Search Results

Search found 39082 results on 1564 pages for 'magic function'.

Page 5/1564 | < Previous Page | 1 2 3 4 5 6 7 8 9 10 11 12  | Next Page >

  • How to pass a member function to a function used in another member function?

    - by Tommaso Ferrari
    I found something about my problem, but I don't already understand very well. I need to do something like this: class T{ double a; public: double b; void setT(double par){ a=par; }; double funct(double par1) { return par1/a; } void exec(){ b=extfunct(funct, 10); } } double extfunct(double (*f)(double),double par2){ return f(par2)+5; } Operation and function are only for example, but the structure is that. The reason of this structure is that I have a precostituited class which finds the minimum of a gived function (it's extfunct in the example). So I have to use it on a function member of a class. I understood the difference between pointer to function and pointer to member function, but I don't understand how to write it. Thanks, and sorry for the poor explanation of the problem.

    Read the article

  • Do I have to 'remove' and 're-attach' Magic Mouse when switching between Snow Leopard and Windows 7?

    - by Teerasej
    I am deciding to buy an Apple Magic Mouse but I am wondering about the method which I have to use with Apple's Mighty Mouse: CASE 1: If I am using OS X and switching to Windows 7: Remove Mighty mouse from Bluetooth device Reboot in Windows 7 Add Mighty mouse as new Bluetooth device in Windows 7 CASE 2: If I need to switch back to OS X: Remove Mighty Mouse from Windows' Bluetooth device list Reboot in OS X Add Mighty mouse as new Bluetooth device in OS X Do I still need to proceed in this way with the Magic Mouse or not?

    Read the article

  • Oracle Hyperion si conferma leader nel Magic Quadrant Gartner 2012

    - by Andrea Cravero
    L'edizione 2012 del Gartner Magic Quadrant for Corporate Performance Management Suites conferma la leadership Oracle Hyperion, che dura ininterrotta dal 2005. Secondo Gartner, "Oracle is a Leader in CPM suites, with one of the most widely distributed solutions in the market. Oracle Hyperion Enterprise Performance Management is recognized by CFOs worldwide. The vendor has a well-established partner channel, with both large and smaller CPM SI specialists. Hyperion skills are also plentiful among the independent consultant community, given the well-established products." "Oracle continues to innovate, bringing incremental improvements across the portfolio as well as new financial close management, disclosure management and predictive planning additions. Furthermore, Oracle has improved integration of Hyperion with the Oracle BI platform, and has improved planning performance, enabling Hyperion Planning to use Oracle Exalytics In-Memory Machine." Il rapporto completo è disponibile qui: Gartner: Magic Quadrant for Corporate Performance Management Suites, 2012 Buona lettura!

    Read the article

  • Gartner: Magic Quadrant for Corporate Performance Management Suites, 2012

    - by Mike.Hallett(at)Oracle-BI&EPM
    Hyperion clearly leads the pack again in Gartner’s analysis of the CPM / EPM market, saying; “Oracle is a Leader in CPM suites, with one of the most widely distributed solutions in the market. Oracle Hyperion Enterprise Performance Management is recognized by CFOs worldwide. The vendor has a well-established partner channel, with both large and smaller CPM SI specialists. Hyperion skills are also plentiful among the independent consultant community, given the well-established products. “ “Oracle continues to innovate, bringing incremental improvements across the portfolio as well as new financial close management, disclosure management and predictive planning additions. Furthermore, Oracle has improved integration of Hyperion with the Oracle BI platform, and has improved planning performance, enabling Hyperion Planning to use Oracle Exalytics In-Memory Machine.” For the full article see here: Gartner: Magic Quadrant for Corporate Performance Management Suites, 2012 And if you missed it, here is also the MQ for BI: Gartner: Magic Quadrant for Business Intelligence Platforms, 2012

    Read the article

  • SQL SERVER – Using RAND() in User Defined Functions (UDF)

    - by pinaldave
    Here is the question I received in email. “Pinal, I am writing a function where we need to generate random password. While writing T-SQL I faced following issue. Everytime I tried to use RAND() function in my User Defined Function I am getting following error: Msg 443, Level 16, State 1, Procedure RandFn, Line 7 Invalid use of a side-effecting operator ‘rand’ within a function. Here is the simplified T-SQL code of the function which I am using: CREATE FUNCTION RandFn() RETURNS INT AS BEGIN DECLARE @rndValue INT SET @rndValue = RAND() RETURN @rndValue END GO I must use UDF so is there any workaround to use RAND function in UDF.” Here is the workaround how RAND() can be used in UDF. The scope of the blog post is not to discuss the advantages or disadvantages of the function or random function here but just to show how RAND() function can be used in UDF. RAND() function is directly not allowed to use in the UDF so we have to find alternate way to use the same function. This can be achieved by creating a VIEW which is using RAND() function and use the same VIEW in the UDF. Here is the step by step instructions. Create a VIEW using RAND function. CREATE VIEW rndView AS SELECT RAND() rndResult GO Create a UDF using the same VIEW. CREATE FUNCTION RandFn() RETURNS DECIMAL(18,18) AS BEGIN DECLARE @rndValue DECIMAL(18,18) SELECT @rndValue = rndResult FROM rndView RETURN @rndValue END GO Now execute the UDF and it will just work fine and return random result. SELECT dbo.RandFn() GO In T-SQL world, I have noticed that there are more than one solution to every problem. Is there any better solution to this question? Please post that question as a comment and I will include it with due credit. Reference: Pinal Dave (http://blog.SQLAuthority.com) Filed under: PostADay, SQL, SQL Authority, SQL Function, SQL Query, SQL Server, SQL Tips and Tricks, T SQL, Technology Tagged: technology

