x30CB, 0x30CC, 0x30CD, 0x30CE, // na ni nu ne no
0x30CF, 0x30D2, 0x30D5, 0x30D8, 0x30DB, // ha hi hu he ho
0x30DE, 0x30DF, 0x30E0, 0x30E1, 0x30E2, // ma mi mu me mo
0x30E4, 0x30E6, 0x30E8, // ya yu yo
0x30E9, 0x30EA, 0x30EB, 0x30EC, 0x30ED, // ra ri ru re ro
0x30EF, 0x30F0, 0x30F1, 0x30F2, // wa (w)i (w)e (w)o
0x30F3 // n
};
#define HIRAGANA_CHARS_SIZE 48
static const PRUnichar gHiraganaChars[HIRAGANA_CHARS_SIZE] =
{
0x3042, 0x3044, 0x3046, 0x3048, 0x304A, // a i u e o
0x304B, 0x304D, 0x304F, 0x3051, 0x3053, // ka ki ku ke ko
0x3055, 0x3057, 0x3059, 0x305B, 0x305D, // sa shi su se so
0x305F, 0x3061, 0x3064, 0x3066, 0x3068, // ta chi tsu te to
0x306A, 0x306B, 0x306C, 0x306D, 0x306E, // na ni nu ne no
0x306F, 0x3072, 0x3075, 0x3078, 0x307B, // ha hi hu he ho
0x307E, 0x307F, 0x3080, 0x3081, 0x3082, // ma mi mu me mo
0x3084, 0x3086, 0x3088, // ya yu yo
0x3089, 0x308A, 0x308B, 0x308C, 0x308D, // ra ri ru re ro
0x308F, 0x3090, 0x3091, 0x3092, // wa (w)i (w)e (w)o
0x3093 // n
};
#define HIRAGANA_IROHA_CHARS_SIZE 47
// Page 94 Writing Systems of The World
static const PRUnichar gHiraganaIrohaChars[HIRAGANA_IROHA_CHARS_SIZE] =
{
0x3044, 0x308D, 0x306F, 0x306B, 0x307B, // i ro ha ni ho
0x3078, 0x3068, 0x3061, 0x308A, 0x306C, // he to chi ri nu
0x308B, 0x3092, 0x308F, 0x304B, 0x3088, // ru (w)o wa ka yo
0x305F, 0x308C, 0x305D, 0x3064, 0x306D, // ta re so tsu ne
0x306A, 0x3089, 0x3080, 0x3046, 0x3090, // na ra mu u (w)i
0x306E, 0x304A, 0x304F, 0x3084, 0x307E, // no o ku ya ma
0x3051, 0x3075, 0x3053, 0x3048, 0x3066, // ke hu ko e te
0x3042, 0x3055, 0x304D, 0x3086, 0x3081, // a sa ki yu me
0x307F, 0x3057, 0x3091, 0x3072, 0x3082, // mi shi (w)e hi mo
0x305B, 0x3059 // se su
};
#define KATAKANA_IROHA_CHARS_SIZE 47
static const PRUnichar gKatakanaIrohaChars[KATAKANA_IROHA_CHARS_SIZE] =
{
0x30A4, 0x30ED, 0x30CF, 0x30CB, 0x30DB, // i ro ha ni ho
0x30D8, 0x30C8, 0x30C1, 0x30EA, 0x30CC, // he to chi ri nu
0x30EB, 0x30F2, 0x30EF, 0x30AB, 0x30E8, // ru (w)o wa ka yo
0x30BF, 0x30EC, 0x30BD, 0x30C4, 0x30CD, // ta re so tsu ne
0x30CA, 0x30E9, 0x30E0, 0x30A6, 0x30F0, // na ra mu u (w)i
0x30CE, 0x30AA, 0x30AF, 0x30E4, 0x30DE, // no o ku ya ma
0x30B1, 0x30D5, 0x30B3, 0x30A8, 0x30C6, // ke hu ko e te
0x30A2, 0x30B5, 0x30AD, 0x30E6, 0x30E1, // a sa ki yu me
0x30DF, 0x30B7, 0x30F1, 0x30D2, 0x30E2, // mi shi (w)e hi mo
0x30BB, 0x30B9 // se su
};
#define LOWER_GREEK_CHARS_SIZE 24
// Note: 0x03C2 GREEK FINAL SIGMA is not used in here....
static const PRUnichar gLowerGreekChars[LOWER_GREEK_CHARS_SIZE] =
{
0x03B1, 0x03B2, 0x03B3, 0x03B4, 0x03B5, // alpha beta gamma delta epsilon
0x03B6, 0x03B7, 0x03B8, 0x03B9, 0x03BA, // zeta eta theta iota kappa
0x03BB, 0x03BC, 0x03BD, 0x03BE, 0x03BF, // lamda mu nu xi omicron
0x03C0, 0x03C1, 0x03C3, 0x03C4, 0x03C5, // pi rho sigma tau upsilon
0x03C6, 0x03C7, 0x03C8, 0x03C9 // phi chi psi omega
};
#define CJK_HEAVENLY_STEM_CHARS_SIZE 10
static const PRUnichar gCJKHeavenlyStemChars[CJK_HEAVENLY_STEM_CHARS_SIZE] =
{
0x7532, 0x4e59, 0x4e19, 0x4e01, 0x620a,
0x5df1, 0x5e9a, 0x8f9b, 0x58ec, 0x7678
};
#define CJK_EARTHLY_BRANCH_CHARS_SIZE 12
static const PRUnichar gCJKEarthlyBranchChars[CJK_EARTHLY_BRANCH_CHARS_SIZE] =
{
0x5b50, 0x4e11, 0x5bc5, 0x536f, 0x8fb0, 0x5df3,
0x5348, 0x672a, 0x7533, 0x9149, 0x620c, 0x4ea5
};
#define HANGUL_CHARS_SIZE 14
static const PRUnichar gHangulChars[HANGUL_CHARS_SIZE] =
{
0xac00, 0xb098, 0xb2e4, 0xb77c, 0xb9c8, 0xbc14,
0xc0ac, 0xc544, 0xc790, 0xcc28, 0xce74, 0xd0c0,
0xd30c, 0xd558
};
#define HANGUL_CONSONANT_CHARS_SIZE 14
static const PRUnichar gHangulConsonantChars[HANGUL_CONSONANT_CHARS_SIZE] =
{
0x3131, 0x3134, 0x3137, 0x3139, 0x3141, 0x3142,
0x3145, 0x3147, 0x3148, 0x314a, 0x314b, 0x314c,
0x314d, 0x314e
};
// Ge'ez set of Ethiopic ordered list. There are other locale-dependent sets.
// For the time being, let's implement two Ge'ez sets only
// per Momoi san's suggestion in bug 102252.
// For details, refer to http://www.ethiopic.org/Collation/OrderedLists.html.
