I will no longer be posting to #think.in and will now be posting to my personal blog I've started building at tarnbarford.net. I've been hinting at doing this for a while and I'm looking forward to being able to post richer content to a much simpler blog.

Much thanks goes to Andrew Brodie who did most of the work getting this site up and running and getting me started blogging. It has been fun. Thanks also to everyone who has stopped by to read my posts and an extra special thanks to those who took the time to leave a comment. I encourage you to come check out my new site and let me know what you think.

Peace out.

I've wanted to post about using Selenium and TeamCity to build and test .NET websites on Windows for a while. Unfortunatley it hasn't been forthcoming as I only really use Windows/.NET at work these days.

I decided I would get my shit together today and write, what may be my last post on .NET for a while, in response to @DavidBurela's Developer Blog Banter topic "How do you test your applications?"

I'm going skip unit-testing (which we do and try to get better at), BDD and JavaScript testing (which I would like to do and get better at) and focus on automated browser testing.

The Build Process

Watching automated test may be fun the first few runs, but it won't last long. Anyway if you want automated browsers tests, surely you want to automate the automated tests, right?

The build process is going to need to automate the browser and run the site the browser is browsing. For this we use IIS which is scriptable with the PowerShell WebAdministration module (maybe I will write another post on .NET).

The problem with having the UI Test step in this sequential process is it very quickly starts taking ages. This sucks; I want feedback as soon as possible. TeamCity provides feedback after each stage, but we don't get our green light.

This could be resolved by putting the UI tests in a nightly build, but I think a staged build or build pipeline is usually better.

The first stage compiles, runs unit tests and produces a deployable website as an artifact, this process is hopefully quite fast. If the first stage is successful the next stage in the pipeline is triggered which runs the less timely integration and UI tests. With a build file that has appropriate build targets setup, this build pipeline can be setup in the TeamCity administration web interface.

If you want to get awesome you can also start using Selenium Grid to distribute test agents or use a cloud based solution like SauceLabs

Tools

Our automated build process consists of a few tools.

TeamCity for our build management and CI server. After years of using CC.Net and a very brief and painful exposure to TFS, I love TeamCity. I also hear great thing about Hudson.

NUnit and included NUnit-Console. It does what I expect a unit-test framework and runner should do.

MSBuild with the MSBuild Community Tasks for our build scripts. I have to have some sort of build script to do continuous integration in .NET when not using TFS. I would like to be using a modern scripting language or DSL instead, but MsBuild is doing the job for now.

Selenium for our browsers automation tests. It has a recorder, a server which does the brower automation and can be controlled from by .NET client library.

Creating Tests

Selenium IDE is a Firefox plug-in for recording and running tests.

Selenium IDE is pretty slick and importantly it's a tool non-developer-testers-types can quickly get their head round and start producing some tests.

What we really want though, is to have our tests in code so we can run on them on build machine and version them in a repository.

Test Scripts

That should generate something like this.

This generates test files you can pretty much just add to your project as NUnit tests. This also means automation tests can be run by the nunit-console just like unit tests. The .NET client drivers are in the Selenium Core download.

It may be worth considering if exporting these tests to C# ideal. Keeping them in the standard html table based form means they can be reloaded into the IDE, and there are also alternate languages that you can export to.

Selenium Server

Selenium server runs on the JVM - Yep, suck it up. TeamCity does too :)

Any standard version of Java should be fine. If you've installed Java correctly it should be in your path and you should be able to type this in a console.

c:\>java -version

java version "1.6.0_18"
Java(TM) SE Runtime Environment (build 1.6.0_18-b07)
Java HotSpot(TM) Client VM (build 16.0-b13, mixed mode, sharing)

This should output the Java version and build information. If so you have successfully installed a java runtime. Yay!

You need to download the Selenium RC and extract the selenium-server.jar file.

Once you have Java and Selenium Server you should be able to start the server from the command-line.

java -jar selenium-server.jar

Selenium Server as a Service

I wanted to run SeleniumRC (the same Jar) as a service. Console windows on servers are not so cool. Another, probably better, option is to have the build script itself start and take down the server. I went with the service option.

I found a solution for running a .jar as a service, but it didn't work on 2k8. Luckily that lead me to the Non Sucking Service Manager which has a great name and worked fine.