    Read the article

  • Vim: Custom Folding function done, custom highlighting required

    - by sixtyfootersdude
    I have defined a function in vim to properly indent folds. Ie so they look like this: Unfolded this is text also text indented text indented text not indented text folded with default function this is text also text +-- 2 lines: indented text ---------------------------- not indented text folded with my new function this is text also text ++- 2 lines: indented text ---------------------------- not indented text The only problem is the the highlighting is still like this: folded with my new function (highlighting shown with tag) this is text also text <hi> ++- 2 lines: indented text ----------------------------</hi> not indented text I would like the highlighting to start at the ++ and not at the beginning of the line. I have looked in the vim manual but could not find anything like that. One so-so solution I found was to make the background black. highlight Folded ctermbg=black ctermfg=white cterm=bold But this make folds less visible. I have tried several variations of: syn keyword Folded lines syn region Folded ... But I don't think that this is the way that folds are selected. Can anyone offer a suggestion? By the way this is my function to indent the folds: set foldmethod=indent function! MyFoldText() let lines = 1 + v:foldend - v:foldstart let ind = indent(v:foldstart) let spaces = '' let i = 0 while i < ind let i = i+1 let spaces = spaces . ' ' endwhile let linestxt = 'lines' if lines == 1 linestxt = 'line' endif return spaces . '+' . v:folddashes . ' '. lines . ' ' . linestxt . ': ' . getline(v:foldstaendfunction endfunction au BufWinEnter,BufRead,BufNewFile * set foldtext=MyFoldText() By the way thanks to njd for helping me get this function setup.

    Read the article

  • Why isn't the compiler smarter in this const function overloading problem?

    - by Frank
    The following code does not compile: #include <iostream> class Foo { std::string s; public: const std::string& GetString() const { return s; } std::string* GetString() { return &s; } }; int main(int argc, char** argv){ Foo foo; const std::string& s = foo.GetString(); // error return 0; } I get the following error: const1.cc:11: error: invalid initialization of reference of type 'const std::string&' from expression of type 'std::string* It does make some sense because foo is not of type const Foo, but just Foo, so the compiler wants to use the non-const function. But still, why can't it recognize that I want to call the const GetString function, by looking at the (type of) variable I assign it to? I found this kind of surprising.

    Read the article

  • How to convert a lambda to an std::function using templates

    - by retep998
    Basically, what I want to be able to do is take a lambda with any number of any type of parameters and convert it to an std::function. I've tried the following and neither method works. std::function([](){});//Complains that std::function is missing template parameters template <typename T> void foo(function<T> f){} foo([](){});//Complains that it cannot find a matching candidate The following code does work however, but it is not what I want because it requires explicitly stating the template parameters which does not work for generic code. std::function<void()>([](){}); I've been mucking around with functions and templates all evening and I just can't figure this out, so any help would be much appreciated. As mentioned in a comment, the reason I'm trying to do this is because I'm trying to implement currying in C++ using variadic templates. Unfortunately, this fails horribly when using lambdas. For example, I can pass a standard function using a function pointer. template <typename R, typename...A> void foo(R (*f)(A...)) {} void bar() {} int main() { foo(bar); } However, I can't figure out how to pass a lambda to such a variadic function. Why I'm interested in converting a generic lambda into an std::function is because I can do the following, but it ends up requiring that I explicitly state the template parameters to std::function which is what I am trying to avoid. template <typename R, typename...A> void foo(std::function<R(A...)>) {} int main() { foo(std::function<void()>([](){})); }

    Read the article

  • Pass a function as parameter in jQuery?

    - by thedp
    Hello, I would like to pass to a jQuery function a regular function, instead of the usual anonymous function, but I'm not sure how such a thing could be done. Instead of this: function setVersion(feature) { $.post("some.php", { abc:"abc" }, function(data){ // do something here }, "json"); } I would like to do this: function foo(data){ // do something here } function setVersion(feature) { $.post("some.php", { abc:"abc" }, foo, "json"); } Thank you.

    Read the article

  • GDL Presents: Make Web Magic | Part I

    GDL Presents: Make Web Magic | Part I Using the latest open web technologies, the developers creating some of the most inspired Chrome Experiments showcase their latest web experiments and discuss how they are making the web faster, more fun, and more open in this 3-episode hangout. Happy experimenting. Host: Paul Irish, Developer Advocate, Chrome Guest: Michael Deal From: GoogleDevelopers Views: 115 2 ratings Time: 31:44 More in Science & Technology

    Read the article

  • GDL Presents: Make Web Magic | Part II

    GDL Presents: Make Web Magic | Part II Using the latest open web technologies, the developers creating some of the most inspired Chrome Experiments showcase their latest web experiments and discuss how they are making the web faster, more fun, and open in this 3-episode hangout. Host: Paul Irish, Developer Advocate, Chrome Guest: Mark Danks From: GoogleDevelopers Views: 2 0 ratings Time: 17:41 More in Science & Technology

    Read the article

  • Function arguments VBA

    - by user1068249
    I have these three functions: When I run the first 2 functions, There's no problem, but when I run the last function (LMTD), It says 'Division by zero' yet when I debug some of the arguments have values, some don't. I know what I have to do, but I want to know why I have to do it, because it makes no sense to me. Tinn-function doesn't have Tut's arguments, so I have to add them to Tinn-function's arguments. Same goes for Tut, that doesn't know all of Tinn's arguments, and LMTD has to have both of Tinn and Tut's arguments. If I do that, it all runs smoothly. Why do I have to do this? Public Function Tinn(Tw, Qw, Qp, Q, deltaT) Tinn = (((Tw * Qw) + (Tut(Q, fd, mix) * Q)) / Qp) + deltaT End Function Public Function Tut(Q, fd, mix) Tut = Tinn(Tw, Qw, Qp, Q, deltaT) - (avgittEffektAiUiLMTD() / ((Q * fd * mix) / 3600)) End Function Public Function LMTD(Tsjo) LMTD = ((Tinn(Tw, Qw, Qp, Q, deltaT) - Tsjo) - (Tut(Q, fd, mix) - Tsjo)) / (WorksheetFunction.Ln ((Tinn(Tw, Qw, Qp, Q, deltaT) - Tsjo) / (Tut(Q, fd, mix) - Tsjo))) End Function

    Read the article

  • How does dereferencing of a function pointer happen?

    - by eSKay
    Why and how does dereferencing a function pointer just "do nothing"? This is what I am talking about: #include<stdio.h> void hello() { printf("hello"); } int main(void) { (*****hello)(); } From a comment over here: function pointers dereference just fine, but the resulting function designator will be immediately converted back to a function pointer And from an answer here: Dereferencing (in way you think) a function's pointer means: accessing a CODE memory as it would be a DATA memory. Function pointer isn't suppose to be dereferenced in that way. Instead, it is called. I would use a name "dereference" side by side with "call". It's OK. Anyway: C is designed in such a way that both function name identifier as well as variable holding function's pointer mean the same: address to CODE memory. And it allows to jump to that memory by using call () syntax either on an identifier or variable. How exactly does dereferencing of a function pointer work?

    Read the article

  • Does the concept of "magic number" change from language to language?

    - by Gerardo Marset
    Take the following code in C/C++, for example: int foo[] = {0, 0, 0, 0}; No magic numbers, right? Now, the Python "equivalent" of that would be: foo = [0, 0, 0, 0] Still no magic numbers. However, in Python, that same thing can be written like this: foo = [0] * 4 And now we DO have a magic number. Or do we? I'm guessing this and other similar things are present on these and other languages.

    Read the article

  • Node.js Adventure - When Node Flying in Wind

    - by Shaun
    In the first post of this series I mentioned some popular modules in the community, such as underscore, async, etc.. I also listed a module named “Wind (zh-CN)”, which is created by one of my friend, Jeff Zhao (zh-CN). Now I would like to use a separated post to introduce this module since I feel it brings a new async programming style in not only Node.js but JavaScript world. If you know or heard about the new feature in C# 5.0 called “async and await”, or you learnt F#, you will find the “Wind” brings the similar async programming experience in JavaScript. By using “Wind”, we can write async code that looks like the sync code. The callbacks, async stats and exceptions will be handled by “Wind” automatically and transparently.   What’s the Problem: Dense “Callback” Phobia Let’s firstly back to my second post in this series. As I mentioned in that post, when we wanted to read some records from SQL Server we need to open the database connection, and then execute the query. In Node.js all IO operation are designed as async callback pattern which means when the operation was done, it will invoke a function which was taken from the last parameter. For example the database connection opening code would be like this. 1: sql.open(connectionString, function(error, conn) { 2: if(error) { 3: // some error handling code 4: } 5: else { 6: // connection opened successfully 7: } 8: }); And then if we need to query the database the code would be like this. It nested in the previous function. 1: sql.open(connectionString, function(error, conn) { 2: if(error) { 3: // some error handling code 4: } 5: else { 6: // connection opened successfully 7: conn.queryRaw(command, function(error, results) { 8: if(error) { 9: // failed to execute this command 10: } 11: else { 12: // records retrieved successfully 13: } 14: }; 15: } 16: }); Assuming if we need to copy some data from this database to another then we need to open another connection and execute the command within the function under the query function. 1: sql.open(connectionString, function(error, conn) { 2: if(error) { 3: // some error handling code 4: } 5: else { 6: // connection opened successfully 7: conn.queryRaw(command, function(error, results) { 8: if(error) { 9: // failed to execute this command 10: } 11: else { 12: // records retrieved successfully 13: target.open(targetConnectionString, function(error, t_conn) { 14: if(error) { 15: // connect failed 16: } 17: else { 18: t_conn.queryRaw(copy_command, function(error, results) { 19: if(error) { 20: // copy failed 21: } 22: else { 23: // and then, what do you want to do now... 24: } 25: }; 26: } 27: }; 28: } 29: }; 30: } 31: }); This is just an example. In the real project the logic would be more complicated. This means our application might be messed up and the business process will be fragged by many callback functions. I would like call this “Dense Callback Phobia”. This might be a challenge how to make code straightforward and easy to read, something like below. 1: try 2: { 3: // open source connection 4: var s_conn = sqlConnect(s_connectionString); 5: // retrieve data 6: var results = sqlExecuteCommand(s_conn, s_command); 7: 8: // open target connection 9: var t_conn = sqlConnect(t_connectionString); 10: // prepare the copy command 11: var t_command = getCopyCommand(results); 12: // execute the copy command 13: sqlExecuteCommand(s_conn, t_command); 14: } 15: catch (ex) 16: { 17: // error handling 18: }   What’s the Problem: Sync-styled Async Programming Similar as the previous problem, the callback-styled async programming model makes the upcoming operation as a part of the current operation, and mixed with the error handling code. So it’s very hard to understand what on earth this code will do. And since Node.js utilizes non-blocking IO mode, we cannot invoke those operations one by one, as they will be executed concurrently. For example, in this post when I tried to copy the records from Windows Azure SQL Database (a.k.a. WASD) to Windows Azure Table Storage, if I just insert the data into table storage one by one and then print the “Finished” message, I will see the message shown before the data had been copied. This is because all operations were executed at the same time. In order to make the copy operation and print operation executed synchronously I introduced a module named “async” and the code was changed as below. 1: async.forEach(results.rows, 2: function (row, callback) { 3: var resource = { 4: "PartitionKey": row[1], 5: "RowKey": row[0], 6: "Value": row[2] 7: }; 8: client.insertEntity(tableName, resource, function (error) { 9: if (error) { 10: callback(error); 11: } 12: else { 13: console.log("entity inserted."); 14: callback(null); 15: } 16: }); 17: }, 18: function (error) { 19: if (error) { 20: error["target"] = "insertEntity"; 21: res.send(500, error); 22: } 23: else { 24: console.log("all done."); 25: res.send(200, "Done!"); 26: } 27: }); It ensured that the “Finished” message will be printed when all table entities had been inserted. But it cannot promise that the records will be inserted in sequence. It might be another challenge to make the code looks like in sync-style? 1: try 2: { 3: forEach(row in rows) { 4: var entity = { /* ... */ }; 5: tableClient.insert(tableName, entity); 6: } 7:  8: console.log("Finished"); 9: } 10: catch (ex) { 11: console.log(ex); 12: }   How “Wind” Helps “Wind” is a JavaScript library which provides the control flow with plain JavaScript for asynchronous programming (and more) without additional pre-compiling steps. It’s available in NPM so that we can install it through “npm install wind”. Now let’s create a very simple Node.js application as the example. This application will take some website URLs from the command arguments and tried to retrieve the body length and print them in console. Then at the end print “Finish”. I’m going to use “request” module to make the HTTP call simple so I also need to install by the command “npm install request”. The code would be like this. 1: var request = require("request"); 2:  3: // get the urls from arguments, the first two arguments are `node.exe` and `fetch.js` 4: var args = process.argv.splice(2); 5:  6: // main function 7: var main = function() { 8: for(var i = 0; i < args.length; i++) { 9: // get the url 10: var url = args[i]; 11: // send the http request and try to get the response and body 12: request(url, function(error, response, body) { 13: if(!error && response.statusCode == 200) { 14: // log the url and the body length 15: console.log( 16: "%s: %d.", 17: response.request.uri.href, 18: body.length); 19: } 20: else { 21: // log error 22: console.log(error); 23: } 24: }); 25: } 26: 27: // finished 28: console.log("Finished"); 29: }; 30:  31: // execute the main function 32: main(); Let’s execute this application. (I made them in multi-lines for better reading.) 