A few days back there was a big debate about when to do unit testing and how much should be tested. A bunch of the programmers I really admire like Greg Young, Jimmy Bogadt and Uncle Bob Martin weighed in. The debate was about what you would expect: Uncle Bob was prosthelytizing his belief that testing is key and that the quest should be for 100% code coverage.

New Blog. The Start-Up Trap. http://t.co/341M4wNDh2

“” Uncle Bob Martin (@unclebobmartin) March 5, 2013

On the other side of the debate the more pragmatic Greg young and Jimmy Bogard were suggesting that blindly testing everything is a naive approach. One should pick and choose the code to test. How much code should be tested? Well it depends. I’ve been in the same boat as Jimmy, testing nothing, testing everything and testing something in-between. What’s the right mixture? I have no idea.

To me it comes down to: what is of value to your business. What’s the consequenceto failure?

If you’re a bank then the consequence of failure is pretty bad. If you’re a service like twitter then it isn’t too bad. For a long time twitter was super unreliable, but they got through it and became highly successful.

Remember this guy?

Test the portions of your application which would have the greatest impact if they were broken. Are you reliant on processing invoices? Then invoice processing should be the focus of your testing. Are you a service which authenticates users? Then that should be the focus of your testing.

Another approach is to test the parts of your application which change the most frequently. The idea being that the parts which change the most frequently are unlikely to have the same level of exposure to user testing as the rest of the site. Changing stable code is necessary but likely to introduce problems.

I don’t know what the solution is to how much testing needs to be done. As with all difficult problems I feel like a single solution is over simplifying the problem.

No testing? Test everything? Somewhere in the middle.

When you’re getting started with typescript you’re probably going to run into a bunch of warnings and errors. Pay attention to the messages, I’ve found them to be almostentirelycorrect. However you will see errors of the form

The name $ does not exist in the current scope

Why that’s jQuery, why doesn’t typescript know about jQuery? Well typescript doesn’t know about anything other than what’s in the current file. It is a dumb as a muffin. A tasty, tasty cranberry orange muffin. Maybe it has thatstreuseltoping on it”¦ sorry, I really like muffins. What can you do to solve this problem? Well that’s where definition files come into play. Definition files are, in effect, C include files. They define the public interfaces of the libraries you’re trying to use. Libraries like jQuery and d3.js. Libraries like your own libraries.

If you’re referencing your own typescript file then you don’t need to create a declarationfile. Typescript can read its own typescript files so there is no need to generate declarationfor them unless you’re distributing the declarationto other developers to develop against. To instruct the transcompiler to make use of either an external typescript file or declarationfile you can include a reference in your typescript file.

/// ///

• *You can see here that I pulled in both a declarationfile for jQuery and a typescript file I created.

If you’re looking for a declarationfile for apubliclydistributed library then there are some great options. You can manually download it from Boris Yankov’s github repo athttps://github.com/borisyankov/DefinitelyTyped. Or you can grab a package off nuget. If you’re using node then you can install the typescirpt definition tool fromhttp://www.tsdpm.com/and install packages using.

tsd install node

To generate your own definition files from a typescript file you can pass ““declaration to the transcompiler. This will,unfortunately only generate from typescript files. If you want to generate declarations from pure javascript then you’ll need to do it by hand.

Sometimes I go to a bootcamp style gym and theprescribedworkout is something insane like 100 pullups, 200 pushup and 300 squats. On the surface this seems impossible. On a good day I can do perhaps 20 pullups in a row which is only 20% of the number needed here. During the workout, which took me a little over an hour last time the muscle rich coaches yell platitudes like something from a Simpsons episode. One of the favorite chants is “Chip away at it, chip away”.

With ripped hands and believing that I would never again be able to inhale like a normal person again I have very little interest in hearing the chants. (My favorite one of the last week was “burpies should be your rest, they’re easy”) However the chip away comment I find to be pretty applicable to software development. On any project of any size and of any age there is going to be a great deal of technical debt. Trying to pay down technical debt all at once is impossible. You’ll never convince management that all new development needs to stop so that you can make invisible improvements to the code base. Instead you should make as small a changes as possible while you’re doing other development.

My rule is that if I’ve opened a file, even if I’m just reading code, then I need to make an improvement. It could be as simple as removing unused using statements or changing a stringconcatenationto a string.format. These don’t seem like they’re going to do much to pay down your technical debt but their effects arecumulative. Eventually you run out of trivial changes to make in your files and you start making slightly larger changes. These slightly larger changes add up to larger changes. Before you know it your code base has been improved.

I followed this idea at a job once. When I started it took us 2 months to do a realease of our software. We identified the biggest pain point and automated it. Then we repeated this process until we actually ran out of things to automate. It took a couple of years but we got builds down so that they were running every night and producing full release packages. I automated myself out of a job. At least I would have if we didn’t end up responsible for two dozen additional products. If we hadn’t worked away at it wewouldhave been working 90 hour weeks just to stay on top of things.

Paying down technical debt is like paying a mortgage. If you just pay a little bit more each week then your debt will be paid down much sooner.

Chip away at it.

