Recent comparisons of browsers focus on JavaScript speed. There are many ways to measure browser performance, including image load time, reloading from cache, start time, and css rendering speed. While Opera was in the lead a few years ago, it has been branded as the fastest browser again.

You should notice the ad on the left for Chrome. So which one is the fastest? I ran the tests on Linux and Windows.

Benchmark Method

There’s nothing special about using JavaScript to detect when onload is fired by the browser, as long as the browser follows the convention that it is fired after the page is loaded. This used to be an issue I noted to the point of just using Firefox and ignoring other browsers. According to this article, it was fixed 2 years ago. So I took the benchmark from the site and ran it. Not surprisingly, Firefox came out highest on the score. Actually, that’s a negative score. Ideally, the page takes no time to load.

Linux

Google Chrome for Linux is different from the one for Windows, as the benchmark would load the first site for the first time, but could not time it nor reload it. Opera got stuck on MySpace once and kept adding more elements to the page, possibly due to an ad that is not usually loaded locally. I ran these tests on Sabayon Linux with kernel 2.6.30 (I expect 2.6.33 to be faster, since it has been patched with Con Koliva’s kernel enhancements). An interesting note here, not seen in other parts of the result set, is that Arora took longer to load pages on the first time, but was faster on all subsequent reloads. The total score for Arora comes second to Opera. On another note, Firefox with the same extensions, ran faster on Linux than Windows.

Windows

Not surprisingly, the winners on Windows were 32 bit browsers. Aside from the small speed increase due to smaller pointer sizes in 32 bit applications, I think Opera and Chrome are faster browsers, as they advertise. A surprising result is that 64 bit IE ran faster than 64 bit Firefox. I noticed that a while ago, but decided to stick with Firefox because it has add-ons. Iron is a stripped down version of Chrome compiled from source. It should be slightly faster, with the slimmer binary and no personal tracking features. I ran this on Windows 7 Pro 64 bit Version 6.1 (Build 7600).

Conclusion

Opera beat all other browsers, just with its default settings. A little more tuning of redraw rate and memory use could improve the score. I expect the real results when browsing to deviate. Chrome and Firefox has DNS prefetching, Firefox and Opera have pipelining. To improve that DNS fetch speed in Opera, you can set your system to use OpenDNS to resolve domain names.

After the Benchmark (you should decide which browser to use)

I measured the memory use!

It looks Chrome and IE were designed for really cheap laptops. (They can’t run on old computers with Windows 2000.)

I noticed that the icons for search jump had different sizes when I loaded it in Google Chrome.

When I investigated the issue, I found the icons had 16×16, 32×32, 64×64 image sizes embedded. Chrome happened to load the larger image sizes.

The final trick was to get Chrome to reload it. I tried restarting the browser and clearing the cache. It finally worked when I uninstalled it from the extensions menu and reinstalled it from the web site.

Minimal SearchJump is available for Chrome, Firefox, and Opera.

While working on the tree part of the coursetree project, I ran into the question of how to display the course dependencies. I have written a recursive function that returns a recursive list of course prerequisites:

I wanted to turn it into a diagram of course prerequisites and decided 2 flat lists could retain the data:

• a list for the levels
• a list for connections between courses

For example,

would translate to

and

Separating the Levels

We start off with a recursive list representing the tree structure:

To tackle the problem, I first solved a similar one: flattening a list.
This can easily be done in Scheme, as syntax and type declarations do not get in the way.

Here’s the same code translated into Python:

Unfortunately, flatten in Python produces a hypernested structure for flat lists, as (1 2 3) in Scheme is equibalent to (cons 1 (cons 2 (cons 3 empty))) or (1 . (2 . 3)).

