|
// This code is originally from Algorithms, 4th Edition: http://algs4.cs.princeton.edu/home/ |
|
// Source: http://introcs.cs.princeton.edu/java/stdlib/StdIn.java.html |
|
|
|
/************************************************************************* |
|
* Compilation: javac StdIn.java |
|
* Execution: java StdIn (interactive test of basic functionality) |
|
* |
|
* Reads in data of various types from standard input. |
|
* |
|
*************************************************************************/ |
|
|
|
import java.util.ArrayList; |
|
import java.util.InputMismatchException; |
|
import java.util.Locale; |
|
import java.util.Scanner; |
|
import java.util.regex.Pattern; |
|
|
|
/** |
|
* The <tt>StdIn</tt> class provides static methods for reading strings |
|
* and numbers from standard input. See |
|
* <a href="http://introcs.cs.princeton.edu/15inout">Section 1.5</a> of |
|
* <i>Introduction to Programming in Java: An Interdisciplinary Approach</i> |
|
* by Robert Sedgewick and Kevin Wayne. |
|
* <p> |
|
* For uniformity across platforms, this class uses <tt>Locale.US</tt> |
|
* for the locale and <tt>"UTF-8"</tt> for the character-set encoding. |
|
* The English language locale is consistent with the formatting conventions |
|
* for Java floating-point literals, command-line arguments |
|
* (via {@link Double#parseDouble(String)}) and standard output. |
|
* <p> |
|
* Like {@link Scanner}, reading a <em>token</em> also consumes preceding Java |
|
* whitespace; reading a line consumes the following end-of-line |
|
* delimeter; reading a character consumes nothing extra. |
|
* <p> |
|
* Whitespace is defined in {@link Character#isWhitespace(char)}. Newlines |
|
* consist of \n, \r, \r\n, and Unicode hex code points 0x2028, 0x2029, 0x0085; |
|
* see <tt><a href="http://www.docjar.com/html/api/java/util/Scanner.java.html"> |
|
* Scanner.java</a></tt> (NB: Java 6u23 and earlier uses only \r, \r, \r\n). |
|
* <p> |
|
* See {@link In} for a version that handles input from files, URLs, |
|
* and sockets. |
|
* <p> |
|
* Note that Java's UTF-8 encoding does not recognize the optional byte-order |
|
* mask. If the input begins with the optional byte-order mask, <tt>StdIn</tt> |
|
* will have an extra character <tt>uFEFF</tt> at the beginning. |
|
* For details, see http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=4508058. |
|
* |
|
* @author David Pritchard |
|
* @author Robert Sedgewick |
|
* @author Kevin Wayne |
|
*/ |
|
public final class StdIn { |
|
|
|
// it doesn't make sense to instantiate this class |
|
private StdIn() { } |
|
|
|
private static Scanner scanner; |
|
|
|
/*** begin: section (1 of 2) of code duplicated from In to StdIn */ |
|
|
|
// assume Unicode UTF-8 encoding |
|
private static final String CHARSET_NAME = "UTF-8"; |
|
|
|
// assume language = English, country = US for consistency with System.out. |
|
private static final Locale LOCALE = Locale.US; |
|
|
|
// the default token separator; we maintain the invariant that this value |
|
// is held by the scanner's delimiter between calls |
|
private static final Pattern WHITESPACE_PATTERN = Pattern.compile("\\p{javaWhitespace}+"); |
|
|
|
// makes whitespace characters significant |
|
private static final Pattern EMPTY_PATTERN = Pattern.compile(""); |
|
|
|
// used to read the entire input |
|
private static final Pattern EVERYTHING_PATTERN = Pattern.compile("\\A"); |
|
|
|
/*** end: section (1 of 2) of code duplicated from In to StdIn */ |
|
|
|
/*** begin: section (2 of 2) of code duplicated from In to StdIn, |
|
* with all methods changed from "public" to "public static" ***/ |
|
|
|
/** |
|
* Is the input empty (except possibly for whitespace)? Use this |
|
* to know whether the next call to {@link #readString()}, |
|
* {@link #readDouble()}, etc will succeed. |
|
* @return true if standard input is empty (except possibly |
|
* for whitespae), and false otherwise |
|
*/ |
|
public static boolean isEmpty() { |
|
return !scanner.hasNext(); |
|
} |
|
|
|
/** |
|
* Does the input have a next line? Use this to know whether the |
|
* next call to {@link #readLine()} will succeed. <p> Functionally |
|
* equivalent to {@link #hasNextChar()}. |
|
* @return true if standard input is empty, and false otherwise |
|
*/ |
|
public static boolean hasNextLine() { |
|
return scanner.hasNextLine(); |
|
} |
|
|
|
/** |
|
* Is the input empty (including whitespace)? Use this to know |
|
* whether the next call to {@link #readChar()} will succeed. |
|
* <p>Functionally equivalent to {@link #hasNextLine()}. |