The parameters I used were "-Xrs -jar [path/to/selenium-server.jar]"

Discussion

I went down the path of browsers automation because I was finding none of our other testing efforts, while valuable, were not actually testing the website would work.

Despite our best efforts we were still have problems where a page or an entire site would return a HTTP 500 Server Error for many reasons.

• Code problems a unit-test would ensure the fix.
• IOC or ORM/Database configuration.
• Compilation error at runtime on a JIT compiled view page.
• JavaScript Ajax.

Anyway it happened for all types of reasons. What we needed was an end-to-end integration test. Browser automation provides these tests, and it has helped.

I actually don't see the tests themselves as being particularly fragile. No doubt they can be, but I think it also encourages clean markup. If the hierarchy or attributes change then UI tests are not the only thing that will break. Stylesheets, scripts and some server logic also depend on it.

While I have seen good value from browser automation testing, I feel that mocking out the entire web framework (like NoseGAE for AppEngine) or testing at HTTP level (which I think is what WebRat does for Rails) can test almost the same thing without involving browser autmation at all.

First up a disclaimer, this is how I like to role. I enjoy it. I'm not saying you shouldn't use a mouse or I'll never use one again.

I spend a lot of time programming on my netbook development environment because I enjoy it. Without all the visual tooling I feel can focus on working with the machine, learning its many tongues and tricks. I've run ubuntu on it for about a year and have learned heaps about linux. There have been many languages, tools and skills I would not have otherwise been able to learn.

One thing I don't like about it is using the crappy mouse pad. It has long been a pain point, even doing trivial things. Recently I saw a tweet by @TheColonial listing some mostly unfamiliar names and adding #themouseisdead. This was exciting, I had recently adopted vim and was loving it.

Last weekend I was a bit under the weather and not thinking clearly enough to write code. So I used to time to set-up and learn a few new tools. I put together a powerful, mouse-less development environment on my tiny netbook, which I think is pretty fun to use.

xmonad

xmonad is a great window manager (~1000 lines of haskell) which seems to plug into most X systems. There are other ways of managing a windowed environment, I never new existed! It is all keyboard driven, it works out non-overlapping window layouts, you can move between multiple desktops and move windows between them.

Here are some typical tasks in a desktop with a few windows using xmonad:

• Switch between window layouts, generated by haskell algorithms (mod-space).
• Select between windows on screen (mod-tab).
• Throw the selected window to another desktop (shift-mod-2).
• Change to the desktop (mod-2).

On Ubuntu it installed with:

sudo apt-get install xmonad

Which installed its haskell dependencies and configured itself. It also added an xmonad session option on the login screen, I wanted to use this and keep the normal session intact.

When you start an xmonad session you are presented with a completely empty screen except for a background picture. You can start terminal session with some keys (mod-shift-enter), but a menu would be nice.

dmenu

dmenu is a generic dynamic menu for X systems, it allows menu items to be selected efficiently with a keyboard.

It installed with:

sudo apt-get install dwm-tools

This adds another session to the login screen, now anything can be opened from the xmonad session by bringing up dmenu (mod-p) and typing the first few letters of the desired application.

Vim

I'm surprised and a little annoyed I didn't start using it sooner really, it is a very powerful text editor. I still fumble through it but feel I'm learning to increase my productivity and reduced my frustration working with code.

sudo apt-get install vim

Now vim is available from the terminal. The vim-tutor is the best place to start, then there is a large ecosystem of plugins.

Vimperator

Vimperator is a Firefox extension that provides command from the keyboard, with vim idioms. It provides deep control of firefox, a basic browsing scenario might work like:

You can open a page in a new tab

:tabopen theage.com.au

Then to open a link to a story in a new tab

shift-f

This adds unique two digit numbers to all links. Now any link can be opened by typing the digits or the link text.

To go to the new tab:

gt

Once read the tab can be deleted

d

TTYtter

Anything typed that doesn't start with "/" is a tweet (the -verify and -slowpost options can help with potential problems that might cause) The forward slash is used to invoke commands like "whois", "replies", "reply", "dm", "follow", "thread" etc. It's built on curl, so we need that first.

apt-get install curl

Then download the perl script from the website, make it executable and move it to the /usr/bin folder.

I like everything about this app down to the ascii art in the menus. The dude has some pretty cool stuff including a network of gopher servers.

