[Don Pickett]

Originally Posted by Eug 

When surfing to new pages in Safari, why does the new page sometimes start in the left upper corner and then expand to fill the page?

When surfing to new pages in Safari, why does the old page sometimes shrink to the bottom right corner?

is there any way to stop this behaviour?

Once again, never seen anything like that.

[Eug]

Somehow I think it's intentional, perhaps a by product of swipe gestures that I've inadvertently activated in Safari on my Magic Mouse.

However, I can't replicate it, nor can I find any setting to turn it on/off. It just happens randomly. Ultimately, it's just annoying.

---

And yeah, spell check is pissing me off too.

[lpkmckenna]

Originally Posted by besson3c 

Safari flies until you start loading a bunch of tabs, then it completely and utterly sucks right now due to bugs.

This is certainly true in Snow Leopard. Safari 5.1 will crap out hard now. It will beachball for a long time, while at the same time the CPU monitor won't indicate any activity.

[Don Pickett]

Originally Posted by lpkmckenna 

This is certainly true in Snow Leopard. Safari 5.1 will crap out hard now. It will beachball for a long time, while at the same time the CPU monitor won't indicate any activity.

Something's wrong with your installs, boys and girls. Never seen anything like that.

[lpkmckenna]

Originally Posted by Don Pickett 

Something's wrong with your installs, boys and girls. Never seen anything like that.

You could be right. I think some Safari extensions might be causing problems. I'm gonna prune them.

But Safari's performance with many tabs open is much poorer than Chrome. If Chrome wasn't so feature-poor, I would have switched. Now I just try to keep the number of tabs low.

[besson3c]

Based on my perceptions it seems that more people have had problems with Webkit 2 than not. Consider yourself lucky don...

Lmpckenna: extensions don't seem to have an impact - at least not for me

[lpkmckenna]

@Don Pickett: John Gruber has noted the sluggishness of Safari 5.1 in this recent post. It's not an isolated issue.

[Don Pickett]

Originally Posted by lpkmckenna 

@Don Pickett: John Gruber has noted the sluggishness of Safari 5.1 in this recent post. It's not an isolated issue.

It may not be isolated, but it may be the result of issues outside of Safari, like extensions. I only run three.

[besson3c]

Originally Posted by Don Pickett 

It may not be isolated, but it may be the result of issues outside of Safari, like extensions. I only run three.

It is reproducible without any extensions, I think several people have confirmed this in addition to myself.

[Don Pickett]

Originally Posted by besson3c 

It is reproducible without any extensions, I think several people have confirmed this in addition to myself.

Then it's related to something else. I haven't seen any weirdness with Safari since upgrading to 10.7. So, either I have the magical, unicorn-enhanced version of Safari which never has any problems, or there's an issue here outside of Safari.

[besson3c]

Originally Posted by Don Pickett 

Then it's related to something else. I haven't seen any weirdness with Safari since upgrading to 10.7. So, either I have the magical, unicorn-enhanced version of Safari which never has any problems, or there's an issue here outside of Safari.

Or you are not visting sites that give it fits?

[Don Pickett]

Originally Posted by besson3c 

Or you are not visting sites that give it fits?

No idea. But, when in doubt, always go with unicorns.

[besson3c]

Originally Posted by Don Pickett 

No idea. But, when in doubt, always go with unicorns.

That's good general life advice


What concerns me more than anything is not so much that there might be bugs or memory usage problems or whatever that cause problems, but that when this happens it brings down all of Safari giving you that dialog that not only forces you to reload the page you are on like Chrome does, but reload every page you have loaded, even the ones that have been idling for some considerable amount of time. I mean, I thought the whole point of Webkit 2 was to sandbox each individual page so that there is nothing that can happen that will cause the entire app to implode like that?

Regardless of whether or not there is something specific that is triggering these problems, dialogs prompting you to reload every page you have open are just not cool. At all.

[lpkmckenna]

Originally Posted by besson3c 

I thought the whole point of Webkit 2 was to sandbox each individual page so that there is nothing that can happen that will cause the entire app to implode like that?

Webkit 2 doesn't do this. Pages are not sandboxed from each other. The page rendering is now separated from the rest of the app functions, but that's it.

[chabig]

I don't have big problems but I'd agree that some things are slower on 10.7 than on 10.6. Still, I like Lion and I'm going to put my trust in 10.7.2.

[besson3c]

Originally Posted by lpkmckenna 

Webkit 2 doesn't do this. Pages are not sandboxed from each other. The page rendering is now separated from the rest of the app functions, but that's it.

Are you sure?

WebKit2 is a new API layer for WebKit designed from the ground up to support a split process model, where the web content (JavaScript, HTML, layout, etc) lives in a separate process from the application UI. This model is very similar to what Google Chrome offers, with the major difference being that we have built the process split model directly into the framework, allowing other clients of WebKit to use it.

