gecko/content/base/public/nsContentUtils.h

1789 lines
66 KiB
C++

/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*-
*
* ***** BEGIN LICENSE BLOCK *****
* Version: MPL 1.1/GPL 2.0/LGPL 2.1
*
* The contents of this file are subject to the Mozilla 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/MPL/
*
* 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, released
* March 31, 1998.
*
* 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 of 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 MPL, 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 MPL, the GPL or the LGPL.
*
* ***** END LICENSE BLOCK ***** */
/* A namespace class for static content utilities. */
#ifndef nsContentUtils_h___
#define nsContentUtils_h___
#include "jsprvtd.h"
#include "jsnum.h"
#include "nsAString.h"
#include "nsIStatefulFrame.h"
#include "nsIPref.h"
#include "nsINodeInfo.h"
#include "nsNodeInfoManager.h"
#include "nsContentList.h"
#include "nsDOMClassInfoID.h"
#include "nsIClassInfo.h"
#include "nsIDOM3Node.h"
#include "nsDataHashtable.h"
#include "nsIScriptRuntime.h"
#include "nsIScriptGlobalObject.h"
#include "nsIDOMEvent.h"
#include "nsTArray.h"
#include "nsTextFragment.h"
#include "nsReadableUtils.h"
struct nsNativeKeyEvent; // Don't include nsINativeKeyBindings.h here: it will force strange compilation error!
class nsIDOMScriptObjectFactory;
class nsIXPConnect;
class nsINode;
class nsIContent;
class nsIDOMNode;
class nsIDOMKeyEvent;
class nsIDocument;
class nsIDocumentObserver;
class nsIDocShell;
class nsINameSpaceManager;
class nsIScriptSecurityManager;
class nsIJSContextStack;
class nsIThreadJSContextStack;
class nsIParserService;
class nsIIOService;
class nsIURI;
class imgIContainer;
class imgIDecoderObserver;
class imgIRequest;
class imgILoader;
class imgICache;
class nsIPrefBranch;
class nsIImageLoadingContent;
class nsIDOMHTMLFormElement;
class nsIDOMDocument;
class nsIConsoleService;
class nsIStringBundleService;
class nsIStringBundle;
class nsIContentPolicy;
class nsILineBreaker;
class nsIWordBreaker;
class nsIJSRuntimeService;
class nsIEventListenerManager;
class nsIScriptContext;
class nsIRunnable;
class nsIInterfaceRequestor;
template<class E> class nsCOMArray;
class nsIPref;
struct JSRuntime;
class nsICaseConversion;
class nsIUGenCategory;
class nsIWidget;
class nsIDragSession;
class nsPIDOMWindow;
class nsPIDOMEventTarget;
class nsIPresShell;
class nsIXPConnectJSObjectHolder;
#ifdef MOZ_XTF
class nsIXTFService;
#endif
#ifdef IBMBIDI
class nsIBidiKeyboard;
#endif
class nsIMIMEHeaderParam;
extern const char kLoadAsData[];
enum EventNameType {
EventNameType_None = 0x0000,
EventNameType_HTML = 0x0001,
EventNameType_XUL = 0x0002,
EventNameType_SVGGraphic = 0x0004, // svg graphic elements
EventNameType_SVGSVG = 0x0008, // the svg element
EventNameType_HTMLXUL = 0x0003,
EventNameType_All = 0xFFFF
};
struct EventNameMapping {
PRUint32 mId;
PRInt32 mType;
};
struct nsShortcutCandidate {
nsShortcutCandidate(PRUint32 aCharCode, PRBool aIgnoreShift) :
mCharCode(aCharCode), mIgnoreShift(aIgnoreShift)
{
}
PRUint32 mCharCode;
PRBool mIgnoreShift;
};
class nsContentUtils
{
public:
static nsresult Init();
// You MUST pass the old ownerDocument of aContent in as aOldDocument and the
// new one as aNewDocument. aNewParent is allowed to be null; in that case
// aNewDocument will be assumed to be the parent. Note that at this point
// the actual ownerDocument of aContent may not yet be aNewDocument.
// XXXbz but then if it gets wrapped after we do this call but before its
// ownerDocument actually changes, things will break...
static nsresult ReparentContentWrapper(nsIContent *aNode,
nsIContent *aNewParent,
nsIDocument *aNewDocument,
nsIDocument *aOldDocument);
/**
* Get a scope from aOldDocument and one from aNewDocument. Also get a
* context through one of the scopes, from the stack or the safe context.
*
* @param aOldDocument The document to get aOldScope from.
* @param aNewDocument The document to get aNewScope from.
* @param aCx [out] Context gotten through one of the scopes, from the stack
* or the safe context.
* @param aOldScope [out] Scope gotten from aOldDocument.
* @param aNewScope [out] Scope gotten from aNewDocument.
*/
static nsresult GetContextAndScopes(nsIDocument *aOldDocument,
nsIDocument *aNewDocument,
JSContext **aCx, JSObject **aOldScope,
JSObject **aNewScope);
/**
* When a document's scope changes (e.g., from document.open(), call this
* function to move all content wrappers from the old scope to the new one.
*/
static nsresult ReparentContentWrappersInScope(nsIScriptGlobalObject *aOldScope,
nsIScriptGlobalObject *aNewScope);
static PRBool IsCallerChrome();
static PRBool IsCallerTrustedForRead();
static PRBool IsCallerTrustedForWrite();
/**
* Check whether a caller is trusted to have aCapability. This also
* checks for UniversalXPConnect in addition to aCapability.
*/
static PRBool IsCallerTrustedForCapability(const char* aCapability);
/**
* Do not ever pass null pointers to this method. If one of your
* nsIContents is null, you have to decide for yourself what
* "IsDescendantOf" really means.
*
* @param aPossibleDescendant node to test for being a descendant of
* aPossibleAncestor
* @param aPossibleAncestor node to test for being an ancestor of
* aPossibleDescendant
* @return PR_TRUE if aPossibleDescendant is a descendant of
* aPossibleAncestor (or is aPossibleAncestor). PR_FALSE
* otherwise.
*/
static PRBool ContentIsDescendantOf(nsINode* aPossibleDescendant,
nsINode* aPossibleAncestor);
/**
* Similar to ContentIsDescendantOf except it crosses document boundaries.
*/
static PRBool ContentIsCrossDocDescendantOf(nsINode* aPossibleDescendant,
nsINode* aPossibleAncestor);
/*
* This method fills the |aArray| with all ancestor nodes of |aNode|
* including |aNode| at the zero index.
*/
static nsresult GetAncestors(nsIDOMNode* aNode,
nsTArray<nsIDOMNode*>* aArray);
/*
* This method fills |aAncestorNodes| with all ancestor nodes of |aNode|
* including |aNode| (QI'd to nsIContent) at the zero index.
* For each ancestor, there is a corresponding element in |aAncestorOffsets|
* which is the IndexOf the child in relation to its parent.
*
* This method just sucks.
*/
static nsresult GetAncestorsAndOffsets(nsIDOMNode* aNode,
PRInt32 aOffset,
nsTArray<nsIContent*>* aAncestorNodes,
nsTArray<PRInt32>* aAncestorOffsets);
/*
* The out parameter, |aCommonAncestor| will be the closest node, if any,
* to both |aNode| and |aOther| which is also an ancestor of each.
* Returns an error if the two nodes are disconnected and don't have
* a common ancestor.
