Smiling Smith

Silverlight, Library Caching and blank screens

2012-01-10T09:45:00.000Z

Here's the scenario: You have a Silverlight 4 application with a splash screen. You business is growing and so is your Silverlight application's xap size. What can you do about it? Wouldn't it be nice if you could somehow split the xap into smaller and reusable files? This is exactly what the "Library Caching" feature is supposed to do: your reusable libraries are packaged into zip files and stored in your browser's cache, reducing the initial download bandwidth.

Theoretically all works well, of course. Until your customers start complaining about blank screens after the xap is downloaded on IE9. To get the application to start, they have to press F5 once (sometimes more). A brief internet search shows that this is a known and reported issue, but as you can see it's solved because Microsoft cannot reproduce it.

So now I have to go back and recompile the xap as a monolith to avoid this nice little feature that only "works" on IE9. How dumb is this?

The skewed accelerometer

2011-02-12T22:24:00.000Z

What can you do if your WP7 accelerometer is not correctly calibrated? As I found out, there seems to be no OS mechanism to calibrate the accelerometer. So if you have a production device you will have to return it for repair or replacement. But what do you do if you have a loaner preproduction device like I do? Well, you have to fix the accelerometer the only way you can: through software.

A correctly calibrated accelerometer should return the vector (0, 0, -1) when you place your device with the screen up on a flat surface (like your office floor). If you have a skewed accelerometer like I do, then you should be able to compensate the difference to the reference vector, once you know what your device is reporting when placed on the floor with the screen up.

To do just that, I wrote a very simple Silverlight application for WP7 that polls the accelerometer every second and displays the vector components, both the sampled value and an average value in order to smooth out the natural variations in the sampling.

You can get the sample application with source code here. After running this app on the floor, I noticed that the Y component of the accelerometer vector was slightly skewed to -0.15, while all other components reported the expected values. After adding 0.15 to the Y component, I got back a calibrated accelerometer.

Now I can get back to my XNA gravity ball code and make it work like it should.

MediaElement and downloaded sound

2011-02-11T14:12:00.000Z

In a nutshell: when you download a sound file to play on a MediaElement, make sure you set the AutoPlay property to True if you call SetSource with the download stream. This stream does not support seeks, so it must play immediately. If you want to store the sound to play it later, store it in a MemoryStream (it supports seeks). I found this out the hard way...

The Tabukiniberu bug

2011-02-08T22:49:00.000Z

This was first presented to me as the "Marshall islands bug": our Silverlight application was opening the Google map in the middle of the Pacific, where it should be showing the mid-Atlantic. Looking closely to the map, I found this extraordinary name: Tabukiniberu, an island in the Pacific near the Equator (the Marshall islands are farther to the north). So what on Earth (literally) was my code doing to spin half the globe and show this (hopefully) beautiful island?

The SL4 app I've been working for in the last 6 months uses both Bing and Google maps AJAX controls due to the different quality of the map content. Google maps are way better on some regions where our business is growing, like Africa and Eastern Europe. On the code side, Bing seems to be better.

When there is nothing to display on the map, the application usually centers on point (0, 0) with a zoom level of 2, providing the "classic" (for my European patterns) wold view with Europe on the top right and the Americas to the left. When there is something to display, the code pans and zooms in to the area of interest so that all pushpins are shown. This particular bug occurs in Google maps code when you call fitBounds on an empty LatLngBounds variable (I mean, initialized but without any added bounds). Instead of showing the mid-Atlantic, you are thrown on a 180 degree spin around the globe to a place very close to Tabukiniberu. Well, at least on the map...

Back to mobile!

2011-01-31T09:43:00.000Z

I'm back to writing mobile code. After seeing renewed my MVP status, I was able to get my hands on a loaner developer Windows Phone 7 device. I have to personally thank Cristina Gonzales Herrero from Microsoft Spain for both (MVP and the phone).

Having a WP7 phone on my hands was just the excuse I needed to get back to mobile development. The first hurdle I had to overcome was the App Hub registration. My first attempt was to use my own company account to enroll but the registration process requires two people from the company. So what happens to companies of one like mine? Well, I turned back and took the individual path only to be greeted by payment form (yes, you have to pay USD $99 to deploy to the phone), that assumed I was a US citizen. After talking to some friends, I understood that the only way to go about this was to create a live.com.pt (I live in Portugal) account, but, before I could do that, I had to change my language preferences on IE. You see, I just cannot use a development machine with a Portuguese OS. The keyboard is OK, but not the OS "translations". I started developing on computers that could hardly "speak Portuguese" so I got used to the English language for development, and this also includes the language preferences on the browser. Incidentally, if you don't have Portuguese as your first browser language selection, you cannot create a live.com.pt account. Sigh! This should me much easier, Microsoft!

After going through all these hurdles, I managed to get my developer account and wrote my first mini-app: an XNA application that explores the device's accelerometer. It merely uses the accelerometer information to move a ball around the screen as you tilt the device (something that you have already seen). The accelerometer input in not treated in any special way so the movement you get is sluggish, just as if the ball is moving through molasses...

You can get the first sample from here.

The next version of the code will include more realistic ball movement, so there is some kinetics a dynamics stuff to review.

To help me in writing this tiny piece of code, I read the following articles:

Using the Accelerometer on Windows Phone 7

Orientation and rotation on Windows Phone

Persisting Silverlight DataGrid sort descriptions

2011-01-19T09:27:00.000Z

I came across this issue recently: how to persist Silverlight DataGrid sort descriptions between changes of the ItemsSource property? The DataGrid on version 4.0 does not expose a direct interface to manipulate the column sort descriptions although it somehow manages them internally. If you want to explicitly manage the column sort descriptions, you must use a PagedCollectionView container, through the SortDescriptions collection.

If you feed a PagedCollectionView to a DataGrid and then let the user change the column sort descriptions through the user interface (by clicking the column headers), the SortDescriptions collection will store the user changes. After realizing this, I devised a very simple strategy to persist the column sort descriptions whenever I needed to refresh the DataGrid:

If the data source is a PagedViewCollection, I just make a copy of the old SortDescriptions collection and apply it to the new PagedViewCollection object.
If the data source is just an IEnumerable, I create a new PagedViewCollection to wrap it and do the same as the previous point.

To make my life a bit easier, I created a very simple helper class with two private members:

private DataGrid dataGrid = null;
private PagedCollectionView pagedView = null;

The constructor simply stores the target DataGrid object reference:

public PersistDataGridSort(DataGrid dataGrid)

{

this.dataGrid = dataGrid;

}

The bulk of the work is done in the SetItemsSource function that takes an object as parameter. This must be either an IEnumerable or a PagedCollectionView:

public void SetItemsSource(object collection)

{

SortDescriptionCollection sortDescriptions = null;

if(collection is PagedCollectionView)

{

if(pagedView != null)

{

sortDescriptions = new SortDescriptionCollection();

foreach(SortDescription sortDescr in pagedView.SortDescriptions)

{

sortDescriptions.Add(sortDescr);

}

pagedView = collection as PagedCollectionView;

pagedView.SortDescriptions.Clear();

}

else if(collection is IEnumerable)

{

IEnumerable enumerable = collection as IEnumerable;

if(pagedView != null)

sortDescriptions = pagedView.SortDescriptions;

pagedView = new PagedCollectionView(enumerable);

}

else

throw new ArgumentException("Collection must be either a PagedCollectionView or an IEnumerable.");

// Set the data grid's source

dataGrid.ItemsSource = pagedView;

// Reinstate the old sort descriptions

if(sortDescriptions != null)

{

foreach(SortDescription sortDescr in sortDescriptions)

pagedView.SortDescriptions.Add(sortDescr);

}

To use this class, just create an instance using the target DataGrid and set its data source by calling this class' SetItemsSource method. Have fun!

FireFox, WCF callbacks and HtmlPage.Window.Alert

2010-11-30T13:52:00.000Z

This seems to be a deadly mix: calling HtmlPage.Window.Alert on a WCF web service callback on FireFox. While you get the correct behavior in IE and Chrome, FireFox just hangs miserably. Beware...

Templated Silverlight ListBox

2010-11-21T19:15:00.001Z

Whenever you use a Silverlight ListBox element with templated items, it is a good idea to force the item content to stretch horizontally. This way each item’s content will occupy the full width of the ListBox, leaving no “dead” areas. I found this to be particularly useful when handling mouse event on the item container because on shorter items you would get an area to the right that is blind to the mouse events. The solution is quite simple (I found it elsewhere on the net, but post it here because I find it so useful):

<ListBox.ItemContainerStyle>
    <Style TargetType="ListBoxItem">
        <Setter Property="HorizontalContentAlignment" Value="Stretch" />
    </Style>
</ListBox.ItemContainerStyle>

Happy silverlighting!

From DataTable to DataContract

2010-11-07T19:34:00.001Z

For the last four months, I have been very busy porting a large SaaS application from the traditional ASP .NET + Ajax platform to Silverlight 4. This application has a quite standard architecture with a data access layer that talks to the database and exposes a simplified interface to the business level. When a piece of data is required, you just need to call a given method with the appropriate parameters and you usually get back a DataTable object with all the data you need. This is a very flexible scenario because you can directly bind to an ASP .NET control or, if you need to further massage the data, transform the data in memory before binding it.

This architecture became a bit of a problem when I started to port the application to Silverlight because it just does not understand ADO .NET, so it is useless to try to marshal a DataTable object through a WCF web service. Because we needed to keep the "old" ASP .NET application running, there was no point in rewriting the old DAL code so a decision was made to keep it and reuse it as a data source for the new WCF web services. This meant writing a lot of boring code to implement the new DataContract classes. When you create a Silverlight-enabled WCF web service, you must send and receive all your data through strongly-typed classes that are very neatly marshalled over between the two endpoints.

I tackled this issue by starting to write all the new DataContract code by hand... until I got bored. The process is incredibly boring, time-consuming and error-prone. After a few weeks at it, I found out that writing all the WCF plumbing was taking a large portion of every ported feature, so this needed a closer look. What could I do to help me bridge the gap between the existing, tried and true DataTables and the new DataContracts?

A code generator.

The idea is quite simple: plug in a piece of code that takes either the freshly-created DataTable (or DataView) and generates a text file containing C# code that declares two DataContracts: one for each row and another for the whole list of rows. Oddly, this code must run on your web server, so it is best suited for your development environment only. To make my life easy, I created a single extension method to the DataTable and DataView classes that you use like this:

dataTable.WriteListDataContract("SingleRowClassName", @"C:\Path", "ID");

The first parameter is the row class name: the extension will automatically generate the corresponding list class name by appending the "List" string to this class name. The second parameter contains the path of the generated file name (based on the first parameter). The last parameter is an optional string containing a comma seperated list of columns that you want to force as non-nullable (the code will throw an exception when trying to convert a DBNull.Value).

Once you have the generated code files, you can comment out the line.

Using the generated code is very straightforward: just paste it in your WCF source code file. Reuse your old DAL code and, when you have the DataTable or DataView ready, create a list class and call its Load method. The generated code reads in the data and converts it into a generic list that you can immediately return to your Silverlight application.

Here's the source code: DataTableExtensions.cs

Google Maps V3 tip: fitBounds

2010-10-28T10:17:00.003+01:00

Never call the fitBounds method on your Google Maps V3 map when the supporting div is hidden. This will result in an incorrectly calculated zoom level because (apparently) the JavaScript code cannot determine the div dimensions. Always make sure the div is visible before calling fitBounds.

A ReadyBoost tip

2010-10-25T13:43:00.002+01:00

I own a relatively old laptop (a once-glorious ASUS Lamborghini VX2) which is limited to 2 GB of RAM. The box is now showing its age, especially when using VS 2010. (Not to mention the lack of a WDDM 1.1 driver for the NVidia Ge Force GO 7700, which makes developing for WP7 a horrible experience). To make things run smoother (or at least to try to make them run smoother), I removed all the “Visual experience” checkboxes in VS 2010 (under Tools / Options) and bought an 8 GB USB flash drive for ReadyBoost.
With the standard FAT formatted file system, ReadyBoost is limited to a 4 GB cache. It does make things a little better but barely worth the effort. Recently I read an MSDN article on this subject that stated that this limitation goes up to 32 GB if you format the flash drive using NTFS. After doing so, I got a 7.7 GB ReadyBoost cache (not the full 8 GB because the manufacturers measure capacity in a different but interesting way) and a noticeable improvement for my old box. Now VS 2010 does load a bit faster and the whole thing does feel a bit more agile.
No luck with the NVidia driver, though…

Silverlight 4, FireFox and Accented Characters

2010-10-06T18:17:00.000+01:00

My mother gave me a very common portuguese name: João. You can translate it to 'John', 'Jean', 'Juan' (take your pick). The odd thing about my first name are the three consecutive vowels and the tilde character on top of the 'a'. I get all sorts of funny jokes from my anglo-saxon friends due to this personal oddity. Why is this relevant?

I have been working with Silverlight 4 for a web application (I'm essentally rewriting a big HTML SaaS application) and met a very strange behavior with Silverlight 4 when running on FireFox (and Chrome): I just could not write my name on a TextBox! This is a situation that has been reported by some users but, interestingly, it's not a very common complaint. Of course IE works OK with this.

When searching for a solution for this problem, I found a sample where this issue did not happen. After a bit of research I found the culprit: the "windowless" parameter on the HTML object element (my code had it while the sample did not). If you set this parameter to "true", FireFox will not compose the '~' and the 'a' to produce the dreaded 'ã', so beware!

This issue is irrelevant for english-speaking (and writing) users.

Windows Phone Developer Tools Beta

2010-07-12T21:03:00.000+01:00

The Windows Phone Developer Tools Beta are now available for download from the Microsoft Download Center.

SQL Compact 4.0 CTP1 Released

2010-07-07T14:08:00.001+01:00

Microsoft has just released SQL Compact 4.0 CTP1. You can download it from the Microsoft download center now.

And now for something completely different...

2010-07-03T15:52:00.000+01:00

After long months of having my laptop running way too hot (sometimes it would shut down to prevent something from melting inside), I decided to open it up and use the household vacuum cleaner around the cooler fan, air ducts and vents. The result? Around 10 degrees (C) cooler...

Enumerating SQL Compact Tables, Columns, Constraints and Indexes

2010-06-14T21:36:00.002+01:00

Now that you know how to enumerate tables and columns, it's fairly easy to enumerate all the other schema elements: indexes, primary keys, unique constraints and foreign keys. Interstingly both primary keys and unique constraints are implemented as indexes and you list them all through the INFORMATION_SCHEMA.INDEXES view (I'm omitting the prefix from now on). This view will tell you immediately if an index is the table's primary key, but will not tell you if you are looking at an unique constraint. To check that, you must take a look at the TABLE_CONSTRAINTS view.

Foreign keys are a bit different: you must know what table they refer to (through a primary key, unique index or unique constraint), what columns are mapped and what are the update and delete rules. You get some of this information from the REFERENTIAL_CONSTRAINTS view, but you must complement it using the KEY_COLUMN_USAGE view.

You can see this mechanism at work on the sample code's updated SqlCeTable.LoadSchema method. Note that I kept the piece-wise loading mechanism in order to make things work faster. Alternatively you could devise an alternative schema loading mechanism using a DataSet to store all the data and then query it appropriately. Unfortunately this will imply a performance penalty if you port this code to a mobile device.

Now that we have a simple way to load and query the database schema, we can widen our horizons a bit and start extracting more useful information like scripts. Stay tuned for the next episodes...

Sample code: SchemaTree2.zip

Windows Phone 7 Development Videos from TechEd 2010

2010-06-11T12:06:00.001+01:00

Here are a couple of videos from TechEd 2010 North America on how to develop Silverlight applications for Windows Phone 7:
An In-Depth View of Building Applications for Windows Phone 7 with Microsoft Silverlight (Part 1)
An In-Depth View of Building Applications for Windows Phone 7 with Microsoft Silverlight (Part 2)

You can follow the full Windows Phone track here.

Enumerating SQL Compact Tables and Columns

2010-06-10T11:47:00.000+01:00

Today I'm starting a small series of blog posts where I will show how you can enumerate a SQL Compact database schema using the INFORMATION_SCHEMA views. Let's start with the basic stuff: tables and columns.

Table enumeration is performed by querying the TABLES view, most specifically the TABLE_NAME column. I implemented the new SqlCeSchema class (see sample code) that you can use for this purpose. Instances of this class are created with a reference to a SqlCeConnection object and you populate the Tables collection by calling the Load method (make sure the connection is open). Note that this operation merely loads the table names and creates empty SqlCeTable objects. To retrieve the column information for each table, you must call the SqlCeTable.LoadSchema method.

I designed the code this way in order to avoid long loading times for large databases. This way you can load only the schema for a particular table without the need to load the entire database schema.

You can see this at work in the sample code: it displays a simple tree containing the list of tables in a given database and the column enumeration is requested per table when you expand the table node.

Right now the code is limited to table and some of the column schema. I will be updating this sample in future posts to include more column schama details, indexes and constraints.

Sample code: SchemaTree1.zip

SQL Compact Table Export

2010-05-31T22:48:00.001+01:00

This is a very simple variation on my last blog post: instead of copying the table to the same database, you can now copy it to another database (although with the same table name, but you can easily change that). The table data is exported in the Export method of the SqlCeTableTool class using a SqlCeDataReader for reading and a SqlCeResultSet for writing. Simple stuff, really...

Sample code: CopyTableSample2.zip

SQL Compact Table Copy

2010-05-30T13:08:00.000+01:00

How do you make a copy of a SQL Compact table? Version 3.5 does not support the SELECT INTO command that allows you to easily do just that. Instead, it provides the INSERT INTO SELECT FROM command that allows you to copy existing table data into another existing table. So we are only missing one part: how to create a replica of the existing table structure.

If you want to stick to managed code only, you can read all the database structure information from the INFORMATION_SCHEMA views. To get information ot of these views, you just have to issue a SELECT command. For instance:

SELECT * FROM INFORMATION_SCHEMA.TABLES

returns all the existing tables in the database. In our case, we need to get information from an existing table and build a CREATE TABLE command out of it in order to create a replica of the original table. For this purpose, you can issue the following SQL command:

SELECT * FROM INFORMATION_SCHEMA.COLUMNS WHERE TABLE_NAME='TableName' ORDER BY ORDINAL_POSITION

With a little work you can read from this data and build a CREATE TABLE command. If you look at the returned data, you will see that you have all the information you need. In fact, you just need to set up a StringBuilder in your code, and start adding the columns and their types. You can fin this in the sample code (see below), although the code still does not support the numeric data type.

After creating the table, you can now copy the data, making sure that you execute a SET IDENTITY_INSERT command on the target table if it has an identity column.

The sample code has a very simple UI that allows you to specify a connection string, the original and target table names and a button to perform the copy.

Sample code: CopyTableSample.zip

The Remote HTC G Sensor Client Application

2010-05-18T09:41:00.000+01:00

Without further ado, here's the final version of the desktop HTC G Sensor client. This very simple .NET 3.5 forms application reads the stream of data from the HTC G Sensor via RAPI and displays it on the screen. Obviously, you can adapt this to meet your particular needs.

The main project in on the HTCGConsumer folder. Before trying to use this sample application, please make sure that the HTCClient.dll file is on your output directory, otherwise you will get an exception. This is the desktop client DLL that connects to the device via RAPI and is consumed by the HTC.GSensor.dll assembly through the HTC.GSensor.RemoteClient class.

Finally, don't forget to remove the RAPI restrictions on your device if you are running a WM6.x device. Enjoy.

RAPI2 article on CodeProject

2010-05-14T14:45:00.001+01:00

Thanks to Alberto's head-up, here's an interesting article on RAPI and its lesser-known COM interface: RAPI2: The Friend You Never Knew You Had

Flashback

2010-05-09T18:48:00.001+01:00

"I'll buy it if it works on my Pocket PC with a MIPS CPU..."

This was the rough message that a potential customer of my products wrote in a recent email. This is not unusual - some of my customers use the products because they still work with SQL CE 2.0. Apparently for business customers, a ten year old technology seems quite fitting for their daily work.

The device component that ships with my products goes all the way back to Pocket PC 2002, when Microsoft decided to support only ARM-based CPUs and drop support for MIPS and other brands. I remember that this was a bit of a nuisance back then because there were some non-ARM devices on the market that suddenly lost the ability to upgrade to version 2002 (something similar is happening now, isn't it?).

What this all means is that this particular person wants me to build the device DLL using the Pocket PC 2000 SDK. To my big disgrace, I found that my VMs were not properly configured so I set out to install a fresh XP VM with the Embedded Visual Tools 3.0 (2002 edition) plus the Pocket PC 2000 SDK. Fortunately, I kept copies of these museum-grade pieces of software because you cannot find them anymore from official Microsoft sources (namely MSDN).

So I started to install the thing and was prompted to enter the product key. Where did I save it? Damn... Nowhere to be found! I do have the eVC4.0 one, but not the one for version 3.0. Or do I? Actually, if you are going down memory lane like I did this afternoon, remember that the key for eVC3.0 is the same as for eVC4.0 (someone told me long ago that the product would validate any key, but I did not try that). So look for a valid download souce for eVC4.0 (like CNET) and use the key they provide.

Now that all is set up, let's enter the time machine...

The HTC G Sensor RAPI Client

2010-05-08T18:02:00.000+01:00

Now that we have a device RAPI extension DLL to capture the HTC G Sensor data it's time to write the PC client DLL. I wrote this as a native code DLL because it made my life simpler when interfacing with RAPI. It exposes five C-style functions that can be readily P/Invoked from .NET:

HTCGSensor_New - Creates a new native client object
HTCGSensor_Delete - Destroys the native client object
HTCGSensor_Open - Opens the remote HTC G Sensor server
HTCGSensor_Close - Closes the remote HTC G Sensor server
HTCGSensor_Read - Reads the HTC G Sensor data

The last three should be familiar to you right now, but why do we need the first two? Believe me that I learned this one quite painfully: when exposing native objects to managed code, always provide their memory management functions. The reason is quite simple: if you are allocating the native objects from the native C++ heap, you must delete them using the C++ memory management library. That's why you see a "new" and a "delete".

So what do these functions look like? They are actually very simple:

HTCGCLIENT_API void* HTCGSensor_New()
{
    HTCGSensorClient* pClient = new HTCGSensorClient();
    return pClient;
}

Why the void* return type? These exported functions are meant to be used by the P/Invoke mechanism that should know nothing about the pointer type: it will be mapped to an IntPtr type and treated as an opaque data type. Likewise, there's no need to check the data type on input:

HTCGCLIENT_API int HTCGSensor_Open(void* pClient)
{
    HTCGSensorClient* client = (HTCGSensorClient*)pClient;
    if(client)
        return client->Open();
    return E_FAIL;
}

So these functions are there to act as conduits between the a C++ object of type HTCGSensorClient, where the bulk of the work is done, and the calling .NET assembly. It should come as no surprise that this class also implements the Open, Close and Read methods, besides the standard constructor and destructor. Things get a bit more interesting here because we must open a connection to the device server through RAPI (the sample code uses an updated version of the CRemoteAPI class that I published ages ago). In a nutshell, the Open function (see the sample code) initializes RAPI and remotely invokes the HTCGSensorStream function on the HTCGServer.dll file (by default it should be placed on the device's \Windows folder, but you might want to change this). Note that a reference to a client-side version of the IRAPIStream object is kept and that's how the Read function gets the G sensor data. This function is implemented in terms of a write and a read operation. The write tells the server to fetch the G sensor data and then waits on the read for the data to arrive (that's why this might prove slow for high-frequency sampling: for each sample there must be a round-trip to the server). The data is copied to an HTCGSensorData variable passed in by pointer reference. Closing the client implies writing a command to the server and closing the RAPI connection.

On my next post I will wrap this small project up by publishing the high-level .NET assemblies that consume the HTC G Sensor data.

