I had been hosting this site on a 256MB VPS from Slicehost for just over 2 years in their Dallas datacenter. Overall, it was a great service (and a much needed upgrade from Dreamhost), with only 3 unscheduled reboots over those years. However, my weekly monitoring reports from Browsermob never hit 100% uptime across their global servers. A typical report looked like this:

Slicehost (Dallas) VPS weekly uptime

Location	Response Time	Checks	Failures	Availability
Amsterdam	5.94 secs	192	26	86.5%
Dallas, TX	1.04 secs	166	0	100%
Dublin, Ireland	2.28 secs	140	0	100%
New York City	6.01 secs	196	33	83.2%
SF Bay Area, CA	1.3 secs	163	0	100%
Singapore	3.83 secs	170	0	100%
Washington, DC	1.1 secs	174	0	100%

Over 4 months of monitoring, between 1 and 3 locations would fail to reach 100% availability each week. Out of curiosity, I wanted to see if any other VPS service could do any better. I signed up for a Linode 512 plan and ported over this site, which had the following relevant configuration:

• CentOS 5.3
• Apache 2.2
• MySQL 5
• WordPress 2.1

The change in the Browsermob profile was immediate. The first full week of running on Linode in their Fremont datacenter looked like this:

Linode (Fremont) VPS weekly uptime

Location	Response Time	Checks	Failures	Availability
Amsterdam	2.06 secs	84	0	100%
Dallas, TX	1.05 secs	84	0	100%
Dublin, Ireland	1.9 secs	83	0	100%
New York City	1.27 secs	84	0	100%
SF Bay Area, CA	672 ms	84	0	100%
Singapore	2.37 secs	84	0	100%
Tokyo	1.58 secs	83	0	100%
Washington, DC	1.15 secs	84	0	100%

(The drop in number of checks is due to a drop in the daily allowance for free Browsermob accounts, in addition to an increase of one new location in Tokyo.)

Diving into the data some more, I found that the range of response times across the globe were also substantially improved. The response time chart in my profile showed a clear improvement at the time of the switch on July 4:

	Slicehost	Linode	% improvement
Average response time (ms)	1889	1513	19.91%
Response time stdev	1161	638.7	44.95%
Coefficent of variation	0.6141	0.4221	31.26%

So, for the same price ($20/month), Linode was able to deliver almost 20% improvement in average global response time along with a much tighter standard deviation.

At the time that I switched, the 256 Slicehost plan cost the same as the 512 Linode plan. Slicehost has since changed their pricing to almost match Linode.

I was hiking up in Yosemite when at the top of North Dome the mirror in my Canon 5D came unglued and started bouncing around in my camera! I took it to a local camera shop, who quoted a $350 repair (which involved replacing the entire mirror housing), but luckily a bit of interneting yield the following service notice from Canon's astoundingly-difficult-to-bookmark support site:

Service Notice: EOS 5D: Main Mirror Detachment

Thank you for using Canon products.

We have discovered that, in rare instances, the main mirror of some EOS 5D Digital SLR cameras may detach due to deterioration in the strength of the adhesive. Accordingly, we would like to convey the details and our service policy concerning this phenomenon.

We offer our sincerest apologies to those customers who have been inconvenienced by this issue. Canon always strives to provide the highest quality products to our customers and we will spare no effort in our quality management to make sure our customers can use our products with confidence. We hope our efforts will earn your understanding.

Phenomenon: The main mirror of the camera detaches and images cannot be viewed through the viewfinder.

Affected products: EOS 5D Digital SLR cameras whose main mirror has detached.

User Support: We will repair and reinforce the mirror portion of the affected products free of charge. If you own one of the affected products, please contact our Customer Support Center.

We appreciate your patience, and we offer our sincerest apologies to the customers using these products who have been inconvenienced by this issue.

This information is for residents of the United States and Puerto Rico only. If you do not reside in the USA or Puerto Rico, please contact the Canon Customer Support Center in your region.

