Closed Bug 78622 Opened 25 years ago Closed 25 years ago

DOMParser and XMLHttpRequest failures when parser blocked (xml-stylesheet PI etc. in data)

Categories

(Core :: XML, defect, P2)

defect

Tracking

()

VERIFIED FIXED
mozilla0.9.5

People

(Reporter: pwilson, Assigned: hjtoi-bugzilla)

Details

(Keywords: regression, Whiteboard: [fixed on trunk and 0.9.4 branch], PDT+)

Attachments

(1 file, 3 obsolete files)

After parsing an XUL document using domParser = this.getComponent("@mozilla.org/xmlextras/domparser;1", "nsIDOMParser"); dom = domParser.parseFromString(xml, "text/xml") The method dom.documentElement returns null. This only happens when stylesheet processing instruction are present (or possibly any PI). Removing the PI's from the document yields a correct element value. This is unfortunate as an easy way to check for a failed parse is dom.documentElement.localName == "PARSERERROR" I believe the problem lies in XULContentSink.
not bookmarks...
Assignee: ben → waterson
Component: Bookmarks → DOM Mozilla Extensions
QA Contact: claudius → lchiang
The XUL document object doesn't really know what children are, this would be solved if we'd merge it with nsDocument...
I believe I have an explanation of the problem which might also explain bugs 75844 and 76039. When parsing an XUL document (and possibly HTML - but not tried). I believe an asynchronous parse of the referenced stylesheet/overlay/javascript documents is invoked. Is it possible that parseFromStream/String is returning while this parsing is continuing? In effect I would be examining a partially completed document. hence the missing element id. in getElementById (Bug 76039) and the missing documentElement in this bug. This would also explain why removing the styleshhet PI enables the parse to complete earlier and successfully. Bug 75844 might also be explained.
Seee bug 83477 where this bug is incidentaly present. The test files and comments explain further.
Status: UNCONFIRMED → ASSIGNED
Component: DOM Mozilla Extensions → XP Toolkit/Widgets: XUL
Ever confirmed: true
Priority: -- → P4
Target Milestone: --- → mozilla1.0
I think this works now, please test.
Whiteboard: worksforme?
I have retested this. Initially I thought it was still broken. For an XUL document the following line: debug("srcDOM:" + srcDOM + " docElem:" + srcDOM.documentElement); gives: >>srcDOM:[object Document] docElem:null<< Then I ran it again several times (from a new invocation of Mozilla each time) and got a different result about one time in five: gives: >>srcDOM:[object Document] docElem:[object XULElement]<< Ughh! a timing related bug. Is some asynchronous thread building the document?
Assignee: waterson → heikki
Status: ASSIGNED → NEW
Target Milestone: mozilla1.0 → mozilla0.9.5
Looking at the code, it does seem the document is built asynchronously. I'll take this to investigate (might be XML Extras bug). Peter, if you could provide a self contained minimal testcase it would be great.
Oh boy, this is a mess... I haven't yet tested this for DOMParser, but I am pretty sure I dissected this from the code with harishd yesterday, and I now checked that this really happens with XMLHttpRequest. Testcase for XMLHttpRequest is: file:///C:/builds/mozilla/extensions/xmlextras/tests/getdisplay.html You see the bug if "responseXML serialized" has no result. Just by commenting out the xml-stylesheet PI in the XML file that it tries to load the sample works fine. If you have xml-stylesheet PI or any other construct that blocks the parser, we fire the load event too soon (XMLHttpRequest) or return too soon (DOMParser). With DOMParser, it looks like our call to OnDataAvailable() can return with message that the parser is blocked (while it is retrieving stylesheet or some such), but DOMParser thinks the parsing is complete. DOMParser should probably be fixed to be a load listener for the document and only return when the load event has been received. With my quite recent change to XMLHttpRequest, if the parser is blocked we can receive OnStopRequest() before the document actually loaded, meaning we fire the load event too soon (because we do it from OnStopRequest()). XMLHttpRequest is more complicated to fix. For non-wellformed documents we must fire the load event from OnStopRequest() - for other cases the load event must be fired when the document informs it has loaded. This is not XUL-specific, so I have changed the component and summary, as well as some other fields.
Severity: normal → major
Status: NEW → ASSIGNED
