The experience of bad coffee

“Two cappuccinos please”.

I see the operator pick up two cardboard cups. They have a terrible pattern on them. So I say,

“Sugar in one please”. I’m hedging my bets. “Brown.”

The ground coffee comes out of the hopper without a whiff. There is a light tamp, if you can call it that, when it comes at a 15 degree angle from a small plastic roundel, attached at thorax height to the hopper. I suspect a force of 20N at best.

Now it starts.

The portafilter is screwed in place, the cardboard cups are put straight under the nozzles and the button is pressed - the little button, which could have served admirable duty not ten seconds ago, flushing tepid water from the front of the boiler tube – the little button, with the two silhouettes of shot glasses on it, trying its best to remind the operator which vessels to extract espresso into.

“How many shots do you put in your cappuccino?”, I ask, smiling.

“It’s up to you”, is the reply, as the operator picks up a dirty milk jug.

“Oh. Well, how many normally?” I pursue.

There is a pause, before the operator says,

“We normally fill half coffee, half milk”. This starts a shiver of panic.

The steam valve is turned and the banshee is unleashed. The wail is awful to hear, high-pitched one moment, low and gurgling the next. I wonder how many times this same, poor batch of milk has been exhaled. Steam rises and there is a smell.

The pour passes by without incident. The operator’s back is turned and I am staring at my croissants. I am given the two cardboard cups with thin, white plastic lids, on a recycled cardboard tray.

I like to take my first sip from a cappuccino with the lid off, and cover my tongue in silky milk foam. I take the lid off the coffee with the sugar in it. There is a surprisingly gentle-looking white layer, powdered with brown. I sip.

I can’t be sure which hits me first – the bitterness or the scalding heat. As my eyelids close in recoil, I think about saying something, but of course, I don’t. I put my ¬£3.80 down on the counter. Ten minutes later, I drop both cardboard coffee cups into a nearby bin.


What software companies can learn from drug dealers

Sauce Labs bannerLast week, I opened an email from Sauce Labs, which described some upcoming new features of their service, including introducing a free account with a cap on usage. Sauce Labs is an online service for running browser tests on different operating system and browser combinations (using Selenium, if you’re interested).

I really like what Sauce are doing – amongst other things, they automatically video the tests, and they’re making it really easy to use their hosted service from behind your company firewall. What I didn’t like up until now was their pricing model. Sauce offered a 30-day free trial, which one would think would be ample to figure out whether you could get value out of their service.

However, this “free trial” model misses the mark for me. I find that my interest in (and therefore my signup to) a new service is piqued several weeks or months before I am actually in a position to use it. And once I do start to test something new out, my usage is faltering and tentative to begin with, only ramping up once I have accrued experience, reflected on it and figured out how the service fits in with everything else I use.

Big congratulations to Sauce, then, for twigging that there is another way to bag new customers – offering a small, but effective amount of their product, which you can use at your leisure. You could call this the “drug dealer” model.

The drug dealer model makes a lot of sense for online services, where the incremental cost of adding a new account is minimal, and the cost of usage is easily measured. In Sauce’s case, their free account comes with a cap of 200 minutes of testing per month – easily enough for me to run a few tests, see if they help my development cycle and build up the use of the tool – at which point, I expect I’ll be hooked.

I am not saying I had anything to do with the decision to offer a capped free account, but…!/hugs/status/31074705414819841 ūüėČ

Kurt Vonnegut & Destructive Testing

I’m reading Kurt Vonnegut’s fantastic Breakfast of Champions at the moment. It’s the kind of book you can sail through, and reading it feels like unboxing a load of toys.

About half way through (where I am now), he has one of the lead characters, a man called Dwayne, who is going crazy, tell a story of a Pontiac car manufacturer and their testing labs:

“We were given a tour of all the research facilities”, he said. The thing that impressed him most, he said, was a series of laboratories and out-of-doors test areas where various parts of automobiles and even entire automobiles were destroyed. Pontiac scientists set upholstery on fire, threw gravel at windshields, snapped crankshafts and driveshafts, staged head-on collisions, tore gearshift levers out by the roots, ran engines at high speeds with almost no lubrication, opened and closed glove compartment doors a hundred times a minute for days, cooled dashboard clocks to within a few degrees of absolute zero, and so on.

