Mozilla 0.9.6 Released

bluephone writes: "Yessireebob. mozilla.org has released the 0.9.6 milestone. Here are Release Notes and a link of files on the FTP server. For milestones 0.9.7 and 0.9.8, the focus is on performace enhancingment, and stability of the Mail/News end of the suite. And boy, is it getting good..."

Comments

Cross-platform performance.

[reaper20]

Very nice release so far, mail/news seems to be "catching up" to the browser function.

The tabbed browsing is almost up to galeon-level, though the speed is still slow, and its missing an (X) to close individual tabs. Use ctrl-w to close tabs in the meantime. This feature is quickly becoming my favorite.

One thing that continually bugs me is the total lack of performance of the linux builds compared to the windows builds. On windows, moz is FAST, and getting faster, and I don't mean just the turbo-load stuff ... does anyone have a reasonable explanation on why the performance is so radically different between linux/win.

From my daily usage, mozilla on windows is "done" as far as for what I need to do, on linux, it still has a long way to go.

What is making mozilla slow on linux?

go Mozilla!

Re:Cross-platform performance.

[rdean400]

One thing I've always been curious about is the effect of the compiler on performance. VC++ produces fairly quick code. How does GCC code compare?

Re:Cross-platform performance.

[Anonymous Coward]

While linux's implementation of pthreads can be considered lacking against other POSIX implementations, it provides for that which makes sense. Additionally, as threads and processes are the same from linux's internal perspective (both created via clone()), you can't really blame it on pthreads. From tests that I've run, pthread_create is actually faster then fork on UP systems, so unless you have a more detailed analysis of why pthreads are the problem, I'm going to call you on this one.

Standards smandards, pthreads under linux is in no way fully compliant w/ the POSIX standard (but then again, who really wants that sorta insanity on their disks).

Re:Cross-platform performance.

[Anders]

One thing I've always been curious about is the effect of the compiler on performance. VC++ produces fairly quick code. How does GCC code compare?

Generally speaking, compiler optimizations do not matter. For tight, number crunching loops, they might - but most desktop software spends its time in system libraries (when they do not wait for user input).

To optimize, it does not help to call ten library functions slightly faster, the functions will still require the same time; what helps is a redesign where you bring down the need to call functions (for example by caching values). Turning on compiler optimizations will not help you with that, though.

Re:Cross-platform performance.

[greenrd]

Both ways are buggy at the the moment. The former sometimes closes the wrong tab. The latter I just tried and it worked, but it also pasted the clipboard contents into the URL bar and went there - which I did not want!

Re:Cross-platform performance.

[Hornsby]

I'm not sure if this just changed with the latest release or not, but --enable-optimize no longer accepts an argument of the compiler optimizations as described in the parent to this comment. The ./configure will croak if you put the ac_add_options --enable-optimize in your .mozconfig. The proper way to set this option is to export the MOZ_OPTIMIZE_FLAGS variable before you run configure, setting it to whatever flags you want passed to the compiler during the build. You also need to include the ac_add_options --enable-optimize line in your .mozconfig with NO arguments, or it will ignore any flags that have been specified in the MOZ_OPTIMIZE_FLAGS variable. Also, seeing as how many of us are using the most modern x86 processors, it would be advisable to set -march and -mcpu to i686 rather than pentiumpro. I'm in the process of rebuilding the lizard now, and I'm quite curious as to what kind of empirical performance increases I'm going to see.

So...

[Anonymous Coward]

Is the location field long enough now? I can barely see more than http:// in my current version.

Mozilla

[1155]

Mozilla seems to really be coming along. As soon as it is streamlined more or less, it should run smoothly on most setups. Perhaps the most evident note on this suite though, is that it is still in the 0.*.* mode in developement. People are making decisions on it before it is fully developed, or before it hits the 1.0 mark.

If you tried it a while back, wait until about 1.5.*, I am sure it will all work perfectly then. This build is a lot better than previous versions though. I would defitely recommend it for the end user who has a p2 300 or above.

Spell Checker

[Malc]

I've found every release better than the last, except for 0.95, which seemed to have gained a few more crashes. I'm excited to see how this one goes. Can anybody give me instructions on how to integrate the Netscape spell checker and change the language settings to en_GB? I tried following instructions for the spell checker before (installed the .xpi), but I couldn't figure out how to actually use it (were the UI bits removed?)