Component: XP Toolkit/Widgets: XUL → XML
Keywords: nsbranch+
OS: Windows 2000 → All
Priority: P4 → P2
Hardware: PC → All
Summary: xuldocument.documentElement returns null if PI's present → DOMParser and XMLHttpRequest failures when parser blocked (xml-stylesheet PI etc. in data)
Whiteboard: worksforme?
For XMLHttpRequest this is a regression, adding keyword. I am fixing both objects under this one bug since the cause is the same and the fix very similar for each case.
Keywords: regression
Attachment #49049 - Attachment is obsolete: true
There, I basically copied the way we do synchronous loading from XMLHttpRequest into DOMParser. In DOMParser we now wait for the document load event before returning from parseFromString/Stream. Harish, can you r=?
Whiteboard: [fixinhand]
+ if (modalEventQueue) { + eventQService->PopThreadEventQueue(modalEventQueue); + } Not sure why that check in necessary. Other than that r=harishd.
Attachment #49139 - Flags: review+
Currently nsEventQueueServiceImpl::PopThreadEventQueue(nsIEventQueue *aQueue) will crash if !aQueue. Dunno if we gan get that situation, but... All of that stuff is basically copied from XMLHttpRequest where it has worked for a year(?) without crash problems as is. Eventually the whole event queue thing must be replaced with something better, but since there does not seem to be a simple way to make a simple event loop a la Windows this is what we have now...
The latest patch is otherwise identical to the previous, except that at the end of parseFromStream() in DOMParser we PopThreadEventQueue() even if we have no mChromeWindow (which happens if the parser was not blocked). Good catch from vidur!
Comment on attachment 49272 [details] [diff] [review] Make sure to PopThreadEventQueue() even if no mChromeWindow The latest patch will call ExitModalEventLoop even if we haven't entered a modal event loop. Not ideal, but is at least safe?
Ok, in the 4th patch we have: +// See if we have a modal event loop +static inline PRBool IsModal(nsIWebBrowserChrome *aWindow) +{ + if (aWindow) { + PRBool isModal; + if (NS_SUCCEEDED(aWindow->IsWindowModal(&isModal))) { + return isModal; + } + } + + return PR_FALSE; +} + and we call that to check if we should ExitModal (like this): +nsDOMParser::Load(nsIDOMEvent* aEvent) +{ + if (IsModal(mChromeWindow)) { + mChromeWindow->ExitModalEventLoop(NS_OK); + } + + mChromeWindow = nsnull; + + return NS_OK; +} The IDL comment for exitModalEventLoop() implies (in my opinion) but does not say for sure that it would be safe to call it even if there is no modal event loop. Some of the implementations are safe, but I am not so sure about all of them, so the new patch should take care of that.
Comment on attachment 49281 [details] [diff] [review] Call ExitModal only if we are modal sr=vidur
Attachment #49281 - Flags: superreview+
Checked in on the trunk, leaving open to track situation with 0.9.4 branch.
Whiteboard: [fixinhand] → [fixed on trunk]
Pls give this to the DIG guys to test, and make sure it doesn't affect them. - PDT
Attachment #49281 - Flags: review+
Attachment #49139 - Attachment is obsolete: true
Attachment #49272 - Attachment is obsolete: true
I was in contact with the DIG people, got some instructions on how to test, and I tested this on an online version of their app - no difference in behaviour. The DIG group also felt that based on the description of this bug this is unlikely to affect them.
check it in - PDT+
Whiteboard: [fixed on trunk] → [fixed on trunk], PDT+
Checked in on the 0.9.4 branch as well, marking fixed.
Status: ASSIGNED → RESOLVED
Closed: 25 years ago
Resolution: --- → FIXED
Whiteboard: [fixed on trunk], PDT+ → [fixed on trunk and 0.9.4 branch], PDT+
pwilson@gorge.net (Peter Wilson)- would you mind verifying that this if fine for you on the branch 0.9.4 and trunk builds?
QA Contact: lchiang → stummala
I have sucessfully loaded my forms a. The same form repeatedly. b. a variety of differnet forms with different xml includes. No problems detected. This appears to be fixed for the nightly build of 9/22/2001 on win32. I am unable to test 0.9.4 as the most recent release, 9/13/2001, pre-dates the check-in for this bug.
as per the reporters comments, looks like this is fixed on trunk. peter can u please verify on branch. (the fix landed on branch too)
the fix landed on both branch and trunk as per developer's comments. the reporter verified this bug on trunk. marking as verified
Status: RESOLVED → VERIFIED
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: