From de6069d3d60eb2da7ee945c38fbe8d7249c66a6a Mon Sep 17 00:00:00 2001 From: Sonia Keys Date: Tue, 13 Aug 2013 13:52:13 -0400 Subject: Go first draft --- go.html.markdown | 423 +++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 423 insertions(+) create mode 100644 go.html.markdown diff --git a/go.html.markdown b/go.html.markdown new file mode 100644 index 00000000..9888b37d --- /dev/null +++ b/go.html.markdown @@ -0,0 +1,423 @@ +--- +name: Go +category: language +language: Go +filename: learngo.go +contributors: + - ["Sonia Keys", "https://github.com/soniakeys"] +--- + +Go was created out of the need to get work done. It's not the latest trend +in computer science, but it is the newest fastest way to solve real-world +problems. + +It has familiar concepts of imperative languages with static typing. +It's fast to compile and fast to execute, it adds easy-to-understand +concurrency to leverage today's multi-core CPUs, and has features to +help with large-scale programming. + +Go comes with a great standard library and an enthusiastic community. + +```Go +// Single line comment +/* Multi- + line comment */ + +// A package clause starts every source file. +// Main is a special name declaring an executable rather than a library. +package main + +// An import declaration comes next. It declares library packages referenced +// in this file. The list must be exactly correct! Missing or unused packages +// are errors, not warnings. +import ( + "fmt" // A package in the Go standard library + "net/http" // Yes, a web server! + "strconv" // String conversions +) + +// A function definition. Main is special. It is the entry point for the +// executable program. Love it or hate it, Go uses brace brackets. +func main() { + // Println is a function that outputs a line to stdout. It can be + // called here because fmt has been imported and the function name + // "Println" is upper case. Symbols starting with an upper case letter + // are publicly visible. No other special syntax is needed to export + // something from a package. + // To call Println, qualify it with the package name, fmt. + fmt.Println("Hello world!") + + // Call another function within this package. + beyondHello() +} + +// Idiomatic Go uses camel case. Functions have parameters in parentheses. +// If there are no parameters, empty parens are still required. +func beyondHello() { + var x int // Variable declaration. Variables must be declared before use. + x = 3 // Variable assignment. + // "Short" declarations use := syntax to declare and assign, infering the + // type from the right hand side as much as possible and using some + // defaults where the rhs could be interpreted different ways. + // Idiomatic Go uses short declarations in preference to var keyword. + y := 4 + sum, prod := learnMultiple(x, y) // function returns two values + fmt.Println("sum:", sum, "prod:", prod) // simple output + learnTypes() // < y minutes, learn more! +} + +// Functions can have parameters and (multiple!) return values. +// In declarations, the symbol precedes the type, and the type does not have +// to be repeated if it is the same for multiple symbols in a row. +func learnMultiple(x, y int) (sum, prod int) { + return x + y, x * y // return two values +} + +// Some built-in types and literals. +func learnTypes() { + // Short declaration usually gives you what you want. + s := "Learn Go!" // string type + + s2 := `A "raw" string literal +can include line breaks.` // same string type + + // non-ASCII literal. Go source is UTF-8. + g := 'Σ' // rune type, an alias for uint32, holds a UTF-8 code point + + f := 3.14195 // float64, an IEEE-754 64-bit floating point number + c := 3 + 4i // complex128, represented internally with two float64s + + // You can use var syntax with an initializer if you want + // something other than the default that a short declaration gives you. + var u uint = 7 // unsigned, but implementation dependent size as with int + var pi float32 = 22. / 7 + + // Or more idiomatically, use conversion syntax with a short declaration. + n := byte('\n') // byte is an alias for uint8 + + // Arrays have size fixed at compile time. + var a4 [4]int // an array of 4 ints, initialized to all 0 + a3 := [...]int{3, 1, 5} // an array of 3 ints, initialized as shown + + // Slices have dynamic size. Arrays and slices each have advantages + // but use cases for slices are much more common. + s3 := []int{4, 5, 9} // compare to a3. no ellipsis here + s4 := make([]int, 4) // allocates slice of 4 ints, initialized to all 0 + var d2 [][]float64 // declaration only, nothing allocated here + bs := []byte("a slice") // type conversion syntax + + p, q := learnMemory() // A little side bar. + // Did you read it? This short declaration declares p and q to be of + // type pointer to int. P is now pointing into a block of of 20 ints, but + // the only one accessible is the one that p is pointing at. There is + // no p++ to get at the next one. + fmt.Println(*p, *q) // * follows a pointer. This prints two ints. + + // Maps are a dynamically growable associative array type, like the + // hash or dictionary types of some other languages. + m := map[string]int{"three": 3, "four": 4} + m["one"] = 1 + + // Unused variables are an error in Go. + // The underbar lets you "use" a variable but discard its value. + _, _, _, _, _, _, _, _, _ = s2, g, f, u, pi, n, a3, s4, bs + // Output of course counts as using a variable. + fmt.Println(s, c, a4, s3, d2, m) + + learnFlowControl() // back in the flow +} + +// Go is fully garbage collected. It has pointers but no pointer arithmetic. +// You can make a mistake with a nil pointer, but not by incrementing a pointer. +func learnMemory() (p, q *int) { + // Named return values p and q have type pointer to int. They are + // initialized to nil at this point. Evaluating *p or *q here would cause + // a panic--a run time error. + p = new(int) // built-in function new allocates memory. + // The allocated int is initialized to 0, p is no longer nil. + s := make([]int, 20) // allocate 20 ints as a single block of memory + s[3] = 7 // assign one of them + r := -2 // declare another local variable + return &s[3], &r // Oh my. + // The line above returns two values, yes, and both of the expressions + // are valid. & takes the address of an object. Elements of a slice are + // addressable, and so are local variables. Built-in functions new and + // make explicitly allocate memory, but local objects can be allocated + // as needed. Here memory for r will be still be referenced after the + // function returns so it will be allocated as well. The int allocated + // with new on the other hand will no longer be referenced and can be + // garbage collected as needed by the Go runtime. The memory allocated + // with make will still be referenced at that one element, and so it + // cannot be garbage collected. All 20 ints remain in memory because + // one of them is still referenced. +} + +func expensiveComputation() int { + return 1e6 +} + +func learnFlowControl() { + // If statements require brace brackets, and do not require parens. + if true { + fmt.Println("told ya") + } + // This is how we format the brace brackets. Formatting is standardized + // by the command line command "go fmt." Everybody does it. You will + // suffer endless disparaging remarks until you conform as well. + if false { + // pout + } else { + // gloat + } + // If statements can be chained of course, but it's idiomatic to use + // the handy switch statement instead. + x := 1 + switch x { + case 0: + case 1: + // cases don't "fall through" + case 2: + // unreached + } + // Like if, for doesn't use parens either. The scope of a variable + // declared in the first clause of the for statement is the statement + // and block. This x shadows the x declared above, but goes out of + // scope after the for block. + for x := 0; x < 3; x++ { // ++ is a statement + fmt.Println("iteration", x) + } + // x == 1 here. + + // For is the only loop statement in Go, but it has alternate forms. + for { // infinite loop + break // just kidding + continue // unreached + } + // The initial assignment of the for statement is handy enough that Go + // if statements can have one as well. Just like in the for statement, + // the := here means to declare and assign y first, then test y > x. + // The scope of y is limited to the if statement and block. + if y := expensiveComputation(); y > x { + x = y + } + // Functions are first class objects and function literals are handy. + // Function literals are closures. + xBig := func() bool { + return x > 100 // references x declared above switch statement. + } + fmt.Println("xBig:", xBig()) // true (we last assigned 1e6 to x) + x /= 1e5 // this makes it == 10 + fmt.Println("xBig:", xBig()) // false now + + // When you need it, you'll love it. Actually Go's goto has been reformed + // a bit to avoid indeterminate states. You can't jump around variable + // declarations and you can't jump into blocks. + goto love +love: + + learnInterfaces() // Good stuff coming up! +} + +// An interface is a list of functionality that a type supports. Notably +// missing from an interface definition is any declaration of which types +// implement the interface. Types simply implement an interface or they don't. +// +// An interface can have any number of methods, but it's actually common +// for an interface to have only single method. It is idiomatic in this +// case for the single method to be named with some action, and for the +// interface name to end in "er." +// +// An interface definition is one kind of a type definition. Interface is +// a built in type. Stringer is defined here as an interface type with one +// method, String. +type Stringer interface { + String() string +} + +// Struct is another built in type. A struct aggregates "fields." +// Pair here has two fields, ints named x and y. +type pair struct { + x, y int +} + +// User defined types can have "methods." These are functions that operate +// in the context of an instance of the user defined type. The instance +// is called the "receiver" and is identified with a declaration just in front +// of the method name. The receiver here is "p." In most ways the receiver +// works just like a function parameter. +// +// This String method has the same name and return value as the String method +// of the Stringer interface. Further, String is the only method of Stringer. +// The pair type thus implements all methods of the Stringer interface and +// we say simply that pair implements Stringer. No other syntax is needed. +func (p pair) String() string { + // Sprintf is another public function in package fmt. + // Dot syntax references fields of p. + return fmt.Sprintf("(%d, %d)", p.x, p.y) +} + +func learnInterfaces() { + // Brace syntax is a "struct literal." It evaluates to an initialized + // struct. The := syntax declares and initializes p to this struct. + p := pair{3, 4} + fmt.Println(p.String()) // call String method of p, of type pair. + var i Stringer // declare i of type Stringer. + i = p // valid because pair implements Stringer + // Call String method of i, of type Stringer. Output same as above. + fmt.Println(i.String()) + // It gets more interesting now. We defined Stringer in this file, + // but the same interface happens to be defined in package fmt. + // Pair thus implements fmt.Stringer as well, and does so with no + // declaration of the fact. The definition of pair doesn't mention + // any interfaces at all, and of course the authors of fmt.Stringer + // had no idea that we were going to define pair. + // + // Functions in the fmt package know how to print some standard built in + // types, and beyond that, they see if a type implements fmt.Stringer. + // If so, they simply call the String method to ask an object for a + // printable representation of itself. + fmt.Println(p) // output same as above. Println calls String method. + fmt.Println(i) // output same as above + + learnErrorHandling() +} + +func learnErrorHandling() { + // Sometimes you just need to know if something worked or not. Go has + // a ", ok" idiom for that. Something, a map expression here, but commonly + // a function, can return a boolean value of ok or not ok as a second + // return value. + m := map[int]string{3: "three", 4: "four"} + if x, ok := m[1]; !ok { // , ok is optional but see how useful it is. + fmt.Println("no one there") + } else { + fmt.Print(x) + } + // An error value communicates not just "ok" but more about the problem. + if _, err := strconv.Atoi("non-int"); err != nil { // _ discards value + // prints "strconv.ParseInt: parsing "non-int": invalid syntax" + fmt.Println(err) + } + // error is a built in type. It is an interface with a single method, + // defined internally as, + // + // type error interface { + // Error() string + // } + // + // The string returned by the Error method is conventionally a printable + // error message. You can define your own error types by simply adding + // an Error method. Your type then automatically implements the error + // interface. We've seen two interfaces now, fmt.Stringer and error. + + // We'll revisit interfaces a little later. Meanwhile, + learnConcurrency() +} + +// Go has concurrency support in the language definition. The element of +// concurrent execution is called a "goroutine" and is similar to a thread +// but "lighter." Goroutines are multiplexed to operating system threads +// and a running Go program can have far more goroutines than available OS +// threads. If a machine has multiple CPU cores, goroutines can run in +// parallel. +// +// Go "Channels" allow communication between goroutines in a way that is +// both powerful and easy to understand. Channel is a type in Go and objects +// of type channel are first class objects--they can be assigned to variables, +// passed around to functions, and so on. A channel works conceptually much +// like a Unix pipe. You put data in at one end and it comes out the other. +// Channel "send" and "receive" operations are goroutine-safe. No locks +// or additional synchronization is needed. + +// Inc increments a number, and sends the result on a channel. The channel +// operation makes this function useful to run concurrently with other +// goroutines. There is no special declaration though that says this function +// is concurrent. It is an ordinary function that happens to have a +// parameter of channel type. +func inc(i int, c chan int) { + c <- i + 1 // <- is the "send" operator when a channel appears on the left. +} + +// We'll use inc to increment some numbers concurrently. +func learnConcurrency() { + // Same make function used earlier to make a slice. Make allocates and + // initializes slices, maps, and channels. + c := make(chan int) + // Start three concurrent goroutines. Numbers will be incremented + // concurrently, perhaps in parallel if the machine is capable and + // properly configured. All three send to the same channel. + go inc(0, c) // go is a statement that starts a new goroutine. + go inc(10, c) + go inc(-805, c) + // Read three results from the channel and print them out. + // There is no telling in what order the results will arrive! + fmt.Println(<-c, <-c, <-c) // channel on right, <- is "receive" operator. + + cs := make(chan string) // another channel, this one handles strings. + cc := make(chan chan string) // a channel of channels. + go func() { c <- 84 }() // start a new goroutine just to send a value + go func() { cs <- "wordy" }() // again, for cs this time + // Select has syntax like a switch statement but is doing something + // pretty different. Each case involves a channel operation. In rough + // terms, a case is selected at random out of the cases that are ready to + // communicate. If none are ready, select waits for one to become ready. + select { + case i := <-c: // the value received can be assigned to a variable + fmt.Println("it's a", i) + case <-cs: // or the value received can be discarded + fmt.Println("it's a string") + case <-cc: // empty channel, not ready for communication. + fmt.Println("didn't happen.") + } + // At this point a value was taken from either c or cs. One of the two + // goroutines started above has completed, the other will remain blocked. + + learnWebProgramming() // Go does it. You want to do it too. +} + +// A simple web server can be created with a single function from the standard +// library. ListenAndServe, in package net/http, listens at the specified +// TCP address and uses an object that knows how to serve data. "Knows how" +// means "satisfies an interface." The second parameter is of type interface, +// specifically http.Handler. http.Handler has a single method, ServeHTTP. +func learnWebProgramming() { + err := http.ListenAndServe(":8080", pair{}) + // Error returns are ubiquitous in Go. Always check error returns and + // do something with them. Often it's enough to print it out as an + // indication of what failed. Of course there are better things to do + // in production code: log it, try something else, shut everything down, + // and so on. + fmt.Println(err) +} + +// You can make any type into an http.Hander by implementing ServeHTTP. +// Lets use the pair type we defined earlier, just because we have it +// sitting around. ServeHTTP has two parameters. The request parameter +// is a struct that we'll ignore here. http.ResponseWriter is yet another +// interface! Here it is an object supplied to us with the guarantee that +// it implements its interface, which includes a method Write. +// We call this Write method to serve data. +func (p pair) ServeHTTP(w http.ResponseWriter, r *http.Request) { + w.Write([]byte("You learned Go in Y minutes!")) +} + +// And that's it for a proof-of-concept web server! If you run this program +// it will print out all the lines from the earlier parts of the lesson, then +// start this web server. To hit the web server, just point a browser at +// localhost:8080 and you'll see the message. (Then you can probably press +// ctrl-C to kill it.) +``` + +## Further Reading + +The root of all things Go is the [official Go web site](http://golang.org/). +There you can follow the tutorial, play interactively, and read lots. + +The language definition itself is highly recommended. It's easy to read +and amazingly short (as language definitions go these days.) + +On the reading list for students of Go is the source code to the standard +library. Comprehensively documented, it demonstrates the best of readable +and understandable Go, Go style, and Go idioms. Click on a function name +in the documentation and the source code comes up! + -- cgit v1.2.3 From a73d5c83c162569d0def8e481d85ece8d517b06a Mon Sep 17 00:00:00 2001 From: Sonia Keys Date: Tue, 13 Aug 2013 17:12:54 -0400 Subject: slashed comments --- go.html.markdown | 198 +++++++++++-------------------------------------------- 1 file changed, 38 insertions(+), 160 deletions(-) diff --git a/go.html.markdown b/go.html.markdown index 9888b37d..e7b35926 100644 --- a/go.html.markdown +++ b/go.html.markdown @@ -27,9 +27,7 @@ Go comes with a great standard library and an enthusiastic community. // Main is a special name declaring an executable rather than a library. package main -// An import declaration comes next. It declares library packages referenced -// in this file. The list must be exactly correct! Missing or unused packages -// are errors, not warnings. +// Import declaration declares library packages referenced in this file. import ( "fmt" // A package in the Go standard library "net/http" // Yes, a web server! @@ -39,27 +37,20 @@ import ( // A function definition. Main is special. It is the entry point for the // executable program. Love it or hate it, Go uses brace brackets. func main() { - // Println is a function that outputs a line to stdout. It can be - // called here because fmt has been imported and the function name - // "Println" is upper case. Symbols starting with an upper case letter - // are publicly visible. No other special syntax is needed to export - // something from a package. - // To call Println, qualify it with the package name, fmt. + // Println outputs a line to stdout. + // Qualify it with the package name, fmt. fmt.Println("Hello world!") // Call another function within this package. beyondHello() } -// Idiomatic Go uses camel case. Functions have parameters in parentheses. +// Functions have parameters in parentheses. // If there are no parameters, empty parens are still required. func beyondHello() { var x int // Variable declaration. Variables must be declared before use. x = 3 // Variable assignment. - // "Short" declarations use := syntax to declare and assign, infering the - // type from the right hand side as much as possible and using some - // defaults where the rhs could be interpreted different ways. - // Idiomatic Go uses short declarations in preference to var keyword. + // "Short" declarations use := to infer the type, declare, and assign. y := 4 sum, prod := learnMultiple(x, y) // function returns two values fmt.Println("sum:", sum, "prod:", prod) // simple output @@ -67,8 +58,6 @@ func beyondHello() { } // Functions can have parameters and (multiple!) return values. -// In declarations, the symbol precedes the type, and the type does not have -// to be repeated if it is the same for multiple symbols in a row. func learnMultiple(x, y int) (sum, prod int) { return x + y, x * y // return two values } @@ -87,12 +76,11 @@ can include line breaks.` // same string type f := 3.14195 // float64, an IEEE-754 64-bit floating point number c := 3 + 4i // complex128, represented internally with two float64s - // You can use var syntax with an initializer if you want - // something other than the default that a short declaration gives you. + // Var syntax with an initializers. var u uint = 7 // unsigned, but implementation dependent size as with int var pi float32 = 22. / 7 - // Or more idiomatically, use conversion syntax with a short declaration. + // Conversion syntax with a short declaration. n := byte('\n') // byte is an alias for uint8 // Arrays have size fixed at compile time. @@ -106,12 +94,8 @@ can include line breaks.` // same string type var d2 [][]float64 // declaration only, nothing allocated here bs := []byte("a slice") // type conversion syntax - p, q := learnMemory() // A little side bar. - // Did you read it? This short declaration declares p and q to be of - // type pointer to int. P is now pointing into a block of of 20 ints, but - // the only one accessible is the one that p is pointing at. There is - // no p++ to get at the next one. - fmt.Println(*p, *q) // * follows a pointer. This prints two ints. + p, q := learnMemory() // declares p, q to be type pointer to int. + fmt.Println(*p, *q) // * follows a pointer. This prints two ints. // Maps are a dynamically growable associative array type, like the // hash or dictionary types of some other languages. @@ -130,26 +114,13 @@ can include line breaks.` // same string type // Go is fully garbage collected. It has pointers but no pointer arithmetic. // You can make a mistake with a nil pointer, but not by incrementing a pointer. func learnMemory() (p, q *int) { - // Named return values p and q have type pointer to int. They are - // initialized to nil at this point. Evaluating *p or *q here would cause - // a panic--a run time error. + // Named return values p and q have type pointer to int. p = new(int) // built-in function new allocates memory. // The allocated int is initialized to 0, p is no longer nil. s := make([]int, 20) // allocate 20 ints as a single block of memory s[3] = 7 // assign one of them r := -2 // declare another local variable - return &s[3], &r // Oh my. - // The line above returns two values, yes, and both of the expressions - // are valid. & takes the address of an object. Elements of a slice are - // addressable, and so are local variables. Built-in functions new and - // make explicitly allocate memory, but local objects can be allocated - // as needed. Here memory for r will be still be referenced after the - // function returns so it will be allocated as well. The int allocated - // with new on the other hand will no longer be referenced and can be - // garbage collected as needed by the Go runtime. The memory allocated - // with make will still be referenced at that one element, and so it - // cannot be garbage collected. All 20 ints remain in memory because - // one of them is still referenced. + return &s[3], &r // & takes the address of an object. } func expensiveComputation() int { @@ -161,16 +132,13 @@ func learnFlowControl() { if true { fmt.Println("told ya") } - // This is how we format the brace brackets. Formatting is standardized - // by the command line command "go fmt." Everybody does it. You will - // suffer endless disparaging remarks until you conform as well. + // Formatting is standardized by the command line command "go fmt." if false { // pout } else { // gloat } - // If statements can be chained of course, but it's idiomatic to use - // the handy switch statement instead. + // Use switch in preference to chained if statements. x := 1 switch x { case 0: @@ -179,10 +147,7 @@ func learnFlowControl() { case 2: // unreached } - // Like if, for doesn't use parens either. The scope of a variable - // declared in the first clause of the for statement is the statement - // and block. This x shadows the x declared above, but goes out of - // scope after the for block. + // Like if, for doesn't use parens either. for x := 0; x < 3; x++ { // ++ is a statement fmt.Println("iteration", x) } @@ -193,14 +158,11 @@ func learnFlowControl() { break // just kidding continue // unreached } - // The initial assignment of the for statement is handy enough that Go - // if statements can have one as well. Just like in the for statement, - // the := here means to declare and assign y first, then test y > x. - // The scope of y is limited to the if statement and block. + // As with for, := in an if statement means to declare and assign y first, + // then test y > x. if y := expensiveComputation(); y > x { x = y } - // Functions are first class objects and function literals are handy. // Function literals are closures. xBig := func() bool { return x > 100 // references x declared above switch statement. @@ -209,48 +171,25 @@ func learnFlowControl() { x /= 1e5 // this makes it == 10 fmt.Println("xBig:", xBig()) // false now - // When you need it, you'll love it. Actually Go's goto has been reformed - // a bit to avoid indeterminate states. You can't jump around variable - // declarations and you can't jump into blocks. + // When you need it, you'll love it. goto love love: learnInterfaces() // Good stuff coming up! } -// An interface is a list of functionality that a type supports. Notably -// missing from an interface definition is any declaration of which types -// implement the interface. Types simply implement an interface or they don't. -// -// An interface can have any number of methods, but it's actually common -// for an interface to have only single method. It is idiomatic in this -// case for the single method to be named with some action, and for the -// interface name to end in "er." -// -// An interface definition is one kind of a type definition. Interface is -// a built in type. Stringer is defined here as an interface type with one -// method, String. +// Define Stringer as an interface type with one method, String. type Stringer interface { String() string } -// Struct is another built in type. A struct aggregates "fields." -// Pair here has two fields, ints named x and y. +// Define pair as a struct with two fields, ints named x and y. type pair struct { x, y int } -// User defined types can have "methods." These are functions that operate -// in the context of an instance of the user defined type. The instance -// is called the "receiver" and is identified with a declaration just in front -// of the method name. The receiver here is "p." In most ways the receiver -// works just like a function parameter. -// -// This String method has the same name and return value as the String method -// of the Stringer interface. Further, String is the only method of Stringer. -// The pair type thus implements all methods of the Stringer interface and -// we say simply that pair implements Stringer. No other syntax is needed. -func (p pair) String() string { +// Define a method on type pair. Pair now implements Stringer. +func (p pair) String() string { // p is called the "receiver" // Sprintf is another public function in package fmt. // Dot syntax references fields of p. return fmt.Sprintf("(%d, %d)", p.x, p.y) @@ -261,21 +200,13 @@ func learnInterfaces() { // struct. The := syntax declares and initializes p to this struct. p := pair{3, 4} fmt.Println(p.String()) // call String method of p, of type pair. - var i Stringer // declare i of type Stringer. + var i Stringer // declare i of interface type Stringer. i = p // valid because pair implements Stringer // Call String method of i, of type Stringer. Output same as above. fmt.Println(i.String()) - // It gets more interesting now. We defined Stringer in this file, - // but the same interface happens to be defined in package fmt. - // Pair thus implements fmt.Stringer as well, and does so with no - // declaration of the fact. The definition of pair doesn't mention - // any interfaces at all, and of course the authors of fmt.Stringer - // had no idea that we were going to define pair. - // - // Functions in the fmt package know how to print some standard built in - // types, and beyond that, they see if a type implements fmt.Stringer. - // If so, they simply call the String method to ask an object for a - // printable representation of itself. + + // Functions in the fmt package call the String method to ask an object + // for a printable representation of itself. fmt.Println(p) // output same as above. Println calls String method. fmt.Println(i) // output same as above @@ -283,57 +214,23 @@ func learnInterfaces() { } func learnErrorHandling() { - // Sometimes you just need to know if something worked or not. Go has - // a ", ok" idiom for that. Something, a map expression here, but commonly - // a function, can return a boolean value of ok or not ok as a second - // return value. + // ", ok" idiom used to tell if something worked or not. m := map[int]string{3: "three", 4: "four"} - if x, ok := m[1]; !ok { // , ok is optional but see how useful it is. + if x, ok := m[1]; !ok { // ok will be false because 1 is not in the map. fmt.Println("no one there") } else { - fmt.Print(x) + fmt.Print(x) // x would be the value, if it were in the map. } // An error value communicates not just "ok" but more about the problem. if _, err := strconv.Atoi("non-int"); err != nil { // _ discards value // prints "strconv.ParseInt: parsing "non-int": invalid syntax" fmt.Println(err) } - // error is a built in type. It is an interface with a single method, - // defined internally as, - // - // type error interface { - // Error() string - // } - // - // The string returned by the Error method is conventionally a printable - // error message. You can define your own error types by simply adding - // an Error method. Your type then automatically implements the error - // interface. We've seen two interfaces now, fmt.Stringer and error. - // We'll revisit interfaces a little later. Meanwhile, learnConcurrency() } -// Go has concurrency support in the language definition. The element of -// concurrent execution is called a "goroutine" and is similar to a thread -// but "lighter." Goroutines are multiplexed to operating system threads -// and a running Go program can have far more goroutines than available OS -// threads. If a machine has multiple CPU cores, goroutines can run in -// parallel. -// -// Go "Channels" allow communication between goroutines in a way that is -// both powerful and easy to understand. Channel is a type in Go and objects -// of type channel are first class objects--they can be assigned to variables, -// passed around to functions, and so on. A channel works conceptually much -// like a Unix pipe. You put data in at one end and it comes out the other. -// Channel "send" and "receive" operations are goroutine-safe. No locks -// or additional synchronization is needed. - -// Inc increments a number, and sends the result on a channel. The channel -// operation makes this function useful to run concurrently with other -// goroutines. There is no special declaration though that says this function -// is concurrent. It is an ordinary function that happens to have a -// parameter of channel type. +// c is a channel, a concurrency-safe communication object. func inc(i int, c chan int) { c <- i + 1 // <- is the "send" operator when a channel appears on the left. } @@ -357,10 +254,9 @@ func learnConcurrency() { cc := make(chan chan string) // a channel of channels. go func() { c <- 84 }() // start a new goroutine just to send a value go func() { cs <- "wordy" }() // again, for cs this time - // Select has syntax like a switch statement but is doing something - // pretty different. Each case involves a channel operation. In rough - // terms, a case is selected at random out of the cases that are ready to - // communicate. If none are ready, select waits for one to become ready. + // Select has syntax like a switch statement but each case involves + // a channel operation. It selects a case at random out of the cases + // that are ready to communicate. select { case i := <-c: // the value received can be assigned to a variable fmt.Println("it's a", i) @@ -375,37 +271,19 @@ func learnConcurrency() { learnWebProgramming() // Go does it. You want to do it too. } -// A simple web server can be created with a single function from the standard -// library. ListenAndServe, in package net/http, listens at the specified -// TCP address and uses an object that knows how to serve data. "Knows how" -// means "satisfies an interface." The second parameter is of type interface, -// specifically http.Handler. http.Handler has a single method, ServeHTTP. +// A single function from package http starts a web server. func learnWebProgramming() { + // ListenAndServe first parameter is TCP address to listen at. + // Second parameter is an interface, specifically http.Handler. err := http.ListenAndServe(":8080", pair{}) - // Error returns are ubiquitous in Go. Always check error returns and - // do something with them. Often it's enough to print it out as an - // indication of what failed. Of course there are better things to do - // in production code: log it, try something else, shut everything down, - // and so on. - fmt.Println(err) + fmt.Println(err) // don't ignore errors } -// You can make any type into an http.Hander by implementing ServeHTTP. -// Lets use the pair type we defined earlier, just because we have it -// sitting around. ServeHTTP has two parameters. The request parameter -// is a struct that we'll ignore here. http.ResponseWriter is yet another -// interface! Here it is an object supplied to us with the guarantee that -// it implements its interface, which includes a method Write. -// We call this Write method to serve data. +// Make pair an http.Handler by implementing its only method, ServeHTTP. func (p pair) ServeHTTP(w http.ResponseWriter, r *http.Request) { + // Serve data with a method of http.ResponseWriter w.Write([]byte("You learned Go in Y minutes!")) } - -// And that's it for a proof-of-concept web server! If you run this program -// it will print out all the lines from the earlier parts of the lesson, then -// start this web server. To hit the web server, just point a browser at -// localhost:8080 and you'll see the message. (Then you can probably press -// ctrl-C to kill it.) ``` ## Further Reading -- cgit v1.2.3