860492 - Viewing big text files hangs the browser for a long time

Status: RESOLVED WORKSFORME

(Core :: Layout: Text and Fonts, defect)

Description

[Ernst Sjostrand]

User Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:23.0) Gecko/20130410 Firefox/23.0
Build ID: 20130410065939

Steps to reproduce:

I often look at build logs of around 10 MB or sometimes more. Loading those takes very long time in Firefox, but worst of all is that it hangs the entire UI completely for a long time.
Here's a bit smaller example: https://launchpadlibrarian.net/136191756/buildlog_ubuntu-raring-amd64.firefox_20.0%2Bbuild1-0ubuntu1_UPLOADING.txt.gz
plain/text of html with a big <pre> or similar doesn't seem to matter, they're both slow.
I actually expect these to load instantaneously and without hangs, just like a good text editor would do. They contain no layout, DOM, html etc so it should be possible to do a lot of optimizations.
It's a little bit better in Nightly than on Aurora.


Actual results:

Here's the trace from about:telemetry

gfxSkipCharsIterator::IsOriginalCharSkipped(int *) (in xul.pdb)
nsSkipCharsRunIterator::NextRun() (in xul.pdb)
PropertyProvider::GetHyphenationBreaks(unsigned int,unsigned int,bool *) (in xul.pdb)
gfxTextRun::BreakAndMeasureText(unsigned int,unsigned int,bool,double,gfxTextRun::PropertyProvider *,bool,double *,gfxFont::RunMetrics *,gfxFont::BoundingBoxType,gfxContext *,bool *,unsigned int *,bool,gfxBreakPriority *) (in xul.pdb)
nsTextFrame::ReflowText(nsLineLayout &,int,nsRenderingContext *,bool,nsHTMLReflowMetrics &,unsigned int &) (in xul.pdb)
nsLineLayout::ReflowFrame(nsIFrame *,unsigned int &,nsHTMLReflowMetrics *,bool &) (in xul.pdb)
nsBlockFrame::ReflowInlineFrame(nsBlockReflowState &,nsLineLayout &,nsLineList_iterator,nsIFrame *,LineReflowStatus *) (in xul.pdb)
nsBlockFrame::DoReflowInlineFrames(nsBlockReflowState &,nsLineLayout &,nsLineList_iterator,nsFlowAreaRect &,int &,nsFloatManager::SavedState *,bool *,LineReflowStatus *,bool) (in xul.pdb)
nsBlockFrame::ReflowInlineFrames(nsBlockReflowState &,nsLineList_iterator,bool *) (in xul.pdb)
nsBlockFrame::ReflowDirtyLines(nsBlockReflowState &) (in xul.pdb)
nsBlockFrame::Reflow(nsPresContext *,nsHTMLReflowMetrics &,nsHTMLReflowState const &,unsigned int &) (in xul.pdb)
nsBlockReflowContext::ReflowBlock(nsRect const &,bool,nsCollapsingMargin &,int,bool,nsLineBox *,nsHTMLReflowState &,unsigned int &,nsBlockReflowState &) (in xul.pdb)
nsBlockFrame::ReflowBlockFrame(nsBlockReflowState &,nsLineList_iterator,bool *) (in xul.pdb)
nsBlockFrame::ReflowDirtyLines(nsBlockReflowState &) (in xul.pdb)
nsBlockFrame::Reflow(nsPresContext *,nsHTMLReflowMetrics &,nsHTMLReflowState const &,unsigned int &) (in xul.pdb)
nsBlockReflowContext::ReflowBlock(nsRect const &,bool,nsCollapsingMargin &,int,bool,nsLineBox *,nsHTMLReflowState &,unsigned int &,nsBlockReflowState &) (in xul.pdb)
nsBlockFrame::ReflowBlockFrame(nsBlockReflowState &,nsLineList_iterator,bool *) (in xul.pdb)
nsBlockFrame::ReflowDirtyLines(nsBlockReflowState &) (in xul.pdb)
