Dunno what the score is with confirmation of Enhancements, but this seems sensible, so confirming. It touches a lot of stuff, so leaving in Browser-General. Gerv
infinite reflow is related to 31574, and that is a crasher as well. Marking dupe, as solving that should fix this. *** This bug has been marked as a duplicate of 31574 ***
ok, I was too quick on marking this a dupe. moving to layout, think that might be the correct component.
sounds like what is needed here is some sort of popup like IE's that says something to the effect of: "A script on this page is taking a long time to complete execution. [ Continue ] [Stop running script ]" (with better wording than that...)
I think it would be better if mozilla could remain responsive (able to navigate, close, etc.) while the script is running. Otherwise, malicious page authors are likely to find a way around the code that determines whether the script is hogging resources or preventing user interaction.
I guess I'll just throw in another suggestion: don't allow the user to interact with the page while a script/alert box is running (edit text boxes, click on links, activate mouseovers - essentially disabling the page UI itself, perhaps with a transparent event sink like a glasspane over it), but *do* enable the stop/back/forward buttons. They can scroll through the page, but not modify anything on it - something that could possibly screw up the script. If the action would invalidate the currently running script(hitting back/stop/forward, entering a URL), pop up a warning saying so.
Adding myself to CC: list.
Adding crash keyword.
Moving to DOM Level 1 and reassigning. I think this is a duplicate, though.
> A better solution to this problem would be to add code in strategic places, > such as reflow iterations, that force the browser to check if any user input > (such as an attempt to close the browser) would make continuing the > infinite/slow reflow moot. I completely agree. You should always insert calls to the event loop after some ms. It is not even hard to implement, if you e.g. call it once during a reflow and in any central place in the compiled JS code. It should be done soon, though, because this might have some side-effects, so needs testing. Changing SUMMARY to davidr's suggested one. Assuming davidr's claims are correct (especially that |while(1)| in a web script or infinite reflows cause a freeze), I don't think, this is an enhancement request, but a design bug, changing SEVERITY to major. I suggest to file separate bugs for layout, JS and so on.
This bug has been marked "future" because the original netscape engineer workingon this is over-burdened. If you feel this is an error, that you or another known resource will be working on this bug,or if it blocks your work in some way -- please attach your concern to the bug for reconsideration.
Bug 45576 - Hang on joining #mozillazine I'm concerned. I've marked helpwanted. If you don't like this bug I'll assign it to nobody. Added keywords, nothing in this world gets done w/o nsbeta3. Bug reassigned to default victim. Again, if you don't like this bug please send it somewhere. nobody is asking for it. This is a catchall, but Ben's right, we should at the very least be able to do time counts on threads. init does stuff like this, if it needs to respawn the console tty too many times it throttles it [odd case: no user is avail to prompt because interface is being throttled, but ...].
> Ben's right, we should at the very least be able to do time counts on threads What I meant was no counter, but to make sure, the event loop is called by just inserting calls at the right place. With only that done (rightTM), the infinite loops would still occur, but the app wouldn't freeze, which is already a progress. More importantly, even in normal operation, the app were more responsive. Maybe, my suggestion was orthogonal to this bug, dunno.
I don't know how appropriate it is to say that this bug "blocks" various freeze bugs. Having this feature implemented well would make mozilla remain responsive during freezes, so, for example, you could close a slow-rendering webpage without killing mozilla. Most of the individual frezze bugs should still be fixed, however, because they might be triggered by non-malicious webpages that we actually want to display completely. Instead of being critical/crash bugs, they would become major or normal bugs (with the perf keyword?), but they'd still be there.
Seems like no (or too few) event loop calls are done during the frame construction.
minusing per PDT rules.
I don't think this is a crash bug.
*** Bug 51945 has been marked as a duplicate of this bug. ***
*** Bug 98087 has been marked as a duplicate of this bug. ***
I have one comment: doing recursive calls to event handling can cause hard to find/reproduce bugs. Has anyone considered the possibility to run XUL in one thread and the HTML in another one? They should be largely separate worlds because of security if nothing else. (No sharing of JS contexts or anything.)
I want to add in general that keeping the browser responsive at all times should become a major priority, not just to deal with infinite loops, but to deal with all sorts of delays. For example, I have broadband and I typically set my Google preferences set to return 100 items per page --- this greatly reduces the time spent hitting the "next" button when I am looking for content. However, using Mozilla it can take several seconds before the browser decides to allow me to click on a link in the search. What's even more annoying is that when I hit the "back" button to return to the search page, again the page is unresponsive for a few seconds. This on a Pentium II 400 laptop with 192 megs of RAM. Suffice it to say that IE performs much better in this respect, and this is one of the main subjective performance issues that differentiates IE from Mozilla. This occurs not only with Google searches but with many pages: links are not active, it is impossible to even scroll a page, etc. There are far too many situations where potentially long processes that should be handled in the background are completely tying up the main user interface thread. Fixing this problem might require a great deal of work, I don't know, but I think it ought to be a priority.
Mitsu, Increasing performance and responsiveness is our top priority, as a matter of fact. Thanks for your comments. However, this discussion is not relevant to this particular bug - we try to keep discussion in bugs focused on the specific issue at hand.
Sorry --- I had originally thought to enter a new bug, but I figured it would be better to find a related bug and enter the comment there. It seemed to me (perhaps I am wrong) that in general, a single general approach would solve both the problem of keeping the browser responsive during infinite loops and during any extended operations. I.e., the general principle being that nothing should hold up the user interface thread for more than a very brief interval, and all operations should be thread-safe. If there is more appropriate bug or RFE which is already open where I should place these comments, I'd be happy to do so.
See also bug 128666 about showing the display during tight loops.
*** Bug 128666 has been marked as a duplicate of this bug. ***
This bug is very visible on large doxygen output with the tree browser. I've created an example at <http://apt.slamb.org:8080/~slamb/std-doxygen/> (the standard C++ library as included with g++ 3.0.4). It takes 270 seconds to load on my 1GHz Athlon with 768MB of RAM. During that time, Mozilla is completely unresponsive - it doesn't even handle expose events.
Scott: that's a general problem with loading content from a hard drive, from a fast server, or from the cache. See bug 129640, "Incremental rendering is busted on fast connections".
Sorry Scott, you're right.
*** Bug 150593 has been marked as a duplicate of this bug. ***
Moving all threading bugs to XPCOM. See bug 160356.
By the definitions on <http://bugzilla.mozilla.org/bug_status.html#severity> and <http://bugzilla.mozilla.org/enter_bug.cgi?format=guided>, crashing and dataloss bugs are of critical or possibly higher severity. Only changing open bugs to minimize unnecessary spam. Keywords to trigger this would be crash, topcrash, topcrash+, zt4newcrash, dataloss.