#define ETHIOPIC_HALEHAME_CHARS_SIZE 26
static const PRUnichar gEthiopicHalehameChars[ETHIOPIC_HALEHAME_CHARS_SIZE] =
{
0x1200, 0x1208, 0x1210, 0x1218, 0x1220, 0x1228,
0x1230, 0x1240, 0x1260, 0x1270, 0x1280, 0x1290,
0x12a0, 0x12a8, 0x12c8, 0x12d0, 0x12d8, 0x12e8,
0x12f0, 0x1308, 0x1320, 0x1330, 0x1338, 0x1340,
0x1348, 0x1350
};
#define ETHIOPIC_HALEHAME_AM_CHARS_SIZE 33
static const PRUnichar gEthiopicHalehameAmChars[ETHIOPIC_HALEHAME_AM_CHARS_SIZE] =
{
0x1200, 0x1208, 0x1210, 0x1218, 0x1220, 0x1228,
0x1230, 0x1238, 0x1240, 0x1260, 0x1270, 0x1278,
0x1280, 0x1290, 0x1298, 0x12a0, 0x12a8, 0x12b8,
0x12c8, 0x12d0, 0x12d8, 0x12e0, 0x12e8, 0x12f0,
0x1300, 0x1308, 0x1320, 0x1328, 0x1330, 0x1338,
0x1340, 0x1348, 0x1350
};
#define ETHIOPIC_HALEHAME_TI_ER_CHARS_SIZE 31
static const PRUnichar gEthiopicHalehameTiErChars[ETHIOPIC_HALEHAME_TI_ER_CHARS_SIZE] =
{
0x1200, 0x1208, 0x1210, 0x1218, 0x1228, 0x1230,
0x1238, 0x1240, 0x1250, 0x1260, 0x1270, 0x1278,
0x1290, 0x1298, 0x12a0, 0x12a8, 0x12b8, 0x12c8,
0x12d0, 0x12d8, 0x12e0, 0x12e8, 0x12f0, 0x1300,
0x1308, 0x1320, 0x1328, 0x1330, 0x1338, 0x1348,
0x1350
};
#define ETHIOPIC_HALEHAME_TI_ET_CHARS_SIZE 34
static const PRUnichar gEthiopicHalehameTiEtChars[ETHIOPIC_HALEHAME_TI_ET_CHARS_SIZE] =
{
0x1200, 0x1208, 0x1210, 0x1218, 0x1220, 0x1228,
0x1230, 0x1238, 0x1240, 0x1250, 0x1260, 0x1270,
0x1278, 0x1280, 0x1290, 0x1298, 0x12a0, 0x12a8,
0x12b8, 0x12c8, 0x12d0, 0x12d8, 0x12e0, 0x12e8,
0x12f0, 0x1300, 0x1308, 0x1320, 0x1328, 0x1330,
0x1338, 0x1340, 0x1348, 0x1350
};
// We know cjk-ideographic need 31 characters to display 99,999,999,999,999,999
// georgian needs 6 at most
// armenian needs 12 at most
// hebrew may need more...
#define NUM_BUF_SIZE 34
static PRBool CharListToText(PRInt32 ordinal, nsString& result, const PRUnichar* chars, PRInt32 aBase)
{
PRUnichar buf[NUM_BUF_SIZE];
PRInt32 idx = NUM_BUF_SIZE;
if (ordinal < 1) {
DecimalToText(ordinal, result);
return PR_FALSE;
}
do {
ordinal--; // a == 0
PRInt32 cur = ordinal % aBase;
buf[--idx] = chars[cur];
ordinal /= aBase ;
} while ( ordinal > 0);
result.Append(buf+idx,NUM_BUF_SIZE-idx);
return PR_TRUE;
}
static const PRUnichar gCJKIdeographicDigit1[10] =
{
0x96f6, 0x4e00, 0x4e8c, 0x4e09, 0x56db, // 0 - 4
0x4e94, 0x516d, 0x4e03, 0x516b, 0x4e5d // 5 - 9
};
static const PRUnichar gCJKIdeographicDigit2[10] =
{
0x96f6, 0x58f9, 0x8cb3, 0x53c3, 0x8086, // 0 - 4
0x4f0d, 0x9678, 0x67d2, 0x634c, 0x7396 // 5 - 9
};
static const PRUnichar gCJKIdeographicDigit3[10] =
{
0x96f6, 0x58f9, 0x8d30, 0x53c1, 0x8086, // 0 - 4
0x4f0d, 0x9646, 0x67d2, 0x634c, 0x7396 // 5 - 9
};
static const PRUnichar gCJKIdeographicUnit1[4] =
{
0x000, 0x5341, 0x767e, 0x5343
};
static const PRUnichar gCJKIdeographicUnit2[4] =
{
0x000, 0x62FE, 0x4F70, 0x4EDF
};
static const PRUnichar gCJKIdeographic10KUnit1[4] =
{
0x000, 0x842c, 0x5104, 0x5146
};
static const PRUnichar gCJKIdeographic10KUnit2[4] =
{
0x000, 0x4E07, 0x4ebf, 0x5146
};
static const PRUnichar gCJKIdeographic10KUnit3[4] =
{
0x000, 0x4E07, 0x5104, 0x5146
};
static const PRBool CJKIdeographicToText(PRInt32 ordinal, nsString& result,
const PRUnichar* digits,
const PRUnichar *unit,
const PRUnichar* unit10k)
{
// In theory, we need the following if condiction,
// However, the limit, 10 ^ 16, is greater than the max of PRUint32
// so we don't really need to test it here.