nsBlockFrame::Reflow(nsPresContext *,nsHTMLReflowMetrics &,nsHTMLReflowState const &,unsigned int &) (in xul.pdb)
nsContainerFrame::ReflowChild(nsIFrame *,nsPresContext *,nsHTMLReflowMetrics &,nsHTMLReflowState const &,int,int,unsigned int,unsigned int &,nsOverflowContinuationTracker *) (in xul.pdb)
nsCanvasFrame::Reflow(nsPresContext *,nsHTMLReflowMetrics &,nsHTMLReflowState const &,unsigned int &) (in xul.pdb)
nsContainerFrame::ReflowChild(nsIFrame *,nsPresContext *,nsHTMLReflowMetrics &,nsHTMLReflowState const &,int,int,unsigned int,unsigned int &,nsOverflowContinuationTracker *) (in xul.pdb)
nsHTMLScrollFrame::ReflowScrolledFrame(ScrollReflowState *,bool,bool,nsHTMLReflowMetrics *,bool) (in xul.pdb)
nsHTMLScrollFrame::ReflowContents(ScrollReflowState *,nsHTMLReflowMetrics const &) (in xul.pdb)
nsHTMLScrollFrame::Reflow(nsPresContext *,nsHTMLReflowMetrics &,nsHTMLReflowState const &,unsigned int &) (in xul.pdb)
nsContainerFrame::ReflowChild(nsIFrame *,nsPresContext *,nsHTMLReflowMetrics &,nsHTMLReflowState const &,int,int,unsigned int,unsigned int &,nsOverflowContinuationTracker *) (in xul.pdb)
ViewportFrame::Reflow(nsPresContext *,nsHTMLReflowMetrics &,nsHTMLReflowState const &,unsigned int &) (in xul.pdb)
PresShell::DoReflow(nsIFrame *,bool) (in xul.pdb)
PresShell::ProcessReflowCommands(bool) (in xul.pdb)
PresShell::FlushPendingNotifications(mozilla::ChangesToFlush) (in xul.pdb)
PresShell::FlushPendingNotifications(mozFlushType) (in xul.pdb)
nsDocumentViewer::LoadComplete(tag_nsresult) (in xul.pdb)
nsDocShell::EndPageLoad(nsIWebProgress *,nsIChannel *,tag_nsresult) (in xul.pdb)
nsDocShell::OnStateChange(nsIWebProgress *,nsIRequest *,unsigned int,tag_nsresult) (in xul.pdb)
nsDocLoader::DoFireOnStateChange(nsIWebProgress * const,nsIRequest * const,int &,tag_nsresult) (in xul.pdb)
nsDocLoader::doStopDocumentLoad(nsIRequest *,tag_nsresult) (in xul.pdb)
nsDocLoader::DocLoaderIsEmpty(bool) (in xul.pdb)
nsDocLoader::OnStopRequest(nsIRequest *,nsISupports *,tag_nsresult) (in xul.pdb)
nsLoadGroup::RemoveRequest(nsIRequest *,nsISupports *,tag_nsresult) (in xul.pdb)
nsDocument::DoUnblockOnload() (in xul.pdb)
nsDocument::UnblockOnload(bool) (in xul.pdb)
nsBindingManager::DoProcessAttachedQueue() (in xul.pdb)
nsRunnableMethodImpl<void ( nsBindingManager::*)(void),1>::Run() (in xul.pdb)
nsThread::ProcessNextEvent(bool,bool *) (in xul.pdb)
NS_ProcessNextEvent(nsIThread *,bool) (in xul.pdb)
mozilla::ipc::MessagePump::Run(base::MessagePump::Delegate *) (in xul.pdb)
MessageLoop::RunHandler() (in xul.pdb)
MessageLoop::Run() (in xul.pdb)
nsBaseAppShell::Run() (in xul.pdb)
nsAppShell::Run() (in xul.pdb)
nsAppStartup::Run() (in xul.pdb)
XREMain::XRE_mainRun() (in xul.pdb)
XREMain::XRE_main(int,char * * const,nsXREAppData const *) (in xul.pdb)
XRE_main (in xul.pdb)
do_main (in firefox.pdb)
wmain (in firefox.pdb)
__tmainCRTStartup (in firefox.pdb)
BaseThreadInitThunk (in wkernel32.pdb)
__RtlUserThreadStart (in wntdll.pdb)
_RtlUserThreadStart (in wntdll.pdb)

Comment 1

[Ernst Sjostrand]

Same behavior on Linux.

