Martian Intuition - Work

on Ruby stack trace limits

2009-07-29T02:46:00Z

LoadErrors raised from Ruby on Rails' ActiveSupport may look something like:

/usr/local/lib/ruby/1.8/gems/activesupport-2.3.3/lib/active_support/dependencies.rb:426:in `load_missing_constant':
Expected /path/to/the/rails/root/app/models/bacon.rb to define Bacon (LoadError)
        from /usr/local/lib/ruby/1.8/gems/activesupport-2.3.3/lib/active_support/dependencies.rb:80:in `const_missing'
        from /usr/local/lib/ruby/1.8/gems/activesupport-2.3.3/lib/active_support/dependencies.rb:92:in `const_missing'
        from /usr/local/lib/ruby/1.8/gems/activesupport-2.3.3/lib/active_support/dependencies.rb:437:in `load_missing_constant'
        from /usr/local/lib/ruby/1.8/gems/activesupport-2.3.3/lib/active_support/dependencies.rb:80:in `const_missing'
        from /usr/local/lib/ruby/1.8/gems/activesupport-2.3.3/lib/active_support/dependencies.rb:437:in `load_missing_constant'
        from /usr/local/lib/ruby/1.8/gems/activesupport-2.3.3/lib/active_support/dependencies.rb:80:in `const_missing'
        from /usr/local/lib/ruby/1.8/gems/activesupport-2.3.3/lib/active_support/dependencies.rb:437:in `load_missing_constant'
        from /usr/local/lib/ruby/1.8/gems/activesupport-2.3.3/lib/active_support/dependencies.rb:80:in `const_missing'
         ... 26 levels...
        from /usr/local/lib/ruby/1.8/irb/init.rb:250:in `load_modules'
        from /usr/local/lib/ruby/1.8/irb/init.rb:21:in `setup'
        from /usr/local/lib/ruby/1.8/irb.rb:54:in `start'
        from /usr/local/bin/irb:13

The folded/clipped/omitted middle of the stack trace leads to a very frustrating debugging experience.

To expand the backtrace & see the full stack, load the application environment manually in irb with a liberal trace limit:

%  cd /path/to/the/rails/root
%  export RAILS_ENV=development
%  irb --back-trace-limit 1000
irb(main):001:0> require 'config/boot.rb'
irb(main):002:0> require 'config/environment.rb'

You will hopefully see the exception raised + the complete stack trace.

svn Segmentation Fault

2009-01-11T01:35:00Z

I recently tried to do a Subversion copy from a remote repository into my local working copy, but I got a segmentation fault, like this:

% svn cp http://xxxxx.xxxxx.com/xxxxx/xxxxx/xxxxx
Segmentation fault

It leaves behind an unversioned .tmp copy of the file being copied, & svn cleanup must be run to release the lock. (I'm using OS X 10.5.6 Leopard [Darwin]).

Here's the info for my local working copy...

% svn info
Path: .
URL: http://mars@xxxxx.xxxxx.com/xxxxx/xxxxx
Repository Root: http://mars@xxxxx.xxxxx.com/xxxxx
Repository UUID: xxxxx-xxxxx-xxxxx
Revision: 11327
Node Kind: directory
Schedule: normal
Last Changed Author: mars
Last Changed Rev: 11327
Last Changed Date: 2009-01-06 15:46:57 -0800 (Tue, 06 Jan 2009)

The Fix

The working copy URL has a username prefixmars@ that the svn cp URL is missing. So switch the local copy URL to drop the username prefix:

% svn switch --relocate http://mars@xxxxx.xxxxx.com/xxxxx/xxxxx http://xxxxx.xxxxx.com/xxxxx/xxxxx

Now the svn cp ... will work without segfaulting.

Epilog

My favorite GUI Subversion client Versions.app checks out local copies with this "username@" URL convention. Fortunately, switching the URL seems to have no effect on Versions' ability to use the working copy.

work-around for OS X Sierra Wireless crashes

2008-03-04T05:16:00Z

If you're trying to use a Sierra AirCard 597E on OS X 10.5 Leopard via the bundled Sierra Wireless Watcher Lite software, then you're probably frustrated like me.

Since my original review, Sierra did release a new version 1.1.3 of the beleaguered PowerPC binary (no it's not Universal/Intel-native). While the new version starts less slowly than 1.1.1, it still mysteriously crashes, sometimes taking the connection with it.

The fix? Simply use Apple's built-in modem dialer, leaving Watcher Lite dormant unless its device provisioning or activation functions are required.

Step 1: In the Network Preferences Panel, check "Show modem status in menu bar"

Step 2: Enjoy the nicely integrated, crash-free EV-DO connectivity

to the West, to San Francisco

2007-12-20T21:39:00Z

On December 9th, after over twelve-years living in Austin, ten of them working as a pro web developer, I accepted a web dev position in San Francisco with Scout Labs. January 7th I start the job: front-end Ruby on Rails web development.

