Search Results

Search found 5044 results on 202 pages for 'clausal logic'.

Page 142/202 | < Previous Page | 138 139 140 141 142 143 144 145 146 147 148 149  | Next Page >

  • DataContractSerializer: type is not serializable because it is not public?

    - by Michael B. McLaughlin
    I recently ran into an odd and annoying error when working with the DataContractSerializer class for a WP7 project. I thought I’d share it to save others who might encounter it the same annoyance I had. So I had an instance of  ObservableCollection<T> that I was trying to serialize (with T being a class I wrote for the project) and whenever it would hit the code to save it, it would give me: The data contract type 'ProjectName.MyMagicItemsClass' is not serializable because it is not public. Making the type public will fix this error. Alternatively, you can make it internal, and use the InternalsVisibleToAttribute attribute on your assembly in order to enable serialization of internal members - see documentation for more details. Be aware that doing so has certain security implications. This, of course, was malarkey. I was trying to write an instance of MyAwesomeClass that looked like this: [DataContract] public class MyAwesomeClass { [DataMember] public ObservableCollection<MyMagicItemsClass> GreatItems { get; set; }   [DataMember] public ObservableCollection<MyMagicItemsClass> SuperbItems { get; set; }     public MyAwesomeClass { GreatItems = new ObservableCollection<MyMagicItemsClass>(); SuperbItems = new ObservableCollection<MyMagicItemsClass>(); } }   That’s all well and fine. And MyMagicItemsClass was also public with a parameterless public constructor. It too had DataContractAttribute applied to it and it had DataMemberAttribute applied to all the properties and fields I wanted to serialize. Everything should be cool, but it’s not because I keep getting that “not public” exception. I could tell you about all the things I tried (generating a List<T> on the fly to make sure it wasn’t ObservableCollection<T>, trying to serialize the the Collections directly, moving it all to a separate library project, etc.), but I want to keep this short. In the end, I remembered my the “Debug->Exceptions…” VS menu option that brings up the list of exception-related circumstances under which the Visual Studio debugger will break. I checked the “Thrown” checkbox for “Common Language Runtime Exceptions”, started the project under the debugger, and voilà: the true problem revealed itself. Some of my properties had fairly elaborate setters whose logic I wanted to ignore. So for some of them, I applied an IgnoreDataMember attribute to them and applied the DataMember attribute to the underlying fields instead. All of which, in line with good programming practices, were private. Well, it just so happens that WP7 apps run in a “partial trust” environment and outside of “full trust”-land, DataContractSerializer refuses to serialize or deserialize non-public members. Of course that exception was swallowed up internally by .NET so all I ever saw was that bizarre message about things that I knew for certain were public being “not public”. I changed all the private fields I was serializing to public and everything worked just fine. In hindsight it all makes perfect sense. The serializer uses reflection to build up its graph of the object in order to write it out. In partial trust, you don’t want people using reflection to get at non-public members of an object since there are potential security problems with allowing that (you could break out of the sandbox pretty quickly by reflecting and calling the appropriate methods and cause some havoc by reflecting and setting the appropriate fields in certain circumstances. The fact that you cannot reflect your own assembly seems a bit heavy-handed, but then again I’m not a compiler writer or a framework designer and I have no idea what sorts of difficulties would go into allowing that from a compilation standpoint or what sorts of security problems allowing that could present (if any). So, lesson learned. If you get an incomprehensible exception message, turn on break on all thrown exceptions and try running it again (it might take a couple of tries, depending) and see what pops out. Chances are you’ll find the buried exception that actually explains what was going on. And if you’re getting a weird exception when trying to use DataContractSerializer complaining about public types not being public, chances are you’re trying to serialize a private or protected field/property.

    Read the article

  • Flow-Design Cheat Sheet &ndash; Part II, Translation

    - by Ralf Westphal
    In my previous post I summarized the notation for Flow-Design (FD) diagrams. Now is the time to show you how to translate those diagrams into code. Hopefully you feel how different this is from UML. UML leaves you alone with your sequence diagram or component diagram or activity diagram. They leave it to you how to translate your elaborate design into code. Or maybe UML thinks it´s so easy no further explanations are needed? I don´t know. I just know that, as soon as people stop designing with UML and start coding, things end up to be very different from the design. And that´s bad. That degrades graphical designs to just time waste on paper (or some designer). I even believe that´s the reason why most programmers view textual source code as the only and single source of truth. Design and code usually do not match. FD is trying to change that. It wants to make true design a first class method in every developers toolchest. For that the first prerequisite is to be able to easily translate any design into code. Mechanically, without thinking. Even a compiler could do it :-) (More of that in some other article.) Translating to Methods The first translation I want to show you is for small designs. When you start using FD you should translate your diagrams like this. Functional units become methods. That´s it. An input-pin becomes a method parameter, an output-pin becomes a return value: The above is a part. But a board can be translated likewise and calls the nested FUs in order: In any case be sure to keep the board method clear of any and all business logic. It should not contain any control structures like if, switch, or a loop. Boards do just one thing: calling nested functional units in proper sequence. What about multiple input-pins? Try to avoid them. Replace them with a join returning a tuple: What about multiple output-pins? Try to avoid them. Or return a tuple. Or use out-parameters: But as I said, this simple translation is for simple designs only. Splits and joins are easily done with method translation: All pretty straightforward, isn´t it. But what about wires, named pins, entry points, explicit dependencies? I suggest you don´t use this kind of translation when your designs need these features. Translating to methods is for small scale designs like you might do once you´re working on the implementation of a part of a larger design. Or maybe for a code kata you´re doing in your local coding dojo. Instead of doing TDD try doing FD and translate your design into methods. You´ll see that way it´s much easier to work collaboratively on designs, remember them more easily, keep them clean, and lessen the need for refactoring. Translating to Events [coming soon]

    Read the article

  • New R Interface to Oracle Data Mining Available for Download

    - by charlie.berger
      The R Interface to Oracle Data Mining ( R-ODM) allows R users to access the power of Oracle Data Mining's in-database functions using the familiar R syntax. R-ODM provides a powerful environment for prototyping data analysis and data mining methodologies. R-ODM is especially useful for: Quick prototyping of vertical or domain-based applications where the Oracle Database supports the application Scripting of "production" data mining methodologies Customizing graphics of ODM data mining results (examples: classification, regression, anomaly detection) The R-ODM interface allows R users to mine data using Oracle Data Mining from the R programming environment. It consists of a set of function wrappers written in source R language that pass data and parameters from the R environment to the Oracle RDBMS enterprise edition as standard user PL/SQL queries via an ODBC interface. The R-ODM interface code is a thin layer of logic and SQL that calls through an ODBC interface. R-ODM does not use or expose any Oracle product code as it is completely an external interface and not part of any Oracle product. R-ODM is similar to the example scripts (e.g., the PL/SQL demo code) that illustrates the use of Oracle Data Mining, for example, how to create Data Mining models, pass arguments, retrieve results etc. R-ODM is packaged as a standard R source package and is distributed freely as part of the R environment's Comprehensive R Archive Network (CRAN). For information about the R environment, R packages and CRAN, see www.r-project.org. R-ODM is particularly intended for data analysts and statisticians familiar with R but not necessarily familiar with the Oracle database environment or PL/SQL. It is a convenient environment to rapidly experiment and prototype Data Mining models and applications. Data Mining models prototyped in the R environment can easily be deployed in their final form in the database environment, just like any other standard Oracle Data Mining model. What is R? R is a system for statistical computation and graphics. It consists of a language plus a run-time environment with graphics, a debugger, access to certain system functions, and the ability to run programs stored in script files. The design of R has been heavily influenced by two existing languages: Becker, Chambers & Wilks' S and Sussman's Scheme. Whereas the resulting language is very similar in appearance to S, the underlying implementation and semantics are derived from Scheme. R was initially written by Ross Ihaka and Robert Gentleman at the Department of Statistics of the University of Auckland in Auckland, New Zealand. Since mid-1997 there has been a core group (the "R Core Team") who can modify the R source code archive. Besides this core group many R users have contributed application code as represented in the near 1,500 publicly-available packages in the CRAN archive (which has shown exponential growth since 2001; R News Volume 8/2, October 2008). Today the R community is a vibrant and growing group of dozens of thousands of users worldwide. It is free software distributed under a GNU-style copyleft, and an official part of the GNU project ("GNU S"). Resources: R website / CRAN R-ODM

    Read the article

  • VirtualBox 3.2 is released! A Red Letter Day?