Google Chrome's scheme is setup so that there are many processes open when you have many pages open. I don't know if it is one per tab or what the scheme is, but it is definitely more than two (one for UI, one for rendering engine). Plugins get their own process too. Maybe the Javascript engine gets one too?

I probably don't understand the technical details, but my understanding is that the main objective is to sandbox as much as possible? AFAIK, the only thing that got sandboxed in Webkit 1 was plugins. To Webkit 2's credit, I've noticed that the same more pages open = everything is slower relationship doesn't exist, but when more pages are open this is where the bugs surface the most often, so I'm optimistic that these problems will be ironed out in due time. Maybe having to reload every page that is open when there is a problem is also a bug.

[Spheric Harlot]

Activity Viewer is currently showing one process called "Safari" (12 threads) and one called "Safari Webinhalt" ("Safari Web Content") (45 threads).

I have 26 tabs open.

[Don Pickett]

Originally Posted by Spheric Harlot 

I have 26 tabs open.

There aren't 26 things I want to read.

[besson3c]

Originally Posted by Don Pickett 

There aren't 26 things I want to read.

I guess it is interesting to know how many tabs people keep open on a regular basis, but like Spheric I keep a lot of tabs open too since I"m a web developer, so you'll have to take our word for it that these problems exist, and that there are people who need/want to have all of those tabs open.

[besson3c]

Originally Posted by Spheric Harlot 

Activity Viewer is currently showing one process called "Safari" (12 threads) and one called "Safari Webinhalt" ("Safari Web Content") (45 threads).

I have 26 tabs open.

Ahh, so Webkit 2 works by spawning new threads in a single process, while Chrome actually spawns additional processes?

[Don Pickett]

Originally Posted by besson3c 

I guess it is interesting to know how many tabs people keep open on a regular basis, but like Spheric I keep a lot of tabs open too since I"m a web developer, so you'll have to take our word for it that these problems exist, and that there are people who need/want to have all of those tabs open.

This is the internet. I'm not sure any of you actually exist.

[Spheric Harlot]

Originally Posted by Don Pickett 

There aren't 26 things I want to read.

Work research doesn't figure into your daily routine.

[lpkmckenna]

Originally Posted by Don Pickett 

There aren't 26 things I want to read.

This is my usual routine: I'll go to a page, say MacNN new threads, then open every new thread in its own tab, then read thru each tab one by one. Or I'll go to Google News and open every story I want to read in its own tab. See?

[lpkmckenna]

Originally Posted by besson3c 

Ahh, so Webkit 2 works by spawning new threads in a single process, while Chrome actually spawns additional processes?

Exactly.

[Don Pickett]

Originally Posted by lpkmckenna 

This is my usual routine: I'll go to a page, say MacNN new threads, then open every new thread in its own tab, then read thru each tab one by one. Or I'll go to Google News and open every story I want to read in its own tab. See?

Why not read a thread, use that handy back button, read another one, etc.?

[The Final Dakar]

Originally Posted by Don Pickett 

Why not read a thread, use that handy back button, read another one, etc.?

Loading times.

[P]

On older machines, also the rendering times (JS etc) - I do the same thing. For these forums, I middle-click all the subforums I'd like to read, and then close the current tab. After that, I middle-click each topic I like and close the front tab and move on like this. Eventually (depending on the tab opening scheme of the browser I use, they vary a bit) I usually have one tab per topic, which I then close as I like. It's quite convenient: Left hand on Command-W, moving to ctrl-tab if I want to keep one tab open for one reason (actually even more convenient on Windows, where it's ctrl-W to close). also works well when I type an answer - I click Post and then tab on, leaving the server to catch up when it feels like it.

[Don Pickett]

You people think about this way too much.

Anyway, 'nuff said.

[Spheric Harlot]

Originally Posted by Don Pickett 

You people think about this way too much.

Anyway, 'nuff said.

heh. I use the "back" and "forward" buttons/swipes for the forums. I'm on a 100Mbit pipe, so it's pretty instant.

11 of my currently open tabs are a research topic, another 9 are a tutorial series, and the rest is current sites/forums/news etc.

[besson3c]

Originally Posted by lpkmckenna 

Exactly.

What did Apple hope to gain designing Webkit2 this way? I mean, a Unix OS kernel is all about resource allocation, so why did Apple reinvent the wheel?

[P]

By splitting the web content from the browser? Security - a bug in the renderer or Javascript engine will not be able to affect anything outside its very limited sandbox.

The threading is of course to make better use of multicore processors.

[besson3c]

Originally Posted by P 

By splitting the web content from the browser? Security - a bug in the renderer or Javascript engine will not be able to affect anything outside its very limited sandbox.

Nor will it in Chrome where it is a separate process

The threading is of course to make better use of multicore processors.