// if( ordinal > 9999999999999999)
// {
// PR_snprintf(cbuf, sizeof(cbuf), "%ld", ordinal);
// result.Append(cbuf);
// }
// else
// {
if (ordinal < 0) {
DecimalToText(ordinal, result);
return PR_FALSE;
}
PRUnichar c10kUnit = 0;
PRUnichar cUnit = 0;
PRUnichar cDigit = 0;
PRUint32 ud = 0;
PRUnichar buf[NUM_BUF_SIZE];
PRInt32 idx = NUM_BUF_SIZE;
PRBool bOutputZero = ( 0 == ordinal );
do {
if(0 == (ud % 4)) {
c10kUnit = unit10k[ud/4];
}
PRInt32 cur = ordinal % 10;
cDigit = digits[cur];
if( 0 == cur)
{
cUnit = 0;
if(bOutputZero) {
bOutputZero = PR_FALSE;
if(0 != cDigit)
buf[--idx] = cDigit;
}
}
else
{
bOutputZero = PR_TRUE;
cUnit = unit[ud%4];
if(0 != c10kUnit)
buf[--idx] = c10kUnit;
if(0 != cUnit)
buf[--idx] = cUnit;
if((0 != cDigit) &&
( (1 != cur) || (1 != (ud%4)) || ( ordinal > 10)) )
buf[--idx] = cDigit;
c10kUnit = 0;
}
ordinal /= 10;
++ud;
} while( ordinal > 0);
result.Append(buf+idx,NUM_BUF_SIZE-idx);
// }
return PR_TRUE;
}
#define HEBREW_THROSAND_SEP 0x0020
#define HEBREW_GERESH 0x05F3
#define HEBREW_GERSHAYIM 0x05F4
static const PRUnichar gHebrewDigit[22] =
{
// 1 2 3 4 5 6 7 8 9
0x05D0, 0x05D1, 0x05D2, 0x05D3, 0x05D4, 0x05D5, 0x05D6, 0x05D7, 0x05D8,
// 10 20 30 40 50 60 70 80 90
0x05D9, 0x05DB, 0x05DC, 0x05DE, 0x05E0, 0x05E1, 0x05E2, 0x05E4, 0x05E6,
// 100 200 300 400
0x05E7, 0x05E8, 0x05E9, 0x05EA
};
static PRBool HebrewToText(PRInt32 ordinal, nsString& result)
{
if (ordinal < 0) {
DecimalToText(ordinal, result);
return PR_FALSE;
}
if (ordinal == 0) {
// This one is treated specially
#ifdef IBMBIDI
static const PRUnichar hebrewZero[] = { 0x05D0, 0x05E4, 0x05E1 };
#else
static const PRUnichar hebrewZero[] = { 0x05E1, 0x05E4, 0x05D0 };
#endif // IBMBIDI
result.Append(hebrewZero);
return PR_TRUE;
}
PRBool outputSep = PR_FALSE;
PRUnichar buf[NUM_BUF_SIZE];
#ifdef IBMBIDI
// Changes: 1) don't reverse the text; 2) don't insert geresh/gershayim.
PRInt32 idx = 0;
#else
PRInt32 idx = NUM_BUF_SIZE;
#endif // IBMBIDI
PRUnichar digit;
do {
PRInt32 n3 = ordinal % 1000;
if(outputSep)
#ifdef IBMBIDI
buf[idx++] = HEBREW_THROSAND_SEP; // output thousand separator
#else
buf[--idx] = HEBREW_THROSAND_SEP; // output thousand separator
#endif // IBMBIDI
outputSep = ( n3 > 0); // request to output thousand separator next time.
PRInt32 d = 0; // we need to keep track of digit got output per 3 digits,
// so we can handle Gershayim and Gersh correctly
// Process digit for 100 - 900
for(PRInt32 n1 = 400; n1 > 0; )
{
if( n3 >= n1)
{
n3 -= n1;
digit = gHebrewDigit[(n1/100)-1+18];
if( n3 > 0)
{
#ifdef IBMBIDI
buf[idx++] = digit;
#else
buf[--idx] = digit;
#endif // IBMBIDI
++d;
} else {
// if this is the last digit
#ifdef IBMBIDI
buf[idx++] = digit;
#else
if (d > 0)
{
buf[--idx] = HEBREW_GERSHAYIM;
buf[--idx] = digit;
} else {
buf[--idx] = digit;
buf[--idx] = HEBREW_GERESH;
} // if
#endif // IBMBIDI
} // if
} else {
n1 -= 100;
} // if
} // for
// Process digit for 10 - 90
PRInt32 n2;
if( n3 >= 10 )
{
// Special process for 15 and 16
if(( 15 == n3 ) || (16 == n3)) {
// Special rule for religious reason...
// 15 is represented by 9 and 6, not 10 and 5
// 16 is represented by 9 and 7, not 10 and 6
n2 = 9;
digit = gHebrewDigit[ n2 - 1];
} else {
n2 = n3 - (n3 % 10);
digit = gHebrewDigit[(n2/10)-1+9];
} // if
n3 -= n2;
if( n3 > 0) {
#ifdef IBMBIDI
buf[idx++] = digit;
#else
buf[--idx] = digit;
#endif // IBMBIDI
++d;
} else {
// if this is the last digit
#ifdef IBMBIDI
buf[idx++] = digit;
#else
if (d > 0)
{
buf[--idx] = HEBREW_GERSHAYIM;
buf[--idx] = digit;
} else {
buf[--idx] = digit;
buf[--idx] = HEBREW_GERESH;
} // if
#endif // IBMBIDI
} // if
} // if
// Process digit for 1 - 9
if ( n3 > 0)
{
digit = gHebrewDigit[n3-1];
// must be the last digit
#ifdef IBMBIDI
buf[idx++] = digit;
#else
if (d > 0)
{
buf[--idx] = HEBREW_GERSHAYIM;
buf[--idx] = digit;
} else {
buf[--idx] = digit;
buf[--idx] = HEBREW_GERESH;
} // if
#endif // IBMBIDI
} // if
ordinal /= 1000;
} while (ordinal >= 1);
#ifdef IBMBIDI
result.Append(buf, idx);
#else
result.Append(buf+idx,NUM_BUF_SIZE-idx);
#endif // IBMBIDI
return PR_TRUE;
}
static PRBool ArmenianToText(PRInt32 ordinal, nsString& result)
{
// XXXbz this system goes out to a lot further than 9999... we should fix
// that. This algorithm seems broken in general. There's this business of
// "7000" being special and then there's the combining accent we're supposed
// to be using...
if (ordinal < 1 || ordinal > 9999) { // zero or reach the limit of Armenian numbering system
DecimalToText(ordinal, result);
return PR_FALSE;
}
PRUnichar buf[NUM_BUF_SIZE];
PRInt32 idx = NUM_BUF_SIZE;
PRInt32 d = 0;
do {
PRInt32 cur = ordinal % 10;
if (cur > 0)
{
PRUnichar u = 0x0530 + (d * 9) + cur;
buf[--idx] = u;
}
++d;
ordinal /= 10;
} while (ordinal > 0);
result.Append(buf + idx, NUM_BUF_SIZE - idx);
return PR_TRUE;
}
static const PRUnichar gGeorgianValue [ 37 ] = { // 4 * 9 + 1 = 37
// 1 2 3 4 5 6 7 8 9
0x10A0, 0x10A1, 0x10A2, 0x10A3, 0x10A4, 0x10A5, 0x10A6, 0x10C1, 0x10A7,
// 10 20 30 40 50 60 70 80 90
0x10A8, 0x10A9, 0x10AA, 0x10AB, 0x10AC, 0x10C2, 0x10AD, 0x10AE, 0x10AF,
// 100 200 300 400 500 600 700 800 900
0x10B0, 0x10B1, 0x10B2, 0x10B3, 0x10C3, 0x10B4, 0x10B5, 0x10B6, 0x10B7,
// 1000 2000 3000 4000 5000 6000 7000 8000 9000
0x10B8, 0x10B9, 0x10BA, 0x10BB, 0x10BC, 0x10BD, 0x10BE, 0x10C4, 0x10C5,
// 10000
0x10BF
};
static PRBool GeorgianToText(PRInt32 ordinal, nsString& result)
{
if (ordinal < 1 || ordinal > 19999) { // zero or reach the limit of Georgian numbering system
DecimalToText(ordinal, result);
return PR_FALSE;
}
PRUnichar buf[NUM_BUF_SIZE];
PRInt32 idx = NUM_BUF_SIZE;
PRInt32 d = 0;
do {
PRInt32 cur = ordinal % 10;
if (cur > 0)
{
PRUnichar u = gGeorgianValue[(d * 9 ) + ( cur - 1)];
buf[--idx] = u;
}
++d;
ordinal /= 10;
} while (ordinal > 0);
result.Append(buf + idx, NUM_BUF_SIZE - idx);
return PR_TRUE;
}
// Convert ordinal to Ethiopic numeric representation.