1: node fetch.js 2: "http://www.igt.com/us-en.aspx" 3: "http://www.igt.com/us-en/games.aspx" 4: "http://www.igt.com/us-en/cabinets.aspx" 5: "http://www.igt.com/us-en/systems.aspx" 6: "http://www.igt.com/us-en/interactive.aspx" 7: "http://www.igt.com/us-en/social-gaming.aspx" 8: "http://www.igt.com/support.aspx" Below is the output. As you can see the finish message was printed at the beginning, and the pages’ length retrieved in a different order than we specified. This is because in this code the request command, console logging command are executed asynchronously and concurrently. Now let’s introduce “Wind” to make them executed in order, which means it will request the websites one by one, and print the message at the end.   First of all we need to import the “Wind” package and make sure the there’s only one global variant named “Wind”, and ensure it’s “Wind” instead of “wind”. 1: var Wind = require("wind");   Next, we need to tell “Wind” which code will be executed asynchronously so that “Wind” can control the execution process. In this case the “request” operation executed asynchronously so we will create a “Task” by using a build-in helps function in “Wind” named Wind.Async.Task.create. 1: var requestBodyLengthAsync = function(url) { 2: return Wind.Async.Task.create(function(t) { 3: request(url, function(error, response, body) { 4: if(error || response.statusCode != 200) { 5: t.complete("failure", error); 6: } 7: else { 8: var data = 9: { 10: uri: response.request.uri.href, 11: length: body.length 12: }; 13: t.complete("success", data); 14: } 15: }); 16: }); 17: }; The code above created a “Task” from the original request calling code. In “Wind” a “Task” means an operation will be finished in some time in the future. A “Task” can be started by invoke its start() method, but no one knows when it actually will be finished. The Wind.Async.Task.create helped us to create a task. The only parameter is a function where we can put the actual operation in, and then notify the task object it’s finished successfully or failed by using the complete() method. In the code above I invoked the request method. If it retrieved the response successfully I set the status of this task as “success” with the URL and body length. If it failed I set this task as “failure” and pass the error out.   Next, we will change the main() function. In “Wind” if we want a function can be controlled by Wind we need to mark it as “async”. This should be done by using the code below. 1: var main = eval(Wind.compile("async", function() { 2: })); When the application is running, Wind will detect “eval(Wind.compile(“async”, function” and generate an anonymous code from the body of this original function. Then the application will run the anonymous code instead of the original one. In our example the main function will be like this. 1: var main = eval(Wind.compile("async", function() { 2: for(var i = 0; i < args.length; i++) { 3: try 4: { 5: var result = $await(requestBodyLengthAsync(args[i])); 6: console.log( 7: "%s: %d.", 8: result.uri, 9: result.length); 10: } 11: catch (ex) { 12: console.log(ex); 13: } 14: } 15: 16: console.log("Finished"); 17: })); As you can see, when I tried to request the URL I use a new command named “$await”. It tells Wind, the operation next to $await will be executed asynchronously, and the main thread should be paused until it finished (or failed). So in this case, my application will be pause when the first response was received, and then print its body length, then try the next one. At the end, print the finish message.   Finally, execute the main function. The full code would be like this. 1: var request = require("request"); 2: var Wind = require("wind"); 3:  4: var args = process.argv.splice(2); 5:  6: var requestBodyLengthAsync = function(url) { 7: return Wind.Async.Task.create(function(t) { 8: request(url, function(error, response, body) { 9: if(error || response.statusCode != 200) { 10: t.complete("failure", error); 11: } 12: else { 13: var data = 14: { 15: uri: response.request.uri.href, 16: length: body.length 17: }; 18: t.complete("success", data); 19: } 20: }); 21: }); 22: }; 23:  24: var main = eval(Wind.compile("async", function() { 25: for(var i = 0; i < args.length; i++) { 26: try 27: { 28: var result = $await(requestBodyLengthAsync(args[i])); 29: console.log( 30: "%s: %d.", 31: result.uri, 32: result.length); 33: } 34: catch (ex) { 35: console.log(ex); 36: } 37: } 38: 39: console.log("Finished"); 40: })); 41:  42: main().start();   Run our new application. At the beginning we will see the compiled and generated code by Wind. Then we can see the pages were requested one by one, and at the end the finish message was printed. Below is the code Wind generated for us. As you can see the original code, the output code were shown. 1: // Original: 2: function () { 3: for(var i = 0; i < args.length; i++) { 4: try 5: { 6: var result = $await(requestBodyLengthAsync(args[i])); 7: console.log( 8: "%s: %d.", 9: result.uri, 10: result.length); 11: } 12: catch (ex) { 13: console.log(ex); 14: } 15: } 16: 17: console.log("Finished"); 18: } 19:  20: // Compiled: 21: /* async << function () { */ (function () { 22: var _builder_$0 = Wind.builders["async"]; 23: return _builder_$0.Start(this, 24: _builder_$0.Combine( 25: _builder_$0.Delay(function () { 26: /* var i = 0; */ var i = 0; 27: /* for ( */ return _builder_$0.For(function () { 28: /* ; i < args.length */ return i < args.length; 29: }, function () { 30: /* ; i ++) { */ i ++; 31: }, 32: /* try { */ _builder_$0.Try( 33: _builder_$0.Delay(function () { 34: /* var result = $await(requestBodyLengthAsync(args[i])); */ return _builder_$0.Bind(requestBodyLengthAsync(args[i]), function (result) { 35: /* console.log("%s: %d.", result.uri, result.length); */ console.log("%s: %d.", result.uri, result.length); 36: return _builder_$0.Normal(); 37: }); 38: }), 39: /* } catch (ex) { */ function (ex) { 40: /* console.log(ex); */ console.log(ex); 41: return _builder_$0.Normal(); 42: /* } */ }, 43: null 44: ) 45: /* } */ ); 46: }), 47: _builder_$0.Delay(function () { 48: /* console.log("Finished"); */ console.log("Finished"); 