The more serious problem is that it exposes a quirk in Python:

Yes, that means the list disappears after being extended with an empty list . . . one of those unpleasant surprises.
So let’s redefine a flat list in Python:

There is a similarity between this approach and the recursive one: different actions are taken for a list node and other nodes. I used a combined functional and imperative approach to solve the problem:

Connecting the Nodes

We start off with a recursive list representing the tree structure, slightly different from the list for separating the levels:

Again, a mix of recursion and iteration easily solves the problem:

Hopefully, you’ve had fun reading this article and meanwhile came up with a better way to represent the tree as a flat structure.

Now you will notice that the icons load a lot faster and all at the same time if you are using either the minimal version or the plain version.

Note: the new version has a new namespace, so the old version must be uninstalled manually from Greasemonkey

• instead of clicking twice, the panel hides itself after a single click.
• by default, the script is only active on Google
• an icon has been added for Clusty
• the event listener for toggling the panel is only applied to the hide link
• visual aspects are more balanced

I’ve had the idea of making a self-updating, navigable tree of Waterloo courses. This is the first step. (Actually, not the first step for me. It started with Django, which had to do with my last work report’s comparison to Zen Cart. Some credit goes to Thomas Dimson for inspiration. He made the Course Qualifier.) The main idea for this step is to gather all the information to be stored in a database. With that (the idea and plan) begins the coding phase:

I wanted to gather the course (“SE 101”), name (“Introduction to Methods of Software Engineering”), desc (“An introduction …”), prereq (“Software Engineering students only”), and offered (“F”) each as separate fields

In order to do that, I wrote a spider to crawl the page:

There are several things to note:

• The prereq field here cannot identify “For Software Engineering students only”. The regular expression only matches the course code.
• Offered, unlike other fields, can contain more than one item
• Prereq may be empty

Finally, the spider pipes its results to an output format. CSV format meets the requirements, as it can be inserted into a database.

Two gotchas:

• Because prereq might be empty, there needs to be an exception handler
• Offered may be variable length. The list needs to be joined to output all of the terms the course is offered.

This part of the project was done in 2 hours with Scrapy. The project can be found in the downloads section.

Recently, I was assigned to fix bugs on a large project done by 2 previous co-op students. I came across the Debugging section of Steve McConnell’s Code Complete 2 while looking up maintenance in the index. He suggests using the scientific method, summarized here:

1. Stabilize the error
2. Gather data
3. Analyze the data
4. Brainstorm hypothesises
5. Carry out the test plan
6. Did the experiment prove or disprove the hypothesis?

I applied this to a recent debugging problem. The bug is as follows, reported by a tester:

I started by trying to understand the problem, which I interpreted as:

Understanding the problem in terms of the business goals of the application opens up many possibilities, I chose to make the next and previous buttons save the input and take the user to a different page. This is where the debugging begins. First, I checked the HTML output for the save button and looked up its id in php scripts using grep. Finding the correct place to edit the code was easy. Making the right modification was the hard part. The code looked like this:

Fortunately, I knew enough Javascript and functional programming to understand what the code was doing, but only on the surface. When the button is clicked, the function is executed. It sets a hidden form field to submitted status so that php stores the input in the database. So I wrote what looked like innocent lines of code:

It worked immediately in Firefox. I thought I was done. Next, I got a report from one of the testers that it did not work for them. I tried it again in IE with the expected result. At this point, superstition came into play. I had previous experience with IE where a javascript error prevented an independent section of code from working. I hypothesized it could be the if statement, because it may not be allowed with jQuery. I tested my hypothesis by taking out the if statement. As no progress was made, I checked the error console in Firefox. It gave an error about $(‘#next’).click on object which does not exist. So I moved the script down below the area where the next link was created. It still did not work in IE. I decided the brute force approach was to learn jQuery and understand exactly what the code was doing. The tutorial was surprisingly short. I made sure my code used the correct jQuery syntax. When I read the documentation on the click method, an idea came to me that IE went away from the page without executing the registered event. There was other evidence supporting this hypothesis in Firefox, clicking next rather than save took many times longer. At this point, I doubted mousedown would work, as I already tried onclick. Luckily, I did look up documentation on mousedown. It looked like it was the correct way to prove or disprove my hypothesis. Switching from click to mousedown did verify my hypothesis. To my surprise, hitting next saved data in IE, with the same speed as hitting save.