In my mind one of the most abused bits of functionality in ASP.net MVC is the routing engine. I don’t think this is a problem unique to ASP.net MVC as other MVC frameworks like Rails also make heavy use of routing.

In a nutshell routing allows you to map the URL(or URI if you want to beentirelycorrect) to a file or action on the server. For instance if you look at the URL for this post

http://blog.simontimms.com/2013/03/04/why-you-shouldnt-be-bothering-with-routes

You can see that there is some information buit into it. I don’t know the internals of wordpress but you can be pretty much assured that there isn’t actually a file on disk calledwhy-you-shouldnt-be-bothering-with-routes in the directory/2013/03/04/. Instead this URL is used by a routing engine to passappropriateparameters to a script which looks up information in a database. The combination of the domain and the name of the blog is probably sufficient to identify the record in the database. The date information in there is just a hint to readers so they know when the blog was written.

The only thing is: its wrong. I’m writing this blog on the 9th of March and it is scheduled to be published on the 13th. The reason that date is there is that I created a draft on the 4th. That I created a draft on the 4th isimmaterial. I have some drafts from 9 months ago. Heck, I have a half written rant about an architectural failure at Backblaze which is so out of date that it will likely never be published. The draft date is not important in the least.Fortunatelynobody looks at the URL to gather this information.

URLs are not meant to convey useful information to people, they’re there as instructions to computers. From time to time it is useful to have a friendly URL that people can remember but this is typically only for allowing them to type it in from a piece of paper. URLshorteningservices are fantastic for that.

Embedding information in the URL might look nice but it has very limited utility. In the .net world routing is provided by System.Web.Routing and this is typically configured in the startup of an MVC application. I have seen a number of MVC applications which have dozens, even hundreds of routes defined. Even Phil Haacked has contributed to the madness by providing a route debugger. Stop the madness! Use the default routes!

Having complex routes makes it very difficult to figure out which controller it is that is causing you trouble. It is also a boat load of extra code you need to test and understand. The fact that a route debugger is necessary should be a big hint that your logic is too complicated. The default route is sufficient for 99% of the requests coming to your site.

The only legitimate uses I can think of for routing requests are:

1. You need to do some optimization for search engines. Apparently google place some emphasis on the structure of a URL. It is difficult to tell because google keep pretty quiet about how they do page ranking. Optimizing for search engines is a prettysleazybusiness anyway so you’re likely better of not even trying.

2. Mapping old URLs. This is a really good call. It sucks when there is a change to the underlying engine behind a website and now none of your links work. By setting up a mapping you can avoid a lot of 404s in your error logs.

If these two reasons don’t match your use case then don’t even bother adding custom routes. It will save you headaches in the future.

It seems like I’m writing a lot of fizzbizz examples at the moment. It is kind of fun experimenting with different languages. There are always different constructs for looping and recursing. I’m also super happy that there are a lot of languages to get through before I have to write it in prolog. I have prettytraumaticmemories of prolog from university. Today’s language is shell script.

Of course shell script isn’t just one language, it depends on which shell you’re using. When I worked a lot with unix derivatives I mostly worked with bash scripting. Unless it was aparticularlyold or odd OS in which case we would end up on plain sh. There are a bunch of other shells out there and I can remember a time when both csh and ksh were also popular. I can also remember when druids sacrificed goats. Is there a link between ksh’s stupidarcanesyntax and goat slaughter? I can’t proveconclusivelythat there is but there are no dead goats here and no ksh syntax either. Draw your own conclusions.

I thought I would try using zsh style scripting for fizz bizz. Zsh is a newer shell which has many of the features of bash and also borrows from other shells. Now I say “newer” but it still dates to 1990.

The script starts with a sha-bang which instructs the program loader that to run the script it should execute /bin/zsh which is where zsh lives on my machine. It might be better to replace it with #!/usr/bin/env zsh which instructs the program loader to launch env which searches for zsh. This allows for searching of the path for zsh. There is an increased security risk with doing so as an alternate zsh might be selected. However this risk is probably worth it for increased portability.

I was hoping that it was possible to putmathematicalexpressions in the case statements but that’s not possible. Instead I took advantage of case statements here and the fact that we can take the remainder modulo 6 to do most of the fizz bizz heavy lifting. On line 3 there is a bit of a syntactic oddity. Zsh is not really designed for doing a lot of math so arithmetic operations need to live inside double parenthesis.

Gosh, I can’t wait for the next interview where I’m asked about fizz bizz. I am going to kill that question.

If you read this blog with any sort of frequency you’ll know that I’m somewhere in the range of 15-18% about typescript. In my travels the other day I created a new web project on a machine without typescript installed(go on, ask me how many computers I have at home). When I went to add typescript to the project it didn’t compile and I squinted my third hardest squint. Digging around on the web I found that one might need to add a target to the .csproj file. The examples I found all pointed to the typescript found in C:\program files.

I hate that.

Doing so means that you have to install typescript, in the default location, for the compile to work. That’s too much friction for setting up the project on a new machine. Instead I copied the contents of that directory into a tools directory in my project and checked it in. Now when people compile it they won’t even notice that it is building their typescript for them and the friction for a new developer is 0.