49: return _builder_$0.Normal(); 50: }) 51: ) 52: ); 53: /* } */ })   How Wind Works Someone may raise a big concern when you find I utilized “eval” in my code. Someone may assume that Wind utilizes “eval” to execute some code dynamically while “eval” is very low performance. But I would say, Wind does NOT use “eval” to run the code. It only use “eval” as a flag to know which code should be compiled at runtime. When the code was firstly been executed, Wind will check and find “eval(Wind.compile(“async”, function”. So that it knows this function should be compiled. Then it utilized parse-js to analyze the inner JavaScript and generated the anonymous code in memory. Then it rewrite the original code so that when the application was running it will use the anonymous one instead of the original one. Since the code generation was done at the beginning of the application was started, in the future no matter how long our application runs and how many times the async function was invoked, it will use the generated code, no need to generate again. So there’s no significant performance hurt when using Wind.   Wind in My Previous Demo Let’s adopt Wind into one of my previous demonstration and to see how it helps us to make our code simple, straightforward and easy to read and understand. In this post when I implemented the functionality that copied the records from my WASD to table storage, the logic would be like this. 1, Open database connection. 2, Execute a query to select all records from the table. 3, Recreate the table in Windows Azure table storage. 4, Create entities from each of the records retrieved previously, and then insert them into table storage. 5, Finally, show message as the HTTP response. But as the image below, since there are so many callbacks and async operations, it’s very hard to understand my logic from the code. Now let’s use Wind to rewrite our code. First of all, of course, we need the Wind package. Then we need to include the package files into project and mark them as “Copy always”. Add the Wind package into the source code. Pay attention to the variant name, you must use “Wind” instead of “wind”. 1: var express = require("express"); 2: var async = require("async"); 3: var sql = require("node-sqlserver"); 4: var azure = require("azure"); 5: var Wind = require("wind"); Now we need to create some async functions by using Wind. All async functions should be wrapped so that it can be controlled by Wind which are open database, retrieve records, recreate table (delete and create) and insert entity in table. Below are these new functions. All of them are created by using Wind.Async.Task.create. 1: sql.openAsync = function (connectionString) { 2: return Wind.Async.Task.create(function (t) { 3: sql.open(connectionString, function (error, conn) { 4: if (error) { 5: t.complete("failure", error); 6: } 7: else { 8: t.complete("success", conn); 9: } 10: }); 11: }); 12: }; 13:  14: sql.queryAsync = function (conn, query) { 15: return Wind.Async.Task.create(function (t) { 16: conn.queryRaw(query, function (error, results) { 17: if (error) { 18: t.complete("failure", error); 19: } 20: else { 21: t.complete("success", results); 22: } 23: }); 24: }); 25: }; 26:  27: azure.recreateTableAsync = function (tableName) { 28: return Wind.Async.Task.create(function (t) { 29: client.deleteTable(tableName, function (error, successful, response) { 30: console.log("delete table finished"); 31: client.createTableIfNotExists(tableName, function (error, successful, response) { 32: console.log("create table finished"); 33: if (error) { 34: t.complete("failure", error); 35: } 36: else { 37: t.complete("success", null); 38: } 39: }); 40: }); 41: }); 42: }; 43:  44: azure.insertEntityAsync = function (tableName, entity) { 45: return Wind.Async.Task.create(function (t) { 46: client.insertEntity(tableName, entity, function (error, entity, response) { 47: if (error) { 48: t.complete("failure", error); 49: } 50: else { 51: t.complete("success", null); 52: } 53: }); 54: }); 55: }; Then in order to use these functions we will create a new function which contains all steps for data copying. 1: var copyRecords = eval(Wind.compile("async", function (req, res) { 2: try { 3: } 4: catch (ex) { 5: console.log(ex); 6: res.send(500, "Internal error."); 7: } 8: })); Let’s execute steps one by one with the “$await” keyword introduced by Wind so that it will be invoked in sequence. First is to open the database connection. 1: var copyRecords = eval(Wind.compile("async", function (req, res) { 2: try { 3: // connect to the windows azure sql database 4: var conn = $await(sql.openAsync(connectionString)); 5: console.log("connection opened"); 6: } 7: catch (ex) { 8: console.log(ex); 9: res.send(500, "Internal error."); 10: } 11: })); Then retrieve all records from the database connection. 1: var copyRecords = eval(Wind.compile("async", function (req, res) { 2: try { 3: // connect to the windows azure sql database 4: var conn = $await(sql.openAsync(connectionString)); 5: console.log("connection opened"); 6: // retrieve all records from database 7: var results = $await(sql.queryAsync(conn, "SELECT * FROM [Resource]")); 8: console.log("records selected. count = %d", results.rows.length); 9: } 10: catch (ex) { 11: console.log(ex); 12: res.send(500, "Internal error."); 13: } 14: })); After recreated the table, we need to create the entities and insert them into table storage. 1: var copyRecords = eval(Wind.compile("async", function (req, res) { 2: try { 3: // connect to the windows azure sql database 4: var conn = $await(sql.openAsync(connectionString)); 5: console.log("connection opened"); 6: // retrieve all records from database 7: var results = $await(sql.queryAsync(conn, "SELECT * FROM [Resource]")); 8: console.log("records selected. count = %d", results.rows.length); 9: if (results.rows.length > 0) { 10: // recreate the table 11: $await(azure.recreateTableAsync(tableName)); 12: console.log("table created"); 13: // insert records in table storage one by one 14: for (var i = 0; i < results.rows.length; i++) { 15: var entity = { 16: "PartitionKey": results.rows[i][1], 17: "RowKey": results.rows[i][0], 18: "Value": results.rows[i][2] 19: }; 20: $await(azure.insertEntityAsync(tableName, entity)); 21: console.log("entity inserted"); 22: } 23: } 24: } 25: catch (ex) { 26: console.log(ex); 27: res.send(500, "Internal error."); 28: } 29: })); Finally, send response back to the browser. 