The company's product, the Scout, already received a revealing write-up by TechCrunch. The possibilities of the Scout are just beginning, and I'm ecstatic to become part of the team building this phenomenally useful tool.

Over the past three days, I visited the city to hunt for an apartment, when ideally I would be wrapping up my life in Austin. With little time to ponder the possibilities, much less explore and find vacancies in the neighborhoods of SF, I homed-in on the SOMA/South Beach area. SOMA gives me walking proximity to the new office, the Embarcadero, Whole Foods, & many public transit possibilities (Caltrain & Muni blocks away; BART Powell Station a 15-minute walk away). My beloved bicycle, already my primary transit in Austin, now becomes my only personal vehicle.

Thanks to a reference from a friend, I found an ideal one-bedroom with a view & quickly decided to take it.

Now here I am, flying back to Texas for two short weeks of project-wrapping, car-selling, move-out-prepping, traveling to Dallas for family Christmas, & saying "see ya later" to many loved friends & beautiful connections.

California, here I come!

a Mobile Broadband Review: AirCard 597E + Sprint

2007-11-05T22:35:00Z

How does Sprint's Mobile Broadband service perform in Austin? Is it worth a twenty-four month commitment totaling over $1500?

In an effort to answer this question, here's a summary of my experiences & observations during this first month of going mobile.

Bandwidth & Latency

During an ssh terminal session—where characters typed are sent across the network and echoed back for display—the cellular radio link makes the network feel springy. When transmission delays occur, responses eventually snap back with a quickness.

Here's an example of a typical session with a 4-bar signal:

For 125MB of Amazon's MP3 downloads, a single-connection download started out strong, bursting up to 153KBps:

and then wobbled around 90KBps, continuing more slowly than those initial bursts.

Concurrent connections saturating the bandwidth (e.g. the iTunes Music Store default three-connection downloads) tend to maintain ~200KBps combined. Even during download-heavy traffic, the connection remained responsive for other network traffic.

Stuck in a spot with weak signal, 2-bars or less? As expected, connection latency rises & throughput drops; still usable but hardly "broadband." Concurrent connections can falter at 1-bar. I observed ssh/scp sessions stalling indefinitely with such little signal.

Stability

In some locations with either weak signal or from within a building, the PPP connection will be "ended by the remote host", within the first 20-seconds. That's a symptom of my signal being to noisy to reach the base-station. Landline/analog modems drop calls with this error when the line signal is to noisy.

Disconnecting & reconnecting can yield a dramatic improvement in connection quality. Once connected, sessions that start out as only 1xRTT (that's one 144Kbps CDMA2000 channel) may rev up to EV-DO.

Once established, the connection is reliable & stable. Not once was my connection dropped during up to 3-hour sessions. YMMV when in a moving vehicle.

Power Consumption

The official specs state:

Voltage: 3.3V
Typical Current - 370mA (1x), 470 mA (EV-DO)
Maximum Current - 1000 mA (EV-DO)

That means typical EV-DO usage will drain between 1.5 & 3.3 Watts. For comparison, the power-hungry hard drive in the MacBook Pro consumes between 0.6 & 1.9 Watts.

This extra battery drain is quite noticeable, cutting an hour from my normal 3:15 of work time on a full charge. For comparison, the Mac's built-in AirPort (802.11 a|b|g|n wireless) makes no noticeable difference on battery-powered work time.

Network Address

The client IP address is dynamically allocated. It changes on every connection.

The Verdict

I'm sticking with the service. The advantages of a private, mobile internet connection outweigh the admittedly expected downsides. The uncertainty: can Sprint hang in there? It's tough times in pin-drop land. Maybe I'll get an early break from my two-year contract!

an Obnoxiously Inconsistent Adobe Short-cut

2007-10-19T23:00:00Z

Cycle through open documents using Command + ` in Illustrator & InDesign; that's the OS X system-standard short-cut.

But that window switcher function is Ctrl + Tab in Photoshop.

Why Adobe, why?

on The Unbinding of Work & Locality

2007-10-05T19:44:00Z

After years of working out of an office & local cafes with WiFi, I've been jonesing for the ability to do my web work anywhere; especially under a shady tree in the park. I love loving my work. So, after comparing the only cellular broadband options available here in Austin TX (just two: Verizon & Sprint), I just picked up a cellular broadband internet card & plan with Sprint.

Here's my initial speed report using SpeakEasy's Speed Test...