Vimium

Vimium is not as powerful as Vimperator, but provides some vim-like keyboard controls for Chrome. Easily installed as a chrome add-in.

Rock'n'roll time?

I can now manage windows over multiple virtual desktops. Use a fast, effective menu to opening applications. Have extensive control of firefox and most importantly, can tweet about it with TTYtter. I am excited and looking forward to working with some code.

My set-up still isn't right yet though; I need a status bar and decent start script (the plain xmonad session just a blank canvas, an internet connection, battery power monitor and a clock would be useful). I'm going to try dzen for this.

And my keyboard skills on the special keys is way worse than than I'd like it to be. The best way to fix that is to keep my hands on the keyboard, right?

Finally thanks to all the people responsible for the OSS platform, languages and many tools. They are a pleasure to use.

I've been enjoying spending some time playing with the Reactive Extensions for Javascript and wanted to write about it before I lost my netbook in bar. Sadly this blog makes it difficult present the content and examples together, so before fixing that, I put it here. 

I'm doing a presentation on non-relational databases at DDD Melbourne this weekend where I am going to demonstrate a map-reduce example with MongoDB and server side Javascript. I've been interested in both independently recently and it's been fun getting them to working together with some Javascript TDD to boot.

I needed a good example to demonstrate map-reduce and decided finding word occurrences across a series of documented seemed a simple enough scenario that is suited to being solved by a map-reduce query.

Below is an example of how we might solve this in plain C#

using System;
using System.Collections.Generic;

using System.Linq;
using System.Text;

class wordCounts {

    static void Main(string[] args) {

        // Setup some data

        List<string> lines = new List<string>() 
            { 
              "Peter Piper picked a peck of pickled peppers", 
              "A peck of pickled peppers Peter Piper picked",
              "If Peter Piper picked a peck of pickled peppers",
              "Where's the peck of pickled peppers Peter Piper picked?"

            };

        // select all words, group, count
        var wordCounts = lines.SelectMany(m => m.Split())
                              .GroupBy(m => m.ToLower())
                              .Select(m => new KeyValuePair<string, int>(m.Key, m.Count()));

        // Print out the results

        foreach (var wordCount in wordCounts) 
        {
            Console.WriteLine(string.Format("{0} {1}", wordCount.Value, wordCount.Key));     
        }
    }
}

This prints each different word in all the lines and the number of the times it occurs. The collection of strings is isomorphic to a collection of documents in the MongoDB for this example.

The SelectMany flattens lists of words from each line to a single list of words and the Group provides keys for each word, this is very similar to what the map function in the map-reduce query does.

The Select function is similar to the reduce function, but as we will see some additional considerations need to be made to allow it to be distributed.

I saw a good diagram ayande published on his blog but I didn't understand why he had multiple instance of the same document being mapped.

I created my own low key diagram to help demonstrate how a functional map-reduce could be distributed. The diagram shows the initial items can be split in half and reduced completely independently. This is interesting as it means our query can be distributed, but it also means we have to handle reducing a little differently.

It's also worth noting that this example shows a balanced tree, but it could be unbalanced and even introduce some redundancy.

MongoDB allows clients to send JavaScript map and reduce functions that will get eval'd and run on the server. Here is the map function.

function wordMap() {

    // try find words in document text
    var words = this.text.match(/\w+/g);

    if (words === null) { 
        return;
    }

    // loop every word in the document 

    for (var i = 0; i < words.length; i++) {
        // emit every word, with count of one

        emit(words[i], { count : 1 });
    }

}

The misunderstood Javascript "this" will be the context from which the function is called. Mongo will call function each document in the collection we are querying, and we can call it from a test context. Unlike the SelectMany the map function doesn't return a list, instead it calls an emit function which it expects to be defined.