“Everything you’re not supposed to do to a car, they did to a car,” Dwayne said to Francine. “And I’ll never forget the sign on the front door of the building where all that torture went on.” Here was the sign Dwayne described to Francine:

© Kurt Vonnegut Jr 1973

The detailed list of destructions made me think about testing web apps like cars. So I had a Google for ‘destructive testing web’ and to be honest, apart from a research paper and a very expensive-looking testing suite. Not the glut I was expecting.

So, to any web app testers out there – what’s the word?

International JavaScript show ‘n’ tell @ Async

The Full Frontal JavaScript conference is tomorrow, so this is a very international audience.

A bit of a show ‘n’ tell tonight at the fortnightly Async JavaScript meetup, hosted at The Skiff in Brighton… notes below.

Links for the talks ought to be added to the Async blog post ( by the speakers.

The next Async features Kyran Dale (session 7 below) on module loading (‘requireJS’), in two weeks time, also @ The Skiff.

1. Mike de Boer (from – Cloud9 IDE

Written in JS (as a node app) for dev’ing JS. Nice. “npm install cloud9”

Mounts filesystem/WebDAV server.

You can write a node app, which is run as a separate app and has the output logged to a console.¬†They’ll have an online version up soon.

Does live debugging, and you can variables real-time via breakpoints. You can also edit the code when execution is paused. i.e. hot-code swapping. He mentioned node inspector (what’s that?).

The code editor can cope with 134k lines of code. Which textmate can’t. Nice.

You get a command-line too, for using git (other version control systems coming later).

2. Glenn Jones (Madgex) – Draggables

HTML5 experiment for dragging data between systems.

Lets you drag across browsers, which is cool. Also you can drag html files (and zipped html files!).

Really nice little thing with a bookmarklet injecting a draggable icon into a page (well, on Firefox).

Glenn is also working on a microformats parser that works cross-browser, which he will open-source.

3. Phil Hawksworth (The Team) – nodejs & ndistro – “multiple projects. quickly.”

I.e. how Phil works.

Problem with node is deployment (i.e. build) can be hard. Plus, the codebase is changing frequently. So how to cope?

ndistro – a node distribution toolkit (

Note: installing involves running install script as root (sudo su Phil recommends)

You write ndistro files for each project, where you specify the version of node and any modules from github

Phil like ejs as a templating language

Phil has some very nice presentation techniques – videos of terminals and a funny, eased, scrolling of a webpage. Guess keynote == good?

Seeing as ndistro files are just bash scripts, you can nest another running of ndistro inside the ndistro file

expressjs = “almost” MVC for node

After a Q: Not sure if npm would play nicely with this.

4. Prem (@premasagar) & Graeme Sutherland (@grasuth) – Sqwidget

The problem: widget developers have a really big problem developing something that runs in other people’s environments (which is like cross-browser development, but worse).

One solution: iFrame

Their solution: sqwidget.

It sandboxes CSS and JavaScript. This is cool, because you can do things like lightboxes, which you can’t do with iFramed widgets.

The widget is embedded using a DIV and its data attributes to do settings

Q: option to swap in different templating? A: yes. Built-in one is a hand-built library called Tim.

Q: licence? A: MIT

5. Mark Wubben, from Copenhagen (who needs a place to live in London) – Chrome2iPad

Problem: how do I get my tabs I have open in Chrome into my iPad?

Solution: Chrome2iPad.

There is a Chrome extension using (a node module) which hotlinks to a webpage on the iPad to the Chrome browser via a Chrome extension

Not released on github yet. When it is available, it will be on

6. Paul Downey (@psd, Osmosoft) – TiddlySpace

TiddlyWiki is a single-page HTML app.

I’m writing these notes in a TiddlyWiki…

TiddlyWiki is interesting because it can save itself when on a file URI.

All the plugins are written in JavaScript.

TiddlySpace is like TiddlyWikis in the sky. It’s slightly different from file-based TiddlyWikis, since you can follow people.

TiddlySpace is based on TiddlyWeb, which is an open-source hosting platform for tiddlers, with a RESTful API.

A tiddler in a TiddlyWiki is a div; a tiddler on TiddlySpace can be HTML, JavaScript, SVG – anything.