Sprint Mobile Broadband (with Sierra Wireless AirCard 597E; EV-DO Rev. A capable)

Download Speed: 2289 kbps (286.1 KB/sec transfer rate)
Upload Speed: 486 kbps (60.8 KB/sec transfer rate)

Compare to my connection at home:

Time Warner Road Runner (urban cable modem)

Download Speed: 6817 kbps (852.1 KB/sec transfer rate)
Upload Speed: 374 kbps (46.8 KB/sec transfer rate)

In the next 29-days before the trial period ends, I'll evaluate if the service merits the $60/month + two-year term that Sprint requires. I'm particularly curious about how my routine ssh sessions will flow on the cellular WAN.

Streaming Radio & Pay-per-Download: a Professional Mashup

2007-09-27T19:57:00Z

What do you get when you combine a Last.fm profile and Amazon's MP3 Downloads?

You get personally tuned, streaming radio (my favorite is "My Neighborhood" in Discovery Mode) with inexpensive, non-DRM, high-quality (256Kbps VBR) MP3 downloads just a few clicks away.

Wow. I just suddenly realized how much the iTunes Music Store missed the Web 2.0 boat. Good luck Apple; I'll still buy your computers & OS.

Here's a quick demo of how the Last.fm-Amazon combo works...

Listening to a Last.fm stream:

Jump directly to the Amazon album page:

Preview and buy any tracks or the whole album:

The barrier to entry?

You must create a Last.fm profile. That's a user account + installation of Last.fm's audio scrobbler app, which acts as an iTunes bridge & streaming radio player.

Also, an Amazon customer account with credit card on-file is required to make MP3 purchases.

The Many-Colored HTTP Error 406

2007-09-21T04:39:00Z

Response code "406" makes it look like a web server error, but if you're troubled in Windows Internet Explorer 6, it may very well be a client-side error.

Client-side Causes

loading a behavior in CSS, like img.transparent { behavior: url(iepngfix.htc); }, where the URL actually returns an error (self-diagnosed in Windows Internet Explorer 6)
broken meta tags (source)

Server-side Causes

Accept-headers that do not declare any types that the server can provide (source)
over-zealous mod_security (source)

WSDL for Zimbra (The Beginning)

2007-08-04T03:21:00Z

Here's a WSDL file for Zimbra's Admin SOAP API.

Caveats

The following WSDL for Zimbra does not fulfill WSDL's true purpose: to make Zimbra's SOAP services immediately consumable. Ideally, WSDL enables mapping remote services to native objects without writing adapter code.

Zimbra's SOAP API does not elegantly fit WSDL semantics

Every message body is wrapped in an operationNameRequest or operationNameResponse element. Yet the WSDL specs say,

Multiple part elements are used if the message has multiple logical units

Zimbra goes the route of monolithic messages. Though not functionally critical, this extra level in the SOAP message causes mapping to unique complexType *Request & *Response objects in the client, instead of directly to part-level simpleType & complexType elements. Direct mapping to usable message parts enables very slim adapter code (if any) in the client. Also, the Zimbra's method names follow the pattern operationNameRequest instead of the normalized, simple operationName, adding further complexity to the WSDL.

Zimbra's web API is not RESTful

A sessionId must be propagated through each request & updated if changed in a response. Also, an initial AuthRequest gets an AuthToken, and subsequent requests' headers must contain that token.

Just give me the file

Download: zimbra_admin-example.wsdl
This is a work-in-progress, folks. Not a working solution.

Operations covered thus far:

AuthRequest
DelegateAuthRequest
CreateAccountRequest
GetAccountRequest
GetAllAdminAccountsRequest
ModifyAccountRequest
RenameAccountRequest
DeleteAccountRequest
SetPasswordRequest
AddAccountAliasRequest
RemoveAccountAliasRequest
GetAllAccountsRequest
CreateDomainRequest (added: 22 August 2007)
GetDomainRequest
GetAllDomainsRequest
ModifyDomainRequest
DeleteDomainRequest
CreateCosRequest
GetAllCosRequest
ModifyCosRequest
DeleteCosRequest

Tested with soap4r 1.5.7

I developed this file to work with soap4r's wsdl2ruby.rb script, I cannot vouch for it's performance with other WSDL consumers.

The shell command I used is:

1234	ruby -d \ /usr/local/lib/ruby/gems/1.8/gems/soap4r-1.5.7/bin/wsdl2ruby.rb \ --wsdl zimbra_admin-example.wsdl --type client \ --module_path ZimbraAdmin --force

Note: If using soap4r, make sure to actually run its install.rb in addition to gem install. Otherwise, Ruby's older built-in wsdl lib will override soap4r.

To Do

usage with a Ruby on Rails app (in progress), eventually releasing a Rails plugin
ponder developing a new, RESTful web services interface for Zimbra, on top of it's shell utilities, like zmprov

Conversion from Typo to Mephisto

2007-07-29T22:06:00Z

Just switched this site/blog from Typo to Mephisto, a process that's taken months to accomplish in spare moments of personal web productivity.

The conversion process didn't work so well at first. It took digging in to the Mephisto Converter code to make it work going from Typo 4.1.1 to Mephisto trunk.