Contact Information for Inquiries
Customer Support Center
1-866-422-2965 (toll free)
8:00 a.m. - Midnight, EST (M-F)
10:00 a.m. - 8:00 p.m., EST (Sat.)
Email: carecenter@cits.canon.com

The Canon customer support rep was far more friendly than their website. The timeline went like this:

• August 6: Canon sent over a UPS ground shipping label to the closest repair facility in Irvine, CA
• August 11: I shipped out my 5D body
• August 18: Canon sent an email confirming that they have inspected the camera and would be starting repairs
• August 22: Canon sent an email confirming that the repair was finished and a Fedex tracking number

So within 2 weeks, I got my camera back, freshly cleaned and sporting a reinforced mirror mount (highlighted in green below) total free of charge:

According to Google, I am the second-most popular "purveyor of [insert genre here]" in the world, bested only by the purveyor of the world's finest teas, Upton Tea Import. Being second in this list is lamentable, but under the circumstances not a terrible position considering that I have a better rank than the leading purveyor of fine needlework and supplies, and the purveyor of EarthBalls and Giant Globes. Gloating aside, how the moniker "purveyor of" came into being merits some discussion. C.M. recently asked,

You use the line "purveyor of fine words." Before commandeering this line, did you look into its etymology? For example, what is correct "fine purveyor of..." or "purveyor of fine..."? Oddly, there is not much online by way of a discussion. There are of course several instances of people using the phrases both ways. I did come across a book about the history of purveyance and it talked about "fine purveyors" as those who procured better cuts of meat or poultry, as opposed to the "coarse purveyors." However, these days, everyone claims to be a "purveyor of fine something". I just wonder if they are interchangeable or if one is more correct than the other. For obvious reasons, you seemed like a good person to ask, being a self-titled "purveyor of fine words" and all.

Well, I chose the tagline 'purveyor of fine words' as a response to the typical self-deprecating blog name that is so common these days -- ones that mix and match words like 'rambling', 'thoughts', 'random', 'drivel', 'brain farts'. I subscribe to one blog that is titled, "Continuing Intermittent Incoherency", which sounds like the author picked up some kind of Mad-Lib for blog names for inspiration. "Randomised nonsense" and "The Solipsistic Sayings of a Random Infidel" also seem to have been derived from the same template.

Perhaps these titles are a byproduct of today's disclaimer-ridden society, where consumers are too moronic to realize that a cup of coffee contains scalding hot liquid, or that a pack of peanuts "may contain nuts", or that power tool enthusiasts should not "attempt to stop a chainsaw with [their] hand". In the online world, this warning zealotry translates into prefacing statements with redundant acronyms like FWIW or IMHO, which authors use to ostensibly indemnify themselves against criticism. "IMHO, you're nothing but a fucktard and the best part of you ran down the crack of your momma's ass", becomes a quaint jest I suppose. In order to buck this trend, I opted to go big instead and inflate myself to gourmet proportions, and thus I promoted myself to a purveyor of fine words.

In response to C.M.'s question, I don't have any more insight into the etymology of the phrase, as mine merely parodies Dean & Deluca's tag line of "Purveyors of Fine Foods and Kitchenware". I would say that "purveyors of fine..." is much more prevalent than "fine purveyors..." insofar as it's difficult to explain the difference between a "purveyor" and a "fine purveyor" (maybe the purveyor is very attractive?), whereas the difference between "food" and "fine food" immediately conjures up contrasting images of corn dogs and Iranian caviar.

When developing dynamically generated forms, you often want to attach a single event handler to the main form object, and have that handle the events generated by the form elements, thus saving you the trouble of constantly attaching event handlers to newly generated elements. However, IE 6 and 7 do not bubble the onchange event beyond the originating select element, meaning that you have to explicitly attach an onchange handler to every select you generate. All other current browsers bubble the event properly.

Here is a test form for checking if your browser registers the onchange event beyond the firing select element. Changing the select options should trigger an alert dialog box.

onchange listener attached to parent <div> node

onchange listener attached to parent <form> node

option1 option2 option3

onchange listener attached to actual <select> node

option1 option2 option3

Adblock is the single most useful Firefox plugin available today. Just like watching sitcoms with automatic commercial-skip, adblock's banner ad supression system elicits a smug sense of satisfaction even after browsing through your 10,000th ad-free web page. However, a huge barrier to adoption seems to be the lack of a default filter set, so when you first install adblock, nothing happens.