    - by Fat Bloke
    Big news today! A new release of VirtualBox packed full of innovation and improvements. Over the next few weeks we'll take a closer look at some of these new features in a lot more depth, but today we'll whet your appetite with the headline descriptions. To start with, we should point out that this is the first Oracle-branded version which makes today a real Red-letter day ;-)  Oracle VM VirtualBox 3.2 Version 3.2 moves VirtualBox forward in 3 main areas ( handily, all beginning with "P" ) : performance, power and supported guest operating system platforms.  Let's take a look: Performance New Latest Intel hardware support - Harnessing the latest in chip-level support for virtualization, VirtualBox 3.2 supports new Intel Core i5 and i7 processor and Intel Xeon processor 5600 Series support for Unrestricted Guest Execution bringing faster boot times for everything from Windows to Solaris guests; New Large Page support - Reducing the size and overhead of key system resources, Large Page support delivers increased performance by enabling faster lookups and shorter table creation times. New In-hypervisor Networking - Significant optimization of the networking subsystem has reduced context switching between guests and host, increasing network throughput by up to 25%. New New Storage I/O subsystem - VirtualBox 3.2 offers a completely re-worked virtual disk subsystem which utilizes asynchronous I/O to achieve high-performance whilst maintaining high data integrity; New Remote Video Acceleration - The unique built-in VirtualBox Remote Display Protocol (VRDP), which is primarily used in virtual desktop infrastructure deployments, has been enhanced to deliver video acceleration. This delivers a rich user experience coupled with reduced computational expense, which is vital when servers are running hundreds of virtual machines; Power New Page Fusion - Traditional Page Sharing techniques have suffered from long and expensive cache construction as pages are scrutinized as candidates for de-duplication. Taking a smarter approach, VirtualBox Page Fusion uses intelligence in the guest virtual machine to determine much more rapidly and accurately those pages which can be eliminated thereby increasing the capacity or vm density of the system; New Memory Ballooning- Ballooning provides another method to increase vm density by allowing the memory of one guest to be recouped and made available to others; New Multiple Virtual Monitors - VirtualBox 3.2 now supports multi-headed virtual machines with up to 8 virtual monitors attached to a guest. Each virtual monitor can be a host window, or be mapped to the hosts physical monitors; New Hot-plug CPU's - Modern operating systems such Windows Server 2008 x64 Data Center Edition or the latest Linux server platforms allow CPUs to be dynamically inserted into a system to provide incremental computing power while the system is running. Version 3.2 introduces support for Hot-plug vCPUs, allowing VirtualBox virtual machines to be given more power, with zero-downtime of the guest; New Virtual SAS Controller - VirtualBox 3.2 now offers a virtual SAS controller, enabling it to run the most demanding of high-end guests; New Online Snapshot Merging - Snapshots are powerful but can eat up disk space and need to be pruned from time to time. Historically, machines have needed to be turned off to delete or merge snapshots but with VirtualBox 3.2 this operation can be done whilst the machines are running. This allows sophisticated system management with minimal interruption of operations; New OVF Enhancements - VirtualBox has supported the OVF standard for virtual machine portability for some time. Now with 3.2, VirtualBox specific configuration data is also stored in the standard allowing richer virtual machine definitions without compromising portability; New Guest Automation - The Guest Automation APIs allow host-based logic to drive operations in the guest; Platforms New USB Keyboard and Mouse - Support more guests that require USB input devices; New Oracle Enterprise Linux 5.5 - Support for the latest version of Oracle's flagship Linux platform; New Ubuntu 10.04 ("Lucid Lynx") - Support for both the desktop and server version of the popular Ubuntu Linux distribution; And as a man once said, "just one more thing" ... New Mac OS X (experimental) - On Apple hardware only, support for creating virtual machines run Mac OS X. All in all this is a pretty powerful release packed full of innovation and speedups. So what are you waiting for?  -FB 

    Read the article

  • What's wrong with this turn to face algorithm?

    - by Chan
    I implement a torpedo object that chases a rotating planet. Specifically, it will turn toward the planet each update. Initially my implement was: void move() { vector3<float> to_target = target - get_position(); to_target.normalize(); position += (to_target * speed); } which works perfectly for torpedo that is a solid sphere. Now my torpedo is actually a model, which has a forward vector, so using this method looks odd because it doesn't actually turn toward but jump toward. So I revised it a bit to get, double get_rotation_angle(vector3<float> u, vector3<float> v) const { u.normalize(); v.normalize(); double cosine_theta = u.dot(v); // domain of arccosine is [-1, 1] if (cosine_theta > 1) { cosine_theta = 1; } if (cosine_theta < -1) { cosine_theta = -1; } return math3d::to_degree(acos(cosine_theta)); } vector3<float> get_rotation_axis(vector3<float> u, vector3<float> v) const { u.normalize(); v.normalize(); // fix linear case if (u == v || u == -v) { v[0] += 0.05; v[1] += 0.0; v[2] += 0.05; v.normalize(); } vector3<float> axis = u.cross(v); return axis.normal(); } void turn_to_face() { vector3<float> to_target = (target - position); vector3<float> axis = get_rotation_axis(get_forward(), to_target); double angle = get_rotation_angle(get_forward(), to_target); double distance = math3d::distance(position, target); gl_matrix_mode(GL_MODELVIEW); gl_push_matrix(); { gl_load_identity(); gl_translate_f(position.get_x(), position.get_y(), position.get_z()); gl_rotate_f(angle, axis.get_x(), axis.get_y(), axis.get_z()); gl_get_float_v(GL_MODELVIEW_MATRIX, OM); } gl_pop_matrix(); move(); } void move() { vector3<float> to_target = target - get_position(); to_target.normalize(); position += (get_forward() * speed); } The logic is simple, I find the rotation axis by cross product, the angle to rotate by dot product, then turn toward the target position each update. Unfortunately, it looks extremely odds since the rotation happens too fast that it always turns back and forth. The forward vector for torpedo is from the ModelView matrix, the third column A: MODELVIEW MATRIX -------------------------------------------------- R U A T -------------------------------------------------- 1 0 0 0 0 1 0 0 0 0 1 0 0 0 0 1 -------------------------------------------------- Any suggestion or idea would be greatly appreciated.

    Read the article

  • Isis Finally Rolls Out

    - by David Dorf
    Google has rolled their wallet out for several chains; I see the NFC readers in Walgreen's when I'm sent their for milk.  But Isis has been relatively quiet until now.  As of last week they have finally launched in their two test cities: Austin, and Salt Lake City.  Below are the supported carriers and phones as of now, but more phones will be added later. Normal 0 false false false EN-US X-NONE X-NONE MicrosoftInternetExplorer4 /* Style Definitions */ table.MsoNormalTable {mso-style-name:"Table Normal"; mso-tstyle-rowband-size:0; mso-tstyle-colband-size:0; mso-style-noshow:yes; mso-style-priority:99; mso-style-qformat:yes; mso-style-parent:""; mso-padding-alt:0in 5.4pt 0in 5.4pt; mso-para-margin:0in; mso-para-margin-bottom:.0001pt; mso-pagination:widow-orphan; font-size:11.0pt; font-family:"Calibri","sans-serif"; mso-ascii-font-family:Calibri; mso-ascii-theme-font:minor-latin; mso-fareast-font-family:"Times New Roman"; mso-fareast-theme-font:minor-fareast; mso-hansi-font-family:Calibri; mso-hansi-theme-font:minor-latin; mso-bidi-font-family:"Times New Roman"; mso-bidi-theme-font:minor-bidi;} AT&T supports: HTC One™ X, LG Escape™, Samsung Galaxy Exhilarate™, Samsung Galaxy S® III, Samsung Galaxy Rugby Pro™ T-Mobile supports: Samsung Galaxy S® II, Samsung Galaxy S® III, Samsung Galaxy S® Relay 4G Verizon supports: Droid Incredible 4G LTE. Of course iPhone owners have no wallet since Apple didn't included an NFC chip. To start using Isis, you have to take your NFC-capable phone to your carrier's store to get the SIM replaced with a more sophisticated one that has a secure element configured for Isis.  The "secure element" is the cryptographic logic that secures mobile payments.  Carriers like the secure element in the SIM while non-carriers (like Google) prefer the secure element in the phone's electronics. (I'm not entirely sure if you could support both Isis and Google Wallet on the same phone.  Anybody know?) Then you can download the Isis app from Google Play and load your cards.  Most credit cards are supported, and there's a process to verify the credit cards are valid.  Then you can select from the list of participating retailers to "follow."  Selecting a retailer allows that retailer to give you offers via the app. The app is well done and easy to use.  You can select a default payment type and also switch between them easily.  When the phone is tapped on the reader, there are two exchanges of information.  The payment information is transferred, and then the Isis "SmartTap" information which includes optional loyalty number and digital coupons.  Of course the value of mobile wallets comes from the ease of handling all three data types (i.e. payment, loyalty, offers). There are several advertisements for Isis running now, and my favorite is below.