Here's the code that made it WFM.

First patch, to avoid stupid migration failure:

Index: db/migrate/060_move_theme_path.rb
===================================================================
--- db/migrate/060_move_theme_path.rb   (revision 2936)
+++ db/migrate/060_move_theme_path.rb   (working copy)
@@ -23,5 +23,6 @@
       end
       FileUtils.rmdir new_path
     end
+  rescue
   end
 end

Then, copy Typo's categorizations table to a new table named articles_categories. ~~(See what happens when you ignore Rails' database conventions?!)~~

Next, create the file vendor/plugins/mephisto_converters/lib/converters/typo/feedback.rb:

module Typo
  class Feedback < ActiveRecord::Base
    establish_connection configurations['typo']
    set_table_name 'feedback'
    belongs_to :text_filter, :class_name => 'Typo::TextFilter'
    def filter
      self.text_filter ? self.text_filter.name : nil
    end
  end
end

Two patches to switch over to use the new Feedback class:

Index: vendor/plugins/mephisto_converters/lib/converters/typo/comment.rb
===================================================================
--- vendor/plugins/mephisto_converters/lib/converters/typo/comment.rb   (revision 2936)
+++ vendor/plugins/mephisto_converters/lib/converters/typo/comment.rb   (working copy)
@@ -1,5 +1,5 @@
 module Typo
-  class Comment < Content
+  class Comment < Feedback
 #    belongs_to :article, :foreign_key => 'parentid', :class_name => 'Typo::Comment'
   end
 end

Index: vendor/plugins/mephisto_converters/lib/converters/typo.rb
===================================================================
--- vendor/plugins/mephisto_converters/lib/converters/typo.rb   (revision 2936)
+++ vendor/plugins/mephisto_converters/lib/converters/typo.rb   (working copy)
@@ -1,6 +1,7 @@
 require 'converters/typo/content'
 require 'converters/typo/page'
 require 'converters/typo/article'
+require 'converters/typo/feedback'
 require 'converters/typo/comment'
 require 'converters/typo/category'
 require 'converters/typo/user'
@@ -107,4 +108,4 @@
     sec
   end
 
-end
\ No newline at end of file
+end

Finally, start her up:

12	% rake db:bootstrap RAILS_ENV=production% script/runner --trace "Mephisto.convert_from :typo" -e production

RMagick install woes on OS X

2007-07-05T23:50:00Z

Well this is quickly becoming Mars' notes from building software on OS X blog!

Anyway, after following the official RMagick install, gem install rmagick fails with an error like:

/usr/local/lib/ruby/gems/1.8/gems/rmagick-1.15.7/./lib/rvg/misc.rb:321:
in `get_type_metrics': unable to read font `/Library/Fonts/Verdana' (Magick::ImageMagickError)

It seems that the examples fail because of a spurious font choice in the documented examples.

All that is needed to build the RMagick gem successfully at this point is:

sudo gem install rmagick -- --disable-htmldoc

And voila, the examples aren't built, thereby avoiding the problem.

RMagick!

"Stack to Deep" Ruby Error on OS X

2007-06-05T20:38:00Z

Say for instance you just applied a software update to your Mac OS X Server and rebooted.

Suddenly a Rails app that was running fine spits out an error when ActionPack does a redirect_to. In the Rails production.log:

SystemStackError (stack level too deep)

Well my friend, before you set off to debugging the Ruby install on your OS X machine, try clearing the Rails sessions. At the command line:

rake tmp:sessions:clear RAILS_ENV=production

Everyone must login again, but alas the problem is solved! [for me]

Is it because we're using Apple's built-in Apache 1.3 web server with source-compiled FastCGI support? Hmmm.

Serious Web Design with PNG

2007-05-30T14:17:00Z

Want to try implementing a design with PNG graphics?

The glaring problem of PNG images is synchronizing color appearance with GIF/JPEG images & CSS colors. This is a well documented problem.

PNG vs JPEG/GIF color problems can be avoided by using PNG for all of the graphics in a site that must perfectly match.

Otherwise, the best path to PNG bliss is a shell command tool that strips out all that advanced color/gamma data (since it confuses different browsers in different ways) and tries each of the 130+ compression modes to find the most efficient one.

pngcrush -rem gAMA -rem cHRM -rem iCCP -rem sRGB -brute infile.png outfile.png

Get it at the pngcrush development site or as I did (for Mac OS X) via Darwin Ports.

Lone Star Ruby

2007-05-19T12:12:00Z

Texas' premier Ruby Conference web site is now officially live.

The site is powered by RadiantCMS running on two Mongrels balanced by Nginx.

There's not much info yet, just a pre-registration form to help the plan the event. Plan to come on down to the Ruby Conference in Austin. It might actually drop under 100-degrees by September 7th!