Only comment so far on the latest build: it polls all of the news groups and servers in my Netscape profile when the news/mail client starts. This is bad as I have a load of crap in there, and a load that are only accessible from when I switch internet connections. I have to click cancel on a lot of dialogs before I can get going :(

Re:Mozilla is a great browser if...

[jacoplane]

There's also a similar project for Mac OS X, qbati, which seems to be just getting underway.

It's nearly on par with IE5.5

[gruntvald]

Though I haven't checked 6.0 yet. If the Mozilla team can straighten out some of the plug in problems (for example, it takes some voodoo before java actually works), or at least come up with a definitive install procedure, we'll be rockin'. The browser is solid, but I don't want to have to be asked what MIME type an m3u file (winamp playlist) is. Heck, I don't actually know! I'm so used to it being taken care of. This kind of "plays nice with others" is something we take for granted - even if it's fake in Bill Gates' case!

Re:These are the days

[Dr. Sp0ng]

* A way to delete the contents of the URL bar without destroying the contents of my clipboard. Right now, I copy a URL from somewhere else, then click in the URL bar and hit delete, just to have the contents of the URL bar copied to my clipboard.

That would be nice, but Moz (and NS4) let you simply middle-click on the HTML display area and it'll go to the URL in your clipboard. Nice timesaver, except when your mouse spazzes and you end up hitting a link instead of empty space and pop up a new window instead :P

I agree 100% with your other points... those are pretty much my main gripes too. About the $EDITOR thing, yes! Every place I *ever* have to enter text should understand vi keys if I want it that way. I already have tcsh and every decently configurable program I run using vi keys, there's no reason Mozilla and all GTK/Qt apps shouldn't be able to as well :P

Oh well, the situation is better than it is on Windows; my vi-trained fingers have a tendancy to hit escape after typing a bunch of text, and in IE this resets the text area to the default value. Quite irritating.

Fast connection to ftp.mozilla.org

[rjamestaylor]

Yes, I just completed installing a nightly build (from the "stock" 0.9.5 build) and reloaded slashdot to see the announcement. According to my reading of MozillaZine earlier I wasn't expecting 0.9.6 until later this week.

But I'm curious as to why the connection to the ftp server was so solid and fast: is it a great example of load balancing ftp?, a sign that people are happy with pre-0.9.6 versions and aren't rushing to upgrade?, or is it (*gulp*) that people aren't interested in Mozilla anymore?

I'm not anti-Mozilla at all. I'm using it for browsing, email, IRC, etc. There are things I like about Konqueror, but I depend on Mozilla. Even my biggest "Internet Explorer"-only client is asking about recasting IE-specific development in Moz-compatible terms. Its just that the server is so fast it doesn't feel like the days of M15 - M18 when I had to fight for a connection...

As an aside: it's perplexing to observe MSFT dropping the ball on browser development. They've got the market wrapped up, but they don't seem to have capitalized on this lead (except the recent MSN fiasco). Or perhaps I'm not giving proper credit to Mozilla developers for pressing ahead with features and usefulness... With the licensing pain with MSFT and the maturation of Mozilla+{Gnome|KDE}+Linux it's getting more and more palpable to switch the enterprise away from the child-settlers.

I do my part for bugtesting

[Compact Dick]

Immediately after downloading the latest nightly, I visit Netscape's German sige and click on the latest Stripshow gallery. Have to make sure those characters display right. Or something. :)

Slashdot crashes mozilla ?

[stefanlasiewski]

Just upgraded to 0.96, and now I see that Slashdot articles with large number of responses crash mozila (I already sent in reports with that crash feedback thingy). This is Win98 , celeron 366 with 512M of ram.

Only seems to happen on articles with a large number of responses (I'm a moderator and I'm trying to browse at -1 , but I can't).

Constantly crashes on Geek Gift Ideas 2001 and Microsoft Would Settle For The Children

I just uninstalled and reinstalled mozilla, and the crashes still happen.

Threads and Processes

[Carnage4Life]

Microsoft's answer to this failing was to make threading as fast as possible, and to push multithreaded programming as a hack around a fundemental OS problem.

Many OS purists think that using multiple processes is a hack around understanding multithreaded programming especially since traditionally there is a context/address switch cost from process to process versus when using different threads. Linux merely legitimizes this hack by implementing the clone system call and copy on write semantics for pages shared amongst processes which makes the worst problems with using multiple processes instead of multiple threads dissappear.

So, now Linux has both faster processes and threads, but thread performance still sucks.