    Read the article

  • Oracle Retail Mobile Point-of-Service

    - by David Dorf
    When most people discuss mobile in retail, they immediately go to shopping applications.  While I agree the consumer side of mobile is huge, I believe its also important to arm store associates with mobile tools.  There are around a dozen major roll-outs of mobile POS to chain retailers, and all have been successful.  This does not, however, signal the demise of traditional registers.  Retailers will adopt mobile POS slowly and reduce the number of fixed registers over time, but there's likely to be a combination of both for the foreseeable future.  Even Apple retains at least one fixed register in every store, you just have to know where to look. The business benefits for mobile POS are pretty straightforward: 1. Faster checkout.  Walmart's CFO recently reported that for every second they shave off the average transaction time, they can potentially save $12M a year in labor.  I think its more likely that labor will be redeployed to enhance the customer experience. 2. Smarter associates.  The sales associates on the floor need the same access to information that consumers have, if not more.  They need ready access to product details, reviews, inventory, etc. to meet consumer expectations.  In a recent study, 40% of consumers said a savvy store associate can impact their final product selection more than a website. 3. Lower costs.  Mobile POS hardware (iPod touch + sled) costs about a fifth of fixed registers, not to mention the reclaimed space that can be used for product displays. But almost all Mobile POS solutions can claim those benefits equally.  Where there's differentiation is on the technical side.  Oracle recently announced availability of the Oracle Retail Mobile Point-of-Service, and it has three big technology advantages in the market: 1. Portable. We used a popular open-source component called PhoneGap that abstracts the app from the underlying OS and hardware so that iOS, Android, and other platforms could be supported.  Further, we used Web technologies such as HTML5 and JavaScript, which are commonly known by many programmers, as opposed to ObjectiveC which is more difficult to find.  The screen can adjust to different form-factors and sizes, just like you see with browsers.  In the future when a new, zippy device gets released, retailers will have the option to move to that device more easily than if they used a native app. 2. Flexible.  Our Mobile POS is free with the Oracle Retail Point-of-Service product.  Retailers can use any combination of fixed and mobile registers, and those ratios can change as required.  Perhaps start with 1 mobile and 4 fixed per store, then transition over time to 4 mobile and 1 fixed without any additional software licenses.  Our scalable solution supports lots of combinations. 3. Consistent.  Because our Mobile POS is fully integrated to our traditional POS, the same business logic is reused.  Third-party Mobile POS solutions often handle pricing, promotions, and tax calculations separately leading to possible inconsistencies within the store.  That won't happen with Oracle's solution. For many retailers, Mobile POS can lower costs, increase customer service, and generally enhance a consumer's in-store experience.  Apple led the way, but lots of other retailers are discovering the many benefits of adding mobile capabilities in their stores.  Just be sure to examine both the business and technology benefits so you get the most value from your solution for the longest period of time.

    Read the article

  • Find Rules and Defaults using the PowerShell for SQL Server 2008 Provider

    - by BuckWoody
    I ran into an issue the other day where I couldn't set up some features in SQL Server 2008 because they ddon't support the use of Rules or Defaults. Let me explain a little more about that. In older versions of SQL Server, you could decalre a "Rule" or "Default" just like you do with a Table Constraint today. You would then "bind" these rules or defaults to the tables you wanted them to apply to. Sure, there are advantages and disadvantages to this approach, but it certainly isn't standard Data Definition Language (DDL), so they are deprecated and many features don't work with them any more. Honestly, it's been so long since I've seen them in use I had forgotten to even check for them. My suspicion is that this was a new database created with an older script. Nevertheless, the feature failed when it ran into one. Immediately I thought that I had better build some logic into my process to try and catch those - but how? Lots of choices here, but since I was using PowerShell to do the rest of the work, I thought I would investigate how easy it would be just to do it there. And using the SQL Server 2008 provider, this could not be simpler. I won't show all of the scrupt here, because I was testing for these as a condition and then bailing out of the script and sending a notification, but all it is using is the DIR command! Here's an example on my "UNIVAC" computer for the "pubs" database: Find Rules using PowerShell: dir SQLSERVER:\SQL\UNIVAC\DEFAULT\Databases\pubs\Rulesdir SQLSERVER:\SQL\UNIVAC\DEFAULT\Databases\pubs\Defaults And this one will look in all databases:  #All Databases:dir SQLSERVER:\SQL\UNIVAC\DEFAULT\Databases | select-object -property Name, Rules, Defaults Awesome. Love me some PowerShell. Script Disclaimer, for people who need to be told this sort of thing: Never trust any script, including those that you find here, until you understand exactly what it does and how it will act on your systems. Always check the script on a test system or Virtual Machine, not a production system. Yes, there are always multiple ways to do things, and this script may not work in every situation, for everything. It’s just a script, people. All scripts on this site are performed by a professional stunt driver on a closed course. Your mileage may vary. Void where prohibited. Offer good for a limited time only. Keep out of reach of small children. Do not operate heavy machinery while using this script. If you experience blurry vision, indigestion or diarrhea during the operation of this script, see a physician immediately.       Share this post: email it! | bookmark it! | digg it! | reddit! | kick it! | live it!

    Read the article

  • How would I handle input with a Game Component?

    - by Aufziehvogel
    I am currently having problems from finding my way into the component-oriented XNA design. I read an overview over the general design pattern and googled a lot of XNA examples. However, they seem to be right on the opposite site. In the general design pattern, an object (my current player) is passed to InputComponent::update(Player). This means the class will know what to do and how this will affect the game (e.g. move person vs. scroll text in a menu). Yet, in XNA GameComponent::update(GameTime) is called automatically without a reference to the current player. The only XNA examples I found built some sort of higher-level Keyboard engine into the game component like this: class InputComponent: GameComponent { public void keyReleased(Keys); public void keyPressed(Keys); public bool keyDown(Keys); public void override update(GameTime gameTime) { // compare previous state with current state and // determine if released, pressed, down or nothing } } Some others went a bit further making it possible to use a Service Locator by a design like this: interface IInputComponent { public void downwardsMovement(Keys); public void upwardsMovement(Keys); public bool pausedGame(Keys); // determine which keys pressed and what that means // can be done for different inputs in different implementations public void override update(GameTime); } Yet, then I am wondering if it is possible to design an input class to resolve all possible situations. Like in a menu a mouse click can mean "click that button", but in game play it can mean "shoot that weapon". So if I am using such a modular design with game components for input, how much logic is to be put into the InputComponent / KeyboardComponent / GamepadComponent and where is the rest handled? What I had in mind, when I heard about Game Components and Service Locator in XNA was something like this: use Game Components to run the InputHandler automatically in the loop use Service Locator to be able to switch input at runtime (i.e. let player choose if he wants to use a gamepad or a keyboard; or which shall be player 1 and which player 2). However, now I cannot see how this can be done. First code example does not seem flexible enough, as on a game pad you could require some combination of buttons for something that is possible on keyboard with only one button or with the mouse) The second code example seems really hard to implement, because the InputComponent has to know in which context we are currently. Moreover, you could imagine your application to be multi-layered and let the key-stroke go through all layers to the bottom-layer which requires a different behaviour than the InputComponent would have guessed from the top-layer. The general design pattern with passing the Player to update() does not have a representation in XNA and I also cannot see how and where to decide which class should be passed to update(). At most time of course the player, but sometimes there could be menu items you have to or can click I see that the question in general is already dealt with here, but probably from a more elobate point-of-view. At least, I am not smart enough in game development to understand it. I am searching for a rather code-based example directly for XNA. And the answer there leaves (a noob like) me still alone in how the object that should receive the detected event is chosen. Like if I have a key-up event, should it go to the text box or to the player?

    Read the article

  • SQL SERVER – How to Get SQL Server Restart Notification?