Sample code: HTCGClient.zip

The HTC G Sensor RAPI Server

2010-05-04T23:00:00.000+01:00

RAPI is one of the most useful features in the Windows CE OS because it allows you to remotely access your device while connected via ActiveSync or WMDC. It's like having a generic socket server on your device that is always listening for a connection from the connected PC and that, besides implementing a nuber of very useful APIs (Registry and File System to name a few) it also allows you to extend it by using special DLLs. Unfortunately this feature also is a security issue because it allows any connected PC to scan your device's file for instance. This issue was solved with the Windows Mobile 5 version of the OS by enforcing security policies that vary in degree of strictness. For Smatphones, the security policy has always been much stricter than for "Pocket PCs" (the main distinction between both is whether or not the screen is touch-sensitive), with their "two-tiered" security model. Pocket PCs have a single-tier security model and are a bit kinder on security. In a nutshell, you usually get these devices either partially or totally locked for RAPI connections. If your device is completely locked, then there's no hope for you and you can stop reading this right now. On the other hand, if you have a partially locked device you can either use VS2008's Device Secutity Manager tool to crack open the device's security or use the SDK cerificates to sign your DLL (you don't want to spend money on this by purchasing Geotrust signing events, do you?). Anyway, if you plan to sell your DLL as part of an application's setup file, you will have to sign it (and do some other crazy stuff like marking the DLL file with the system attribute bit and adding a bit of XML to the cab).

Writing the DLL is way simpler than all of the above. Open VS2008 and create a new project and select Visual C++ / Smart Device / Win32 Smart Device Project. Name this project "HTCGServer". When the Win32 Smart Device Project Wizard shows up, click the Next button to select the desired SDKs. Next, select the "DLL" radio button and the "Export symbols" check box.

Because this is a RAPI extension library, we need to make a reference to the RAPI header files which ship
with the SDKs. On my system the header file is on this folder:

C:\Program Files\Windows CE Tools\wce500\Windows Mobile 5.0 Pocket PC SDK\Activesync\Inc

For the Windows Mobile 6.0 SDK, you can find the same files here:

C:\Program Files\Windows Mobile 6 SDK\Activesync\inc

Go to your project's properties and select the "C/C++" option on the left tree and add one of the above paths (please validate these paths against your system first) to the "Additional Include Directories" property. This is where you add paths for header files other than the ones that ship with the SDK and that are not in your project's source path. Make sure that both the "Release" and "Debug" configurations have these paths set before you go on.

Now, let's take a look at the HTCGServer.h and HTCGServer.cpp fils that the Project Wizard created for us. Both files contain samples of DLL exportable symbols: a variable, a function and a class. All of them are marked with the HTCGSERVER_API macro that is defined at the top of HTCGServer.h:

#ifdef HTCGSERVER_EXPORTS

#define HTCGSERVER_API __declspec(dllexport)
#else
#define HTCGSERVER_API __declspec(dllimport)
#endif

This works in a very simlple way: when you compile this header file from within the DLL project, the HTCGSERVER_EXPORTS macro is defined, so HTCGSERVER_API resolves to __declspec(dllexport), meaning that you want to export this symbol from the DLL. You can confirm this by opening your project's property pages and inspecting the C/C++ / Preprocessor / Preprocessor Definitions property: you will find the HTCGSSERVER_EXPORTS definition there (done for you again by the Project Wizard).

Now erase everything below these lines and write the following:

#include <rapi.h>

extern "C"
{
HTCGSERVER_API int HTCGSensorStream(DWORD cbInput, BYTE* pInput, DWORD* pcbOutput, BYTE** ppOutput, IRAPIStream* pStream);
};

These lines include the RAPI header file (required for the IRAPIStream declaration) and declare a "C" function named HTCGSensorStream. This is the function that will be called by the RAPI runtime and it cannot be exported with the standard C++ linker names (these are mangled in order to include parameter type information for better run-time type checking). The number and types of the parameters are defined by RAPI for DLL extension functions and these can be of two types: blocking and non-blocking. A blocking extension function works like a regular function call: does its work and returns data back to the caller. A non-blocking function (or streamed call) uses the pStream parameter to keep an open communications channel to the caller on the PC: that's exactly what we'll use.

Let's create a stub for this function on the HTCGServer.cpp file:

//
// Starts the HTC G Sensor streaming to the PC via RAPI
//
int HTCGSensorStream(DWORD cbInput, BYTE* pInput, DWORD* pcbOutput, BYTE** ppOutput, IRAPIStream* pStream)
{
  *ppOutput = NULL;
  *pcbOutput = 0;
  LocalFree(pInput);
  pStream->Release();

  return 0;
}

Right now, the function does nothing but return after cleaning up. I'll get back to it after finishing the data server class. Now you can go back to last post's sample code and copy the HTCGSensor.h and HTCGSensor.cpp files to the new poject directory and add them to the project tree. This is all we need to write the final part of the extension DLL: the server class.