The target? Well just paste this into your .csproj file right before the

I put typescript in a tools/typescript directory at the same level as my .sln file. All good to go.

A while back I blogged about how I thought having a FizzBizz like problem on a job interview was a good idea. In that post I mentioned that coming up with novel ways to do fizz bizz should be fun for more senior developers. I though perhaps it might be fun to try out this F# language I’ve been hearing so much about as of late. So I poped over to the Try F# site where they have a nifty online F# interpretor. A bit of paying resulted in

F# is a functional language which has really good support for lists or arrays. Here I used the | operator which is a forward-chaining operator to pass the results from one function to the next. It is reminiscent of shell scripting (incidentally doing fizz bizz in shell scripting would also be fun). As part of F# there is a concept called a filter which acts in sort of the same way as a switch statement. It allows you to match the elements of a list and perform different actions depending on the match. That’s what you see on lines 3-6.

Being pretty new at this F# stuff I went hunting for other people’s solutions online. No better way to learn a language than to see how other people use it. That’s why github is so awesome. Well it is one of the reasons. I found a swell answer on StackOverflow inTomas Petricek“˜s answer. The gist of it seems to be that this is a functional language and we should be attempting to use it functionally. To do that we can declare an active pattern and slot that in instead of the individual modulo checks. This allows us to remove the where from the filter

So there is an F# version of FizzBizz.

There are two schools of thought behind how to construct methods. The first says that your method has one entry point so it should have one exit point. A method like that would look like

With this it is easy to see where the value is returned to the caller. Other than that there is not much good about the structure of the code. There are a lot of brackets and you have to read the whole function to figure out what it is doing. As I understand it this particular structure of code is a throwback to the days of structured coding and was used to avoid the dreaded GOTO. The other method of designing a method is to return early. If we rebuild that last method it would look like

Here we can see that there are a bunch of places which return. As we read through it is easy to see where we return. There is no need to read the entire function as we trace through it. At the top is a guard clause which will return right away if the parameters to the function are incorrect.

Now to be fair I took returning early to an extreme in this example. Everywhere I could retur early I did. If I were to come across this function in a code review it would be questioned. I would probably rewrite it to use a number of functions, something like

It is still a bit messy but that is probably more a function of a poor API for the imaginary client I created.

So the conclusion is that single entry-single entry is an outdated concept which makes code difficult to read. At the same time leaning heavily of return early createsmessycode too. Mixing the two and makingintelligentdecisions on a per method basis is the optimal strategy.

Irritatingly frequently I encounter a request which basically boils down to “I have this amazingly complicated excel spreadsheet I’m using to stop the company from going bankrupt, can you make yourapplicationproduce it?”. So I bravely dive into it and find 96 tabs and graphs and the what not. I am terrified that so much of the business world relies on spreadsheets of this sort but my nightmarish fears are not the subject of this blog. Well not today.

I make use of theabsolutelyfantastic excel library EPPlus. In today’s nightmare I was to create a databar. This is a special sort of conditional formatting which basically creates a bar inside a cell.

The length of the bar is calculated based on the range between the largest and smallest values in the range. Aren’t they pretty?

The latest version of EPPlus adds support for conditional formatting and you too can create these sorts of spreadsheets programatically.

Basically all you do is select the range you want and assign the conditional formatting of type databar to it. As far as I can tell there is no support yet for solid fill bars.

Ten years ago when I wrote JavaScript I only ever had one function and the function was always called validate. The only function for which I used JavaScript was validating for entry. Somewhere along the line JavaScript became awesome and it was possible to build large applications using JavaScript. I personally think the watershed moment was when gmail burst onto the scene. I understand that the first version of gmail waswrittenin Java using aframeworkcalled GWT which transcompiles to JavaScript. However, the point remains that it is possible to build complex applications using JavaScript.

In object oriented languages code is arranged into functions, these functions are grouped together into classes and the classes are placed into namespaces. This hierarchical approach allows for thearrangementof large quantities of code into a logical structure. For the most part this method of arranging code has gone unchallenged for decades(with the exception of AOP, but that’s a whole other post). JavaScript was never envisioned as a language for building large applications so it lacks first class support both classes and namespaces. Just as with Perl it is possible to simulate objects using arrays and even namespaces.Unfortunately,the syntax is prettyarcane.

Typescript to the rescue!

First classes. The syntax is pretty easy

This short snipped of code shows off a lot of different features of the classes. First we see a class level property in the description. Next there is a constructor fo the class. You’ll notice that the propertypublicationDate is prefaced by the keyword public. This is a great little shortcut which turnspublicationDate into a public property on the class. It is the same as if we had created a property like description then assigned a value to it in the constructor. Finally we see the use of the this keyword to denote that we are accessing a class level variable.

Now if we wanted to move this class into a namespace we would use the module directive

Modules can be described in any number of files and the JavaScript engine will combine them together for you.

The beauty part of TypeScript and other languages which compile to JavaScript is that if you decide to abandon the technology and switch back to JavaScript then the exit cost iseffectivelyzero. TypeScript produces this, very readable, JavaScript from our module and class