|
* @return true if standard input is empty, and false otherwise |
|
*/ |
|
public static boolean hasNextChar() { |
|
scanner.useDelimiter(EMPTY_PATTERN); |
|
boolean result = scanner.hasNext(); |
|
scanner.useDelimiter(WHITESPACE_PATTERN); |
|
return result; |
|
} |
|
|
|
|
|
/** |
|
* Reads and returns the next line, excluding the line separator if present. |
|
* @return the next line, excluding the line separator if present |
|
*/ |
|
public static String readLine() { |
|
String line; |
|
try { line = scanner.nextLine(); } |
|
catch (Exception e) { line = null; } |
|
return line; |
|
} |
|
|
|
/** |
|
* Reads and returns the next character. |
|
* @return the next character |
|
*/ |
|
public static char readChar() { |
|
scanner.useDelimiter(EMPTY_PATTERN); |
|
String ch = scanner.next(); |
|
assert (ch.length() == 1) : "Internal (Std)In.readChar() error!" |
|
+ " Please contact the authors."; |
|
scanner.useDelimiter(WHITESPACE_PATTERN); |
|
return ch.charAt(0); |
|
} |
|
|
|
|
|
/** |
|
* Reads and returns the remainder of the input, as a string. |
|
* @return the remainder of the input, as a string |
|
*/ |
|
public static String readAll() { |
|
if (!scanner.hasNextLine()) |
|
return ""; |
|
|
|
String result = scanner.useDelimiter(EVERYTHING_PATTERN).next(); |
|
// not that important to reset delimeter, since now scanner is empty |
|
scanner.useDelimiter(WHITESPACE_PATTERN); // but let's do it anyway |
|
return result; |
|
} |
|
|
|
|
|
/** |
|
* Reads the next token and returns the <tt>String</tt>. |
|
* @return the next <tt>String</tt> |
|
*/ |
|
public static String readString() { |
|
return scanner.next(); |
|
} |
|
|
|
/** |
|
* Reads the next token from standard input, parses it as an integer, and returns the integer. |
|
* @return the next integer on standard input |
|
* @throws InputMismatchException if the next token cannot be parsed as an <tt>int</tt> |
|
*/ |
|
public static int readInt() { |
|
return scanner.nextInt(); |
|
} |
|
|
|
/** |
|
* Reads the next token from standard input, parses it as a double, and returns the double. |
|
* @return the next double on standard input |
|
* @throws InputMismatchException if the next token cannot be parsed as a <tt>double</tt> |
|
*/ |
|
public static double readDouble() { |
|
return scanner.nextDouble(); |
|
} |
|
|
|
/** |
|
* Reads the next token from standard input, parses it as a float, and returns the float. |
|
* @return the next float on standard input |
|
* @throws InputMismatchException if the next token cannot be parsed as a <tt>float</tt> |
|
*/ |
|
public static float readFloat() { |
|
return scanner.nextFloat(); |
|
} |
|
|
|
/** |
|
* Reads the next token from standard input, parses it as a long integer, and returns the long integer. |
|
* @return the next long integer on standard input |
|
* @throws InputMismatchException if the next token cannot be parsed as a <tt>long</tt> |
|
*/ |
|
public static long readLong() { |
|
return scanner.nextLong(); |
|
} |
|
|
|
/** |
|
* Reads the next token from standard input, parses it as a short integer, and returns the short integer. |
|
* @return the next short integer on standard input |
|
* @throws InputMismatchException if the next token cannot be parsed as a <tt>short</tt> |
|
*/ |
|
public static short readShort() { |
|
return scanner.nextShort(); |
|
} |
|
|
|
/** |
|
* Reads the next token from standard input, parses it as a byte, and returns the byte. |
|
* @return the next byte on standard input |
|
* @throws InputMismatchException if the next token cannot be parsed as a <tt>byte</tt> |
|
*/ |
|
public static byte readByte() { |
|
return scanner.nextByte(); |
|
} |
|
|
|
/** |
|
* Reads the next token from standard input, parses it as a boolean, |
|
* and returns the boolean. |
|
* @return the next boolean on standard input |
|
* @throws InputMismatchException if the next token cannot be parsed as a <tt>boolean</tt>: |
|
* <tt>true</tt> or <tt>1</tt> for true, and <tt>false</tt> or <tt>0</tt> for false, |
|
* ignoring case |
|
*/ |
|
public static boolean readBoolean() { |
|
String s = readString(); |
|
if (s.equalsIgnoreCase("true")) return true; |
|
if (s.equalsIgnoreCase("false")) return false; |
|
if (s.equals("1")) return true; |
|
if (s.equals("0")) return false; |
|
throw new InputMismatchException(); |
|
} |
|
|
|
/** |
|
* Reads all remaining tokens from standard input and returns them as an array of strings. |
|
* @return all remaining tokens on standard input, as an array of strings |
|