*/
static nsresult GetCommonAncestor(nsIDOMNode *aNode,
nsIDOMNode *aOther,
nsIDOMNode** aCommonAncestor);
/**
* Returns the common ancestor, if any, for two nodes. Returns null if the
* nodes are disconnected.
*/
static nsINode* GetCommonAncestor(nsINode* aNode1,
nsINode* aNode2);
/**
* Compares the document position of nodes.
*
* @param aNode1 The node whose position is being compared to the reference
* node
* @param aNode2 The reference node
*
* @return The document position flags of the nodes. aNode1 is compared to
* aNode2, i.e. if aNode1 is before aNode2 then
* DOCUMENT_POSITION_PRECEDING will be set.
*
* @see nsIDOMNode
* @see nsIDOM3Node
*/
static PRUint16 ComparePosition(nsINode* aNode1,
nsINode* aNode2);
/**
* Returns true if aNode1 is before aNode2 in the same connected
* tree.
*/
static PRBool PositionIsBefore(nsINode* aNode1,
nsINode* aNode2)
{
return (ComparePosition(aNode1, aNode2) &
(nsIDOM3Node::DOCUMENT_POSITION_PRECEDING |
nsIDOM3Node::DOCUMENT_POSITION_DISCONNECTED)) ==
nsIDOM3Node::DOCUMENT_POSITION_PRECEDING;
}
/**
* Utility routine to compare two "points", where a point is a
* node/offset pair
* Returns -1 if point1 < point2, 1, if point1 > point2,
* 0 if error or if point1 == point2.
* NOTE! If the two nodes aren't in the same connected subtree,
* the result is undefined, but the optional aDisconnected parameter
* is set to PR_TRUE.
*/
static PRInt32 ComparePoints(nsINode* aParent1, PRInt32 aOffset1,
nsINode* aParent2, PRInt32 aOffset2,
PRBool* aDisconnected = nsnull);
/**
* Find the first child of aParent with a resolved tag matching
* aNamespace and aTag. Both the explicit and anonymous children of
* aParent are examined. The return value is not addrefed.
*
* XXXndeakin this should return the first child whether in anonymous or
* explicit children, but currently XBL doesn't tell us the relative
* ordering of anonymous vs explicit children, so instead it searches
* the explicit children first then the anonymous children.
*/
static nsIContent* FindFirstChildWithResolvedTag(nsIContent* aParent,
PRInt32 aNamespace,
nsIAtom* aTag);
/**
* Brute-force search of the element subtree rooted at aContent for
* an element with the given id. aId must be nonempty, otherwise
* this method may return nodes even if they have no id!
*/
static nsIContent* MatchElementId(nsIContent *aContent,
const nsAString& aId);
/**
* Similar to above, but to be used if one already has an atom for the ID
*/
static nsIContent* MatchElementId(nsIContent *aContent,
nsIAtom* aId);
/**
* Given a URI containing an element reference (#whatever),
* resolve it to the target content element with the given ID.
*
* If aFromContent is anonymous XBL content then the URI
* must refer to its binding document and we will return
* a node in the same anonymous content subtree as aFromContent,
* if one exists with the correct ID.
*
* @param aFromContent the context of the reference;
* currently we only support references to elements in the
* same document as the context, so this must be non-null
*
* @return the element, or nsnull on failure
*/
static nsIContent* GetReferencedElement(nsIURI* aURI,
nsIContent *aFromContent);
/**
* Reverses the document position flags passed in.
*
* @param aDocumentPosition The document position flags to be reversed.
*
* @return The reversed document position flags.
*
* @see nsIDOMNode
* @see nsIDOM3Node
*/
static PRUint16 ReverseDocumentPosition(PRUint16 aDocumentPosition);
static PRUint32 CopyNewlineNormalizedUnicodeTo(const nsAString& aSource,
PRUint32 aSrcOffset,
PRUnichar* aDest,
PRUint32 aLength,
PRBool& aLastCharCR);
static PRUint32 CopyNewlineNormalizedUnicodeTo(nsReadingIterator<PRUnichar>& aSrcStart, const nsReadingIterator<PRUnichar>& aSrcEnd, nsAString& aDest);
static nsISupports *
GetClassInfoInstance(nsDOMClassInfoID aID);
static const nsDependentSubstring TrimCharsInSet(const char* aSet,
const nsAString& aValue);
static const nsDependentSubstring TrimWhitespace(const nsAString& aStr,
PRBool aTrimTrailing = PR_TRUE);
/**
* Returns true if aChar is of class Ps, Pi, Po, Pf, or Pe.
*/
static PRBool IsPunctuationMark(PRUint32 aChar);
static PRBool IsPunctuationMarkAt(const nsTextFragment* aFrag, PRUint32 aOffset);
/**
* Returns true if aChar is of class Lu, Ll, Lt, Lm, Lo, Nd, Nl or No
*/
static PRBool IsAlphanumeric(PRUint32 aChar);
static PRBool IsAlphanumericAt(const nsTextFragment* aFrag, PRUint32 aOffset);
/*
* Is the character an HTML whitespace character?
*
* We define whitespace using the list in HTML5 and css3-selectors:
* U+0009, U+000A, U+000C, U+000D, U+0020
*
* HTML 4.01 also lists U+200B (zero-width space).
*/
static PRBool IsHTMLWhitespace(PRUnichar aChar);
static void Shutdown();
/**
* Checks whether two nodes come from the same origin. aTrustedNode is
* considered 'safe' in that a user can operate on it and that it isn't
* a js-object that implements nsIDOMNode.
* Never call this function with the first node provided by script, it
* must always be known to be a 'real' node!
*/
static nsresult CheckSameOrigin(nsIDOMNode* aTrustedNode,
nsIDOMNode* aUnTrustedNode);
// Check if the (JS) caller can access aNode.
static PRBool CanCallerAccess(nsIDOMNode *aNode);
// Check if the (JS) caller can access aWindow.
// aWindow can be either outer or inner window.
static PRBool CanCallerAccess(nsPIDOMWindow* aWindow);
/**
* Get the docshell through the JS context that's currently on the stack.
* If there's no JS context currently on the stack aDocShell will be null.
*
* @param aDocShell The docshell or null if no JS context
*/
static nsIDocShell *GetDocShellFromCaller();
/**
* The two GetDocumentFrom* functions below allow a caller to get at a
* document that is relevant to the currently executing script.
*
* GetDocumentFromCaller gets its document by looking at the last called
* function and finding the document that the function itself relates to.
* For example, consider two windows A and B in the same origin. B has a
* function which does something that ends up needing the current document.
* If a script in window A were to call B's function, GetDocumentFromCaller
* would find that function (in B) and return B's document.
*
* GetDocumentFromContext gets its document by looking at the currently
* executing context's global object and returning its document. Thus,
* given the example above, GetDocumentFromCaller would see that the
* currently executing script was in window A, and return A's document.
*/
/**
* Get the document from the currently executing function. This will return
* the document that the currently executing function is in/from.
*
* @return The document or null if no JS Context.
*/
static nsIDOMDocument *GetDocumentFromCaller();
/**
* Get the document through the JS context that's currently on the stack.
* If there's no JS context currently on the stack it will return null.
* This will return the document of the calling script.
*
* @return The document or null if no JS context
*/
static nsIDOMDocument *GetDocumentFromContext();
// Check if a node is in the document prolog, i.e. before the document
// element.
static PRBool InProlog(nsINode *aNode);
static nsIParserService* GetParserService();
static nsINameSpaceManager* NameSpaceManager()
{
return sNameSpaceManager;
}
static nsIIOService* GetIOService()
{
return sIOService;
}
static imgILoader* GetImgLoader()
{
return sImgLoader;
}
#ifdef MOZ_XTF
static nsIXTFService* GetXTFService();
#endif
#ifdef IBMBIDI
static nsIBidiKeyboard* GetBidiKeyboard();
#endif
/**
* Get the cache security manager service. Can return null if the layout
* module has been shut down.
