Hi dk70,
The bookmarklet of Polonus? Just a demo from the page on Memory Leaking in Wikipedia, patched the piece of demo code there in the Crunchator, and Leak Monitor leaked the results. Nothing to do with risky surfing. Just to demonstrate a point,
polonus
Yeah Leak Monitor works and author is a known figure and contributor to Firefox code/fixes. Think I read some Mozilla Chief Engineer something giving him much credit for 2.0 improvements (bug-hunting I assume). I can tell you many a website/javascript triggered it when I used 1.5.0.x Now on 2.0 it hardly ever happens. Some extensions still make it go crazy. I dont think Mozilla Add-on site should approve extensions which triggers monitor but they obviously do and some extension makers still dont test that much.
I somehow doubt every leak means 500mb used or no one would be able to use Gmail for long but it probably have some negative impact on “memory management”. In some cases may be even stability.
Hi dk70, the leaks are meant to be taken as cumulative leakage over a full 24 hours, not only mem leaks but also idle mem waste.
From your responses, it can be fairly concluded that you are a coder, and you know your way around browsers and appl. quite some bit.
Question to you, why you rank the coders of Flock amongst the crapola crowd? What are the problems with Flock in your opinion.
polonus
PS a tweak to free up some mem in FF:
That should be all it takes. To verify you got it right, load up a few pages in a few tabs. Start up task manager or your platform’s monitoring tool of choice and see how much memory Firefox is using. Now, minimize Firefox…
D
No I have never coded anything except batchfiles.
Did I say anything bad about Flock I should have deleted it. However it share same code as 1.5.0.x and so also the same bugs making it very hard to use a tool like Leak Monitor for a longer periode. Must have been what I meant. Flock is clearly more than just extension pack. http://screencastsonline.com/sco/Shows/files/6bab7806ed46d123ba0175a8a90fcff3-82.html Not for everyone of course. I might use it if it was using 2.0 code - which it will eventually.
If minimize setting dont do anything for used memory you can be sure FF is halfdead. Shows something has gone wrong, can rarely be fixed except by restarting. Should always go down to 10-30mb or so.
1.5 has some hefty max. values for ram cache. Defaults are changed/lowered in 2.0. Why not use those on 1.5/Flock? user_pref(“browser.cache.memory.capacity”, 24576); is what I use user.js with 2gb ram. Is default not my number and naturally helps making 2.0 look better - http://kb.mozillazine.org/Browser.cache.memory.capacity Fast forward/rewind cache I have lowered to 4 http://kb.mozillazine.org/Browser.sessionhistory.max_total_viewers
Have seen some getting better performance by killing this http://kb.mozillazine.org/Browser.chrome.image_icons.max_size I use user_pref(“browser.chrome.image_icons.max_size”, 32); because then I still get favicon but avoid perhaps troublesome thumbnailing.
FF wont ever be notepad no matter what.
nsWindow.cpp 27 Oct 200 04:30:54 -0000
@@ -71,6 +71,7 @@
#include “nsTransform2D.h”
#include “nsIEventQueue.h”
#include <windows.h>
+#include <psapi.h>
// unknwn.h is needed to build with WIN32_LEAN_AND_MEAN
#include <unknwn.h>
@@ -275,6 +276,7 @@
static PRUint32 gLastInputEventTime = 0;
static int gTrimOnMinimize = 2; // uninitialized, but still true
+static int gRestoreSavedWorkingSet = 2; // uninitialized, but still true
#if 0
static PRBool is_vk_down(int vk)
@@ -1540,6 +1542,7 @@
(conveniently created before any visible windows and after
the profile has been initialized) */
gTrimOnMinimize = 1;
gRestoreSavedWorkingSet = 1;
nsCOMPtr prefs = do_GetService(NS_PREFSERVICE_CONTRACTID);
if (prefs) {
nsCOMPtr prefBranch;
@@ -1550,6 +1553,11 @@
&trimOnMinimize))
&& !trimOnMinimize)
gTrimOnMinimize = 0;
PRBool restoreSavedWorkingSet;
if (NS_SUCCEEDED(prefBranch->GetBoolPref("config.restore_saved_workingset",
&restoreSavedWorkingSet))
&& !restoreSavedWorkingSet)
gRestoreSavedWorkingSet = 0;
}
}
}
@@ -3707,6 +3715,113 @@
return TRUE;
}
+static LPDWORD CaptureWorkingSet()
+{
(lpNewBufferData[0] + 1) * sizeof(DWORD) > dwBufferSize) {
HeapFree(GetProcessHeap(), 0, lpNewBufferData);
return NULL;
HeapFree(GetProcessHeap(), 0, lpNewBufferData);
lpResizedBuffer = (LPDWORD)HeapAlloc(GetProcessHeap(), 0, dwBufferSize);
if (!lpResizedBuffer) return NULL;
+static void SaveOrRestoreWorkingSet(BOOL SaveOrRestore)
+{
HeapFree(GetProcessHeap(), 0, lpOldBuffer);
+#if 1
// Capture a fresh copy of the WorkingSet so that we can measure
// how much we're swapping in.
LPDWORD lpTmpBufferData = CaptureWorkingSet();
if (lpTmpBufferData != NULL) {
DWORD dwNumPresent = 0;
DWORD cPages = lpTmpBufferData[0]; // First DWORD is page count
for ( DWORD i = 1; i <= cPages; i++ ) {
DWORD pageAddr = lpTmpBufferData[i] & 0xFFFFF000;
DWORD pageFlags = lpTmpBufferData[i] & 0x00000FFF;
if ((pageFlags & 31) > 0 && (pageFlags & 31) < 16) {
dwNumPresent++;
}
}
HeapFree(GetProcessHeap(), 0, lpTmpBufferData);
lpTmpBufferData = NULL;
char tmpstatus[200];
_snprintf(tmpstatus, sizeof(tmpstatus), "There are currently %u pages (%u KB) present.\n", dwNumPresent, 4 * dwNumPresent);
OutputDebugString(tmpstatus);
}
+#endif
+
+
// Start forcing the pages back into memory.
DWORD dwNumForced = 0;
DWORD cPages = lpLocalData[0]; // First DWORD is page count
for ( DWORD i = 1; i <= cPages; i++ ) {
DWORD pageAddr = lpLocalData[i] & 0xFFFFF000;
DWORD pageFlags = lpLocalData[i] & 0x00000FFF;
if ((pageFlags & 31) > 0 && (pageFlags & 31) < 16) {
// Force the page to be swapped in.
IsBadReadPtr((VOID*)pageAddr, 4096);
dwNumForced++;
}
}
// Log how many pages were forced back into memory.
+#if 1
char tmpstatus[200];
_snprintf(tmpstatus, sizeof(tmpstatus), "Forced %u pages (%u KB) to become present.\n", dwNumForced, 4 * dwNumForced);
OutputDebugString(tmpstatus);
+#endif
+
// Put this workingset buffer back into the array so that it can
// be used again until the workingset is captured again.
if (InterlockedCompareExchangePointer((PVOID*)&lpBufferData, lpLocalData, NULL) != NULL) {
// the exchange did not occur because a newer buffer was
// already put in its place, so just free the older buffer.
HeapFree(GetProcessHeap(), 0, lpLocalData);
}
// Check for pending paints and dispatch any pending paint
// messages for any nsIWidget which is a descendant of the
// top-level window that this window is embedded within.
@@ -4500,6 +4615,17 @@
::ShowWindow(mWnd, SW_SHOWMINIMIZED);
result = PR_TRUE;
}
if (gRestoreSavedWorkingSet) {
if (wParam == SC_MINIMIZE) {
OutputDebugString("gRestoreSavedWorkingSet is saving the current working set\n");
SaveOrRestoreWorkingSet(TRUE);
OutputDebugString("gRestoreSavedWorkingSet is done saving\n");
} else if (wParam == SC_RESTORE) {
OutputDebugString("gRestoreSavedWorkingSet is restoring the original working set\n");
SaveOrRestoreWorkingSet(FALSE);
OutputDebugString("gRestoreSavedWorkingSet is done restoring\n");
}
}
break;
default:
polonus
Go tell them at Bugzilla - anyone can contribute or at least get into a discussion. Problem is you can file a bug for 1.5.0.x only to be told it is fixed in 2.0 daily whatever = considered fixed by Mozilla though not yet available in official final release. Then there is 3.0 which I have not even tried! So hard for normal people to take part in. Firefox always on the move and often on different levels.
Anyway, I still think it would be nice to have a new version 2.1 or something which only had the goal of optimizing and cleaning. If you check some Mozilla blogs you can see some think about the compromise betweem adding and tuning http://gemal.dk/blog/2005/11/10/nice_checkins_for_the_weekend/ “It seems that reviewers generally like to review new functionality than to review changes/optimizations to current code” Just the way it is - current code is old in 6 months cause then we are targeting 3.0 so why bother!. There is probably some opensource/Mozilla/general market situation logic to this missing cleanup version. Normal users think: it works or it does not - and clearly it does. Remember most people dont install any or only very few extensions and they sure play a big part of memory problems. Possible only the “geek” group are crying that loud…
I need to check out Opera and IE7 some more but they wont be notepad either. We are in 2006 and if Firefox runs sweet on a computer with lets say 512mb ram what is there to complain about? As said before activate leak monitor on 1.5.0.x then try 2.0 - lots of progress, shaving off 1.5555 seconds of startup time is almost irelevant though they claim 2.0 have improved on that as well. Cant say I care or understand why it is so important so some people. Like XP boot time, I lack understanding.
Hi dk70,
Tested Flock with the Leak Gauge script, and these were the results:
Summary:
Leaked 0 out of 0 DOM Windows
Leaked 0 out of 0 documents
Leaked 0 out of 0 docshells
Summary:
Leaked 0 out of 0 DOM Windows
Leaked 0 out of 0 documents
Leaked 0 out of 0 docshells
Not bad, not bad at all,
polonus
Well try Gmail or a Vbulletin forum. I would imagine Yahoo xxxxx as well. Or a site where sitemeter.js is active, part of a statscript. I think opera.com menu script triggers it as well 
Dont know how much Flock coders have changed but at least in regular FF 1.5.0.x there are loads of leaks/problems. The Gmail problems have been known for ages so may Flock have fixed those? Of course these has to be translated and it is wrong to only blaim firefox. However, definitely fewer with 2.0… Has not fixed extensions of course but anyone can email author or file a bug. Must be careful not to judge too quickly because may be X extensions changes environment so the otherwize perfect Y breaks. X is problem not Y. Ideally one has to test an extension on a clean profile to exclude negative outside influence. Send in the debuggers. What is cool is anyone can use Leak Monitor and findings are accepted as genuine and reproducable problems. Why some extension makers dont use it is a good question. Mozilla Add-on site should encourage it more, their “reviews” are poor QA.
Hi dk70,
What I have grasped from this thread discussion is that the problem comes with extension coders not following the same routines to avoid the known mem leaks. AdBlock Plus 0.7 had serious problems, NoScript problems are still there.
The point where I found Flock to leak is when you download programs, but that is understandable. The best solution is certain code routines.
But when an outside extension is really causing problems like race conditions etc. like a certain version of MacAfeeSiteAdvisor, it can force you to nake a fresh reinstall and trouble taking your profiles along. Learn to know what your browser is doing under the hood with TamperData is instructive.
polonus
Tamperdata is beyond what most enjoy. Next is to download source, fix stuff and compile yourself. Mozilla would like world domination so they must deliver. Also Firefox is a simple browser out of the box for a reason. Supposed to be used by Grandma as well as the gee. So problems they solve 8) If users really need to do their own diagnostics I think they have failed somehow. In this case making extension site look like an ice cream bar. They have more or less useless rating system and just about anything gets approved. If you go ask about a problem on Mozillazine it is very likely you will be asked to list your extensions and/or test with new profile. First thing to look into when something is weird. Mozilla encourage extensions uncritically.
Not really much to do about Flash, Java, Windows Media plugins. Think they all have a version/bug log similar to Firefox and most know about leaks/hangups during or after video loading. Latest Flash seem to be a lot better though. I assume Firefox is about not guilty when it comes to perhaps IE optimized webcode filled with media.
Well it was/is the old Adblock 0.5 which have most problems. Adblock Plus 0.7 saved the day with a total recoding. http://adblockplus.org/en/ Sure a few bugs but nothing major - he was one of the first to make use of Leak Monitor btw. Really a super extension, nice forum, actively developed etc. In case you missed it you dont have to use strange regular expressions any more, like in way too fat superG. Make a “dumb” mile long list of what you want to block and Adblock optimize that list internally - which is faster than regular expressions according to author. Practically transparent if you use own personalized list and not that superG. There are other “list-makers” should you want a starting point.
And you are right about what you said in PM. Session Saver is to be avoided (same author who did old Adblock, has gone zzzzzz long ago) and Session Manager https://addons.mozilla.org/firefox/2324/ or Tab Mix Plus to be used. At least on 1.5.0.x, author of Session Manager and Crash Recovery is involved in 2.0 coding for FF own Session handling (about time!) For same reason his extensions wont be updated for 2.0 as I understand it.
Hi dk70,
The forum should be glad, with someone around like you. Well your point about different ways of computing, different ways of looking at it, is valid. The newbies and the recreational computer users like the things that come out of the box, and they also could not survive without basic security that come “as default”. The advanced users know more abou the vulnerability gap, and their multi-layered security measure, make them more or less sniff out where problems may ly ahead, and they aren’t that dependant on others to provide solutions (also so with AV). This are some very general remarks, fed by experience.
FF or Flock made extensions like Christmas tree decoration, and they did it while some like a simple tree and not everybody is in need of the same extensions.
Extensions are not treated with the same programming routines general code that goes into FF or Flock is. Whenever there are problems later, someone get the bug assigned, and you know as well as I do, prevention and safe coding is always better then having to get rid of the bugs later. Then Mozilla is nothing else like in the old days: Internet Now with Netscape has been changed for Internet Now with Firefox.
The average user is not like the old pol, that fires up the old
REC 2.0 - Reverse Engineering Compiler from: http://www.backerstreet.com/rec/rec.htm to see what is there.
But then when I arrive at some insights, I can tell you why, and how, and not based on the authority of someone else (or this person knows what he is talking about, and lucky for us they are still around). Thanks for sharing your views with us in this thread, I think there are people that certainly will benefit from these insights,
polonus (anti-malware activist)