7. Kyran Dale (academia) – JavaScript physics

Impressed by JavaScript’s performance to do simulations.

Shows Craig Reynold’s Boids running 300 objects, drawn in canvas.¬†Chrome on Linux. Real-time variable changing.

Now a robotics demo – 100 robots (Brightenbergs) with light-sensors and lights. They go mad. But without slow-down.

Next a magnetic controller. This one does some weird stuff, but again, in a peformant way.

Q: source? A: will be available, probably on a Mercurial repo (

8. Morgan Roderick (@mrgnrdrck) – PubSub in JavaScript

Writes really large systems in JavaScript.

PubSub is messaging in JavaScript. It is an alternative to the “observer pattern” (see Java, apparently).

Keeps different bits of your system nicely decoupled.

Don’t use PubSub when you have 1:1 communication, or need accurate control.

Q: is is threaded? A: all messages are passed asynchronously. In terms of performance, it’s about 10x faster than jQuery.

Q: how is it different to custom events in jQuery? A: it doesn’t depend on the DOM

Q: how do you wake listeners up? A: They just get called. If a call fails, the exception is caught and it is called again.ÔĽŅ

PaaS: A mid-2010 survey

I’ve said in the past that I am a hands-off type of guy when it comes to servers. To avoid dealing with servers in any traditional sense, I have been keeping an eager eye on how things are going with server-side JavaScript, and the push-to-deploy type of application development (Heroku, Joyent Smart Platform). But there’s something relatively new in the offing, which claims to do away with code entirely – the Platform-as-a-Service (PaaS) company. You might have heard of some – LongJump,, Zoho Creator, QuickBase (there’s a longer list at the bottom of the post).

In general, people who label themselves with “PaaS” want to do two things: reduce your dependency on coders, and give you a shortcut to setting up business online. Reasonably appealing, you might think? I’ve been giving the PaaS sector a bit of a once-over in the last couple of weeks, and this is an attempt to piece together half-formed conclusions about what PaaS is and whether it is useful to me.

Rapid recommendation

If you don’t have time to read the rest of this article, then here’s my advice:

  • if you are at the non-technical end of the scale, with no access to designers and developers, and you don’t care about the way your apps will look, PaaS will help you open up your company information for visualising and editing; changes can be set up to trigger other changes and notifications, and you can schedule reports
  • if you are a professional web designer and developer, you are likely to become frustrated at the unresponsive tools used to create PaaS apps, and the barriers to designing the experience of the people using your app

A motivating problem

This survey was prompted by the familiar need for a pair of hands to help build a system that neither myself nor my partner felt capable of building well. If I give you an outline of what the system roughly looks like, it ought to set my opinions of PaaS in context.

Essentially, we want to make a website handling information about people and the properties they live in. There are public areas for browsing and private areas for people with accounts. Plenty of design has gone into the appearance and function. There are some web services in the mix – PayPal, a booking system I wrote in server-side JavaScript, and EchoSign. Interactions with the website and notifications from foreign systems need to have effects, manipulating the people and property information.

I think that what I have just described sounds like a very generic web application. Some bits require more know-how than we possess to build: how to write and manage workflows, which link events and notifications to effects; what our data stores should look like and where they should be relative to the website and its contents; how to get our system to respond to notifications from foreign systems; and how to serve the product to multiple customers and charge for it.

I don’t recall what prompted me to look at PaaS, but the PaaS companies I first looked at made a big deal about how you can easily model company data structures and create workflows between them. This sounded great. So I went about having a long look at what was out there and whether it would be cheaper and easier to build my app on top of that.

What does PaaS claim to do?

In general, PaaS companies want you to model your business and its processes as objects on their platform, with various properties hanging off the objects (think and its tabs). You create forms that open up the system to let people add and amend data, and you hook workflows on to triggers such as the submission of a form. The example that everyone seems to use is an expenses approval system, with Bob, his manager, his manager’s manager and Lynne from accounts, ploughing through the process of dealing with Bob’s $5000 business trip claim.

Once you’ve configured a PaaS app to model whatever it is you want to model, you can publish the app to the world, often as something that other people can buy and customise for themselves and their own businesses.

Variations around this theme include companies devoted to: creating complex workflows; providing sets of micro-services; pain-free hosting for code.