*/
static nsIScriptSecurityManager* GetSecurityManager()
{
return sSecurityManager;
}
static nsresult GenerateStateKey(nsIContent* aContent,
nsIDocument* aDocument,
nsIStatefulFrame::SpecialStateID aID,
nsACString& aKey);
/**
* Create a new nsIURI from aSpec, using aBaseURI as the base. The
* origin charset of the new nsIURI will be the document charset of
* aDocument.
*/
static nsresult NewURIWithDocumentCharset(nsIURI** aResult,
const nsAString& aSpec,
nsIDocument* aDocument,
nsIURI* aBaseURI);
/**
* Convert aInput (in charset aCharset) to UTF16 in aOutput.
*
* @param aCharset the name of the charset; if empty, we assume UTF8
*/
static nsresult ConvertStringFromCharset(const nsACString& aCharset,
const nsACString& aInput,
nsAString& aOutput);
/**
* Determine whether a buffer begins with a BOM for UTF-8, UTF-16LE,
* UTF-16BE, UTF-32LE, UTF-32BE.
*
* @param aBuffer the buffer to check
* @param aLength the length of the buffer
* @param aCharset empty if not found
* @return boolean indicating whether a BOM was detected.
*/
static PRBool CheckForBOM(const unsigned char* aBuffer, PRUint32 aLength,
nsACString& aCharset, PRBool *bigEndian = nsnull);
/**
* Determine whether aContent is in some way associated with aForm. If the
* form is a container the only elements that are considered to be associated
* with a form are the elements that are contained within the form. If the
* form is a leaf element then all elements will be accepted into this list,
* since this can happen due to content fixup when a form spans table rows or
* table cells.
*/
static PRBool BelongsInForm(nsIDOMHTMLFormElement *aForm,
nsIContent *aContent);
static nsresult CheckQName(const nsAString& aQualifiedName,
PRBool aNamespaceAware = PR_TRUE);
static nsresult SplitQName(nsIContent* aNamespaceResolver,
const nsAFlatString& aQName,
PRInt32 *aNamespace, nsIAtom **aLocalName);
static nsresult LookupNamespaceURI(nsIContent* aNamespaceResolver,
const nsAString& aNamespacePrefix,
nsAString& aNamespaceURI);
static nsresult GetNodeInfoFromQName(const nsAString& aNamespaceURI,
const nsAString& aQualifiedName,
nsNodeInfoManager* aNodeInfoManager,
nsINodeInfo** aNodeInfo);
static void SplitExpatName(const PRUnichar *aExpatName, nsIAtom **aPrefix,
nsIAtom **aTagName, PRInt32 *aNameSpaceID);
static nsAdoptingCString GetCharPref(const char *aPref);
static PRPackedBool GetBoolPref(const char *aPref,
PRBool aDefault = PR_FALSE);
static PRInt32 GetIntPref(const char *aPref, PRInt32 aDefault = 0);
static nsAdoptingString GetLocalizedStringPref(const char *aPref);
static nsAdoptingString GetStringPref(const char *aPref);
static void RegisterPrefCallback(const char *aPref,
PrefChangedFunc aCallback,
void * aClosure);
static void UnregisterPrefCallback(const char *aPref,
PrefChangedFunc aCallback,
void * aClosure);
static void AddBoolPrefVarCache(const char* aPref, PRBool* aVariable);
static nsIPrefBranch *GetPrefBranch()
{
return sPrefBranch;
}
static nsILineBreaker* LineBreaker()
{
return sLineBreaker;
}
static nsIWordBreaker* WordBreaker()
{
return sWordBreaker;
}
static nsICaseConversion* GetCaseConv()
{
return sCaseConv;
}
static nsIUGenCategory* GetGenCat()
{
return sGenCat;
}
/**
* @return PR_TRUE if aContent has an attribute aName in namespace aNameSpaceID,
* and the attribute value is non-empty.
*/
static PRBool HasNonEmptyAttr(nsIContent* aContent, PRInt32 aNameSpaceID,
nsIAtom* aName);
/**
* Method that gets the primary presContext for the node.
*
* @param aContent The content node.
* @return the presContext, or nsnull if the content is not in a document
* (if GetCurrentDoc returns nsnull)
*/
static nsPresContext* GetContextForContent(nsIContent* aContent);
/**
* Method to do security and content policy checks on the image URI
*
* @param aURI uri of the image to be loaded
* @param aContext the context the image is loaded in (eg an element)
* @param aLoadingDocument the document we belong to
* @param aLoadingPrincipal the principal doing the load
* @param aImageBlockingStatus the nsIContentPolicy blocking status for this
* image. This will be set even if a security check fails for the
* image, to some reasonable REJECT_* value. This out param will only
* be set if it's non-null.
* @return PR_TRUE if the load can proceed, or PR_FALSE if it is blocked.
* Note that aImageBlockingStatus, if set will always be an ACCEPT
* status if PR_TRUE is returned and always be a REJECT_* status if
* PR_FALSE is returned.
*/
static PRBool CanLoadImage(nsIURI* aURI,
nsISupports* aContext,
nsIDocument* aLoadingDocument,
nsIPrincipal* aLoadingPrincipal,
PRInt16* aImageBlockingStatus = nsnull);
/**
* Method to start an image load. This does not do any security checks.
* This method will attempt to make aURI immutable; a caller that wants to
* keep a mutable version around should pass in a clone.
*
* @param aURI uri of the image to be loaded
* @param aLoadingDocument the document we belong to
* @param aLoadingPrincipal the principal doing the load
* @param aReferrer the referrer URI
* @param aObserver the observer for the image load
* @param aLoadFlags the load flags to use. See nsIRequest
* @return the imgIRequest for the image load
*/
static nsresult LoadImage(nsIURI* aURI,
nsIDocument* aLoadingDocument,
nsIPrincipal* aLoadingPrincipal,
nsIURI* aReferrer,
imgIDecoderObserver* aObserver,
PRInt32 aLoadFlags,
imgIRequest** aRequest);
/**
* Returns whether the given URI is in the image cache.
*/
static PRBool IsImageInCache(nsIURI* aURI);
/**
* Method to get an imgIContainer from an image loading content
*
* @param aContent The image loading content. Must not be null.
* @param aRequest The image request [out]
* @return the imgIContainer corresponding to the first frame of the image
*/
static already_AddRefed<imgIContainer> GetImageFromContent(nsIImageLoadingContent* aContent, imgIRequest **aRequest = nsnull);
/**
* Method that decides whether a content node is draggable
*
* @param aContent The content node to test.
* @return whether it's draggable
*/
static PRBool ContentIsDraggable(nsIContent* aContent);
/**
* Method that decides whether a content node is a draggable image
*
* @param aContent The content node to test.
* @return whether it's a draggable image
*/
static PRBool IsDraggableImage(nsIContent* aContent);
/**
* Method that decides whether a content node is a draggable link
*
* @param aContent The content node to test.
* @return whether it's a draggable link
*/
static PRBool IsDraggableLink(nsIContent* aContent);
/**
* Convenience method to create a new nodeinfo that differs only by name
* from aNodeInfo.