    - by Pinal Dave
    Few days back my friend called me to know if there is any tool which can be used to get restart notification about SQL in their environment. I told that SQL Server can do it by itself with some configurations. He was happy and surprised to know that he need not spend any extra money. In SQL Server, we can configure stored procedure(s) to run at start-up of SQL Server. This blog would give steps to achieve how to achieve it. There are many situations where this feature can be used. Below are few. Logging SQL Server startup timings Modify data in some table during startup (i.e. table in tempdb) Sending notification about SQL start. Step 1 – Enable ‘scan for startup procs’ This can be done either using T-SQL or User Interface of Management Studio. EXEC sys.sp_configure N'Show Advanced Options', N'1' GO RECONFIGURE WITH OVERRIDE GO EXEC sys.sp_configure N'scan for startup procs', N'1' GO RECONFIGURE WITH OVERRIDE GO Below is the interface to change the setting. We need to go to “Server” > “Properties” and use “Advanced” tab. “Scan for Startup Procs” is the parameter under “Miscellaneous” section as shown below. We need to make value as “True” and hit OK. Step 2 – Create stored procedure It’s important to note that the procedure is executed after recovery is finished for ALL databases. Here is a sample stored procedure. You can use your own logic in the procedure. CREATE PROCEDURE SQLStartupProc AS BEGIN CREATE TABLE ##ThisTableShouldAlwaysExists (AnyColumn INT) END Step 3 – Set Procedure to run at startup We need to use sp_procoption to mark the procedure to run at startup. Here is the code to let SQL know that this is startup proc. sp_procoption 'SQLStartupProc', 'startup', 'true' This can be used only for procedures in master database. Msg 15398, Level 11, State 1, Procedure sp_procoption, Line 89 Only objects in the master database owned by dbo can have the startup setting changed. We also need to remember that such procedure should not have any input/output parameter. Here is the error which would be raised. Msg 15399, Level 11, State 1, Procedure sp_procoption, Line 107 Could not change startup option because this option is restricted to objects that have no parameters. Verification Here is the query to find which procedures is marked as startup procedures. SELECT name FROM sys.objects WHERE OBJECTPROPERTY(OBJECT_ID, 'ExecIsStartup') = 1 Once this is done, I have restarted SQL instance and here is what we would see in SQL ERRORLOG Launched startup procedure 'SQLStartupProc'. This confirms that stored procedure is executed. You can also notice that this is done after all databases are recovered. Recovery is complete. This is an informational message only. No user action is required. After few days my friend again called me and asked – I want to turn this OFF? Use comments section and post the answer for him.  Reference: Pinal Dave (http://blog.sqlauthority.com)Filed under: PostADay, SQL, SQL Authority, SQL Query, SQL Server, SQL Tips and Tricks, SQL Utility, T SQL

    Read the article

  • SQL Server Optimizer Malfunction?

    - by Tony Davis
    There was a sharp intake of breath from the audience when Adam Machanic declared the SQL Server optimizer to be essentially "stuck in 1997". It was during his fascinating "Query Tuning Mastery: Manhandling Parallelism" session at the recent PASS SQL Summit. Paraphrasing somewhat, Adam (blog | @AdamMachanic) offered a convincing argument that the optimizer often delivers flawed plans based on assumptions that are no longer valid with today’s hardware. In 1997, when Microsoft engineers re-designed the database engine for SQL Server 7.0, SQL Server got its initial implementation of a cost-based optimizer. Up to SQL Server 2000, the developer often had to deploy a steady stream of hints in SQL statements to combat the occasionally wilful plan choices made by the optimizer. However, with each successive release, the optimizer has evolved and improved in its decision-making. It is still prone to the occasional stumble when we tackle difficult problems, join large numbers of tables, perform complex aggregations, and so on, but for most of us, most of the time, the optimizer purrs along efficiently in the background. Adam, however, challenged further any assumption that the current optimizer is competent at providing the most efficient plans for our more complex analytical queries, and in particular of offering up correctly parallelized plans. He painted a picture of a present where complex analytical queries have become ever more prevalent; where disk IO is ever faster so that reads from disk come into buffer cache faster than ever; where the improving RAM-to-data ratio means that we have a better chance of finding our data in cache. Most importantly, we have more CPUs at our disposal than ever before. To get these queries to perform, we not only need to have the right indexes, but also to be able to split the data up into subsets and spread its processing evenly across all these available CPUs. Improvements such as support for ColumnStore indexes are taking things in the right direction, but, unfortunately, deficiencies in the current Optimizer mean that SQL Server is yet to be able to exploit properly all those extra CPUs. Adam’s contention was that the current optimizer uses essentially the same costing model for many of its core operations as it did back in the days of SQL Server 7, based on assumptions that are no longer valid. One example he gave was a "slow disk" bias that may have been valid back in 1997 but certainly is not on modern disk systems. Essentially, the optimizer assesses the relative cost of serial versus parallel plans based on the assumption that there is no IO cost benefit from parallelization, only CPU. It assumes that a single request will saturate the IO channel, and so a query would not run any faster if we parallelized IO because the disk system simply wouldn’t be able to handle the extra pressure. As such, the optimizer often decides that a serial plan is lower cost, often in cases where a parallel plan would improve performance dramatically. It was challenging and thought provoking stuff, as were his techniques for driving parallelism through query logic based on subsets of rows that define the "grain" of the query. I highly recommend you catch the session if you missed it. I’m interested to hear though, when and how often people feel the force of the optimizer’s shortcomings. Barring mistakes, such as stale statistics, how often do you feel the Optimizer fails to find the plan you think it should, and what are the most common causes? Is it fighting to induce it toward parallelism? Combating unexpected plans, arising from table partitioning? Something altogether more prosaic? Cheers, Tony.

    Read the article

  • My 2D collision code does not work as expected. How do I fix it?

    - by farmdve
    I have a simple 2D game with a tile-based map. I am new to game development, I followed the LazyFoo tutorials on SDL. The tiles are in a bmp file, but each tile inside it corresponds to an internal number of the type of tile(color, or wall). The game is simple, but the code is a lot so I can only post snippets. // Player moved out of the map if((player.box.x < 0)) player.box.x += GetVelocity(player, 0); if((player.box.y < 0)) player.box.y += GetVelocity(player, 1); if((player.box.x > (LEVEL_WIDTH - DOT_WIDTH))) player.box.x -= GetVelocity(player, 0); if((player.box.y > (LEVEL_HEIGHT - DOT_HEIGHT))) player.box.y -= GetVelocity(player, 1); // Now that we are here, we check for collisions if(touches_wall(player.box)) { if(player.box.x < player.prev_x) { player.box.x += GetVelocity(player, 0); } if(player.box.x > player.prev_x) { player.box.x -= GetVelocity(player, 0); } if(player.box.y < player.prev_y) { player.box.y += GetVelocity(player, 1); } if(player.box.y > player.prev_y) { player.box.y -= GetVelocity(player, 1); } } player.prev_x = player.box.x; player.prev_y = player.box.y; Let me explain, player is a structure with the following contents typedef struct { Rectangle box; //Player position on a map(tile or whatever). int prev_x, prev_y; // Previous positions int key_press[3]; // Stores which key was pressed/released. Limited to three keys. E.g Left,right and perhaps jump if possible in 2D int velX, velY; // Velocity for X and Y coordinate. //Health int health; bool main_character; uint32_t jump_ticks; } Player; And Rectangle is just a typedef of SDL_Rect. GetVelocity is a function that according to the second argument, returns the velocity for the X or Y axis. This code I have basically works, however inside the if(touches_wall(player.box)) if statement, I have 4 more. These 4 if statements are responsible for detecting collision on all 4 sides(up,down,left,right). However, they also act as a block for any other movement. Example: I move down the object and collide with the wall, as I continue to move down and still collide with the wall, I wish to move left or right, which is indeed possible(not to mention in 3D games), but remember the 4 if statements? They are preventing me from moving anywhere. The original code on the LazyFoo Productions website has no problems, but it was written in C++, so I had to rewrite most of it to work, which is probably where the problem comes from. I also use a different method of moving, than the one in the examples. Of course, that was just an example. I wish to be able to move no matter at which wall I collide. Before this bit of code, I had another one that had more logic in there, but it was flawed.