This statement puzzles me greatly. How can Linux threads be faster yet their performance still sucks? Faster than what then?

mostly to support implementing multithreading in userspace (ick).

Huh? How is userland programs being able to create multiple threads a bad idea? Should creating multiple processes the only way to handle multiple tasks at once in an application?

So, the moral of the story is that Linux has a much better core, but seeing that the Linux community actually cares about standards, performance isn't quite up to snuff.

This statement implies that Linux has POSIX compliant threads which the last time I checked is not true especially since the primary kernel hackers (Alan Cox, Linus, etc) are against it. They specifically had issues with the inconsistent way signal handling is suposed to be implemented amongst threads in the same process if memory serves me correctly.

Re:Threads and Processes

[vscjoe]

Many OS purists think that using multiple processes is a hack around understanding multithreaded programming especially since traditionally there is a context/address switch cost from process to process versus when using different threads.

It's not about "understanding". If you program in a language like C or C++ that is so lacking in fault isolation and safety features, you have to use separate processes to keep things out of each other's hair. The alternative, which is known to be more efficient, is to use safe languages and multithreading.

This kind of disconnect is also why Java has such a bad name among Linux users. Java doesn't need separate processes--you can run your whole desktop in a single thread, safely. What Java needs is fast threads.

In any case, as far as Mozilla is concerned, it's slow because it's written that way, not because of anything wrong with Linux. And the Mozilla developers neither seem to understand nor care much about the UNIX and X11 world. The existence of browsers like Opera show that you don't need something as big and slow as Mozilla, and even Opera isn't all that efficient. Rendering HTML is not rocket science, and it is actually kind of amazing how many megabytes people manage to expend on it.

Re:Threads and Processes

[MassacrE]

yes, but Java also has very few things which can execute asyncronously (looking forward to java.nio); which means that you need to make threads for *everything*. Its not uncommon to have two threads per socket to handle reading and writing blocks.

When you count all the synchronization which is happening between all these threads, as well as the sheer amount of context switching, memory usage and the number of processes which have to be scanned to find one capable of being accessed, you see one of many reasons people still use 'unsafe' languages.

Re:Threads and Processes

[yorgasor]

There are two main types of threads, user threads and kernel threads. pthreads, or POSIX threads are an example of user threads, while the clone() call is a kernel level thread.

One of the main drawbacks to user level threads, is that when one thread makes an I/O call, the whole process is blocked from running until the I/O call has completed, even when other threads are available to run since the kernel isn't aware threads are really being used. However, since user level threads can be implemented with a cross platform library, using user level threads like pthreads makes your program very portable.

In kernel threads, when one thread is blocked on an I/O call, another thread can still be run. You can also have multiple threads running concurrently on multiple processors. However, since the application is written for that particular kernel, it is not portable at all.

So the original poster was trying to state that, although Linux has blazing fast kernel threads, its POSIX thread library kind of sucks. Since Mozilla makes heavy use of POSIX threads to maintain cross platform ability, it runs a little more sluggishly on Linux.

Re:Threads and Processes

[vscjoe]

When you count all the synchronization which is happening between all these threads,

Synchronization between Java threads is a lot faster than any of the IPC mechanisms that, say, Gnome, KDE, or Mozilla are using.

as well as the sheer amount of context switching,

If the Linux kernel and thread libraries provided better primitives, you wouldn't need as many system calls to build threaded applications on top of them.

memory usage

If you run a complete desktop inside a single Java process, you end up using a lot less memory than Gnome or KDE: Java applications can share code and data a lot easier and with a lot less overhead inside a single process than C/C++-based systems. In fact, the voracious appetite of C/C++ GUI apps is one reason for abandoning that approach.

and the number of processes which have to be scanned to find one capable of being accessed,

I don't understand what you mean by this. If there is only a single Java process, you don't have to "scan" lots of them.

If you mean that Java using native threads makes lots of kernel threads under Linux and that that has a bit of overhead, you are right. But that's a shortcoming of the Linux threads system and Sun's particular implementation of threads on Linux, not of the Java approach. In a well-designed thread implementation, a thread is no more expensive than an object.

which means that you need to make threads for *everything*. Its not uncommon to have two threads per socket to handle reading and writing blocks.

Java lacks some APIs and Sun's Java implementation has limitations, but that doesn't invalidate the approach. There are systems like Java that have better threading support. Still, a single-process approach based on JDK and Linux, with all the limitations of the JDK and of Linux, is already more efficient than the many process approach of Gnome or KDE and similar systems.