The main issue is that adblock does not have any intelligence as to the content that is included with a webpage; it is just a generic regex-based filter system, so it is only as effective as the filters that you provide. There are plenty of pre-made lists available but they tend to be overly-aggressive in what is supressed, resulting in occasional broken pages and/or pages that dead-end because adblock has removed the "Next" button. The most dangerous public set seems to be the EasyList, which has a 360+ item block list. Evidence that the creators know of its greedy nature is their inclusion of a 20+ item whitelist to manually compensate what was initially blocked. Even more unstable is the EasyElement list that searches through the DOM to remove suspected elements directly from the main document -- a list of 570+ substrings to search for.

Intead of using such a large, reactive list of simple and site-specific string matches that tries to supress 100% of ads, I posit that you only need 2 adblock filters to eliminate 70-80% of ads, and still be confident that legitimate content isn't being flagged as a false positive. By getting into the heads of HTML writers, we can pick out the most common patterns used to include ads and create regex patterns to suppress the ads.

1. /(\b|_)ad(x|s?)(\b|_)/
   This regex looks for any element that contains the string 'ad', 'ads', or 'adx' surrounded by a word boundary, because the vast majority of web sites partition their ads into a single directory or serve them through a single script. The word boundary check is crucial to this filter because just searching for the characters 'ad' is ineffective. Instead, the word boundary restriction means that adblock will supress elements that contain strings like 'ads.server.com' or 'www.server.com/ads/' or 'server.com/ad_server.php', but not 'adobe.com' or 'server.com/adjustment'.
2. /ad.*\d+[xX]\d+/
   This regex exploits the common technique of ad designers to use the image dimensions in their element name, i.e., "server.com/newads.php?location=top&size=468x80". Like the previous rule, we don't just exclude any element that has dimensions, but qualify that by searching for the string 'ad' as well.

At this point, your browsing experience will be significantly improved, but you can bump up your block rate to about 80-90% with a few more simple substring matches. There are many well known ad providers that exist solely to deliver ads, so we can consildate those in composite filter rules:

1. /a(2\.yimg|dserv|dvert|tdmt|twola)/
   This rule collects all the ad serving systems that start with 'a': Yahoo, Atlas, AOLTimeWarner, and generic ad serving systems.
2. /b(anners|logads)/
falkag.net
   These pick up anything labeled with 'banner', the 'blogads' network, or Falk AdSolutions.

Realistically, reducing the ad load by 90% should be more than sufficient for anyone. Chasing that last 10% -- and whitelisting the collateral damage -- will always be a losing battle. Your time is better used reading the content that is on the page you requested in the first place.

Effective immediately, I have a new title at work — actually 6 new titles...

Internets Strategerist

Sr. Tube Developer

The Decider

Guapo

Scrabblista

Assistant to the Regional Manager

Bonus points if you can match all the cards with their respective references:

• The Office
• Senator Ted Stevens
• G.W. Bush & Will Ferrell
• Victor's Taqueria
• G.W. Bush
• My desk at work

When using yum install, sometimes the old GPG keys installed with rpm are obsolete, resulting in an error like the following:

warning: rpmts_HdrFromFdno: Header V3 DSA signature: NOKEY, key ID db42a60e public key not available for autoconf-2.59-5.noarch.rpm Retrieving GPG key from file:///etc/pki/rpm-gpg/RPM-GPG-KEY-fedora The GPG key at file:///etc/pki/rpm-gpg/RPM-GPG-KEY-fedora (0x4F2A6FD2) is already installed but is not the correct key for this package. Check that this is the correct key for the "Fedora Core 4 - i386 - Base" repository.

To fix, just add the new keys to rpm (changing the path for you particular install):

rpm --import /etc/pki/rpm-gpg/RPM*

Some forum posts have suggested disabling GPG (i.e. gpgcheck=0), which can be a foolish thing to do. You want to maintain some level of assurance that what you're downloading is legit.

del.icio.us has changed their API host, which breaks the current direc.tor because of the xmlHTTPRequest's domain security policy. To fix this, the del.icio.us guys have added a couple public pages to allow direc.tor to continuing functioning. Now, you must first browse to https://api.del.icio.us before starting the direc.tor bookmarklet. Let me know if you are having any issues.