    Read the article

  • Efficiently separating Read/Compute/Write steps for concurrent processing of entities in Entity/Component systems

    - by TravisG
    Setup I have an entity-component architecture where Entities can have a set of attributes (which are pure data with no behavior) and there exist systems that run the entity logic which act on that data. Essentially, in somewhat pseudo-code: Entity { id; map<id_type, Attribute> attributes; } System { update(); vector<Entity> entities; } A system that just moves along all entities at a constant rate might be MovementSystem extends System { update() { for each entity in entities position = entity.attributes["position"]; position += vec3(1,1,1); } } Essentially, I'm trying to parallelise update() as efficiently as possible. This can be done by running entire systems in parallel, or by giving each update() of one system a couple of components so different threads can execute the update of the same system, but for a different subset of entities registered with that system. Problem In reality, these systems sometimes require that entities interact(/read/write data from/to) each other, sometimes within the same system (e.g. an AI system that reads state from other entities surrounding the current processed entity), but sometimes between different systems that depend on each other (i.e. a movement system that requires data from a system that processes user input). Now, when trying to parallelize the update phases of entity/component systems, the phases in which data (components/attributes) from Entities are read and used to compute something, and the phase where the modified data is written back to entities need to be separated in order to avoid data races. Otherwise the only way (not taking into account just "critical section"ing everything) to avoid them is to serialize parts of the update process that depend on other parts. This seems ugly. To me it would seem more elegant to be able to (ideally) have all processing running in parallel, where a system may read data from all entities as it wishes, but doesn't write modifications to that data back until some later point. The fact that this is even possible is based on the assumption that modification write-backs are usually very small in complexity, and don't require much performance, whereas computations are very expensive (relatively). So the overhead added by a delayed-write phase might be evened out by more efficient updating of entities (by having threads work more % of the time instead of waiting). A concrete example of this might be a system that updates physics. The system needs to both read and write a lot of data to and from entities. Optimally, there would be a system in place where all available threads update a subset of all entities registered with the physics system. In the case of the physics system this isn't trivially possible because of race conditions. So without a workaround, we would have to find other systems to run in parallel (which don't modify the same data as the physics system), other wise the remaining threads are waiting and wasting time. However, that has disadvantages Practically, the L3 cache is pretty much always better utilized when updating a large system with multiple threads, as opposed to multiple systems at once, which all act on different sets of data. Finding and assembling other systems to run in parallel can be extremely time consuming to design well enough to optimize performance. Sometimes, it might even not be possible at all because a system just depends on data that is touched by all other systems. Solution? In my thinking, a possible solution would be a system where reading/updating and writing of data is separated, so that in one expensive phase, systems only read data and compute what they need to compute, and then in a separate, performance-wise cheap, write phase, attributes of entities that needed to be modified are finally written back to the entities. The Question How might such a system be implemented to achieve optimal performance, as well as making programmer life easier? What are the implementation details of such a system and what might have to be changed in the existing EC-architecture to accommodate this solution?

    Read the article

  • How do you formulate the Domain Model in Domain Driven Design properly (Bounded Contexts, Domains)?

    - by lko
    Say you have a few applications which deal with a few different Core Domains. The examples are made up and it's hard to put a real example with meaningful data together (concisely). In Domain Driven Design (DDD) when you start looking at Bounded Contexts and Domains/Sub Domains, it says that a Bounded Context is a "phase" in a lifecycle. An example of Context here would be within an ecommerce system. Although you could model this as a single system, it would also warrant splitting into separate Contexts. Each of these areas within the application have their own Ubiquitous Language, their own Model, and a way to talk to other Bounded Contexts to obtain the information they need. The Core, Sub, and Generic Domains are the area of expertise and can be numerous in complex applications. Say there is a long process dealing with an Entity for example a Book in a core domain. Now looking at the Bounded Contexts there can be a number of phases in the books life-cycle. Say outline, creation, correction, publish, sale phases. Now imagine a second core domain, perhaps a store domain. The publisher has its own branch of stores to sell books. The store can have a number of Bounded Contexts (life-cycle phases) for example a "Stock" or "Inventory" context. In the first domain there is probably a Book database table with basically just an ID to track the different book Entities in the different life-cycles. Now suppose you have 10+ supporting domains e.g. Users, Catalogs, Inventory, .. (hard to think of relevant examples). For example a DomainModel for the Book Outline phase, the Creation phase, Correction phase, Publish phase, Sale phase. Then for the Store core domain it probably has a number of life-cycle phases. public class BookId : Entity { public long Id { get; set; } } In the creation phase (Bounded Context) the book could be a simple class. public class Book : BookId { public string Title { get; set; } public List<string> Chapters { get; set; } //... } Whereas in the publish phase (Bounded Context) it would have all the text, release date etc. public class Book : BookId { public DateTime ReleaseDate { get; set; } //... } The immediate benefit I can see in separating by "life-cycle phase" is that it's a great way to separate business logic so there aren't mammoth all-encompassing Entities nor Domain Services. A problem I have is figuring out how to concretely define the rules to the physical layout of the Domain Model. A. Does the Domain Model get "modeled" so there are as many bounded contexts (separate projects etc.) as there are life-cycle phases across the core domains in a complex application? Edit: Answer to A. Yes, according to the answer by Alexey Zimarev there should be an entire "Domain" for each bounded context. B. Is the Domain Model typically arranged by Bounded Contexts (or Domains, or both)? Edit: Answer to B. Each Bounded Context should have its own complete "Domain" (Service/Entities/VO's/Repositories) C. Does it mean there can easily be 10's of "segregated" Domain Models and multiple projects can use it (the Entities/Value Objects)? Edit: Answer to C. There is a complete "Domain" for each Bounded Context and the Domain Model (Entity/VO layer/project) isn't "used" by the other Bounded Contexts directly, only via chosen paths (i.e. via Domain Events). The part that I am trying to figure out is how the Domain Model is actually implemented once you start to figure out your Bounded Contexts and Core/Sub Domains, particularly in complex applications. The goal is to establish the definitions which can help to separate Entities between the Bounded Contexts and Domains.

    Read the article

  • Non use of persisted data

    - by Dave Ballantyne
    Working at a client site, that in itself is good to say, I ran into a set of circumstances that made me ponder, and appreciate, the optimizer engine a bit more. Working on optimizing a stored procedure, I found a piece of code similar to : select BillToAddressID, Rowguid, dbo.udfCleanGuid(rowguid) from sales.salesorderheaderwhere BillToAddressID = 985 A lovely scalar UDF was being used,  in actuality it was used as part of the WHERE clause but simplified here.  Normally I would use an inline table valued function here, but in this case it wasn't a good option. So this seemed like a pretty good case to use a persisted column to improve performance. The supporting index was already defined as create index idxBill on sales.salesorderheader(BillToAddressID) include (rowguid) and the function code is Create Function udfCleanGuid(@GUID uniqueidentifier)returns varchar(255)with schemabindingasbegin Declare @RetStr varchar(255) Select @RetStr=CAST(@Guid as varchar(255)) Select @RetStr=REPLACE(@Retstr,'-','') return @RetStrend Executing the Select statement produced a plan of : Nothing surprising, a seek to find the data and compute scalar to execute the UDF. Lets get optimizing and remove the UDF with a persisted column Alter table sales.salesorderheaderadd CleanedGuid as dbo.udfCleanGuid(rowguid)PERSISTED A subtle change to the SELECT statement… select BillToAddressID,CleanedGuid from sales.salesorderheaderwhere BillToAddressID = 985 and our new optimized plan looks like… Not a lot different from before!  We are using persisted data on our table, where is the lookup to fetch it ?  It didnt happen,  it was recalculated.  Looking at the properties of the relevant Compute Scalar would confirm this ,  but a more graphic example would be shown in the profiler SP:StatementCompleted event. Why did the lookup happen ? Remember the index definition,  it has included the original guid to avoid the lookup.  The optimizer knows this column will be passed into the UDF, run through its logic and decided that to recalculate is cheaper than the lookup.  That may or may not be the case in actuality,  the optimizer has no idea of the real cost of a scalar udf.  IMO the default cost of a scalar UDF should be seen as a lot higher than it is, since they are invariably higher. Knowing this, how do we avoid the function call?  Dropping the guid from the index is not an option, there may be other code reliant on it.   We are left with only one real option,  add the persisted column into the index. drop index Sales.SalesOrderHeader.idxBillgocreate index idxBill on sales.salesorderheader(BillToAddressID) include (rowguid,cleanedguid) Now if we repeat the statement select BillToAddressID,CleanedGuid from sales.salesorderheaderwhere BillToAddressID = 985 We still have a compute scalar operator, but this time it wasnt used to recalculate the persisted data.  This can be confirmed with profiler again. The takeaway here is,  just because you have persisted data dont automatically assumed that it is being used.