*/
static nsresult NameChanged(nsINodeInfo *aNodeInfo, nsIAtom *aName,
nsINodeInfo** aResult)
{
nsNodeInfoManager *niMgr = aNodeInfo->NodeInfoManager();
*aResult = niMgr->GetNodeInfo(aName, aNodeInfo->GetPrefixAtom(),
aNodeInfo->NamespaceID()).get();
return *aResult ? NS_OK : NS_ERROR_OUT_OF_MEMORY;
}
/**
* Convenience method to create a new nodeinfo that differs only by prefix
* from aNodeInfo.
*/
static nsresult PrefixChanged(nsINodeInfo *aNodeInfo, nsIAtom *aPrefix,
nsINodeInfo** aResult)
{
nsNodeInfoManager *niMgr = aNodeInfo->NodeInfoManager();
*aResult = niMgr->GetNodeInfo(aNodeInfo->NameAtom(), aPrefix,
aNodeInfo->NamespaceID()).get();
return *aResult ? NS_OK : NS_ERROR_OUT_OF_MEMORY;
}
/**
* Returns the appropriate event argument names for the specified
* namespace and event name. Added because we need to switch between
* SVG's "evt" and the rest of the world's "event", and because onerror
* takes 3 args.
*/
static void GetEventArgNames(PRInt32 aNameSpaceID, nsIAtom *aEventName,
PRUint32 *aArgCount, const char*** aArgNames);
/**
* If aNode is not an element, return true exactly when aContent's binding
* parent is null.
*
* If aNode is an element, return true exactly when aContent's binding parent
* is the same as aNode's.
*
* This method is particularly useful for callers who are trying to ensure
* that they are working with a non-anonymous descendant of a given node. If
* aContent is a descendant of aNode, a return value of PR_FALSE from this
* method means that it's an anonymous descendant from aNode's point of view.
*
* Both arguments to this method must be non-null.
*/
static PRBool IsInSameAnonymousTree(nsINode* aNode, nsIContent* aContent);
/**
* Return the nsIXPConnect service.
*/
static nsIXPConnect *XPConnect()
{
return sXPConnect;
}
/**
* Report a localized error message to the error console.
* @param aFile Properties file containing localized message.
* @param aMessageName Name of localized message.
* @param aParams Parameters to be substituted into localized message.
* @param aParamsLength Length of aParams.
* @param aURI URI of resource containing error (may be null).
* @param aSourceLine The text of the line that contains the error (may be
empty).
* @param aLineNumber Line number within resource containing error.
* @param aColumnNumber Column number within resource containing error.
* @param aErrorFlags See nsIScriptError.
* @param aCategory Name of module reporting error.
*/
enum PropertiesFile {
eCSS_PROPERTIES,
eXBL_PROPERTIES,
eXUL_PROPERTIES,
eLAYOUT_PROPERTIES,
eFORMS_PROPERTIES,
ePRINTING_PROPERTIES,
eDOM_PROPERTIES,
#ifdef MOZ_SVG
eSVG_PROPERTIES,
#endif
eBRAND_PROPERTIES,
eCOMMON_DIALOG_PROPERTIES,
PropertiesFile_COUNT
};
static nsresult ReportToConsole(PropertiesFile aFile,
const char *aMessageName,
const PRUnichar **aParams,
PRUint32 aParamsLength,
nsIURI* aURI,
const nsAFlatString& aSourceLine,
PRUint32 aLineNumber,
PRUint32 aColumnNumber,
PRUint32 aErrorFlags,
const char *aCategory);
/**
* Get the localized string named |aKey| in properties file |aFile|.
*/
static nsresult GetLocalizedString(PropertiesFile aFile,
const char* aKey,
nsXPIDLString& aResult);
/**
* Fill (with the parameters given) the localized string named |aKey| in
* properties file |aFile|.
*/
static nsresult FormatLocalizedString(PropertiesFile aFile,
const char* aKey,
const PRUnichar **aParams,
PRUint32 aParamsLength,
nsXPIDLString& aResult);
/**
* Returns true if aDocument is a chrome document
*/
static PRBool IsChromeDoc(nsIDocument *aDocument);
/**
* Returns true if aDocument is in a docshell whose parent is the same type
*/
static PRBool IsChildOfSameType(nsIDocument* aDoc);
/**
* Get the script file name to use when compiling the script
* referenced by aURI. In cases where there's no need for any extra
* security wrapper automation the script file name that's returned
* will be the spec in aURI, else it will be the spec in aDocument's
* URI followed by aURI's spec, separated by " -> ". Returns PR_TRUE
* if the script file name was modified, PR_FALSE if it's aURI's
* spec.
*/
static PRBool GetWrapperSafeScriptFilename(nsIDocument *aDocument,
nsIURI *aURI,
nsACString& aScriptURI);
/**
* Returns true if aDocument belongs to a chrome docshell for
* display purposes. Returns false for null documents or documents
* which do not belong to a docshell.
*/
static PRBool IsInChromeDocshell(nsIDocument *aDocument);
/**
* Release *aSupportsPtr when the shutdown notification is received
*/
static nsresult ReleasePtrOnShutdown(nsISupports** aSupportsPtr) {
NS_ASSERTION(aSupportsPtr, "Expect to crash!");
NS_ASSERTION(*aSupportsPtr, "Expect to crash!");
return sPtrsToPtrsToRelease->AppendElement(aSupportsPtr) != nsnull ? NS_OK :
NS_ERROR_OUT_OF_MEMORY;
}
/**
* Return the content policy service
*/
static nsIContentPolicy *GetContentPolicy();
/**
* Quick helper to determine whether there are any mutation listeners
* of a given type that apply to this content or any of its ancestors.
* The method has the side effect to call document's MayDispatchMutationEvent
* using aTargetForSubtreeModified as the parameter.
*
* @param aNode The node to search for listeners
* @param aType The type of listener (NS_EVENT_BITS_MUTATION_*)
* @param aTargetForSubtreeModified The node which is the target of the
* possible DOMSubtreeModified event.
*
* @return true if there are mutation listeners of the specified type
*/
static PRBool HasMutationListeners(nsINode* aNode,
PRUint32 aType,
nsINode* aTargetForSubtreeModified);
/**
* This method creates and dispatches a trusted event.
* Works only with events which can be created by calling
* nsIDOMDocumentEvent::CreateEvent() with parameter "Events".
* @param aDoc The document which will be used to create the event.
* @param aTarget The target of the event, should be QIable to
* nsIDOMEventTarget.
* @param aEventName The name of the event.
* @param aCanBubble Whether the event can bubble.
* @param aCancelable Is the event cancelable.
* @param aDefaultAction Set to true if default action should be taken,
* see nsIDOMEventTarget::DispatchEvent.
*/
static nsresult DispatchTrustedEvent(nsIDocument* aDoc,
nsISupports* aTarget,
const nsAString& aEventName,
PRBool aCanBubble,
PRBool aCancelable,
PRBool *aDefaultAction = nsnull);
/**
* This method creates and dispatches a trusted event to the chrome
* event handler.
* Works only with events which can be created by calling
* nsIDOMDocumentEvent::CreateEvent() with parameter "Events".
* @param aDocument The document which will be used to create the event,
* and whose window's chrome handler will be used to
* dispatch the event.
* @param aTarget The target of the event, used for event->SetTarget()
* @param aEventName The name of the event.
* @param aCanBubble Whether the event can bubble.
* @param aCancelable Is the event cancelable.
* @param aDefaultAction Set to true if default action should be taken,
* see nsIDOMEventTarget::DispatchEvent.