General characteristics of PaaS

Without giving what would probably be a boring account of all the companies I looked at, there are several axes along which PaaS seems to range which it is worth looking at. I’ve picked out a couple of example companies that fall along each axis.

Building a business

Several PaaS companies highlight your ability to use their platform to run a business. That can mean several things: you use the app you make to help run your business; you use the app as a part of your own customers’ experience; you run a business selling the app to other companies to use with their customers. Some companies cater for all these scenarios, others specialise in one or two.

If you’re building an application to use with all your customers, it makes sense to make it a multi-tenanted system – that is, the different customers all log-in to the same app, but see whatever is appropriate to them. If you are building an app that other businesses will sell to their customers, you’ll want something a level deeper – multi-tenancy within multi-tenancy (erm, Inception?).

Apps for your own use: Zoho Creator, TrackVia, Iceberg
Build your shop: Caspio Bridge, Wolf Frameworks
Shop-within-a-shop: WorkXpress, Rollbase, LongJump

Technical accessibility

If you’re a no-hoper when it comes to HTML and you’ve never heard of PHP, there are PaaS products for you. They will help you model your business and its processes and do helpful things like sending you emails when your tasks are late.

Equally, some products require someone versed in the vagaries of Microsoft systems integration to get anything out of them. However, if you have the requisite knowledge, you could be integrating with your Active Directory or performing complex database queries to help the system make decisions.

Anyone could do it: TrackVia, WorkXpress
You’ve just come off a web dev course: Zoho Creator, BlankSlate, Iceberg
You’ll need a seriously hardcore dude: Skelta, LongJump

Access to the bare metal

If you’ve ever used Salesforce, you’ll know that almost every Salesforce app looks the same. This is not an adequate level of control over the experience. Ideally, a PaaS product should give you – the app developer – enough power to control your customers’ and their customers’ experience (hopefully whilst providing tools to make this easier than coding from scratch). Some companies have a stab at this (even if they don’t make it very easy), others lock the experience down.

Complete control: BlankSlate, LongJump
Not quite like putty: Zoho Creator, Rollbase, WorkXpress
You do it OUR way!: Skelta, Iceberg

Presentation: AJAX vs. templates

This is not PaaS’ strong point. It seems very few PaaS companies expect there to be a presentation layer beyond tabs and tables, which is a shame.

You can pretty much take it for granted that online products provide API access to read & write your data. PaaS products are online products and that assumption seems to hold good with most of them. This of course means that you can create experiences all of your own by writing AJAZy mashups. This isn’t ideal in practice, as it leaves the non-JavaScript browser without an experience at all, not to mention that AJAZ-heavy apps are unlikely to function appropriately on a mobile; SEO and accessibility are concerns as well. Unfortunately, some PaaS companies actively promote this method.

You could employ an app developer to use the PaaS API’s as a substitute data store, but this approach does seem a little odd given that a stated aim is to take the developers out of the process, but anyway‚Ķ

The PaaS companies who have their heads screwed on correctly, manage your presentation layer using templates, interpreted on the server. However, you are often straitjacketed into using a particular layout. It would be a valid PaaS model to offer you an easy way to hook in a presentation layer of your choice, and since an app’s appeal very often benefits from decent visual presentation, this model could look very attractive.

AJAZ-only is the way: Wolf Frameworks, BlankSlate, Caspio Bridge
Templated goodness: LongJump, Zoho Creator

But does any of this suit?

My overall impression from looking at around twenty-five PaaS companies, and deeper at a dozen, is that everyone wants you to think their way, and that even though webby people are generally supportive of open data, there is something very “lock-in” about the way you put these applications together. This includes the feeling that you have to get into their developers’ heads to understand how you should construct an application; that your application is really just tweaking something that’s been set up in advance, both functionally and visually; that no-one wants to play nicely with any other tool-providers out there.

Then there’s the question of whether the objects/properties/process model is as straightforward a mapping of the real world as is claimed. Even if an entire business can be abstracted to these bare elements, the difference between two businesses is more than skin-deep, and so this abstraction feels hollow. I am troubled by the observation that I struggled to find any examples of PaaS offerings backing good public websites and services. Only BlankSlate had anything attractive and functional to show, and they had integrated with existing websites (BlankSlate are entirely focused on providing APIs).