We can write unit tests for this function by calling the function from a test mock context, calling a mock emit function (using Javascript as our mocking framework, wow).

eval(loadFile("src/js/wordMap.js"));


var emit;
var results;
var context;

testCases(test,

    function setUp() {
        emit = function (key, value) { 
            results.push({ key : key, value : value });
        };
        context = { text : "", map : wordMap };
        results = []; 
    },

    function empty_string_emits_nothing() {
        context.text = "";
        context.map();
        assert.that(results.length, eq(0));
    },

    function single_word_emits_single_word() {
        context.text = "findme";
        context.map();
        assert.that(results.length, eq(1));
        assert.that(results[0].key, eq("findme"));
        assert.that(results[0].value.count, eq(1));
    },

    function two_different_words_emits_twice() {
        context.text = "for bar";
        context.map();
        assert.that(results.length, eq(2));
    },

    function two_same_words_emits_twice() {
        context.text = "test test";
        context.map();
        assert.that(results.length, eq(2));
    },

    function tearDown() {
    }
);

The reduce function must reduce a list of a chosen type to a single value of that same type; it must be transitive so it doesn't matter how the mapped items are grouped.

function wordReduce(key, values) {
        var total = 0;
        for (var i = 0; i < values.length; i++) {
            total += values[i].count;
        }
        return { count : total };
    }

Similarly we can test this method does exactly what we expect it to.

eval(loadFile("src/js/wordReduce.js"));

testCases(test,

    function reduce_one_items_returns_count_of_one() {
        var result = wordReduce("test", [{ count : 1 }]);
        assert.that(result.count, eq(1));
    },

    function reduce_multiple_items_returns_item_count() {
        var result = wordReduce("test", [{ count : 1 }, { count : 1 }, { count : 1 }]);
        assert.that(result.count, eq(3));
    },

    function reduce_sums_counts() {
        var result = wordReduce("test", [{ count : 2 }, { count : 3 }]);
        assert.that(result.count, eq(5));
    },

    function reduce_is_transitive() {
        var result = wordReduce("test", [{ count : 1 }].concat(
                        wordReduce("test", [{ count : 1 }, { count : 1 }]
                     ));
        assert.that(result.count, eq(3));
    }
);

I'm using Rhino to run the Javascript so I used RhinoUnit as a test runner as it also uses the JVM and runs as an ANT scriptdef task, the setup was pretty painless. Here are the relevant ANT script sections

<scriptdef name="rhinounit"

           src="lib/rhinoUnitAnt.js"
           language="javascript">
    <attribute name="options"/>
    <attribute name="ignoredglobalvars"/>

    <attribute name="haltOnFirstFailure"/>
    <attribute name="rhinoUnitUtilPath"/>
    <element name="fileset" type="fileset"/>

</scriptdef>

<target name="javascript-tests">
    <rhinounit options="{verbose:true, stackTrace:true}" 
               haltOnFirstFailure="false" 
               rhinoUnitUtilPath="lib/rhinoUnitUtil.js">

        <fileset dir="test">
            <include name="*.js"/>
        </fileset>
    </rhinounit>

</target>

The word count example recreated in Mongo using a Python client and passing the map/reduce functions to the server.

from pymongo import Connection;
from pymongo.code import Code;


# open connection and connect to 'ddd' database
connection = Connection()
db = connection.ddd

# remove any existing data
db.drop_collection("messages")


# insert some data
lines = open("data/peter_piper.txt").readlines();

for line in lines:
    db.messages.insert( { "text" : line } )


# load map and reduce functions
map = Code(open("src/js/wordMap.js","r").read())
reduce = Code(open("src/js/wordReduce.js","r").read())


# run the map-reduce query
result = db.messages.map_reduce(map, reduce)

# print the results    
for doc in result.find():
    print doc["value"]["count"],doc["_id"]

And it worked! I'd like to run the query on a larger result-set, but there isn't much point on this tiny low-spec'd netbook.

The slides and demo I'll show in my NoSql presentation for tomorrows DevEvenings Melbourne ORM Smackdown. I hope to take the time and write a more considered post of my findings and opinions as I've found it very interesting.

Here is a link to the slides.

First I'll use some Python and the PyMongo module to connect to MongoDB, list databases, insert documents and get them out again.

Python 2.6.4 (r264:75706, Dec  7 2009, 18:45:15) 
[GCC 4.4.1] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> from pymongo import Connection
>>> connection = Connection()
>>> connection.database_names()
[u'files', u'working', u'demo', u'downloads', u'posts', u'local', u'admin']
>>> db = connection.demo
>>> import datetime
>>> db.messages.insert( { 'author' : 'tarn', 'date': datetime.now(), 'message' : 'Hello Mongo' } )
ObjectId('4bbc4beec73d721445000003')
>>> db.messages.find_one()
{u'date': datetime.datetime(2010, 4, 7, 19, 10, 6, 355000), u'message': u'Hello Mongo', 
u'_id': ObjectId('4bbc4beec73d721445000003'), u'author': u'tarn'}