Re:Threads and Processes

[vrai]

If you program in a language like C or C++ that is so lacking in fault isolation and safety features, you have to use separate processes to keep things out of each other's hair.

This is a sweeping generalisation. I have worked on a large number of multithreaded applications (at work and home) of which 90+% were written in c/c++ and all of which used threads.

Without wishing to sound too rude: the myth that production quality multithreaded programming is difficult in c/c++ tends to be propagated by people who are bad c/c++ coders and/or who lack experience developing multithreaded applications.

In the same way bad java coders produce slow code, bad c/c++ coders produce leaky code. This is not a fault of the language, but of the coder (who should take the time to learn the language properly or stick to QBASIC).

Re:Threads and Processes

[roca]

> Rendering HTML is not rocket science

Oh yes it is.

Go to the W3C Web site and ingest the HTML4 and CSS2 specifications. Throw in all the I18N requirements (Bidi, charsets, etc) (Opera doesn't do them BTW). Throw in all the image formats and plugin support. Now throw in hacks to make a million differently broken pages work reasonably. NOW, in case you think you're done, make sure your engine is fully incremental so everything updates smoothly when stuff takes a while to load, or when you resize the window, or when the document modifies itself in arbitrary ways using the DOM (Opera doesn't handle the latter). Now make it all robust and fast for when some fool writes a page with 100 IFRAMEs, or 1000 combo boxes, or 10000 paragraphs all nested inside each other. And make sure you have ZERO buffer overruns or your users are toast.

Sounds easy, huh?

Opera 6.0

[popeyethesailor]

This is not meant as a Troll. If you are a Windows user, checkout the latest beta from Opera, it rocks. Choice of Single/Multiple document interface,new skins, and mouse gestures too! Still retains fastest browser credits. Give it a try.

An MSers take on Mozilla

[Kraft]

I have tried to go over to mozilla as part of my slow conversion from Win to Linux. Thought it might be a good place to start.

When I installed Mozilla at 0.9.5 I was impressed. This app has come a loong way!

However, I have a couple of hickups, which someone might be able to help me with

- Load time Compared to IE, which takes 1-2 secs to load, Mozilla take around 8 secs to load. Not that much extra, but when my short term memory is 5 seconds, I most often choose to load IE, so I don't forget what I wanted to check out.
- Shortcuts I don't know who fucked up the shortcuts, but I must use alt-d over 100 times a day in IE, the shortcut that brings you to the address bar. I had a (not too investigative) look at the Mozilla help, and couldn't find any info on shortcuts, which brings me to
- Help You can't search the help! Hello.
- Search My seconds favorite feature in any program is text search, and I have found the search in Mozilla to be buggy (forgetting last search word and settings, needing to 5 click before it starts, not finding text which is there)

The most important to me is load time. I just don't see myself, only using Mozilla until load time is decreased. But hey, good luck to the dev team, I will hang in there.

Re:huh?

[Yorrike]

Mozilla is actually quite slow for me on my work PC (Athlon 700, 320MB RAM, Redhat 7.2), but my work PC is stuffed (nothing works properly on it).

My home PC (Athlon 650, 512MB RAM Redhat 7.0), on the other hand, runs most things at a rapid pace. Mozilla, once loaded into memory, is lightning fast, though I'm starting to lean towards Galeon for my browsing needs at home and work. Mainly beacuse it can't be beaten for speed or configuration options. I really dig the RPMfind, Google, Freshmeat etc. search bar (talk about software at your finger tips).

Tabbed browsing vs MDI

[EditDroid]

This has puzzled me... Why are people raving about tabbed interfaces, while at the same time ridiculing MDI? Aren't they, for all practical purposes, the same thing?

Re:Tabbed browsing vs MDI

[Explo]

This has puzzled me... Why are people raving about tabbed interfaces, while at the same time ridiculing MDI? Aren't they, for all practical purposes, the same thing?

With MDI applications everything tends to be inside the one master window. For me, the difference between tabs in Mozilla and typical MDI applications is that you can freely mix and match the approaches between one window with everything (MDI) and N windows with one specific thing (SDI). For example, when at work, you can keep your work search engine queries ("what the hell this isolated parameter meant in tc command syntax") in one window tabbed and have another window with more leisure-spirited (eg. slashdot and friends) material tabbed inside, possibly even keeping the windows in different virtual desktops. Or whatever is your ideal preference between pure MDI and SDI (or either extreme, if you wish so).