Isn't this also accommodated with separate processes, or are threads somehow more optimized for this sort of thing?

[Big Mac]

You'd think they'd use separate processes, but either WebKit2 was a stopgap and WebKit3 will spin off processes for each page, or the Safari team found some compelling reason to keep all of them in one process. It seems like the Chrome model is substantially better in that respect. Does Chrome suffer from undo memory or processor hungriness by using separate processes?

[CharlesS]

Originally Posted by P 

By splitting the web content from the browser? Security - a bug in the renderer or Javascript engine will not be able to affect anything outside its very limited sandbox.

I don't think he was asking why the renderer is separate from the app, but rather why separate pages aren't run in their own processes, which would probably be more secure.

Originally Posted by besson3c 

Isn't this also accommodated with separate processes, or are threads somehow more optimized for this sort of thing?

Either threads or processes will take advantage of multicore processors. The one place where threading might be more optimized is if they're using GCD, which is able to determine the best number of threads to run at once based on the available system resources, current load, etc. This could come in handy if you were to reload a really large number of tabs at once, for example, and could help avoid making the app do so much work at once that the CPU bogs down and the UI becomes unresponsive.

Note that I haven't actually looked at the source code, so I don't know for sure if they're actually doing this or not, although it seems like a pretty safe bet.

There's probably also a bit more overhead running all the separate processes, but I don't know whether or not it's enough to make a difference.

[besson3c]

Originally Posted by CharlesS 

I don't think he was asking why the renderer is separate from the app, but rather why separate pages aren't run in their own processes, which would probably be more secure.

Yup..

Either threads or processes will take advantage of multicore processors. The one place where threading might be more optimized is if they're using GCD, which is able to determine the best number of threads to run at once based on the available system resources, current load, etc. This could come in handy if you were to reload a really large number of tabs at once, for example, and could help avoid making the app do so much work at once that the CPU bogs down and the UI becomes unresponsive.

Note that I haven't actually looked at the source code, so I don't know for sure if they're actually doing this or not, although it seems like a pretty safe bet.

There's probably also a bit more overhead running all the separate processes, but I don't know whether or not it's enough to make a difference.

Thanks for this explanation/hypothesis, but I'm wondering if you could clarify your very last comment here?

I'm assuming that in the interest of code portability that Webkit2 does not require any particular kernel level sort of modifications. Therefore, when there is a memory leak or memory otherwise starts to balloon, with multiple processes the kernel can take memory away from those idle pages the Chrome way, whereas the Webkit2 way the kernel just has to assign more memory to Webkit2/Safari as a whole, right? I don't think Webkit2/Safari will claim unused memory from idle pages, will it?

If not, wouldn't it require more overhead to use Webkit2/Safari as memory consumption starts to balloon? Or, was your statement referring to the overhead needed for normal usage when there are no rogue pages causing Webkit2/Safari to go apeshit?

[CharlesS]

The latter. Each process will have a small amount of overhead to launch it and to keep it running, although it may not be enough to be noticeable, and you are certainly correct that if there are large memory leaks, that would easily overshadow it.

[besson3c]

Originally Posted by Big Mac 

You'd think they'd use separate processes, but either WebKit2 was a stopgap and WebKit3 will spin off processes for each page, or the Safari team found some compelling reason to keep all of them in one process. It seems like the Chrome model is substantially better in that respect. Does Chrome suffer from undo memory or processor hungriness by using separate processes?

It could be that Apple thinks that it can do a better job than its own kernel by creating another layer that allocates and deallocates resources to Safari in a single process, but I'm trying to understand if this is really their thinking, and if so what the justification is. AFAIK you are almost always going to get better performance doing something like this at a lower level, i.e. the kernel level. The downside is often the complexity of operating at a lower level, but in this case that work is already done. I don't see how Webkit2 would be a stopgap measure to the Chrome way, because I don't think it is terribly complicated to do what Chrome does in comparison to the Webkit2 way, which seems substantially more complex.

Like I said to Charles, it would also seem like the Chrome way would be less memory/processor intensive since the kernel could deallocate resources from idle pages, whereas with Webkit2 all the kernel recognizes is a single process.

[besson3c]

Originally Posted by CharlesS 

The latter. Each process will have a small amount of overhead to launch it and to keep it running, although it may not be enough to be noticeable, and you are certainly correct that if there are large memory leaks, that would easily overshadow it.

It would also seem like sensible design to better accommodate the worst case scenarios, especially since it is not at all uncommon for quirks mode, Javascript loops/setInterval, Javascript having to fetch and parse large data structures, and whatever else to challenge a browser.

With plugins like Flash in their own process (I think?!) this no doubt helps considerably, but wouldn't it also be more effective for these processes to be children of individual processes rather than a massive parent process?