*/ |
|
public static String[] readAllStrings() { |
|
// we could use readAll.trim().split(), but that's not consistent |
|
// because trim() uses characters 0x00..0x20 as whitespace |
|
String[] tokens = WHITESPACE_PATTERN.split(readAll()); |
|
if (tokens.length == 0 || tokens[0].length() > 0) |
|
return tokens; |
|
|
|
// don't include first token if it is leading whitespace |
|
String[] decapitokens = new String[tokens.length-1]; |
|
for (int i = 0; i < tokens.length - 1; i++) |
|
decapitokens[i] = tokens[i+1]; |
|
return decapitokens; |
|
} |
|
|
|
/** |
|
* Reads all remaining lines from standard input and returns them as an array of strings. |
|
* @return all remaining lines on standard input, as an array of strings |
|
*/ |
|
public static String[] readAllLines() { |
|
ArrayList<String> lines = new ArrayList<String>(); |
|
while (hasNextLine()) { |
|
lines.add(readLine()); |
|
} |
|
return lines.toArray(new String[0]); |
|
} |
|
|
|
/** |
|
* Reads all remaining tokens from standard input, parses them as integers, and returns |
|
* them as an array of integers. |
|
* @return all remaining integers on standard input, as an array |
|
* @throws InputMismatchException if any token cannot be parsed as an <tt>int</tt> |
|
*/ |
|
public static int[] readAllInts() { |
|
String[] fields = readAllStrings(); |
|
int[] vals = new int[fields.length]; |
|
for (int i = 0; i < fields.length; i++) |
|
vals[i] = Integer.parseInt(fields[i]); |
|
return vals; |
|
} |
|
|
|
/** |
|
* Reads all remaining tokens from standard input, parses them as doubles, and returns |
|
* them as an array of doubles. |
|
* @return all remaining doubles on standard input, as an array |
|
* @throws InputMismatchException if any token cannot be parsed as a <tt>double</tt> |
|
*/ |
|
public static double[] readAllDoubles() { |
|
String[] fields = readAllStrings(); |
|
double[] vals = new double[fields.length]; |
|
for (int i = 0; i < fields.length; i++) |
|
vals[i] = Double.parseDouble(fields[i]); |
|
return vals; |
|
} |
|
|
|
/*** end: section (2 of 2) of code duplicated from In to StdIn */ |
|
|
|
|
|
// do this once when StdIn is initialized |
|
static { |
|
resync(); |
|
} |
|
|
|
/** |
|
* If StdIn changes, use this to reinitialize the scanner. |
|
*/ |
|
private static void resync() { |
|
setScanner(new Scanner(new java.io.BufferedInputStream(System.in), CHARSET_NAME)); |
|
} |
|
|
|
private static void setScanner(Scanner scanner) { |
|
StdIn.scanner = scanner; |
|
StdIn.scanner.useLocale(LOCALE); |
|
} |
|
|
|
/** |
|
* Reads all remaining tokens, parses them as integers, and returns |
|
* them as an array of integers. |
|
* @return all remaining integers, as an array |
|
* @throws InputMismatchException if any token cannot be parsed as an <tt>int</tt> |
|
* @deprecated For more consistency, use {@link #readAllInts()} |
|
*/ |
|
public static int[] readInts() { |
|
return readAllInts(); |
|
} |
|
|
|
/** |
|
* Reads all remaining tokens, parses them as doubles, and returns |
|
* them as an array of doubles. |
|
* @return all remaining doubles, as an array |
|
* @throws InputMismatchException if any token cannot be parsed as a <tt>double</tt> |
|
* @deprecated For more consistency, use {@link #readAllDoubles()} |
|
*/ |
|
public static double[] readDoubles() { |
|
return readAllDoubles(); |
|
} |
|
|
|
/** |
|
* Reads all remaining tokens and returns them as an array of strings. |
|
* @return all remaining tokens, as an array of strings |
|
* @deprecated For more consistency, use {@link #readAllStrings()} |
|
*/ |
|
public static String[] readStrings() { |
|
return readAllStrings(); |
|
} |
|
|
|
|
|
/** |
|
* Interactive test of basic functionality. |
|
*/ |
|
public static void main(String[] args) { |
|
|
|
System.out.println("Type a string: "); |
|
String s = StdIn.readString(); |
|
System.out.println("Your string was: " + s); |
|
System.out.println(); |
|
|
|
System.out.println("Type an int: "); |
|
int a = StdIn.readInt(); |
|
System.out.println("Your int was: " + a); |
|
System.out.println(); |
|
|
|
System.out.println("Type a boolean: "); |
|
boolean b = StdIn.readBoolean(); |
|
System.out.println("Your boolean was: " + b); |
|
System.out.println(); |
|
|
|
System.out.println("Type a double: "); |
|
double c = StdIn.readDouble(); |
|
System.out.println("Your double was: " + c); |
|
System.out.println(); |
|
|
|
} |
|
|
|
} |
Seems like the benchmark is dominated by IO. See the results from my edit in https://gist.github.com/pyrtsa/486bd679b326b01b6b85#file-test-run-txt