    Read the article

  • Building Tag Cloud Declarative ADF Component

    - by Arunkumar Ramamoorthy
    When building a website, there could a requirement to add a tag cloud to let the users know the popular tags (or terms) used in the site. In this blog, we would build a simple declarative component to be used as tag cloud in the page. To start with, we would first create the declarative component, which could display the tag cloud. We will do that by creating a new custom application from the new gallery. Give a name for the app and the project and from the new gallery, let us create a new ADF Declarative Component We need to specify the name for the declarative component, attributes in it etc. as follows For displaying the tags as cloud, we need to pass the content to this component. So, we will create an attribute to hold the values for the tag. Let us name it as "value" and make it as java.lang.String  type. Once after this, to hold the component, we need to create a tag library. This can be done by clicking on the Add Tag Library button. Clicking on OK buttons in all the open dialogs would create a declarative component for us. Now, we need to display the tag cloud based on the value passed to the component. To do that, we assume that the value is a Tree Binding and has two attributes in it, say "Name" and "Weight". To make a tag cloud, we would put together the "Name" in a loop and set it's font size based on the "Weight". After putting our logic to work, here is how the source look Attributes added to the declarative components can be retrieved by using #{attrs.<attribute_name>}. Now, we need to deploy this project as ADF Library Jar file, so that this can be distributed to the consuming applications. We'll select ADF Library Jar as type and create the profile. We would be getting the jar file after deployment. To test the functionality, we could create a simple Fusion Web Application. To add our custom component to the consuming application, we can create a file system connection pointing to the location where the jar file is and add it or, add through the project properties of the ViewController project. Now, our custom component has been added to the consuming application. We could test that by creating a VO in the model project with a query like, select 'Faces' as Name,25 as Weight from dual union all select 'ADF', 15 from dual  union all select 'ADFdi', 30 from dual union all select 'BC4J', 20 from dual union all select 'EJB', 40 from dual union all select 'WS', 35 from dual Add this VO to the AppModule, so that it would be exposed to the data control. Then, we could create a jspx page, and add a tree binding to the VO created. We can now see our Tag Cloud declarative component is available in the component palette.  It can be inserted from the component palette to our page and set it's value property to CollectionModel of the tree binding created. Now that we've created the Declarative component and added that to our page successfully, we can run the page to see how it looks. As per the query, the Tags are displayed in different fonts, based on their weight.

    Read the article

  • Doubts about several best practices for rest api + service layer

    - by TheBeefMightBeTough
    I'm going to be starting a project soon that exposes a restful api for business intelligence. It may not be limited to a restful api, so I plan to delegate requests to a service layer that then coordinates multiple domain objects (each of which have business logic local to the object). The api will likely have many calls as it is a long-term project. While thinking about the design, I recalled a few best practices. 1) Use command objects at the controller layer (I'm using Spring MVC). 2) Use DTOs at the service layer. 3) Validate in both the controller and service layer, though for different reasons. I have my doubts about these recommendations. 1) Using command objects adds a lot of extra single-purpose classes (potentially one per request). What exactly is the benefit? Annotation based validation can be done using this approach, sure. What if I have two requests that take the same parameters, but have different validation requirements? I would have to have two different classes with exactly the same members but different annotations? Bleh. 2) I have heard that using DTOs is preferable to parameters because it makes for more maintainable code down the road (say, e.g., requirements change and the service parameters need to be altered). I don't quite understand this. Shouldn't an api be more-or-less set in stone? I would understand that in the early phases of a project (or, especially, an entire company) the domain itself will not be well understood, and thus core domain objects may change along with the apis that manipulate these objects. At this point however the number of api methods should be small and their dependents few, so changes to the methods could easily be tolerated from a maintainability standpoint. In a large api with many methods and a substantial domain model, I would think having a DTO for potentially each domain object would become unwieldy. Am I misunderstanding something here? 3) I see validation in the controller and service layer as redundant in most cases. Why would I validate that parameters are not null and are in general well formed in the controller if the service is going to do exactly the same (and more). Couldn't I just do all the validation in the service and throw a runtime exception with a list of bad parameters then catch that in the controller to make the error messages more presentable? Better yet, couldn't I just make the error messages user-friendly in the service and let the exception trickle up to a global handler (ControllerAdvice in spring, for example)? Is there something wrong with either of these approaches? (I do see a use case for controller validation if the input does not map one-to-one with the service input, but since the controllers are for a rest api and not forms, the api parameters will probably map directly to service parameters.) I do also have a question about unchecked vs checked exceptions. Namely, I'm not really sure why I'd ever want to use a checked exception. Every time I have seen them used they just get wrapped into general exceptions (DomainException, SystemException, ApplicationException, w/e) to reduce the signature length of methods, or devs catch Exception rather than dealing with the App1Exception, App2Exception, Sys1Exception, Sys2Exception. I don't see how either of these practices is very useful. Why not just use unchecked exceptions always and catch the ones you actually do care about? You could just document what unchecked exceptions the method throws.

    Read the article

  • Dont Throw Duplicate Exceptions

    In your code, youll sometimes have write code that validates input using a variety of checks.  Assuming you havent embraced AOP and done everything with attributes, its likely that your defensive coding is going to look something like this: public void Foo(SomeClass someArgument) { if(someArgument == null) { throw new InvalidArgumentException("someArgument"); } if(!someArgument.IsValid()) { throw new InvalidArgumentException("someArgument"); }   // Do Real Work } Do you see a problem here?  Heres the deal Exceptions should be meaningful.  They have value at a number of levels: In the code, throwing an exception lets the develop know that there is an unsupported condition here In calling code, different types of exceptions may be handled differently At runtime, logging of exceptions provides a valuable diagnostic tool Its this last reason I want to focus on.  If you find yourself literally throwing the exact exception in more than one location within a given method, stop.  The stack trace for such an exception is likely going to be identical regardless of which path of execution led to the exception being thrown.  When that happens, you or whomever is debugging the problem will have to guess which exception was thrown.  Guessing is a great way to introduce additional problems and/or greatly increase the amount of time require to properly diagnose and correct any bugs related to this behavior. Dont Guess Be Specific When throwing an exception from multiple code paths within the code, be specific.  Virtually ever exception allows a custom message use it and ensure each case is unique.  If the exception might be handled differently by the caller, than consider implementing a new custom exception type.  Also, dont automatically think that you can improve the code by collapsing the if-then logic into a single call with short-circuiting (e.g. if(x == null || !x.IsValid()) ) that will guarantee that you cant easily throw different information into the message as easily as constructing the exception separately in each case. The code above might be refactored like so:   public void Foo(SomeClass someArgument) { if(someArgument == null) { throw new ArgumentNullException("someArgument"); } if(!someArgument.IsValid()) { throw new InvalidArgumentException("someArgument"); }   // Do Real Work } In this case its taking advantage of the fact that there is already an ArgumentNullException in the framework, but if you didnt have an IsValid() method and were doing validation on your own, it might look like this: public void Foo(SomeClass someArgument) { if(someArgument.Quantity < 0) { throw new InvalidArgumentException("someArgument", "Quantity cannot be less than 0. Quantity: " + someArgument.Quantity); } if(someArgument.Quantity > 100) { throw new InvalidArgumentException("someArgument", "SomeArgument.Quantity cannot exceed 100. Quantity: " + someArgument.Quantity); }   // Do Real Work }   Note that in this last example, Im throwing the same exception type in each case, but with different Message values.  Im also making sure to include the value that resulted in the exception, as this can be extremely useful for debugging.  (How many times have you wished NullReferenceException would tell you the name of the variable it was trying to reference?) Dont add work to those who will follow after you to maintain your application (especially since its likely to be you).  Be specific with your exception messages follow DRY when throwing exceptions within a given method by throwing unique exceptions for each interesting case of invalid state. Did you know that DotNetSlackers also publishes .net articles written by top known .net Authors? We already have over 80 articles in several categories including Silverlight. Take a look: here.

    Read the article

  • Seperation of project responsibilities in new project

    - by dreza
    We have very recently started a new project (MVC 3.0) and some of our early discussion has been around how the work and development will be split amongst the team members to ensure we get the least amount of overlap of work and so help make it a bit easier for each developer to get on and do their work. The project is expected to take about 6 months - 1 year (although not all developers are likely to be on and might filter off towards the end), Our team is going to be small so this will help out a bit I believe. The team will essentially consist of: 3 x developers (1 a slightly more experienced and will be the lead) 1 x project manager / product owner / tester An external company responsbile for doing our design work General project/development decisions so far have included: Develop in an Agile way using SCRUM techniques (We are still very much learning this approach as a company) Use MVVM archectecture Use Ninject and DI where possible Attempt to use as TDD as much as possible to drive development. Keep our controllers as skinny as possible Keep our views as simple as possible During our discussions two approaches have been broached as too how to seperate the workload given our objectives outlined above. OPTION 1: A framework seperation where each person is responsible for conceptual areas with overlap and discussion primarily in the integration areas. The integration areas would the responsibily of both developers as required. View prototypes (**Graphic designer**) | - Mockups | Views (Razor and view helpers etc) & Javascript (**Developer 1**) | - View models (Integration point) | Controllers and Application logic (**Developer 2**) | - Models (Integration point) | Domain model and persistence (**Developer 3**) PROS: Integration points are quite clear and so developers can work without dependencies on others fairly easily Code practices such as naming conventions and style is more easily managed in regards to consistancy as primarily only one developer will be handling an area CONS: Completion of an entire feature becomes a bit grey as no single person is responsible for an entire feature (story?) A person might not have a full appreciation for all areas of the project and so code overlap might be lacking if suddenly that person left. OPTION 2: A more task orientated approach where each person is responsible for the completion of the entire task from view - controller - model. PROS: A person is responsible for one entire feature so it's "complete" state can be clearly defined Code overlap into different areas will occur so each individual has good coverage over the entire application CONS: Overlap of development will occur in all the modules and developers can develop/extend without a true understanding of what the original code owner was intending. This could potentially lead more easily to code bloat? Following a convention might be harder as developers are adding to all areas of the project If a developer sets up a way of doing things would it be harder to enforce the other developers to follow that convention or even build on it (or even discuss it?). Dunno.. Bugs could more easily be introduced into areas not thought about by the developer It's easier to possibly to carry a team member in so far as one member just hacks code together to complete a task whilst another takes time to build a foundation that could be used by others and so help make future tasks easier i.e. starts building a framework? QUESTION: As it might appear I'm more in favor of option 1, however I'm interested to see how others might have approached this or what is the standard or best or preferred way of undertaking a project. Or indeed any different approach to handling this?