// The detail is available at http://www.ethiopic.org/Numerals/
// The algorithm used here is based on the pseudo-code put up there by
// Daniel Yacob <yacob@geez.org>.
// Another reference is Unicode 3.0 standard section 11.1.
#define ETHIOPIC_ONE 0x1369
#define ETHIOPIC_TEN 0x1372
#define ETHIOPIC_HUNDRED 0x137B
#define ETHIOPIC_TEN_THOUSAND 0x137C
static PRBool EthiopicToText(PRInt32 ordinal, nsString& result)
{
nsAutoString asciiNumberString; // decimal string representation of ordinal
DecimalToText(ordinal, asciiNumberString);
if (ordinal < 1) {
result.Append(asciiNumberString);
return PR_FALSE;
}
PRUint8 asciiStringLength = asciiNumberString.Length();
// If number length is odd, add a leading "0"
// the leading "0" preconditions the string to always have the
// leading tens place populated, this avoids a check within the loop.
// If we didn't add the leading "0", decrement asciiStringLength so
// it will be equivalent to a zero-based index in both cases.
if (asciiStringLength & 1) {
asciiNumberString.Insert(NS_LITERAL_STRING("0"), 0);
} else {
asciiStringLength--;
}
// Iterate from the highest digits to lowest
// indexFromLeft indexes digits (0 = most significant)
// groupIndexFromRight indexes pairs of digits (0 = least significant)
for (PRUint8 indexFromLeft = 0, groupIndexFromRight = asciiStringLength >> 1;
indexFromLeft <= asciiStringLength;
indexFromLeft += 2, groupIndexFromRight--) {
PRUint8 tensValue = asciiNumberString.CharAt(indexFromLeft) & 0x0F;
PRUint8 unitsValue = asciiNumberString.CharAt(indexFromLeft + 1) & 0x0F;
PRUint8 groupValue = tensValue * 10 + unitsValue;
PRBool oddGroup = (groupIndexFromRight & 1);
// we want to clear ETHIOPIC_ONE when it is superfluous
if (ordinal > 1 &&
groupValue == 1 && // one without a leading ten
(oddGroup || indexFromLeft == 0)) { // preceding (100) or leading the sequence
unitsValue = 0;
}
// put it all together...
if (tensValue) {
// map onto Ethiopic "tens":
result.Append((PRUnichar) (tensValue + ETHIOPIC_TEN - 1));
}
if (unitsValue) {
//map onto Ethiopic "units":
result.Append((PRUnichar) (unitsValue + ETHIOPIC_ONE - 1));
}
// Add a separator for all even groups except the last,
// and for odd groups with non-zero value.
if (oddGroup) {
if (groupValue) {
result.Append((PRUnichar) ETHIOPIC_HUNDRED);
}
} else {
if (groupIndexFromRight) {
result.Append((PRUnichar) ETHIOPIC_TEN_THOUSAND);
}
}
}
return PR_TRUE;
}
PRBool
nsBulletFrame::GetListItemText(nsIPresContext* aCX,
const nsStyleList& aListStyle,
nsString& result)
{
#ifdef IBMBIDI
const nsStyleVisibility* vis = GetStyleVisibility();
// XXX For some of these systems, "." is wrong! This should really be
// pushed down into the individual cases!
if (NS_STYLE_DIRECTION_RTL == vis->mDirection) {
result.Append(NS_LITERAL_STRING("."));
}
#endif // IBMBIDI
PRBool success = PR_TRUE;
switch (aListStyle.mListStyleType) {
case NS_STYLE_LIST_STYLE_DECIMAL:
case NS_STYLE_LIST_STYLE_OLD_DECIMAL:
default: // CSS2 say "A users agent that does not recognize a numbering system
// should use 'decimal'
success = DecimalToText(mOrdinal, result);
break;
case NS_STYLE_LIST_STYLE_DECIMAL_LEADING_ZERO:
success = DecimalLeadingZeroToText(mOrdinal, result);
break;
case NS_STYLE_LIST_STYLE_LOWER_ROMAN:
case NS_STYLE_LIST_STYLE_OLD_LOWER_ROMAN:
success = RomanToText(mOrdinal, result,
gLowerRomanCharsA, gLowerRomanCharsB);
break;
case NS_STYLE_LIST_STYLE_UPPER_ROMAN:
case NS_STYLE_LIST_STYLE_OLD_UPPER_ROMAN:
success = RomanToText(mOrdinal, result,
gUpperRomanCharsA, gUpperRomanCharsB);
break;
case NS_STYLE_LIST_STYLE_LOWER_ALPHA:
case NS_STYLE_LIST_STYLE_OLD_LOWER_ALPHA:
success = CharListToText(mOrdinal, result, gLowerAlphaChars, ALPHA_SIZE);
break;
case NS_STYLE_LIST_STYLE_UPPER_ALPHA:
case NS_STYLE_LIST_STYLE_OLD_UPPER_ALPHA:
success = CharListToText(mOrdinal, result, gUpperAlphaChars, ALPHA_SIZE);
break;
case NS_STYLE_LIST_STYLE_KATAKANA:
success = CharListToText(mOrdinal, result, gKatakanaChars,
KATAKANA_CHARS_SIZE);
break;
case NS_STYLE_LIST_STYLE_HIRAGANA:
success = CharListToText(mOrdinal, result, gHiraganaChars,
HIRAGANA_CHARS_SIZE);
break;
case NS_STYLE_LIST_STYLE_KATAKANA_IROHA:
success = CharListToText(mOrdinal, result, gKatakanaIrohaChars,
KATAKANA_IROHA_CHARS_SIZE);
break;
case NS_STYLE_LIST_STYLE_HIRAGANA_IROHA:
success = CharListToText(mOrdinal, result, gHiraganaIrohaChars,
HIRAGANA_IROHA_CHARS_SIZE);
break;
case NS_STYLE_LIST_STYLE_LOWER_GREEK:
success = CharListToText(mOrdinal, result, gLowerGreekChars ,
LOWER_GREEK_CHARS_SIZE);
break;
case NS_STYLE_LIST_STYLE_CJK_IDEOGRAPHIC:
case NS_STYLE_LIST_STYLE_MOZ_TRAD_CHINESE_INFORMAL:
success = CJKIdeographicToText(mOrdinal, result, gCJKIdeographicDigit1,
gCJKIdeographicUnit1,
gCJKIdeographic10KUnit1);
break;
case NS_STYLE_LIST_STYLE_MOZ_TRAD_CHINESE_FORMAL:
success = CJKIdeographicToText(mOrdinal, result, gCJKIdeographicDigit2,
gCJKIdeographicUnit2,
gCJKIdeographic10KUnit1);
break;
case NS_STYLE_LIST_STYLE_MOZ_SIMP_CHINESE_INFORMAL:
success = CJKIdeographicToText(mOrdinal, result, gCJKIdeographicDigit1,
gCJKIdeographicUnit1,
gCJKIdeographic10KUnit2);
break;
case NS_STYLE_LIST_STYLE_MOZ_SIMP_CHINESE_FORMAL:
success = CJKIdeographicToText(mOrdinal, result, gCJKIdeographicDigit3,
gCJKIdeographicUnit2,
gCJKIdeographic10KUnit2);
break;
case NS_STYLE_LIST_STYLE_MOZ_JAPANESE_INFORMAL:
success = CJKIdeographicToText(mOrdinal, result, gCJKIdeographicDigit1,
gCJKIdeographicUnit1,
gCJKIdeographic10KUnit3);
break;
case NS_STYLE_LIST_STYLE_MOZ_JAPANESE_FORMAL:
success = CJKIdeographicToText(mOrdinal, result, gCJKIdeographicDigit2,
gCJKIdeographicUnit2,
gCJKIdeographic10KUnit3);
break;
case NS_STYLE_LIST_STYLE_HEBREW:
success = HebrewToText(mOrdinal, result);
break;
case NS_STYLE_LIST_STYLE_ARMENIAN:
success = ArmenianToText(mOrdinal, result);
break;
case NS_STYLE_LIST_STYLE_GEORGIAN:
success = GeorgianToText(mOrdinal, result);
break;
case NS_STYLE_LIST_STYLE_MOZ_ARABIC_INDIC:
success = OtherDecimalToText(mOrdinal, 0x0660, result);
break;
case NS_STYLE_LIST_STYLE_MOZ_PERSIAN:
case NS_STYLE_LIST_STYLE_MOZ_URDU:
success = OtherDecimalToText(mOrdinal, 0x06f0, result);
break;
case NS_STYLE_LIST_STYLE_MOZ_DEVANAGARI:
success = OtherDecimalToText(mOrdinal, 0x0966, result);
break;
case NS_STYLE_LIST_STYLE_MOZ_GURMUKHI:
success = OtherDecimalToText(mOrdinal, 0x0a66, result);
break;
case NS_STYLE_LIST_STYLE_MOZ_GUJARATI:
success = OtherDecimalToText(mOrdinal, 0x0AE6, result);
break;
case NS_STYLE_LIST_STYLE_MOZ_ORIYA:
success = OtherDecimalToText(mOrdinal, 0x0B66, result);
break;
case NS_STYLE_LIST_STYLE_MOZ_KANNADA:
success = OtherDecimalToText(mOrdinal, 0x0CE6, result);
break;
case NS_STYLE_LIST_STYLE_MOZ_MALAYALAM:
success = OtherDecimalToText(mOrdinal, 0x0D66, result);
break;
case NS_STYLE_LIST_STYLE_MOZ_THAI:
success = OtherDecimalToText(mOrdinal, 0x0E50, result);
break;
case NS_STYLE_LIST_STYLE_MOZ_LAO:
success = OtherDecimalToText(mOrdinal, 0x0ED0, result);
break;
case NS_STYLE_LIST_STYLE_MOZ_MYANMAR:
success = OtherDecimalToText(mOrdinal, 0x1040, result);
break;
case NS_STYLE_LIST_STYLE_MOZ_KHMER:
success = OtherDecimalToText(mOrdinal, 0x17E0, result);
break;
case NS_STYLE_LIST_STYLE_MOZ_BENGALI:
success = OtherDecimalToText(mOrdinal, 0x09E6, result);
break;
case NS_STYLE_LIST_STYLE_MOZ_TELUGU:
success = OtherDecimalToText(mOrdinal, 0x0C66, result);
break;
case NS_STYLE_LIST_STYLE_MOZ_TAMIL:
success = TamilToText(mOrdinal, result);
break;
case NS_STYLE_LIST_STYLE_MOZ_CJK_HEAVENLY_STEM:
success = CharListToText(mOrdinal, result, gCJKHeavenlyStemChars,
CJK_HEAVENLY_STEM_CHARS_SIZE);
break;
case NS_STYLE_LIST_STYLE_MOZ_CJK_EARTHLY_BRANCH:
success = CharListToText(mOrdinal, result, gCJKEarthlyBranchChars,
CJK_EARTHLY_BRANCH_CHARS_SIZE);
break;
case NS_STYLE_LIST_STYLE_MOZ_HANGUL:
success = CharListToText(mOrdinal, result, gHangulChars, HANGUL_CHARS_SIZE);
break;
case NS_STYLE_LIST_STYLE_MOZ_HANGUL_CONSONANT:
success = CharListToText(mOrdinal, result, gHangulConsonantChars,
HANGUL_CONSONANT_CHARS_SIZE);
break;
case NS_STYLE_LIST_STYLE_MOZ_ETHIOPIC_HALEHAME:
success = CharListToText(mOrdinal, result, gEthiopicHalehameChars,
ETHIOPIC_HALEHAME_CHARS_SIZE);
break;
case NS_STYLE_LIST_STYLE_MOZ_ETHIOPIC_NUMERIC:
success = EthiopicToText(mOrdinal, result);
break;
case NS_STYLE_LIST_STYLE_MOZ_ETHIOPIC_HALEHAME_AM:
success = CharListToText(mOrdinal, result, gEthiopicHalehameAmChars,
ETHIOPIC_HALEHAME_AM_CHARS_SIZE);
break;
case NS_STYLE_LIST_STYLE_MOZ_ETHIOPIC_HALEHAME_TI_ER:
success = CharListToText(mOrdinal, result, gEthiopicHalehameTiErChars,
ETHIOPIC_HALEHAME_TI_ER_CHARS_SIZE);
break;
case NS_STYLE_LIST_STYLE_MOZ_ETHIOPIC_HALEHAME_TI_ET:
success = CharListToText(mOrdinal, result, gEthiopicHalehameTiEtChars,
ETHIOPIC_HALEHAME_TI_ET_CHARS_SIZE);
break;
}
// XXX For some of these systems, "." is wrong! This should really be
// pushed up into the cases...