The main advantage to the craft of web development and web design is that you are free to imagine any types of interaction and see those brought to life. You design and create both public and private areas, and hand-code the logic that links the two. Normally, this involves a creative and complex response to a client’s problem – a design process.

I think that the experience of use, which is the end result of a design process, is the most important facet of an online system. In my opinion, PaaS companies are neglecting this, and they appear to imagine their customers want to build applications for robots. Even if a PaaS app is only destined to be used inside a business (which it would appear they are in the vast majority of cases), that is no excuse for a bad experience. Haven’t big-company employees been complaining about the state of their corporate IT for decades?

Web development by web developers is something approached with a toolbox. What makes a tool? Something that is self-contained and has its part in the bigger job – a favourite framework, or an image editor, for example. The tools work together to turn stuff into useful stuff. PaaS is supposed to be bringing the experience of creating application development to the people who are coming up with the needs for them – “business” people. Most PaaS products do not look like a toolbox, but a single, infinitely complex tool. And when an app is built, it’s likely the people using it will be doing so through the tool used to build it i.e. itself.

Something’s gone wrong here.

So what do I want?

Here’s my shopping list for PaaS:

I want the typical PaaS setup to resemble a toolbox, full of useful things that you use to create experiences. The output is not just the tool, reconfigured.

I don’t want to fight with my tools – they should be tuned to the context they will be used in and they should respond to my touch (laggy interfaces are a massive turn-off).

The best (web) tools are those you can extend in a modular way with plugins. They play nicely with other tools, exporting and importing files in standard or commonly-used formats. People get passionate about them, because they let you exercise your craft.

Tools should talk JSON (or, if necessary, XML). While we’re on the J-subject, JavaScript is spoken by many, and JavaScript on the server got hot a while ago – if you want to empower less-technical people, start supporting it.

Everything you create should have a URL and behave as a RESTful resource. Then people can do cool things with their resources you didn’t expect.

Notifications over HTTP is the underpinning of workflow. The thinking behind Web Hooks is very appealing – systems talking to little blobs of code on the web. This keeps things nice and open.

Open source used as a learning process: almost everything a person makes is not worth hiding and definitely worth sharing. In an environment where people are working with similar systems, open source will make people make better things (we don’t have to be talking code – process, procedure, example are all good).

Finally, don’t try to do everything. Not everyone needs to write their own process editor or form maker, nor do they need to invent their own cloud or firewall.

The companies

This does not include everyone who thinks they are doing PaaS (Google App Engine and Heroku are not on my list). My investigation was focused on developing applications with less technical skills, but I’m mentioning all the PaaS-esque companies I found in case they are useful to someone. In no particular order…

Generally convincing

Zoho Creator
Wolf Frameworks

Generally unconvincing

Caspio Bridge
Microsoft Dynamics
Google Scripts
Boomi AtomSphere
Informatica Cloud
Cast Iron OmniConnect

Employee effect on the world’s biggest companies

I was wondering on Saturday how you could reasonably compare a large company with a loose band of freelancers, where the word “employee” doesn’t operate. I dug out of Wikipedia some statistics on the world’s biggest companies, figuring that a reasonable comparison would be the amount of money made per employee.

The data is likely to be a bit rough and I used data from different years indifferently, but I was only aiming for a rough idea so I think the conclusions are valid.

Here are the headlines, the data is in a Google Spreadsheet you can play with here.

Employees of the world’s wealthiest companies pull in an incredible amount of cash

The world’s most financially successful companies boast a significant amount of revenue per employee – an average of $1.17m, with Fannie Mae and Freddie Mac (US mortgage megaliths) standing out at front with an amazing $8.7m per employee between them (this dates from 2007, pre-global financial meltdown and government takeover, but still…). Even the lowest earnings per employee I looked at was around $114k, significantly above the US average salary of $32k.

This has made clear to me how much of an economic heavyweight successful companies can be. It is pretty amazing to see that for every single person Fannie Mae, Exxon Mobil, Legal & General or Goldman Sachs employs, millions of dollars are being pumped into the US economy.