    Read the article

  • Seeking advice on tools and technology for my new game [closed]

    - by k.k. slider
    I'm a C# developer who has been programming a game in my spare time using XNA and Visual Studio. The game's logic is mostly done and I've completed a prototype that has most of the functionality of (what I envision to be) the final game. However, having heard about the uncertain future and (possibly) limited audience for XNA games, I'm looking to switch platforms... but I don't know what technology would best suit my needs. Below are some specifics about my game and what exactly I'm looking for, if you're interested: The game is a 2D turn-based tactical RPG (strategy game) for two players. It is a basic sprite and tile based game with animations and sound. 3D capabilities are not necessary. I'd like to allow players to compete with others online, and have a basic ranking/matchmaking system. I will probably need something that can interact with a server and a database (the game is turn-based and has no RNG, so cheating would be easy to detect even if most computation is done client-side and minimal data is sent to the server). Ideally, I would be able to release an early version of the game and have people give feedback as I develop additional features (similar to Minecraft). I'd prefer to have a way to release periodic updates to the game instead of releasing an absolute final product. To reach the widest possible audience, I'd prefer technology that allows me to release on PC, Android, iOS, and (maybe) Mac. This is a game with simple mouse inputs which can fit on a mobile touch screen. The game should be monetizable. If I find success with this game, then I may consider becoming a full-time indie game developer. I have several other game ideas and have learned quite a bit from my first attempt at game development. My first thought was an F2P/microtransaction model, but I'm open to other suggestions. Language isn't a primary concern of mine, since I have a decent amount of experience using several languages to program large projects. I'm willing to spend money (e.g. on a developer's license), but the more expensive it gets, the more hesitant I am to use it. I've looked into the following solutions... there are a LOT of tools out there... if anyone has experience with any of these and would like to recommend/reject any of them, it would be helpful. C#/.NET (XNA/MonoGame/SDL/SlimDX/Xamarin/ExEn/ANX?) HTML5/JS (AppMobi/PhoneGap/Marmalade/FlashCanvas/Cordova/libRocket?) Python (Pyglet/Pygame/Kivy?) Java (JavaFX/libGDX?) Unity/Construct 2/Cocos2D/NME/Corona/other game creation software? I'd like something that can do 2D and isn't limited by being too high-level. Other languages (Lua/LOVE? Moai?) Thanks for answering this rather long and tedious question...

    Read the article

  • Best approach to depth streaming via existing codec

    - by Kevin
    I'm working on a development system (and game) intended for games set mostly in static third-person views. We produce our scenery by CG and photographic techniques. Our background art is rendered off-line by a production-grade renderer. To allow the runtime imagery to properly interact with the background art, I wrote a program to convert from depth output by Mental Ray into a texture, and a pixel shader to draw a quad such that the Z data comes from the texture. This technique is working out very well, but now we've decided that some of the camera angle changes between scenes should be animated. The animation itself is straightforward to produce from our CG models. We intend to encode it to some HD video codec such as H.264. The problem is that in order to maintain our runtime imagery on the screen, the depth buffer will need to be loaded for each video frame. Due to the bandwidth, the video's depth data will need to be compressed efficiently. I've looked into methods for performing temporal compression of depth info and found an interesting research paper here: http://web4.cs.ucl.ac.uk/staff/j.kautz/publications/depth-streaming.pdf The method establishes a mapping between 16-bit depth values and YCbCr values. The mapping is tuned to the properties of existing video codecs in order to maximize precision of the decoded depths after the YCbCr has undergone video compression. It allows an existing, unmodified video codec to be used on the backend. I'm looking at how to pull this off with the least possible work. (This design change was unplanned.) Our game engine itself is native C++, presently for Win32 and DirectX, although we've worked hard to keep platform dependence segregated because we intend other ports. We don't have motion video facilities in the engine yet but will ultimately need that anyway for cinematics. I was planning on using some off-the-shelf motion video solution we can plug into our engine, and haven't chosen one yet. This new added requirement makes selecting one harder since, among other things, we'll now need to bypass colourspace conversion on one of the streams, and also will need to be playing two streams simultaneously in lockstep, on top of in some cases audio on one of them (for the cinematics). I'm also wondering if it's possible (or even useful) to do the conversion from YCbCr to depth in a pixel shader, or if it's better to just do it in CPU and separately load the resulting depth values into a locked tex. The conversion unfortunately does involve branching logic per-pixel. (There are more naive mappings that don't need branching, but they produce inferior results.) It could be reduced to a table lookup but the table would be 32MB. Programming is second-nature to me but I'm not that experienced with pix shaders and have zero knowledge of off-the-shelf video solutions. I'd therefore be interested in advice from others who may have dealt more with depth streaming, pixel shaders, and/or off-the-shelf codecs, regarding how feasible the proposed application is and what off-the-shelf video systems out there would best get along with this usage case.

    Read the article

  • Learn Many Languages

    - by Jeff Foster
    My previous blog, Deliberate Practice, discussed the need for developers to “sharpen their pencil” continually, by setting aside time to learn how to tackle problems in different ways. However, the Sapir-Whorf hypothesis, a contested and somewhat-controversial concept from language theory, seems to hold reasonably true when applied to programming languages. It states that: “The structure of a language affects the ways in which its speakers conceptualize their world.” If you’re constrained by a single programming language, the one that dominates your day job, then you only have the tools of that language at your disposal to think about and solve a problem. For example, if you’ve only ever worked with Java, you would never think of passing a function to a method. A good developer needs to learn many languages. You may never deploy them in production, you may never ship code with them, but by learning a new language, you’ll have new ideas that will transfer to your current “day-job” language. With the abundant choices in programming languages, how does one choose which to learn? Alan Perlis sums it up best. “A language that doesn‘t affect the way you think about programming is not worth knowing“ With that in mind, here’s a selection of languages that I think are worth learning and that have certainly changed the way I think about tackling programming problems. Clojure Clojure is a Lisp-based language running on the Java Virtual Machine. The unique property of Lisp is homoiconicity, which means that a Lisp program is a Lisp data structure, and vice-versa. Since we can treat Lisp programs as Lisp data structures, we can write our code generation in the same style as our code. This gives Lisp a uniquely powerful macro system, and makes it ideal for implementing domain specific languages. Clojure also makes software transactional memory a first-class citizen, giving us a new approach to concurrency and dealing with the problems of shared state. Haskell Haskell is a strongly typed, functional programming language. Haskell’s type system is far richer than C# or Java, and allows us to push more of our application logic to compile-time safety. If it compiles, it usually works! Haskell is also a lazy language – we can work with infinite data structures. For example, in a board game we can generate the complete game tree, even if there are billions of possibilities, because the values are computed only as they are needed. Erlang Erlang is a functional language with a strong emphasis on reliability. Erlang’s approach to concurrency uses message passing instead of shared variables, with strong support from both the language itself and the virtual machine. Processes are extremely lightweight, and garbage collection doesn’t require all processes to be paused at the same time, making it feasible for a single program to use millions of processes at once, all without the mental overhead of managing shared state. The Benefits of Multilingualism By studying new languages, even if you won’t ever get the chance to use them in production, you will find yourself open to new ideas and ways of coding in your main language. For example, studying Haskell has taught me that you can do so much more with types and has changed my programming style in C#. A type represents some state a program should have, and a type should not be able to represent an invalid state. I often find myself refactoring methods like this… void SomeMethod(bool doThis, bool doThat) { if (!(doThis ^ doThat)) throw new ArgumentException(“At least one arg should be true”); if (doThis) DoThis(); if (doThat) DoThat(); } …into a type-based solution, like this: enum Action { DoThis, DoThat, Both }; void SomeMethod(Action action) { if (action == Action.DoThis || action == Action.Both) DoThis(); if (action == Action.DoThat || action == Action.Both) DoThat(); } At this point, I’ve removed the runtime exception in favor of a compile-time check. This is a trivial example, but is just one of many ideas that I’ve taken from one language and implemented in another.