Before

After

Download it here.

I started using Sabayon back in 2007 as it was more fun than Windows XP. I decided on it after trying out many other Linux distributions with Live CDs downloaded and burned at school. Gentoo was the fastest Linux distribution, and Sabayon made it easy to install. At that time, it had automatic graphics card configuration to enable Compiz-Fusion. Compiz-Fusion is like the glassy windows you see in Vista, plus many other effects (the most famous being wobbly windws). I suppose that was the reason so many people tried Sabayon at the time. Unfortunately, on the laptop where I installed Sabayon, the graphics card was not capable of these effects (but I did get lightweight transparent windows, highly useful on a small screen). Anyways, here’s what the wallpaper looked like:

At that time, Sabayon (since it was just a way to install Gentoo) was for power users. The dark red and fossilized penguin footprint certainly leaves the rest of the pack behind. I was able to build a system on an Inspiron 1000 that would load the next web page as soon as I clicked on the link. The Reiser file system did the trick for caches, and compile time optimizations made better use of the Pentium 4 CPU.

Later in 2008, I got my own laptop and dual-booted XP with Sabayon 3.4. It’s hard to say which OS I liked better. Linux had trouble with wireless and crashed whenever I launched a program that used OpenGL. Windows came with all the right drivers, except I can’t do anything besides work and play there. I ended up playing a game called NetHack and spending a lot of wasted effort on an English presentation on Marcus Aurelius’s Meditations. I think I simply used Windows more because I liked the desktop colors:

Sabayon had too many splashes of red here and there, but all that changed with the next release.

This was my first glimpse of the future of Sabayon. It (I mean he) was (all) about dreams (besides a surprise for me). I remember this quite well. It was when Sabayon became a binary distribution instead of just a Gentoo box that came with the box (not exactly out of the box as the wireless required ndiswrapper). Let me emphasize that point again. I got a system I couldn’t boot as soon as I tried hibernate.  Thus, the dreams burned to the ground. So I went back to Windows again, this time with Vista. When I had the chance, I made my computer quad boot with XP, Vista, Sabayon, and Kubuntu. All of them had special characteristics and capabilities. Vista was the only one that could suspend and wake with even cpu frequency scaling on both cores. Kubuntu was the only one that supported a full KDE 4 desktop. Now days, KDE 4 is available for all 4 of them. In the February of 2009, I tried Sabayon 4:

partly hoping those CPU frequency scaling bugs were fixed. However, Dr. Scheme had poorly written C code that required custom compile options. I got my work done in Windows. From May until December I used Windows 7 RC. It was where I was able to do everything in an easy and quick way. Most importantly, I was ahead of the technology curve with the elegant UI (now KDE users are beginning to copy it). I installed Sabayon 5.1 again over the past Christmas break:

This one definitely deserved a version bump as the developers switched from -Os to -O2. -O2 is generally faster, but -Os is always smaller, in terms of compiled programs. This was great, as my CPU fan was used a lot less. I rarely hear it now after a few tweaks. Moreover, this is the first time I used Linux continuously over a month for over a year. Congratulations to Sabayon! More details about each release for those who want to compare what truly matters to a user with what Sabayon has provided through its history:

Sabayon Linux 3.3 x86/x86-64: Press Release

Sabayon Linux x86/x86-64 3.4: Stable Release

Sabayon Linux x86/x86-64 3.5: Stable Release

Sabayon Linux x86/x86-64 4 Revision 1 Rolling Release

Sabayon Linux 5.1-r1 GNOME and KDE: Stable release

SearchJump has been updated with new link for bing. You can download it for Greasemonkey.