Comment 2

[Ernst Sjostrand]

Attachment: Example

Actually my previous statement was false, putting the above build log in a html file with <pre></pre> makes it a LOT worse, at least on Nightly.

Comment 3

[Ernst Sjostrand]

Oh that gz isn't smaller, it's 33 MB uncompressed.

Comment 4

[Matthias Versen [:Matti]]

A profile could help: https://developer.mozilla.org/en-US/docs/Performance/Profiling_with_the_Built-in_Profiler

Comment 5

[Mitza]

Mozilla/5.0 (Windows NT 6.1; WOW64; rv:21.0) Gecko/20100101 Firefox/21.0
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:22.0) Gecko/20130415 Firefox/22.0
Mozilla/5.0 (Windows NT 6.1; Win64; x64; rv:23.0) Gecko/20130410 Firefox/23.0

I was able to reproduce the issue on Firefox 23 (Build ID:20130410031044), Firefox 22 (Build ID:20130415004014) and on Firefox21 Beta2 (Build ID:20130408165307).
It takes a few seconds for loading the page, but after the page is loaded, the scroll works smoothly.
Does this bug should be moved to Core - Document navigation component?

Comment 6

[John Daggett (:jtd)]

We've had a number of these bugs filed but I'm going to single this one out as a proxy for hangs due to large text file rendering.  Not strictly a "font" hang, probably better described as a layout hang.

Comment 7

[Ernst Sjostrand]

After looking at it a bit more I think this could be split into two bugs.
One for text/plain and one for HTML with a huge <pre></pre> tag.
text/plain seems to be slow because of various "BreakAndMeasureText" and Reflow.

For the <pre> case Nightly started taking 4 GBs of memory so there was some kind of fishy regression going on I think...

So should we keep this bug aboug text/plain and file a new one for the <pre> case?

Comment 8

[:Gijs (out for now; he/him)]

(In reply to John Daggett - out 14 - 24 June (:jtd) from comment #6)
> We've had a number of these bugs filed but I'm going to single this one out
> as a proxy for hangs due to large text file rendering.  Not strictly a
> "font" hang, probably better described as a layout hang.

John, can you triage this to a more appropriate component if this is layout-related? It probably shouldn't stay in Firefox in that case...

Comment 10

[John Daggett (:jtd)]

Another testcase, from bug 942832:

http://people.redhat.com/stransky/ff/log.txt.zip

Comment 11

[Ernst Sjostrand]

It must be possible to optimize the layout code to display text/plain with less CPU overhead, no?

Comment 12

[expletyl]

I don't know it's the same problem or not but Firefox has really hard time opening the following URL, but it can be easily opened in Chrome:

http://fa.wikipedia.beta.wmflabs.org/w/index.php?title=%D8%B9%D8%A8%D8%AF%D8%A7%D9%84%DA%A9%D8%B1%DB%8C%D9%85_%D8%B3%D8%B1%D9%88%D8%B4&diff=715&oldid=714

I tried running Firefox in safe mode and disabling JavaScript, but no luck.

Using Firefox 46.0.1 on a Win 10 64-bit machine.

Comment 13

[Martin Stránský [:stransky] (ni? me)]

Still a problem. For instance build-logs from koji:
https://kojipkgs.fedoraproject.org//packages/firefox/47.0/3.fc24/data/logs/x86_64/build.log

Firefox freezes (UI) and takes 100% CPU.

Comment 15

[Wayne Mery (:wsmwk)]

WFM?

Chrome and Firefox load in same amount of time after initial load of https://fa.wikipedia.beta.wmflabs.org/w/index.php?title=%D8%B9%D8%A8%D8%AF%D8%A7%D9%84%DA%A9%D8%B1%DB%8C%D9%85_%D8%B3%D8%B1%D9%88%D8%B4&diff=715&oldid=714

Comment 16

[Ernst Sjostrand]

Yeah WFM. Multi process also makes everything a lot better here.