[CharlesS]

Originally Posted by besson3c 

I don't see how Webkit2 would be a stopgap measure to the Chrome way, because I don't think it is terribly complicated to do what Chrome does in comparison to the Webkit2 way, which seems substantially more complex.

While I agree with most of what you've said, I don't think it's more complex. They could be using GCD, which really does simplify this stuff quite a bit and would likely make it much more simple this way (no need to figure out how many processes to spawn, or for IPC between the processes). It really is easy to use — you just make a block / operation object for each thing you want to do and give them all to the dispatch queue — GCD does all the hard work of managing and running the threads for you.

Of course, the flaw in this hypothesis is that WebKit is cross-platform, which means it will run on other platforms that don't necessarily have GCD. Of course, GCD is open-source, and Apple could conceivably bundle it in for Windows and Linux, but if they're not doing that, then I don't know.

[besson3c]

Originally Posted by CharlesS 

While I agree with most of what you've said, I don't think it's more complex. They could be using GCD, which really does simplify this stuff quite a bit and would likely make it much more simple this way (no need to figure out how many processes to spawn, or for IPC between the processes). It really is easy to use — you just make a block / operation object for each thing you want to do and give them all to the dispatch queue — GCD does all the hard work of managing and running the threads for you.

Of course, the flaw in this hypothesis is that WebKit is cross-platform, which means it will run on other platforms that don't necessarily have GCD. Of course, GCD is open-source, and Apple could conceivably bundle it in for Windows and Linux, but if they're not doing that, then I don't know.

Ahhh... I figured based on what you wrote about GCD that it was a component provided by the OS itself, and not just compiled into the apps that use it? IOW, can GCD be provided on Windows and Linux just by compiling it into Safari?

Maybe I should horse around with Safari 5.1 for Windows to see if I can reproduce these sorts of problems there. It would be amusing if the Windows version didn't exhibit these same problems

[CharlesS]

Originally Posted by besson3c 

Ahhh... I figured based on what you wrote about GCD that it was a component provided by the OS itself, and not just compiled into the apps that use it?

Yep, in Snow Leopard and up.

IOW, can GCD be provided on Windows and Linux just by compiling it into Safari?

Apple released the source a few years ago with the intention of making it a cross-platform standard, so I would assume it could.

[P]

Originally Posted by Big Mac 

You'd think they'd use separate processes, but either WebKit2 was a stopgap and WebKit3 will spin off processes for each page, or the Safari team found some compelling reason to keep all of them in one process. It seems like the Chrome model is substantially better in that respect. Does Chrome suffer from undo memory or processor hungriness by using separate processes?

In general, there is a greater overhead to launching separate processes than spinning off a thread, and it is much harder to communicate significant amounts of data between different processes than between threads in the same processes. This is slightly different between kernels, however - Linux is especially good at keeping the overhead for forking a new process low, while xnu at least used to have a fairly high overhead for that. I know Apple has made improvements, and I don't know the details, but it may simply be that their performance with a single context, multiple thread setup was so much better that they opted to do it that way.

Preview can't stand mosaics.

Another problem I have with Preview: I cannot resize mosaics that are 15000 pixels in width. Try it. Preview crashes!

Has this been fixed in 10.7.2? Or should I not bother with the update? I don't want Apple to make things any worse than 10.7.1.

[Waragainstsleep]

Putting aside my extraordinary rage at the loss of four fingered horizontal swiping to switch apps, does anyone else find it murderously annoying that you can't switch from the Finder to a full screen app using Command-Tab? Its driving me berserk.

[TETENAL]

Originally Posted by Waragainstsleep 

does anyone else find it murderously annoying that you can't switch from the Finder to a full screen app using Command-Tab? Its driving me berserk.

Works for me.

[Eug]

Originally Posted by Eug 

Somehow I think it's intentional, perhaps a by product of swipe gestures that I've inadvertently activated in Safari on my Magic Mouse.

However, I can't replicate it, nor can I find any setting to turn it on/off. It just happens randomly. Ultimately, it's just annoying.

I have now not only seen the page shrink to the bottom right corner, but also the new page fall from the top. I have seen this in 10.7.2 as well. I see no setting for this anywhere. Very weird.

I still wonder if it's a Magic Mouse gestures thing.

Does anyone else keep getting the SCHMUCK error with 10.7.2?

[MrsLarry]

Anyone else totally annoyed that you can't open attachment directly in Mail 5.1?

For example, I have an email with a PDF attached. I'd like to open in in Acrobat, but my only options are viewing it inline, Save or Quick Look (in Preview).

[Demonhood]

It annoys the heck out of me. I liked being able to see the little icons representing the attachments and do whatever I wanted with them. Replacing all of those with a "quick look" button is not sufficient.

[Thorzdad]

Wow. Yet another reason not to upgrade.
Reading about these backward moves in Lion is getting pretty depressing.