Big doesn’t mean rich

Interestingly, the world’s most people-heavy companies were not showing such impressive financial results – whilst the average revenue per employee was up near $200k, the highest didn’t even break $1m and the lowest was just under $12k (although this is Indian Railways who employ 1.6m people – I couldn’t find the average Indian salary but it is likely that $12k is several times bigger).

The big difference between the earning potential of the world’s wealthiest and biggest companies could come down to the difference in industry – those with the most staff tended to be in manufacturing, infrastructure and retail, with the richest companies circulating around the world’s oil and gas, healthcare and finance.

Most companies don’t grow past 500k heads

I wondered whether there was any pattern of companies increasing their revenue per employee up to a certain size, and then becoming more inefficient per head as they continued to grow, so I made a graph for the world’s 38 largest companies:

Graph showing revenue per head as you add more heads

What this shows must be taken with a pinch of salt given that country of origin is not shown. However, we can see that passing half a million employees is very hard; it is also very hard to do that whilst maintaining a high revenue per head.

There appears to be a weak correlation between increasing a company size between 250k and 500k employees and decreasing revenue per head. It would be interesting to expand the data set to include smaller companies.

Graph showing revenue per head up to 500k heads

Ready Steady Vote

"ready steady vote" on Flickr

JavaScript in the cloud

thanks to flowerhouse on flickr

The idea of running JavaScript on the server has been around for a while now (think Jaxer back in 2008), but it recently got a big boost with the featuring of Node.js at JSConf in November 2009. Node.js found immediate fame by demonstrating blinding performance as a web server, and by building on the hotly hyped V8 JavaScript engine Google bundled with Chrome.

I have a big interest in server-side¬†JavaScript¬†(SSJS) because I generally code in the browser and would enjoy being able to make things happen on a server without having to resort to another language. The ideal would be to take client-side code and run it unchanged on a server. The last step is pain-free application hosting – call it “Platform as a service” for want of a less boring name.

Over a year ago, Joyent¬†bought Reasonably Smart, a cloud provider of SSJS, and subsequently transmuted it into¬†Smart Platform, which runs on their distributed infrastructure. This followed the shining example of Heroku and Google App Engine, in that application deployment became a matter of pushing some new code to a Git repository online. For me, this was fantastically attractive, as I avoid touching servers as much as realistically possible. They came out with Smart Platform shortly before¬†the demise of AppJet’s free service, which I had been happily trying for a while, as it did much the same thing.

Nearly a year on, and Smart Platform hasn’t really taken off, if¬†Google Trends is anything to go by. And my ambitions to run my entire web estate on identical server-side and client-side code have rather fallen by the way-side. I am happy to see that Smart Platform now has a permanent full-time coder (@konobi), who is making Smart Platform play nicely with the CommonJS spec for server-side JS, but I don’t yet know when Joyent are going to drop a new release, or where they’re going with it.

Enter Node.js. The interest in Node.js has stayed high since its November launch, with developers building an assortment of frameworks and tools on top of it or with it, in effect legitimising it as a serious development environment for software running on a server.

Of significant interest to me is the¬†jsdom project, which brings the browser environment to Node.js in a similar way to how Env.js brought it to Rhino (and Ruby). Env.js has supported jQuery¬†since the get-go, and there are¬†rumblings of jQuery support in Node.js via jsdom. Both these situations inspire joy as, if they pan out successfully, I’ll be able to use one methodology for writing and testing¬†JavaScript¬†applications, whether they run in the browser or on the server. Unfortunately, I’ve not seen Env.js working on any hosted platform (although Smart Platform¬†may get it soon).

The number of cloud providers for Node.js has increased from 0 to 2 very recently: Heroku has¬†announced a very experimental Node.js stack, and¬†ElusiveHippo is in development, releasing accounts slowly. I should probably also mention that Ryan Dahl (who created Node.js) works for Joyent, so I wouldn’t be suprised if a Smart Platform-style service came out for Node.js.

Now then, if Coda starts¬†supporting Git, or I start doing all my code editing in Bespin, I think with a bit of fairy-dust I’m going to have a pretty slick¬†JavaScript-only¬†setup fairly soon…

Web designers

(I was thrilled to read’s description of a web designer, which inspired this post.)

by Dreamtargets on flickr - /photos/dreamtargets/1085206378/

It’s not every industry where you can be a 30-year old and legitimately claim to have been working in the industry since it’s birth. The field of web design is one of these industries, and the nature of the job has changed significantly since the days when animated gifs were the height of sophistication. I can’t claim any knowledge of what it was like to be a web designer in the late 90’s – the closest I can come is that I had a subscription to .Net magazine. However, I do know what it’s like to be a web designer now and I know that it is a badly named job.