1: var copyRecords = eval(Wind.compile("async", function (req, res) { 2: try { 3: // connect to the windows azure sql database 4: var conn = $await(sql.openAsync(connectionString)); 5: console.log("connection opened"); 6: // retrieve all records from database 7: var results = $await(sql.queryAsync(conn, "SELECT * FROM [Resource]")); 8: console.log("records selected. count = %d", results.rows.length); 9: if (results.rows.length > 0) { 10: // recreate the table 11: $await(azure.recreateTableAsync(tableName)); 12: console.log("table created"); 13: // insert records in table storage one by one 14: for (var i = 0; i < results.rows.length; i++) { 15: var entity = { 16: "PartitionKey": results.rows[i][1], 17: "RowKey": results.rows[i][0], 18: "Value": results.rows[i][2] 19: }; 20: $await(azure.insertEntityAsync(tableName, entity)); 21: console.log("entity inserted"); 22: } 23: // send response 24: console.log("all done"); 25: res.send(200, "All done!"); 26: } 27: } 28: catch (ex) { 29: console.log(ex); 30: res.send(500, "Internal error."); 31: } 32: })); If we compared with the previous code we will find now it became more readable and much easy to understand. It’s very easy to know what this function does even though without any comments. When user go to URL “/was/copyRecords” we will execute the function above. The code would be like this. 1: app.get("/was/copyRecords", function (req, res) { 2: copyRecords(req, res).start(); 3: }); And below is the logs printed in local compute emulator console. As we can see the functions executed one by one and then finally the response back to me browser.   Scaffold Functions in Wind Wind provides not only the async flow control and compile functions, but many scaffold methods as well. We can build our async code more easily by using them. I’m going to introduce some basic scaffold functions here. In the code above I created some functions which wrapped from the original async function such as open database, create table, etc.. All of them are very similar, created a task by using Wind.Async.Task.create, return error or result object through Task.complete function. In fact, Wind provides some functions for us to create task object from the original async functions. If the original async function only has a callback parameter, we can use Wind.Async.Binding.fromCallback method to get the task object directly. For example the code below returned the task object which wrapped the file exist check function. 1: var Wind = require("wind"); 2: var fs = require("fs"); 3:  4: fs.existsAsync = Wind.Async.Binding.fromCallback(fs.exists); In Node.js a very popular async function pattern is that, the first parameter in the callback function represent the error object, and the other parameters is the return values. In this case we can use another build-in function in Wind named Wind.Async.Binding.fromStandard. For example, the open database function can be created from the code below. 1: sql.openAsync = Wind.Async.Binding.fromStandard(sql.open); 2:  3: /* 4: sql.openAsync = function (connectionString) { 5: return Wind.Async.Task.create(function (t) { 6: sql.open(connectionString, function (error, conn) { 7: if (error) { 8: t.complete("failure", error); 9: } 10: else { 11: t.complete("success", conn); 12: } 13: }); 14: }); 15: }; 16: */ When I was testing the scaffold functions under Wind.Async.Binding I found for some functions, such as the Azure SDK insert entity function, cannot be processed correctly. So I personally suggest writing the wrapped method manually.   Another scaffold method in Wind is the parallel tasks coordination. In this example, the steps of open database, retrieve records and recreated table should be invoked one by one, but it can be executed in parallel when copying data from database to table storage. In Wind there’s a scaffold function named Task.whenAll which can be used here. Task.whenAll accepts a list of tasks and creates a new task. It will be returned only when all tasks had been completed, or any errors occurred. For example in the code below I used the Task.whenAll to make all copy operation executed at the same time. 1: var copyRecordsInParallel = eval(Wind.compile("async", function (req, res) { 2: try { 3: // connect to the windows azure sql database 4: var conn = $await(sql.openAsync(connectionString)); 5: console.log("connection opened"); 6: // retrieve all records from database 7: var results = $await(sql.queryAsync(conn, "SELECT * FROM [Resource]")); 8: console.log("records selected. count = %d", results.rows.length); 9: if (results.rows.length > 0) { 10: // recreate the table 11: $await(azure.recreateTableAsync(tableName)); 12: console.log("table created"); 13: // insert records in table storage in parallal 14: var tasks = new Array(results.rows.length); 15: for (var i = 0; i < results.rows.length; i++) { 16: var entity = { 17: "PartitionKey": results.rows[i][1], 18: "RowKey": results.rows[i][0], 19: "Value": results.rows[i][2] 20: }; 21: tasks[i] = azure.insertEntityAsync(tableName, entity); 22: } 23: $await(Wind.Async.Task.whenAll(tasks)); 24: // send response 25: console.log("all done"); 26: res.send(200, "All done!"); 27: } 28: } 29: catch (ex) { 30: console.log(ex); 31: res.send(500, "Internal error."); 32: } 33: })); 34:  35: app.get("/was/copyRecordsInParallel", function (req, res) { 36: copyRecordsInParallel(req, res).start(); 37: });   Besides the task creation and coordination, Wind supports the cancellation solution so that we can send the cancellation signal to the tasks. It also includes exception solution which means any exceptions will be reported to the caller function.   Summary In this post I introduced a Node.js module named Wind, which created by my friend Jeff Zhao. As you can see, different from other async library and framework, adopted the idea from F# and C#, Wind utilizes runtime code generation technology to make it more easily to write async, callback-based functions in a sync-style way. By using Wind there will be almost no callback, and the code will be very easy to understand. Currently Wind is still under developed and improved. There might be some problems but the author, Jeff, should be very happy and enthusiastic to learn your problems, feedback, suggestion and comments. You can contact Jeff by - Email: [email protected] - Group: https://groups.google.com/d/forum/windjs - GitHub: https://github.com/JeffreyZhao/wind/issues   Source code can be download here.   Hope this helps, Shaun All documents and related graphics, codes are provided "AS IS" without warranty of any kind. Copyright © Shaun Ziyan Xu. This work is licensed under the Creative Commons License.