Although PHP has a great library of functions, many of them are not included in the standard build, or haven't been included into the popular package installers like yum or apt-get. The man page doesn't leave you with much instruction, other than something like "compile PHP with the flag --with-pspell[=dir]". At this point you have 2 options:

• Recompile PHP
  Find your existing PHP configure command and append this new flag, and recompile PHP. This takes a while and is generally quite bothersome, if not unacceptable like when you're in a production environment.
• Create a dynamic extension
  Compile a separate file (usually ending in .so) that you copy into a PHP directory, and edit php.ini. If you are running multiple machines on the same OS, you can just copy the file to all those machines as well. Much easier, and you can turn it on and off at will.

Here's how to create the extension for modules that appear in the PHP manual on a linux-based system (for third-party extensions, it's most likely the same).

1. Check that you have the PHP development package, which often comes in a separate package. Yum lists it as php-devel. You'll need its components in a few steps.
2. Download the PHP source code for whatever version you're currently running
3. In the source code, there is an ext/ directory that should contain a subdirectory for the module that you're looking for. Change to that subdirectory, i.e. ext/pspell/.
4. Type phpize
   (This won't work if you didn't verify step 1)
5. Type ./configure --with-pspell=/usr
   Replace the red portion with the text that is specified in the PHP man page for the module you want. For example, MySQL improved would be something like --with-mysqli=/usr/local/mysql/bin/mysql_config. Be aware that the path is sometimes a base directory, and sometimes needs to point to a specific file. Read the PHP docs carefully.
6. Type make
7. When finished, the compiler should tell you where it created the .so file (most likely in the modules/ subdirectory of your current location). Copy the .so file to your PHP extensions directory, i.e. /usr/lib/php/modules. If you don't know this, it's listed in your php.ini file under the extension_dir parameter. You'll need root access to do this.
8. Finally, tell PHP about your new extension by adding one line to php.ini:
   extension=pspell.so
   Alternatively, if you already have a bunch of extensions installed, you can place it in your /etc/php.d directory in it's own ini file for a cleaner installation approach.
9. Restart apache, if you're using it
10. Check phpinfo() to verify that your new module is installed

I've been looking for a nice web Javascript spell checker, and came across a great implementation by Emil that he named LiteSpellChecker. It mimics the spell checker in MS Word by underlining misspelled words and presenting a nice substitute word selection menu. The javascript takes a standard <textarea> element, erases the background, and inserts a shadow <div> underneath that holds the redline segments.

Bug Fixes (based on 2005-7-24 version of LiteSpellChecker)

The current implementation on his site has some bugs, so I've started to tackle some of them:

• Fixed: Redlines get misaligned when scrolling with arrow keys in Firefox
  When scrolling long text using Firefox, the redlines become misaligned as you move towards the bottom of the page. Once they begin to misalign, you have to manually refresh in order get it right again.
• Fixed: Ignore word function breaks if there are non-word segments in the text
  If the text contains characters like ++ or @@, then the ignore words function fails. This problem crops up whenever you have wiki markup syntax.
• Fixed: Numbers are flagged as misspelled
  Any word that contains a number is now ignored by the spell checker.

Performance Improvements

Since many of my users will be working with long documents, performance is key.

• Sped up ignore function by almost 5x
  The ignore word function loops through all words and feeds it back into the spellchecker. In long documents, the lag becomes noticable, and also causes a flicker of all the redlines. Instead of blindly looping through all words, I changed it to a case-insensitive search for the ignored word, and only processed the matching words. Venkman reports a speed up from 278.2ms to 57.95ms on a Powerbook G4.
• Improved responsiveness while editing long text
  Since the spell checker updates the spelling status on every keystroke, the UI becomes unusable when editing long text. I added a spell check delay that suspends the spell checker while typing continuously. You can adjust the sensitivity via the SPELL_CHECK_DELAY variable.

Download my source code. Important: read Emil's demo page before attempting to do anything with these files.

• litespellchecker.js
• spellcheckerbase.js