People in the web industry tend to be labelled either “web designers” or “web developers”, the assumption being that you either draw the pictures or write the code. This may have been a reasonable separation at one point. The truth as I see it now is that the web designer has progressed from writing HTML and CSS to taking responsibility for everything that happens within the browser. Web developers have retreated to the increasingly complex job of moving and processing data, around and between servers.

This means that web design, in practice, is a combination of typography, human-computer interface design, graphic design, rapid prototyping, JavaScript programming, HTML/CSS, agile product development, to name a selection. The modern web designer understands the significance of REST and HTTP, and why they make her job easier. She may even know regular expressions. If there is another word to replace “designer”, which captured the spirit of experimentation, creation and science that goes into modern web design, I am stuck to think of it.

Web design has flourished as a design profession – matured enough for an average project to bear the hallmarks of the kind of “design thinking” you might find amongst architects or industrial designers. This is all for the good. But web design is hard – there are barbed-wire boundaries between the work of web designers and those that surround them: the web developers, the content authors, the artists, the project managers, the business development consultants. Web design is relatively straightforward in isolation; meshed with all the other people involved in a project, it quickly becomes unstable. In my opinion, what holds the industry back from developing beyond a craft – that is, a complex and manual task – is the absence of good tools.

I overheard a conversation recently about why Flash became so popular with designers and has remained so, despite a strong feeling opposing its indiscriminate application. The problem boils down to one of tooling – Flash applications are built in beautiful, assistive development environments, which produce cross-browser and cross-platform applications. Web sites, on the other hand, are usually built in text editors and involve a significant investment of time to tweak and test for all the browsers in which they are likely to be run. Adobe produce compelling videos demonstrating how their tools simplify workflows and allow multiple professions to work together. Web designers generally have to “throw things over the wall” or duck things being chucked in their direction.

Specifically, I’m interested in seeing solutions to these sorts of problems:

  • Why isn’t it easy for content authors to decouple their stuff from the ongoing design of a website?
  • Why do I design a website in a different environment from the one it is going to be deployed to?
  • Why do I have to cut up PhotoShop comps into web pages and then cut up those into templates?
  • Why can’t I change the use of a single colour across a website by changing it once in the stylesheet?
  • Why can’t two web designers work on the same web page without it being a pain to maintain?
  • Why does everyone have to diagnose, fix and learn the same cross-browser tricks?

Testing local TiddlyWeb applications in virtual machines

This is particular to VirtualBox, but I imagine the principles are the same for VMWare, Parallels and all those other virtual doohikies.

I’ve been developing a TiddlyWeb application on a local server on my Mac. This is fine for when I want to test in Firefox, Safari, Chrome, Webkit and others, but not so hot when I want to test in IE6/7/8. Virtualisation makes the general problem a lot easier, and I run VirtualBox for that purpose.

Unfortunately, accessing a locally running web server, rather than plain ol’ Internet, gave me some problems, so I thought I’d write down how I managed it.

Generally, to start up a TiddlyWeb instance, I’ve been using:

twanager server localhost 3000

I can point a browser to http://localhost:3000/ and I’m away. If I want to run on port 80, I can use:

sudo twanager server localhost 80

and then I can point a browser to http://localhost/.

The key thing to getting my virtual machines talking to my local server is to run TiddlyWeb on an IP address, rather than on “localhost”, which I had been doing. The IP address to run the application on is the IP address of your machine on the network. So how do you get this?

Given I’m using a mac, the non-confusing way is to open up System Preferences, hit the “Network” icon and read off the IP address in the “Status” section.
Network options in System Preferences

The more confusing way is to open up a Terminal, type in ifconfig and read off the IP address:

Running ifconfig on the Mac Terminal to get your IP address
With your IP address in hand, you can now use twanager to start up TiddlyWeb on an address accessible to your virtual machines:

twanager server 3000

Then you’ll be in IE testing hell in no time!