*/
static nsresult DispatchChromeEvent(nsIDocument* aDoc,
nsISupports* aTarget,
const nsAString& aEventName,
PRBool aCanBubble,
PRBool aCancelable,
PRBool *aDefaultAction = nsnull);
/**
* Determines if an event attribute name (such as onclick) is valid for
* a given element type. Types are from the EventNameType enumeration
* defined above.
*
* @param aName the event name to look up
* @param aType the type of content
*/
static PRBool IsEventAttributeName(nsIAtom* aName, PRInt32 aType);
/**
* Return the event id for the event with the given name. The name is the
* event name with the 'on' prefix. Returns NS_USER_DEFINED_EVENT if the
* event doesn't match a known event name.
*
* @param aName the event name to look up
*/
static PRUint32 GetEventId(nsIAtom* aName);
/**
* Used only during traversal of the XPCOM graph by the cycle
* collector: push a pointer to the listener manager onto the
* children deque, if it exists. Do nothing if there is no listener
* manager.
*
* Crucially: does not perform any refcounting operations.
*
* @param aNode The node to traverse.
* @param children The buffer to push a listener manager pointer into.
*/
static void TraverseListenerManager(nsINode *aNode,
nsCycleCollectionTraversalCallback &cb);
/**
* Get the eventlistener manager for aNode. If a new eventlistener manager
* was created, aCreated is set to PR_TRUE.
*
* @param aNode The node for which to get the eventlistener manager.
* @param aCreateIfNotFound If PR_FALSE, returns a listener manager only if
* one already exists.
*/
static nsIEventListenerManager* GetListenerManager(nsINode* aNode,
PRBool aCreateIfNotFound);
/**
* Remove the eventlistener manager for aNode.
*
* @param aNode The node for which to remove the eventlistener manager.
*/
static void RemoveListenerManager(nsINode *aNode);
static PRBool IsInitialized()
{
return sInitialized;
}
/**
* Checks if the localname/prefix/namespace triple is valid wrt prefix
* and namespace according to the Namespaces in XML and DOM Code
* specfications.
*
* @param aLocalname localname of the node
* @param aPrefix prefix of the node
* @param aNamespaceID namespace of the node
*/
static PRBool IsValidNodeName(nsIAtom *aLocalName, nsIAtom *aPrefix,
PRInt32 aNamespaceID);
/**
* Creates a DocumentFragment from text using a context node to resolve
* namespaces.
*
* @param aContextNode the node which is used to resolve namespaces
* @param aFragment the string which is parsed to a DocumentFragment
* @param aWillOwnFragment is PR_TRUE if ownership of the fragment should be
* transferred to the caller.
* @param aReturn [out] the created DocumentFragment
*/
static nsresult CreateContextualFragment(nsIDOMNode* aContextNode,
const nsAString& aFragment,
PRBool aWillOwnFragment,
nsIDOMDocumentFragment** aReturn);
/**
* Creates a new XML document, which is marked to be loaded as data.
*
* @param aNamespaceURI Namespace for the root element to create and insert in
* the document. Only used if aQualifiedName is not
* empty.
* @param aQualifiedName Qualified name for the root element to create and
* insert in the document. If empty no root element will
* be created.
* @param aDoctype Doctype node to insert in the document.
* @param aDocumentURI URI of the document. Must not be null.
* @param aBaseURI Base URI of the document. Must not be null.
* @param aPrincipal Prinicpal of the document. Must not be null.
* @param aScriptObject The object from which the context for event handling
* can be got.
* @param aResult [out] The document that was created.
*/
static nsresult CreateDocument(const nsAString& aNamespaceURI,
const nsAString& aQualifiedName,
nsIDOMDocumentType* aDoctype,
nsIURI* aDocumentURI,
nsIURI* aBaseURI,
nsIPrincipal* aPrincipal,
nsIScriptGlobalObject* aScriptObject,
nsIDOMDocument** aResult);
/**
* Sets the text contents of a node by replacing all existing children
* with a single text child.
*
* The function always notifies.
*
* Will reuse the first text child if one is available. Will not reuse
* existing cdata children.
*
* @param aContent Node to set contents of.
* @param aValue Value to set contents to.
* @param aTryReuse When true, the function will try to reuse an existing
* textnodes rather than always creating a new one.
*/
static nsresult SetNodeTextContent(nsIContent* aContent,
const nsAString& aValue,
PRBool aTryReuse);
/**
* Get the textual contents of a node. This is a concatenation of all
* textnodes that are direct or (depending on aDeep) indirect children
* of the node.
*
* NOTE! No serialization takes place and <br> elements
* are not converted into newlines. Only textnodes and cdata nodes are
* added to the result.
*
* @param aNode Node to get textual contents of.
* @param aDeep If true child elements of aNode are recursivly descended
* into to find text children.
* @param aResult the result. Out param.
*/
static void GetNodeTextContent(nsINode* aNode, PRBool aDeep,
nsAString& aResult)
{
aResult.Truncate();
AppendNodeTextContent(aNode, aDeep, aResult);
}
/**
* Same as GetNodeTextContents but appends the result rather than sets it.
*/
static void AppendNodeTextContent(nsINode* aNode, PRBool aDeep,
nsAString& aResult);
/**
* Utility method that checks if a given node has any non-empty
* children.
* NOTE! This method does not descend recursivly into elements.
* Though it would be easy to make it so if needed
*/
static PRBool HasNonEmptyTextContent(nsINode* aNode);
/**
* Delete strings allocated for nsContentList matches
*/
static void DestroyMatchString(void* aData)
{
if (aData) {
nsString* matchString = static_cast<nsString*>(aData);
delete matchString;
}
}
/**
* Unbinds the content from the tree and nulls it out if it's not null.
*/
static void DestroyAnonymousContent(nsCOMPtr<nsIContent>* aContent);
/**
* Keep script object aNewObject, held by aScriptObjectHolder, alive.
*
* NOTE: This currently only supports objects that hold script objects of one
* scripting language.
*
* @param aLangID script language ID of aNewObject
* @param aScriptObjectHolder the object that holds aNewObject
* @param aTracer the tracer for aScriptObject
* @param aNewObject the script object to hold
* @param aWasHoldingObjects whether aScriptObjectHolder was already holding
* script objects (ie. HoldScriptObject was called
* on it before, without a corresponding call to
* DropScriptObjects)
*/
static nsresult HoldScriptObject(PRUint32 aLangID, void* aScriptObjectHolder,
nsScriptObjectTracer* aTracer,
void* aNewObject, PRBool aWasHoldingObjects)
{
if (aLangID == nsIProgrammingLanguage::JAVASCRIPT) {
return aWasHoldingObjects ? NS_OK :
HoldJSObjects(aScriptObjectHolder, aTracer);
}
return HoldScriptObject(aLangID, aNewObject);
}
/**
* Drop any script objects that aScriptObjectHolder is holding.
*
* NOTE: This currently only supports objects that hold script objects of one
* scripting language.
*
* @param aLangID script language ID of the objects that
* @param aScriptObjectHolder the object that holds script object that we want
* to drop
* @param aTracer the tracer for aScriptObject
*/
static nsresult DropScriptObjects(PRUint32 aLangID, void* aScriptObjectHolder,
nsScriptObjectTracer* aTracer)
{
if (aLangID == nsIProgrammingLanguage::JAVASCRIPT) {
return DropJSObjects(aScriptObjectHolder);
}
aTracer->Trace(aScriptObjectHolder, DropScriptObject, nsnull);
return NS_OK;
}
/**
* Keep the JS objects held by aScriptObjectHolder alive.