    Read the article

  • Does Apple Magic Mouse fully work on Windows 7 x86/x64?

    - by Sorin Sbarnea
    I would like to know if Apple Magic Mouse works on Windows 7 (x86/x64) on non-Apple computers. Here are some checklists: x64 compatibility left click right click middle button? vertical scroll horizontal scroll bind additional gestures to keystrokes? are any usage problems? In case it works please advise on how to get the drivers.

    Read the article

  • Friday Fun: Haunted House – Quest for the Magic Book

    - by Asian Angel
    In this week’s game you embark on a quest into a haunted house to search for a magic book…a search that will have to be conducted room by room in order to successfully make your way through the house. Will your quest be successful or will you go home empty handed? What Is the Purpose of the “Do Not Cover This Hole” Hole on Hard Drives? How To Log Into The Desktop, Add a Start Menu, and Disable Hot Corners in Windows 8 HTG Explains: Why You Shouldn’t Use a Task Killer On Android

    Read the article

  • JavaScript setTimeout setInterval within one function

    - by dagoof
    I think I might be overtired but I cannot for the life of me make sense of this, and I think it's due to a lack of knowledge of javascript var itv=function(){ return setInterval(function(){ sys.puts('interval'); }, 1000); } var tout=function(itv){ return setTimeout(function(){ sys.puts('timeout'); clearInterval(itv); }, 5500); } With these two functions I can call a=tout(itv()); and get a looping timer to run for 5.5 seconds and then exit, essentially. By my logic, this should work but it simply is not var dotime=function(){ return setTimeout(function(){ clearInterval(function(){ return setInterval(function(){ sys.puts("interval"); }, 1000); }); }, 5500); } any insight in this matter would be appreciated.

    Read the article

  • jQuery calling a predefined function

    - by Mircea
    I have a big function defined and executed on an element click. Is there a way to execute again that "big" function later, call it somehow, without writing it again? $('#element').click(function(big){ some big function ... ... }); $('#another_element').click(function(){ this should execute the previous "big" function }); Thanx.

    Read the article

  • In javascript, how do I find out the name of a function from within that function?

    - by roryhewitt
    The title should make it clear. Is it possible from within a function to determine the name of that function? Basically, I'm adding some debugging code to a number of functions, and I'd like to be able to simply add the following line inside every function: if (global_debugOn) alert("Processing function " + function-name); How can I get 'function-name'? Yes, obviously I could simply type in the function name (after all I'm typing in the whole alert bit), but that's a hassle to do, especially if there is a nice simple way of retrieving it dynamically. Plus, as function names change during development, I'd like to keep it up-to-date... I hoped that maybe the arguments attribute might hold this (e.g. arguments[0], like in C), but I couldn't get it to work. I'm not even sure if arguments works anyway. Thanks! Rory

    Read the article

  • How to structure javascript callback so that function scope is maintained properly

    - by Chetan
    I'm using XMLHttpRequest, and I want to access a local variable in the success callback function. Here is the code: function getFileContents(filePath, callbackFn) { var xhr = new XMLHttpRequest(); xhr.onreadystatechange = function() { if (xhr.readyState == 4) { callbackFn(xhr.responseText); } } xhr.open("GET", chrome.extension.getURL(filePath), true); xhr.send(); } And I want to call it like this: var test = "lol"; getFileContents("hello.js", function(data) { alert(test); }); Here, test would be out of the scope of the callback function, since only the enclosing function's variables are accessible inside the callback function. What is the best way to pass test to the callback function so the alert(test); will display test correctly?

    Read the article

< Previous Page | 1 2 3 4 5 6 7 8 9 10 11 12  | Next Page >