    Read the article

  • Investigating Strategies For Functional Decomposition

    - by Liam McLennan
    Introducing Functional Decomposition Before I begin I must apologise. I think I am using the term ‘functional decomposition’ loosely, and probably incorrectly. For the purpose of this article I use functional decomposition to mean the recursive splitting of a large problem into increasingly smaller ones, so that the one large problem may be solved by solving a set of smaller problems. The justification for functional decomposition is that the decomposed problem is more easily solved. As software developers we recognise that the smaller pieces are more easily tested, since they do less and are more cohesive. Functional decomposition is important to all scientific pursuits. Once we understand natural selection we can start to look for humanities ancestral species, once we understand the big bang we can trace our expanding universe back to its origin. Isaac Newton acknowledged the compositional nature of his scientific achievements: If I have seen further than others, it is by standing upon the shoulders of giants   The Two Strategies For Functional Decomposition of Computer Programs Private Methods When I was working on my undergraduate degree I was taught to functionally decompose problems by using private methods. Consider the problem of painting a house. The obvious solution is to solve the problem as a single unit: public void PaintAHouse() { // all the things required to paint a house ... } We decompose the problem by breaking it into parts: public void PaintAHouse() { PaintUndercoat(); PaintTopcoat(); } private void PaintUndercoat() { // everything required to paint the undercoat } private void PaintTopcoat() { // everything required to paint the topcoat } The problem can be recursively decomposed until a sufficiently granular level of detail is reached: public void PaintAHouse() { PaintUndercoat(); PaintTopcoat(); } private void PaintUndercoat() { prepareSurface(); fetchUndercoat(); paintUndercoat(); } private void PaintTopcoat() { fetchPaint(); paintTopcoat(); } According to Wikipedia, at least one computer programmer has referred to this process as “the art of subroutining”. The practical issues that I have encountered when using private methods for decomposition are: To preserve the top level API all of the steps must be private. This means that they can’t easily be tested. The private methods often have little cohesion except that they form part of the same solution. Decomposing to Classes The alternative is to decompose large problems into multiple classes, effectively using a class instead of each private method. The API delegates to related classes, so the API is not polluted by the sub-steps of the problem, and the steps can be easily tested because they are each in their own highly cohesive class. Additionally, I think that this technique facilitates better adherence to the Single Responsibility Principle, since each class can be decomposed until it has precisely one responsibility. Revisiting my previous example using class composition: public class HousePainter { private undercoatPainter = new UndercoatPainter(); private topcoatPainter = new TopcoatPainter(); public void PaintAHouse() { undercoatPainter.Paint(); topcoatPainter.Paint(); } } Summary When decomposing a problem there is more than one way to represent the sub-problems. Using private methods keeps the logic in one place and prevents a proliferation of classes (thereby following the four rules of simple design) but the class decomposition is more easily testable and more compatible with the Single Responsibility Principle.

    Read the article

  • WiX, MSDeploy and an appealing configuration/deployment paradigm

    - by alexhildyard
    I do a lot of application and server configuration; I've done this for many years and have tended to view the complexity of this strictly in terms of the complexity of the ultimate configuration to be deployed. For example, specific APIs aside, I would tend to regard installing a server certificate as a more complex activity than, say, copying a file or adding a Registry entry.My prejudice revolved around the idea of a sequential deployment script that not only had the explicit prescription to apply a specific server configuration, but also made the implicit presumption that the server in question was in a good known state. Scripts like this fail for hundreds of reasons -- the Default Website didn't exist; the application had already been deployed; the application had already been partially deployed and failed to rollback fully, and so on. And so the problem is that the more complex the configuration activity, the more scope for error in any individual part of that activity, and therefore the greater the chance the server in question will not end up at exactly the desired configuration level.Recently I was introduced to a completely different mindset, which, for want of a better turn of phrase, I will call the "make it so" mindset. It's extremely simple both to explain and to implement. In place of the head-down, imperative script you used to use, you substitute a set of checks -- much like exception handlers -- around each configuration activity, starting with a check of the current system state. Thus the configuration logic becomes: "IF these services aren't started then start them, and IF XYZ website doesn't exist then create it, and IF these shares don't exist then create them, and IF these shares aren't permissioned in some particular way, then permission them so." This works. Really well, in my experience. Scenario 1: You want to get a system into a good known state; it's already in a good known state; you quickly realise there is nothing to do.Scenario 2: You want to get the system into a good known state; your script is flawed or the system is bust; it cannot be put into that state. You know exactly where (at least part of) the problem is and why.Scenario 3: You want to get the system into a good known state; people are fiddling around with the system just now. That's fine. You do what you can, and later you come back and try it againScenario 4: No one wants to deploy anything; they want you to prove that the previous deployment was successful. So you re-run the deployment script with the "-WhatIf" flag. It reports that there was nothing to change. There's your proof.I mentioned two technologies in the title -- MSI and MSDeploy. I am thinking specifically of the conversation that took place here. Having worked with both technologies, I think Rob Mensching's response is appropriately nuanced, and in essence the difference is this: sometimes your target is either to achieve a specific new server state, or to rollback to a known good one. Then again, your target may be to configure what you can, and to understand what you can't. Implicitly MSDeploy's "rollback" is simply to redeploy the previous version, whereas a well-crafted MSI will actively put your system into that state without further intervention. Either way, if all goes well it will leave you with a system in one of two states, whereas MSDeploy could leave your system in one of many states. The key is that MSDeploy and MSI are complementary technologies; which suits you best depends as much on Operational guidance as your Configuration remit.What I wanted to say was that I have always been for atomic, transactional-based configuration, but having worked with the "make it so" paradigm, I have been favourably impressed by the actual results. I'm tempted to put a more technical post up on this in due course.

    Read the article

  • What is a good design pattern / lib for iOS 5 to synchronize with a web service?

    - by Junto
    We are developing an iOS application that needs to synchronize with a remote server using web services. The existing web services have an "operations" style rather than REST (implemented in WCF but exposing JSON HTTP endpoints). We are unsure of how to structure the web services to best fit with iOS and would love some advice. We are also interested in how to manage the synchronization process within iOS. Without going into detailed specifics, the application allows the user to estimate repair costs at a remote site. These costs are broken down by room and item. If the user has an internet connection this data can be sent back to the server. Multiple photographs can be taken of each item, but they will be held in a separate queue, which sends when the connection is optimal (ideally wifi). Our backend application controls the unique ids for each room and item. Thus, each time we send these costs to the server, the server echoes the central database ids back, thus, that they can be synchronized in the mobile app. I have simplified this a little, since the operations contract is actually much larger, but I just want to illustrate the basic requirements without complicating matters. Firstly, the web service architecture: We currently have two operations: GetCosts and UpdateCosts. My assumption is that if we used a strict REST architecture we would need to break our single web service operations into multiple smaller services. This would make the services much more chatty and we would also have to guarantee a delivery order from the app. For example, we need to make sure that containing rooms are added before the item. Although this seems much more RESTful, our perception is that these extra calls are expensive connections (security checks, database calls, etc). Does the type of web api (operation over service focus) determine chunky vs chatty? Since this is mobile (3G), are we better handling lots of smaller messages, or a few large ones? Secondly, the iOS side. What is the current advice on how to manage data synchronization within the iOS (5) app itself. We need multiple queues and we need to guarantee delivery order in each queue (and technically, ordering between queues). The server needs to control unique ids and other properties and echo them back to the application. The application then needs to update an internal database and when re-updating, make sure the correct ids are available in the update message (essentially multiple inserts and updates in one call). Our backend has a ton of business logic operating on these cost estimates. We don't want any of this in the app itself. Currently the iOS app sends the cost data, and then the server echoes that data back with populated ids (and other data). The existing cost data is deleted and the echoed response data is added to the client database on the device. This is causing us problems, because any photos might not have been sent, but the original entity tree has been removed and replaced. Obviously updating the costs tree rather than replacing it would remove this problem, but I'm not sure if there are any nice xcode libraries out there to do such things. I welcome any advice you might have.

    Read the article

< Previous Page | 138 139 140 141 142 143 144 145 146 147 148 149  | Next Page >