Then have a look at some content I have in the database

>>> db = connection.working
>>> db.posts.count()
106
>>> for post in db.posts.find()[:5]:
...     print post["date"],post["title"],"by",post["author"]
... 
2010-01-25 18:06:00 Python Silverlight/Moonlight 2 Xapping by tarn
2010-03-12 13:53:00 Devevenings Presentation - IOC/Unit Testing/Mocking in ASP.NET MVC by tarn
2010-02-17 19:25:00 Revisiting Modal Binding an Interface, now with DictionaryAdapterFactory by tarn
2009-12-02 18:35:00 Creating Silverlight apps in the browser by tarn
2009-10-02 23:08:00 #.think.in infoDose #43 (11th September - 22nd September) by brodie

File storage using the GridFS class from the gridfs module. Show some files and then write a file out to the file system.

>>> from gridfs import GridFS
>>> fs = GridFS(connection.files)
>>> len(fs.list())
116
>>> for file in fs.list()[:5]:
...     print file
... 
post/debugging-ironpython-with-my-excalibur/image.png
post/debugging-ironpython-with-my-excalibur/image_thumb.png
post/devevenings-presentation---iocunit-testingmocking-in-asp.net-mvc/20102f32fdevevening_presentation.pptx
post/devevenings-presentation---iocunit-testingmocking-in-asp.net-mvc/20102f32fguestbook.zip
post/think.in-infodose-40-5th-august---16th-august/image.png
>>>
>>> with open('image.png','w') as out_file:
...     with fs.open('post/debugging-ironpython-with-my-excalibur/image.png') as in_file:
...             out_file.write(in_file.read())
...

Now to some C# (mono) and a controller class for a basic web application to view the data. It serves files found in the database, but only sends bach the correct MIME type for "image/png". Lazy.

public class HomeController : Controller
{
    BlogRepository _blogRepository;

    public HomeController()
    {
        _blogRepository = new BlogRepository(); 
    }

    public ActionResult Index ()
    {
        ViewData["posts"] = _blogRepository.GetPosts();
        return View ();
    }

    public ActionResult Entry(string id)
    {
        ViewData["post"] = _blogRepository.GetById(id);
        return View ();
    }

    public ActionResult Resource(string slug, string fileName)
    {
        return new FileStreamResult(_blogRepository.GetFile( "post/" + slug + "/" + fileName ), "image/png");
    }

}

This is a very basic repository that provides the data for the demo application. I did only enough with the C# provider to get it working and try to disconnect my connections.

public class BlogRepository
{
    Mongo _mongo;

    public BlogRepository()
    {
        string connstr = ConfigurationManager.AppSettings["connectionString"];
        _mongo = new Mongo(connstr);
    }

    public Stream GetFile(string name)
    {
        try
        {
            _mongo.Connect();
            var db = _mongo["files"];
            var fs = new GridFile(db);
            Stream data = fs.Open(name, FileMode.Open, FileAccess.Read);
            Stream output = new MemoryStream();
            CopyStream(data,output);
            output.Seek(0,SeekOrigin.Begin);
            return output;
        }
        finally
        {
            _mongo.Disconnect();    
        }
    }

    public List<Document> GetPosts()
    {
        try
        {
            _mongo.Connect(); 
            var db = _mongo["working"];
            var posts = db["posts"];
            using(ICursor all = posts.Find(new Document())){
                return all.Documents.ToList();
            }
        }
        finally
        {
            _mongo.Disconnect();
        }
    }

    public Document GetById(string id)
    {   
        try
        {
            _mongo.Connect();
            var db = _mongo["working"];
            var posts = db["posts"];
            Document doc = posts.FindOne( new Document() {{ "_id" , new Oid(id) }} );
            return doc;
        }
        finally
        {
            _mongo.Disconnect();    
        }
    }

    public static void CopyStream(Stream input, Stream output)
    {
        byte[] buffer = new byte[32768];
        while (true)
        {
            int read = input.Read (buffer, 0, buffer.Length);
            if (read <= 0)
                return;
            output.Write (buffer, 0, read);
        }
    }
}

So that's where I got for my demo for the DevEvenings ORM Smackdown. No doubt I will continue looking into MongoDB and other object/document databases.