The server class is actually quite simple: it opens the sensor in the constructor, closes it in the destructor and contains a single function (Run) that cntinually reads data from the stream and writes back the sensor data to the listening PC software. This particular implementation uses a polling mechanism that waits for a command from the client and writes sensor data back. It's quite simple to understand how this works, but it's not the best approach performance-wise. From my own experience I have learned that you get the best performance when you put the stream reading and writing in different threads. This allows you to stream data from the server to the client without waiting for a specific command. Performance improves at the cost of complexity.

HRESULT HTCGSensorServer::Run()

{
HRESULT hr = S_OK;
bool bRun = true;

HTCGSensorData sensorData;

while(bRun)
{
    DWORD dwMsg = 0, dwRead, dwWrite;

    hr = m_pStream->Read(&dwMsg, sizeof(dwMsg),
                         &dwRead);
    if(FAILED(hr) || dwRead != sizeof(dwMsg))
      bRun = false;
    else
    {
      switch(dwMsg)
      {
        case CMD_QUIT:
          bRun = false;
          break;
        case CMD_READ:
          if(m_sensor.Read(&sensorData))
            hr = m_pStream->Write(&sensorData,
                                  sizeof(sensorData),
                                  &dwWrite);
          else
            bRun = false;
          break;
      }
    }
}
return hr;
}

Easy, right? The loop waits for two different commands: one for reading the sensor and the other to stop the loop (thereby exiting the server). Now you just need to call this function from the HTCGSensorStream entry point and your server is ready. Add the following lines to the top of the function:

HTCGSensorServer server(pStream);
server.Run();

Done.

On my next post I will implement the PC client DLL. Meanwhile, here's the sample code: HTCGServer.zip

The HTG G Sensor

2010-04-29T21:09:00.000+01:00

Here is a very simple application that illustrates how to use the HTC G Sensor. This is included in a number of devices (including my own HTC Touch Pro) and it provides a way to determine the spatial position of the phone by providing "tilt" values for the X, Y and Z coordinate axis.

Although HTC did not provide a public API for the G Sensor (shame on them for this), Scott Seligman reverse engineered it which eventually led to a Codeplex project on this subject. The code you see here was ported from the C# version of the Codeplex project back to C++.

The application is quite simple: it merely connects to the G Sensor data source, samples it every 200 ms and displays the tilt values on the screen. As usual, I used WTL to write this simple app and its main features are:

The HTCGSensor class that implements an interface to the HTC G Sensor API;
The view window (CHtcSensorView) that samples the G Sensor data every 200 milliseconds and displays the tilt values;
The frame window (CHtcSensorFrame) where the sampling timer is started (see the OnCreate handler)

The HTCGSensor class has a few points worth noticing: the sensor DLL is dynamically linked through the LoadLibrary API and the data reading function pointer is cached for better performance. Note that access to the sensor data requires acquiring a handle so, proper finalization is required.

Sample code: HtcSensor.zip

The sample application requires an HTC device with a G Sensor like the Diamond. You can use this information in a number of different ways on the device. On a forthcoming post, I will show how you can consume this information on a desktop .NET 3.5 application.

Smiling Smith?

2010-04-27T09:31:00.000+01:00

Why a new blog? Why this name?

First of all, I want to write about all of my development experiences under one single roof. Until recently I was able to frequently write about native code development on Windows Mobile devices, but due to the recent developments (no native code on WP7), this has become a much harder task. Also, I have recently been working on different technologies (like ASP .NET and XNA) and want to write about them, and the older blogs were too narrow in scope for this.

The blog name is taken from a very old nickname that I never forgot. A few hundred years ago there was a very interesting BBC TV series about an IT consultant named Smith that did a lot of cool stuff. The series was particularly successful on the public television, the only one we had back then. At this time I was working on a very informal team of young developers (remember CP/M?) and there was this older guy who named us all "Smith". Due to my high spirits, I was the smiling one. The Smiling Smith.