#ifdef IBMBIDI
if (NS_STYLE_DIRECTION_RTL != vis->mDirection)
#endif // IBMBIDI
result.Append(NS_LITERAL_STRING("."));
return success;
}
#define MIN_BULLET_SIZE 5 // from laytext.c
#define MINMAX(_value,_min,_max) \
((_value) < (_min) \
? (_min) \
: ((_value) > (_max) \
? (_max) \
: (_value)))
void
nsBulletFrame::GetDesiredSize(nsIPresContext* aCX,
const nsHTMLReflowState& aReflowState,
nsHTMLReflowMetrics& aMetrics)
{
const nsStyleList* myList = GetStyleList();
nscoord ascent;
if (myList->mListStyleImage && mImageRequest) {
PRUint32 status;
mImageRequest->GetImageStatus(&status);
if (status & imgIRequest::STATUS_SIZE_AVAILABLE &&
!(status & imgIRequest::STATUS_ERROR)) {
nscoord widthConstraint = NS_INTRINSICSIZE;
nscoord heightConstraint = NS_INTRINSICSIZE;
PRBool fixedContentWidth = PR_FALSE;
PRBool fixedContentHeight = PR_FALSE;
nscoord minWidth, maxWidth, minHeight, maxHeight;
// Determine whether the image has fixed content width
widthConstraint = aReflowState.mComputedWidth;
minWidth = aReflowState.mComputedMinWidth;
maxWidth = aReflowState.mComputedMaxWidth;
if (widthConstraint != NS_INTRINSICSIZE) {
fixedContentWidth = PR_TRUE;
}
// Determine whether the image has fixed content height
heightConstraint = aReflowState.mComputedHeight;
minHeight = aReflowState.mComputedMinHeight;
maxHeight = aReflowState.mComputedMaxHeight;
if (heightConstraint != NS_UNCONSTRAINEDSIZE) {
fixedContentHeight = PR_TRUE;
}
PRBool haveComputedSize = PR_FALSE;
PRBool needIntrinsicImageSize = PR_FALSE;
nscoord newWidth=0, newHeight=0;
if (fixedContentWidth) {
newWidth = MINMAX(widthConstraint, minWidth, maxWidth);
if (fixedContentHeight) {
newHeight = MINMAX(heightConstraint, minHeight, maxHeight);
haveComputedSize = PR_TRUE;
} else {
// We have a width, and an auto height. Compute height from
// width once we have the intrinsic image size.
if (mIntrinsicSize.height != 0) {
newHeight = (mIntrinsicSize.height * newWidth) / mIntrinsicSize.width;
haveComputedSize = PR_TRUE;
} else {
newHeight = 0;
needIntrinsicImageSize = PR_TRUE;
}
}
} else if (fixedContentHeight) {
// We have a height, and an auto width. Compute width from height
// once we have the intrinsic image size.
newHeight = MINMAX(heightConstraint, minHeight, maxHeight);
if (mIntrinsicSize.width != 0) {
newWidth = (mIntrinsicSize.width * newHeight) / mIntrinsicSize.height;
haveComputedSize = PR_TRUE;
} else {
newWidth = 0;
needIntrinsicImageSize = PR_TRUE;
}
} else {
// auto size the image
if (mIntrinsicSize.width == 0 && mIntrinsicSize.height == 0)
needIntrinsicImageSize = PR_TRUE;
else
haveComputedSize = PR_TRUE;
newWidth = mIntrinsicSize.width;
newHeight = mIntrinsicSize.height;
}
mComputedSize.width = newWidth;
mComputedSize.height = newHeight;
#if 0 // don't do scaled images in bullets
if (mComputedSize == mIntrinsicSize) {
mTransform.SetToIdentity();
} else {
if (mComputedSize.width != 0 && mComputedSize.height != 0) {
mTransform.SetToScale(float(mIntrinsicSize.width) / float(mComputedSize.width),
float(mIntrinsicSize.height) / float(mComputedSize.height));
}
}
#endif
aMetrics.width = mComputedSize.width;
aMetrics.height = mComputedSize.height;
aMetrics.ascent = aMetrics.height;
aMetrics.descent = 0;
return;
}
}
const nsStyleFont* myFont = GetStyleFont();
nsCOMPtr<nsIFontMetrics> fm;
aCX->GetMetricsFor(myFont->mFont, getter_AddRefs(fm));
nscoord bulletSize;
float p2t;
float t2p;
nsAutoString text;
switch (myList->mListStyleType) {
case NS_STYLE_LIST_STYLE_NONE:
aMetrics.width = 0;
aMetrics.height = 0;
aMetrics.ascent = 0;
aMetrics.descent = 0;
break;
case NS_STYLE_LIST_STYLE_DISC:
case NS_STYLE_LIST_STYLE_CIRCLE:
case NS_STYLE_LIST_STYLE_BASIC:
case NS_STYLE_LIST_STYLE_SQUARE:
aCX->GetTwipsToPixels(&t2p);
fm->GetMaxAscent(ascent);
bulletSize = NSTwipsToIntPixels(
(nscoord)NSToIntRound(0.8f * (float(ascent) / 2.0f)), t2p);
if (bulletSize < 1) {
bulletSize = MIN_BULLET_SIZE;
}
aCX->GetPixelsToTwips(&p2t);
bulletSize = NSIntPixelsToTwips(bulletSize, p2t);
mPadding.bottom = NSIntPixelsToTwips((nscoord) NSToIntRound((float)ascent / (8.0f * p2t)),p2t);
aMetrics.width = mPadding.right + bulletSize;
aMetrics.height = mPadding.bottom + bulletSize;
aMetrics.ascent = mPadding.bottom + bulletSize;
aMetrics.descent = 0;
break;
default:
case NS_STYLE_LIST_STYLE_DECIMAL_LEADING_ZERO:
case NS_STYLE_LIST_STYLE_DECIMAL:
case NS_STYLE_LIST_STYLE_OLD_DECIMAL:
case NS_STYLE_LIST_STYLE_LOWER_ROMAN:
case NS_STYLE_LIST_STYLE_UPPER_ROMAN:
case NS_STYLE_LIST_STYLE_LOWER_ALPHA:
case NS_STYLE_LIST_STYLE_UPPER_ALPHA:
case NS_STYLE_LIST_STYLE_OLD_LOWER_ROMAN:
case NS_STYLE_LIST_STYLE_OLD_UPPER_ROMAN:
case NS_STYLE_LIST_STYLE_OLD_LOWER_ALPHA:
case NS_STYLE_LIST_STYLE_OLD_UPPER_ALPHA:
case NS_STYLE_LIST_STYLE_KATAKANA:
case NS_STYLE_LIST_STYLE_HIRAGANA:
case NS_STYLE_LIST_STYLE_KATAKANA_IROHA:
case NS_STYLE_LIST_STYLE_HIRAGANA_IROHA:
case NS_STYLE_LIST_STYLE_LOWER_GREEK:
case NS_STYLE_LIST_STYLE_HEBREW:
case NS_STYLE_LIST_STYLE_ARMENIAN:
case NS_STYLE_LIST_STYLE_GEORGIAN:
case NS_STYLE_LIST_STYLE_CJK_IDEOGRAPHIC:
case NS_STYLE_LIST_STYLE_MOZ_SIMP_CHINESE_INFORMAL:
case NS_STYLE_LIST_STYLE_MOZ_SIMP_CHINESE_FORMAL:
case NS_STYLE_LIST_STYLE_MOZ_TRAD_CHINESE_INFORMAL:
case NS_STYLE_LIST_STYLE_MOZ_TRAD_CHINESE_FORMAL:
case NS_STYLE_LIST_STYLE_MOZ_JAPANESE_INFORMAL:
case NS_STYLE_LIST_STYLE_MOZ_JAPANESE_FORMAL:
case NS_STYLE_LIST_STYLE_MOZ_CJK_HEAVENLY_STEM:
case NS_STYLE_LIST_STYLE_MOZ_CJK_EARTHLY_BRANCH:
case NS_STYLE_LIST_STYLE_MOZ_ARABIC_INDIC:
case NS_STYLE_LIST_STYLE_MOZ_PERSIAN:
case NS_STYLE_LIST_STYLE_MOZ_URDU:
case NS_STYLE_LIST_STYLE_MOZ_DEVANAGARI:
case NS_STYLE_LIST_STYLE_MOZ_GURMUKHI:
case NS_STYLE_LIST_STYLE_MOZ_GUJARATI:
case NS_STYLE_LIST_STYLE_MOZ_ORIYA:
case NS_STYLE_LIST_STYLE_MOZ_KANNADA:
case NS_STYLE_LIST_STYLE_MOZ_MALAYALAM:
case NS_STYLE_LIST_STYLE_MOZ_BENGALI:
case NS_STYLE_LIST_STYLE_MOZ_TAMIL:
case NS_STYLE_LIST_STYLE_MOZ_TELUGU:
case NS_STYLE_LIST_STYLE_MOZ_THAI:
case NS_STYLE_LIST_STYLE_MOZ_LAO:
case NS_STYLE_LIST_STYLE_MOZ_MYANMAR:
case NS_STYLE_LIST_STYLE_MOZ_KHMER:
case NS_STYLE_LIST_STYLE_MOZ_HANGUL:
case NS_STYLE_LIST_STYLE_MOZ_HANGUL_CONSONANT:
case NS_STYLE_LIST_STYLE_MOZ_ETHIOPIC_HALEHAME:
case NS_STYLE_LIST_STYLE_MOZ_ETHIOPIC_NUMERIC:
case NS_STYLE_LIST_STYLE_MOZ_ETHIOPIC_HALEHAME_AM:
case NS_STYLE_LIST_STYLE_MOZ_ETHIOPIC_HALEHAME_TI_ER:
case NS_STYLE_LIST_STYLE_MOZ_ETHIOPIC_HALEHAME_TI_ET:
GetListItemText(aCX, *myList, text);
fm->GetHeight(aMetrics.height);
aReflowState.rendContext->SetFont(fm);
aReflowState.rendContext->GetWidth(text, aMetrics.width);
aMetrics.width += mPadding.right;
fm->GetMaxAscent(aMetrics.ascent);
fm->GetMaxDescent(aMetrics.descent);
break;
}
}
NS_IMETHODIMP
nsBulletFrame::Reflow(nsIPresContext* aPresContext,
nsHTMLReflowMetrics& aMetrics,
const nsHTMLReflowState& aReflowState,
nsReflowStatus& aStatus)
{
DO_GLOBAL_REFLOW_COUNT("nsBulletFrame", aReflowState.reason);
DISPLAY_REFLOW(aPresContext, this, aReflowState, aMetrics, aStatus);
PRBool isStyleChange = PR_FALSE;
if (eReflowReason_StyleChange == aReflowState.reason)
isStyleChange = PR_TRUE;
else if (eReflowReason_Incremental == aReflowState.reason) {
// XXXwaterson The reflow had better be targeted at this frame.
nsReflowType type;
aReflowState.path->mReflowCommand->GetType(type);
/* if the style changed, see if we need to load a new url */
if (eReflowType_StyleChanged == type)
isStyleChange = PR_TRUE;
}
if (isStyleChange) {
nsIURI *newURI = GetStyleList()->mListStyleImage;
if (newURI) {
if (!mListener) {
nsBulletListener *listener;
NS_NEWXPCOM(listener, nsBulletListener);
NS_ADDREF(listener);
listener->SetFrame(this);
listener->QueryInterface(NS_GET_IID(imgIDecoderObserver), getter_AddRefs(mListener));
NS_ASSERTION(mListener, "queryinterface for the listener failed");
NS_RELEASE(listener);
}
PRBool needNewRequest = PR_TRUE;
if (mImageRequest) {
// Reload the image, maybe...
nsCOMPtr<nsIURI> oldURI;
mImageRequest->GetURI(getter_AddRefs(oldURI));
if (oldURI) {
PRBool same;
newURI->Equals(oldURI, &same);
if (same) {
needNewRequest = PR_FALSE;
} else {
mImageRequest->Cancel(NS_ERROR_FAILURE);
mImageRequest = nsnull;
}
}
}
if (needNewRequest) {
nsresult rv;
nsCOMPtr<imgILoader> il(do_GetService("@mozilla.org/image/loader;1", &rv));
if (NS_FAILED(rv))
return rv;
nsCOMPtr<nsILoadGroup> loadGroup;
GetLoadGroup(aPresContext, getter_AddRefs(loadGroup));
// Get the document URI for the referrer...
nsIURI* documentURI = nsnull;
nsCOMPtr<nsIDocument> doc;
if (mContent) {
doc = mContent->GetDocument();
if (doc) {
documentURI = doc->GetDocumentURL();
}
}
// XXX: initialDocumentURI is NULL !
il->LoadImage(newURI, nsnull, documentURI, loadGroup, mListener, doc,
nsIRequest::LOAD_NORMAL, nsnull, nsnull,
getter_AddRefs(mImageRequest));
}
}
}
// Get the base size
GetDesiredSize(aPresContext, aReflowState, aMetrics);
// Add in the border and padding; split the top/bottom between the
// ascent and descent to make things look nice
const nsMargin& borderPadding = aReflowState.mComputedBorderPadding;
aMetrics.width += borderPadding.left + borderPadding.right;
aMetrics.height += borderPadding.top + borderPadding.bottom;
aMetrics.ascent += borderPadding.top;
aMetrics.descent += borderPadding.bottom;
if (aMetrics.mComputeMEW) {
aMetrics.mMaxElementWidth = aMetrics.width;
}
aStatus = NS_FRAME_COMPLETE;
NS_FRAME_SET_TRUNCATION(aStatus, aReflowState, aMetrics);
return NS_OK;
}
NS_IMETHODIMP nsBulletFrame::OnStartContainer(imgIRequest *aRequest,
imgIContainer *aImage)
{
if (!aImage) return NS_ERROR_INVALID_ARG;
if (!aRequest) return NS_ERROR_INVALID_ARG;
NS_ENSURE_TRUE(mPresContext, NS_ERROR_UNEXPECTED); // Why are we bothering?