*
* @param aScriptObjectHolder the object that holds JS objects that we want to
* keep alive
* @param aTracer the tracer for aScriptObject
*/
static nsresult HoldJSObjects(void* aScriptObjectHolder,
nsScriptObjectTracer* aTracer);
/**
* Drop the JS objects held by aScriptObjectHolder.
*
* @param aScriptObjectHolder the object that holds JS objects that we want to
* drop
*/
static nsresult DropJSObjects(void* aScriptObjectHolder);
static void PreserveWrapper(nsISupports* aScriptObjectHolder,
nsWrapperCache* aCache)
{
if (!aCache->PreservingWrapper()) {
nsXPCOMCycleCollectionParticipant* participant;
CallQueryInterface(aScriptObjectHolder, &participant);
HoldJSObjects(aScriptObjectHolder, participant);
aCache->SetPreservingWrapper(PR_TRUE);
}
}
static void ReleaseWrapper(nsISupports* aScriptObjectHolder,
nsWrapperCache* aCache)
{
if (aCache->PreservingWrapper()) {
DropJSObjects(aScriptObjectHolder);
aCache->SetPreservingWrapper(PR_FALSE);
}
}
static void TraceWrapper(nsWrapperCache* aCache, TraceCallback aCallback,
void *aClosure)
{
if (aCache->PreservingWrapper()) {
aCallback(nsIProgrammingLanguage::JAVASCRIPT, aCache->GetWrapper(),
aClosure);
}
}
/**
* Convert nsIContent::IME_STATUS_* to nsIWidget::IME_STATUS_*
*/
static PRUint32 GetWidgetStatusFromIMEStatus(PRUint32 aState);
/*
* Notify when the first XUL menu is opened and when the all XUL menus are
* closed. At opening, aInstalling should be TRUE, otherwise, it should be
* FALSE.
*/
static void NotifyInstalledMenuKeyboardListener(PRBool aInstalling);
/**
* Do security checks before loading a resource. Does the following checks:
* nsIScriptSecurityManager::CheckLoadURIWithPrincipal
* NS_CheckContentLoadPolicy
* nsIScriptSecurityManager::CheckSameOriginURI
*
* You will still need to do at least SameOrigin checks before on redirects.
*
* @param aURIToLoad URI that is getting loaded.
* @param aLoadingPrincipal Principal of the resource that is initiating
* the load
* @param aCheckLoadFlags Flags to be passed to
* nsIScriptSecurityManager::CheckLoadURIWithPrincipal
* NOTE: If this contains ALLOW_CHROME the
* CheckSameOriginURI check will be skipped if
* aURIToLoad is a chrome uri.
* @param aAllowData Set to true to skip CheckSameOriginURI check when
aURIToLoad is a data uri.
* @param aContentPolicyType Type \
* @param aContext Context |- to be passed to
* @param aMimeGuess Mimetype | NS_CheckContentLoadPolicy
* @param aExtra Extra /
*/
static nsresult CheckSecurityBeforeLoad(nsIURI* aURIToLoad,
nsIPrincipal* aLoadingPrincipal,
PRUint32 aCheckLoadFlags,
PRBool aAllowData,
PRUint32 aContentPolicyType,
nsISupports* aContext,
const nsACString& aMimeGuess = EmptyCString(),
nsISupports* aExtra = nsnull);
/**
* Trigger a link with uri aLinkURI. If aClick is false, this triggers a
* mouseover on the link, otherwise it triggers a load after doing a
* security check using aContent's principal.
*
* @param aContent the node on which a link was triggered.
* @param aPresContext the pres context, must be non-null.
* @param aLinkURI the URI of the link, must be non-null.
* @param aTargetSpec the target (like target=, may be empty).
* @param aClick whether this was a click or not (if false, this method
* assumes you just hovered over the link).
* @param aIsUserTriggered whether the user triggered the link. This would be
* false for loads from auto XLinks or from the
* click() method if we ever implement it.
*/
static void TriggerLink(nsIContent *aContent, nsPresContext *aPresContext,
nsIURI *aLinkURI, const nsString& aTargetSpec,
PRBool aClick, PRBool aIsUserTriggered);
/**
* Return top-level widget in the parent chain.
*/
static nsIWidget* GetTopLevelWidget(nsIWidget* aWidget);
/**
* Return the localized ellipsis for UI.
*/
static const nsDependentString GetLocalizedEllipsis();
/**
* The routine GetNativeEvent is used to fill nsNativeKeyEvent.
* It's also used in DOMEventToNativeKeyEvent.
* See bug 406407 for details.
*/
static nsEvent* GetNativeEvent(nsIDOMEvent* aDOMEvent);
static PRBool DOMEventToNativeKeyEvent(nsIDOMKeyEvent* aKeyEvent,
nsNativeKeyEvent* aNativeEvent,
PRBool aGetCharCode);
/**
* Get the candidates for accelkeys for aDOMKeyEvent.
*
* @param aDOMKeyEvent [in] the key event for accelkey handling.
* @param aCandidates [out] the candidate shortcut key combination list.
* the first item is most preferred.
*/
static void GetAccelKeyCandidates(nsIDOMKeyEvent* aDOMKeyEvent,
nsTArray<nsShortcutCandidate>& aCandidates);
/**
* Get the candidates for accesskeys for aNativeKeyEvent.
*
* @param aNativeKeyEvent [in] the key event for accesskey handling.
* @param aCandidates [out] the candidate access key list.
* the first item is most preferred.
*/
static void GetAccessKeyCandidates(nsKeyEvent* aNativeKeyEvent,
nsTArray<PRUint32>& aCandidates);
/**
* Hide any XUL popups associated with aDocument, including any documents
* displayed in child frames. Does nothing if aDocument is null.
*/
static void HidePopupsInDocument(nsIDocument* aDocument);
/**
* Retrieve the current drag session, or null if no drag is currently occuring
*/
static already_AddRefed<nsIDragSession> GetDragSession();
/*
* Initialize and set the dataTransfer field of an nsDragEvent.
*/
static nsresult SetDataTransferInEvent(nsDragEvent* aDragEvent);
// filters the drag and drop action to fit within the effects allowed and
// returns it.
static PRUint32 FilterDropEffect(PRUint32 aAction, PRUint32 aEffectAllowed);
/**
* Return true if aURI is a local file URI (i.e. file://).
*/
static PRBool URIIsLocalFile(nsIURI *aURI);
/**
* If aContent is an HTML element with a DOM level 0 'name', then
* return the name. Otherwise return null.
*/
static nsIAtom* IsNamedItem(nsIContent* aContent);
/**
* Get the application manifest URI for this document. The manifest URI
* is specified in the manifest= attribute of the root element of the
* document.
*
* @param aDocument The document that lists the manifest.
* @param aURI The manifest URI.
*/
static void GetOfflineAppManifest(nsIDocument *aDocument, nsIURI **aURI);
/**
* Check whether an application should be allowed to use offline APIs.
*/
static PRBool OfflineAppAllowed(nsIURI *aURI);
/**
* Check whether an application should be allowed to use offline APIs.
*/
static PRBool OfflineAppAllowed(nsIPrincipal *aPrincipal);
/**
* Increases the count of blockers preventing scripts from running.
* NOTE: You might want to use nsAutoScriptBlocker rather than calling
* this directly
*/
static void AddScriptBlocker();
/**
* Decreases the count of blockers preventing scripts from running.
* NOTE: You might want to use nsAutoScriptBlocker rather than calling
* this directly
*
* WARNING! Calling this function could synchronously execute scripts.
*/
static void RemoveScriptBlocker();
/**
* Add a runnable that is to be executed as soon as it's safe to execute
* scripts.
* NOTE: If it's currently safe to execute scripts, aRunnable will be run
* synchronously before the function returns.
*
* @param aRunnable The nsIRunnable to run as soon as it's safe to execute
* scripts. Passing null is allowed and results in nothing
* happening. It is also allowed to pass an object that
* has not yet been AddRefed.
* @return false on out of memory, true otherwise.
*/
static PRBool AddScriptRunner(nsIRunnable* aRunnable);
/**
* Returns true if it's safe to execute content script and false otherwise.
*
* The only known case where this lies is mutation events. They run, and can
* run anything else, when this function returns false, but this is ok.
*/
static PRBool IsSafeToRunScript() {
return sScriptBlockerCount == 0;
}
/**
* Get/Set the current number of removable updates. Currently only
* UPDATE_CONTENT_MODEL updates are removable, and only when firing mutation
* events. These functions should only be called by mozAutoDocUpdateRemover.
* The count is also adjusted by the normal calls to BeginUpdate/EndUpdate.
*/
static void AddRemovableScriptBlocker()
{
AddScriptBlocker();
++sRemovableScriptBlockerCount;
}
static void RemoveRemovableScriptBlocker()
{
NS_ASSERTION(sRemovableScriptBlockerCount != 0,
"Number of removable blockers should never go below zero");
--sRemovableScriptBlockerCount;
RemoveScriptBlocker();
}
static PRUint32 GetRemovableScriptBlockerLevel()
{
return sRemovableScriptBlockerCount;
}
/* Process viewport META data. This gives us information for the scale
* and zoom of a page on mobile devices. We stick the information in
* the document header and use it later on after rendering.
*
* See Bug #436083
*/
static nsresult ProcessViewportInfo(nsIDocument *aDocument,
const nsAString &viewportInfo);
static nsIScriptContext* GetContextForEventHandlers(nsINode* aNode,
nsresult* aRv);
static JSContext *GetCurrentJSContext();
static nsIInterfaceRequestor* GetSameOriginChecker();
static nsIThreadJSContextStack* ThreadJSContextStack()
{
return sThreadJSContextStack;
}
/**
* Get the Origin of the passed in nsIPrincipal or nsIURI. If the passed in
* nsIURI or the URI of the passed in nsIPrincipal does not have a host, the
* origin is set to 'null'.
*
* The ASCII versions return a ASCII strings that are puny-code encoded,
* suitable for for example header values. The UTF versions return strings
* containing international characters.
*
* aPrincipal/aOrigin must not be null.
*/
static nsresult GetASCIIOrigin(nsIPrincipal* aPrincipal,
nsCString& aOrigin);
static nsresult GetASCIIOrigin(nsIURI* aURI, nsCString& aOrigin);
static nsresult GetUTFOrigin(nsIPrincipal* aPrincipal,
nsString& aOrigin);
static nsresult GetUTFOrigin(nsIURI* aURI, nsString& aOrigin);
/**
* This method creates and dispatches "command" event, which implements
* nsIDOMXULCommandEvent.
* If aShell is not null, dispatching goes via
* nsIPresShell::HandleDOMEventWithTarget.
*/
static nsresult DispatchXULCommand(nsIContent* aTarget,
PRBool aTrusted,
nsIDOMEvent* aSourceEvent = nsnull,
nsIPresShell* aShell = nsnull,
PRBool aCtrl = PR_FALSE,
PRBool aAlt = PR_FALSE,
PRBool aShift = PR_FALSE,
PRBool aMeta = PR_FALSE);
/**
* Gets the nsIDocument given the script context. Will return nsnull on failure.
*
* @param aScriptContext the script context to get the document for; can be null
*
* @return the document associated with the script context
*/
static already_AddRefed<nsIDocument>
GetDocumentFromScriptContext(nsIScriptContext *aScriptContext);
/**
* The method checks whether the caller can access native anonymous content.
* If there is no JS in the stack or privileged JS is running, this
* method returns PR_TRUE, otherwise PR_FALSE.
*/
static PRBool CanAccessNativeAnon();
static nsresult WrapNative(JSContext *cx, JSObject *scope,
nsISupports *native, const nsIID* aIID, jsval *vp,
// If non-null aHolder will keep the jsval alive
// while there's a ref to it
nsIXPConnectJSObjectHolder** aHolder = nsnull,
PRBool aAllowWrapping = PR_FALSE);
// Same as the WrapNative above, but use this one if aIID is nsISupports' IID.
static nsresult WrapNative(JSContext *cx, JSObject *scope,
nsISupports *native, jsval *vp,
// If non-null aHolder will keep the jsval alive
// while there's a ref to it
nsIXPConnectJSObjectHolder** aHolder = nsnull,
PRBool aAllowWrapping = PR_FALSE)
{
return WrapNative(cx, scope, native, nsnull, vp, aHolder, aAllowWrapping);
}
private:
static PRBool InitializeEventTable();
static nsresult doReparentContentWrapper(nsIContent *aChild,
JSContext *cx,
JSObject *aOldGlobal,
JSObject *aNewGlobal,
nsIDocument *aOldDocument,
nsIDocument *aNewDocument);
static nsresult EnsureStringBundle(PropertiesFile aFile);
static nsIDOMScriptObjectFactory *GetDOMScriptObjectFactory();
static nsresult HoldScriptObject(PRUint32 aLangID, void* aObject);
static void DropScriptObject(PRUint32 aLangID, void *aObject, void *aClosure);
static PRBool CanCallerAccess(nsIPrincipal* aSubjectPrincipal,
nsIPrincipal* aPrincipal);
static nsIDOMScriptObjectFactory *sDOMScriptObjectFactory;
static nsIXPConnect *sXPConnect;
static nsIScriptSecurityManager *sSecurityManager;
static nsIThreadJSContextStack *sThreadJSContextStack;
static nsIParserService *sParserService;
static nsINameSpaceManager *sNameSpaceManager;
static nsIIOService *sIOService;
#ifdef MOZ_XTF
static nsIXTFService *sXTFService;
#endif
static nsIPrefBranch *sPrefBranch;
static nsIPref *sPref;
static imgILoader* sImgLoader;
static imgICache* sImgCache;
static nsIConsoleService* sConsoleService;
static nsDataHashtable<nsISupportsHashKey, EventNameMapping>* sEventTable;
static nsIStringBundleService* sStringBundleService;
static nsIStringBundle* sStringBundles[PropertiesFile_COUNT];
static nsIContentPolicy* sContentPolicyService;
static PRBool sTriedToGetContentPolicy;
static nsILineBreaker* sLineBreaker;
static nsIWordBreaker* sWordBreaker;
static nsICaseConversion* sCaseConv;
static nsIUGenCategory* sGenCat;
// Holds pointers to nsISupports* that should be released at shutdown
static nsTArray<nsISupports**>* sPtrsToPtrsToRelease;
static nsIScriptRuntime* sScriptRuntimes[NS_STID_ARRAY_UBOUND];
static PRInt32 sScriptRootCount[NS_STID_ARRAY_UBOUND];
static PRUint32 sJSGCThingRootCount;
#ifdef IBMBIDI
static nsIBidiKeyboard* sBidiKeyboard;
#endif
static PRBool sInitialized;
static PRUint32 sScriptBlockerCount;
static PRUint32 sRemovableScriptBlockerCount;
static nsCOMArray<nsIRunnable>* sBlockedScriptRunners;
static PRUint32 sRunnersCountAtFirstBlocker;
static nsIInterfaceRequestor* sSameOriginChecker;
};
#define NS_HOLD_JS_OBJECTS(obj, clazz) \
nsContentUtils::HoldJSObjects(NS_CYCLE_COLLECTION_UPCAST(obj, clazz), \
&NS_CYCLE_COLLECTION_NAME(clazz))
#define NS_DROP_JS_OBJECTS(obj, clazz) \
nsContentUtils::DropJSObjects(NS_CYCLE_COLLECTION_UPCAST(obj, clazz))
class NS_STACK_CLASS nsCxPusher
{
public:
nsCxPusher();
~nsCxPusher(); // Calls Pop();
// Returns PR_FALSE if something erroneous happened.