PRUint32 status;
aRequest->GetImageStatus(&status);
if (status & imgIRequest::STATUS_ERROR) {
return NS_OK;
}
nscoord w, h;
aImage->GetWidth(&w);
aImage->GetHeight(&h);
float p2t;
mPresContext->GetPixelsToTwips(&p2t);
nsSize newsize(NSIntPixelsToTwips(w, p2t), NSIntPixelsToTwips(h, p2t));
if (mIntrinsicSize != newsize) {
mIntrinsicSize = newsize;
// Now that the size is available (or an error occurred), trigger
// a reflow of the bullet frame.
nsCOMPtr<nsIPresShell> shell;
nsresult rv = mPresContext->GetShell(getter_AddRefs(shell));
if (NS_SUCCEEDED(rv) && shell) {
NS_ASSERTION(mParent, "No parent to pass the reflow request up to.");
if (mParent) {
// Reflow the first child of the parent not the bullet frame.
// The bullet frame is not in a line list so marking it dirty
// has no effect. The reflowing of the bullet frame is done
// indirectly.
nsIFrame* frame = nsnull;
mParent->FirstChild(mPresContext, nsnull, &frame);
NS_ASSERTION(frame, "No frame to mark dirty for bullet frame.");
if (frame) {
frame->AddStateBits(NS_FRAME_IS_DIRTY);
mParent->ReflowDirtyChild(shell, frame);
}
}
}
}
return NS_OK;
}
NS_IMETHODIMP nsBulletFrame::OnDataAvailable(imgIRequest *aRequest,
gfxIImageFrame *aFrame,
const nsRect *aRect)
{
if (!aRect) return NS_ERROR_NULL_POINTER;
NS_ENSURE_TRUE(mPresContext, NS_ERROR_UNEXPECTED); // Why are we bothering?
// XXX Should we do anything here if an error occured in the decode?
nsRect r(*aRect);
// XXX what if this frame ever has a padding or border?
float p2t;
mPresContext->GetPixelsToTwips(&p2t);
r.x = NSIntPixelsToTwips(r.x, p2t);
r.y = NSIntPixelsToTwips(r.y, p2t);
r.width = NSIntPixelsToTwips(r.width, p2t);
r.height = NSIntPixelsToTwips(r.height, p2t);
Invalidate(mPresContext, r, PR_FALSE);
return NS_OK;
}
NS_IMETHODIMP nsBulletFrame::OnStopDecode(imgIRequest *aRequest,
nsresult aStatus,
const PRUnichar *aStatusArg)
{
// XXX should the bulletframe do anything if the image failed to load?
// it didn't in the old code...
#if 0
NS_ENSURE_TRUE(mPresContext, NS_ERROR_UNEXPECTED); // Why are we bothering?
nsCOMPtr<nsIPresShell> presShell;
mPresContext->GetShell(getter_AddRefs(presShell));
if (NS_FAILED(aStatus)) {
// We failed to load the image. Notify the pres shell
if (NS_FAILED(aStatus) && (mImageRequest == aRequest || !mImageRequest)) {
imageFailed = PR_TRUE;
}
}
#endif
return NS_OK;
}
NS_IMETHODIMP nsBulletFrame::FrameChanged(imgIContainer *aContainer,
gfxIImageFrame *aNewFrame,
nsRect *aDirtyRect)
{
NS_ENSURE_TRUE(mPresContext, NS_ERROR_UNEXPECTED); // Why are we bothering?
nsRect r(*aDirtyRect);
float p2t;
mPresContext->GetPixelsToTwips(&p2t);
r.x = NSIntPixelsToTwips(r.x, p2t);
r.y = NSIntPixelsToTwips(r.y, p2t);
r.width = NSIntPixelsToTwips(r.width, p2t);
r.height = NSIntPixelsToTwips(r.height, p2t);
Invalidate(mPresContext, r, PR_FALSE);
return NS_OK;
}
void
nsBulletFrame::GetLoadGroup(nsIPresContext *aPresContext, nsILoadGroup **aLoadGroup)
{
if (!aPresContext)
return;
NS_PRECONDITION(nsnull != aLoadGroup, "null OUT parameter pointer");
nsCOMPtr<nsIPresShell> shell;
aPresContext->GetShell(getter_AddRefs(shell));
if (!shell)
return;
nsCOMPtr<nsIDocument> doc;
shell->GetDocument(getter_AddRefs(doc));
if (!doc)
return;
*aLoadGroup = doc->GetDocumentLoadGroup().get(); // already_AddRefed
}
NS_IMPL_ISUPPORTS2(nsBulletListener, imgIDecoderObserver, imgIContainerObserver)
nsBulletListener::nsBulletListener() :
mFrame(nsnull)
{
}
nsBulletListener::~nsBulletListener()
{
}
NS_IMETHODIMP nsBulletListener::OnStartDecode(imgIRequest *aRequest)
{
return NS_OK;
}
NS_IMETHODIMP nsBulletListener::OnStartContainer(imgIRequest *aRequest,
imgIContainer *aImage)
{
if (!mFrame)
return NS_ERROR_FAILURE;
return mFrame->OnStartContainer(aRequest, aImage);
}
NS_IMETHODIMP nsBulletListener::OnStartFrame(imgIRequest *aRequest,
gfxIImageFrame *aFrame)
{
return NS_OK;
}
NS_IMETHODIMP nsBulletListener::OnDataAvailable(imgIRequest *aRequest,
gfxIImageFrame *aFrame,
const nsRect *aRect)
{
if (!mFrame)
return NS_ERROR_FAILURE;
return mFrame->OnDataAvailable(aRequest, aFrame, aRect);
}
NS_IMETHODIMP nsBulletListener::OnStopFrame(imgIRequest *aRequest,
gfxIImageFrame *aFrame)
{
return NS_OK;
}
NS_IMETHODIMP nsBulletListener::OnStopContainer(imgIRequest *aRequest,
imgIContainer *aImage)
{
return NS_OK;
}
NS_IMETHODIMP nsBulletListener::OnStopDecode(imgIRequest *aRequest,
nsresult status,
const PRUnichar *statusArg)
{
if (!mFrame)
return NS_ERROR_FAILURE;
return mFrame->OnStopDecode(aRequest, status, statusArg);
}
NS_IMETHODIMP nsBulletListener::FrameChanged(imgIContainer *aContainer,
gfxIImageFrame *newframe,
nsRect * dirtyRect)
{
if (!mFrame)
return NS_ERROR_FAILURE;
return mFrame->FrameChanged(aContainer, newframe, dirtyRect);
}