PRBool Push(nsPIDOMEventTarget *aCurrentTarget);
// If nothing has been pushed to stack, this works like Push.
// Otherwise if context will change, Pop and Push will be called.
PRBool RePush(nsPIDOMEventTarget *aCurrentTarget);
// If a null JSContext is passed to Push(), that will cause no
// push to happen and false to be returned.
PRBool Push(JSContext *cx);
// Explicitly push a null JSContext on the the stack
PRBool PushNull();
// Pop() will be a no-op if Push() or PushNull() fail
void Pop();
nsIScriptContext* GetCurrentScriptContext() { return mScx; }
private:
// Combined code for PushNull() and Push(JSContext*)
PRBool DoPush(JSContext* cx);
nsCOMPtr<nsIScriptContext> mScx;
PRBool mScriptIsRunning;
PRBool mPushedSomething;
#ifdef DEBUG
JSContext* mPushedContext;
#endif
};
class nsAutoGCRoot {
public:
// aPtr should be the pointer to the jsval we want to protect
nsAutoGCRoot(jsval* aPtr, nsresult* aResult) :
mPtr(aPtr)
{
mResult = *aResult = AddJSGCRoot(aPtr, "nsAutoGCRoot");
}
// aPtr should be the pointer to the JSObject* we want to protect
nsAutoGCRoot(JSObject** aPtr, nsresult* aResult) :
mPtr(aPtr)
{
mResult = *aResult = AddJSGCRoot(aPtr, "nsAutoGCRoot");
}
// aPtr should be the pointer to the thing we want to protect
nsAutoGCRoot(void* aPtr, nsresult* aResult) :
mPtr(aPtr)
{
mResult = *aResult = AddJSGCRoot(aPtr, "nsAutoGCRoot");
}
~nsAutoGCRoot() {
if (NS_SUCCEEDED(mResult)) {
RemoveJSGCRoot(mPtr);
}
}
static void Shutdown();
private:
static nsresult AddJSGCRoot(void *aPtr, const char* aName);
static nsresult RemoveJSGCRoot(void *aPtr);
static nsIJSRuntimeService* sJSRuntimeService;
static JSRuntime* sJSScriptRuntime;
void* mPtr;
nsresult mResult;
};
class nsAutoScriptBlocker {
public:
nsAutoScriptBlocker() {
nsContentUtils::AddScriptBlocker();
}
~nsAutoScriptBlocker() {
nsContentUtils::RemoveScriptBlocker();
}
};
class mozAutoRemovableBlockerRemover
{
public:
mozAutoRemovableBlockerRemover(nsIDocument* aDocument);
~mozAutoRemovableBlockerRemover();
private:
PRUint32 mNestingLevel;
nsCOMPtr<nsIDocument> mDocument;
nsCOMPtr<nsIDocumentObserver> mObserver;
};
#define NS_AUTO_GCROOT_PASTE2(tok,line) tok##line
#define NS_AUTO_GCROOT_PASTE(tok,line) \
NS_AUTO_GCROOT_PASTE2(tok,line)
#define NS_AUTO_GCROOT(ptr, result) \ \
nsAutoGCRoot NS_AUTO_GCROOT_PASTE(_autoGCRoot_, __LINE__) \
(ptr, result)
#define NS_INTERFACE_MAP_ENTRY_CONTENT_CLASSINFO(_class) \
NS_DOM_INTERFACE_MAP_ENTRY_CLASSINFO(_class)
#define NS_INTERFACE_MAP_ENTRY_TEAROFF(_interface, _allocator) \
if (aIID.Equals(NS_GET_IID(_interface))) { \
foundInterface = static_cast<_interface *>(_allocator); \
if (!foundInterface) { \
*aInstancePtr = nsnull; \
return NS_ERROR_OUT_OF_MEMORY; \
} \
} else
/*
* Check whether a floating point number is finite (not +/-infinity and not a
* NaN value). We wrap JSDOUBLE_IS_FINITE in a function because it expects to
* take the address of its argument, and because the argument must be of type
* jsdouble to have the right size and layout of bits.
*
* Note: we could try to exploit the fact that |infinity - infinity == NaN|
* instead of using JSDOUBLE_IS_FINITE. This would produce more compact code
* and perform better by avoiding type conversions and bit twiddling.
* Unfortunately, some architectures don't guarantee that |f == f| evaluates
* to true (where f is any *finite* floating point number). See
* https://bugzilla.mozilla.org/show_bug.cgi?id=369418#c63 . To play it safe
* for gecko 1.9, we just reuse JSDOUBLE_IS_FINITE.
*/
inline NS_HIDDEN_(PRBool) NS_FloatIsFinite(jsdouble f) {
return JSDOUBLE_IS_FINITE(f);
}
/*
* In the following helper macros we exploit the fact that the result of a
* series of additions will not be finite if any one of the operands in the
* series is not finite.
*/
#define NS_ENSURE_FINITE(f, rv) \
if (!NS_FloatIsFinite(f)) { \
return (rv); \
}
#define NS_ENSURE_FINITE2(f1, f2, rv) \
if (!NS_FloatIsFinite((f1)+(f2))) { \
return (rv); \
}
#define NS_ENSURE_FINITE3(f1, f2, f3, rv) \
if (!NS_FloatIsFinite((f1)+(f2)+(f3))) { \
return (rv); \
}
#define NS_ENSURE_FINITE4(f1, f2, f3, f4, rv) \
if (!NS_FloatIsFinite((f1)+(f2)+(f3)+(f4))) { \
return (rv); \
}
#define NS_ENSURE_FINITE5(f1, f2, f3, f4, f5, rv) \
if (!NS_FloatIsFinite((f1)+(f2)+(f3)+(f4)+(f5))) { \
return (rv); \
}
#define NS_ENSURE_FINITE6(f1, f2, f3, f4, f5, f6, rv) \
if (!NS_FloatIsFinite((f1)+(f2)+(f3)+(f4)+(f5)+(f6))) { \
return (rv); \
}
// Deletes a linked list iteratively to avoid blowing up the stack (bug 460444).
#define NS_CONTENT_DELETE_LIST_MEMBER(type_, ptr_, member_) \
{ \
type_ *cur = (ptr_)->member_; \
(ptr_)->member_ = nsnull; \
while (cur) { \
type_ *next = cur->member_; \
cur->member_ = nsnull; \
delete cur; \
cur = next; \
} \
}
class nsContentTypeParser {
public:
nsContentTypeParser(const nsAString& aString);
~nsContentTypeParser();
nsresult GetParameter(const char* aParameterName, nsAString& aResult);
nsresult GetType(nsAString& aResult)
{
return GetParameter(nsnull, aResult);
}
private:
NS_ConvertUTF16toUTF8 mString;
nsIMIMEHeaderParam* mService;
};
#endif /* nsContentUtils_h___ */