summaryrefslogtreecommitdiffstats
path: root/js/src/tests/ecma/LexicalConventions/7.7.4.js
diff options
context:
space:
mode:
Diffstat (limited to 'js/src/tests/ecma/LexicalConventions/7.7.4.js')
-rw-r--r--js/src/tests/ecma/LexicalConventions/7.7.4.js208
1 files changed, 208 insertions, 0 deletions
diff --git a/js/src/tests/ecma/LexicalConventions/7.7.4.js b/js/src/tests/ecma/LexicalConventions/7.7.4.js
new file mode 100644
index 000000000..afbe4c12e
--- /dev/null
+++ b/js/src/tests/ecma/LexicalConventions/7.7.4.js
@@ -0,0 +1,208 @@
+/* -*- indent-tabs-mode: nil; js-indent-level: 2 -*- */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+
+/**
+ File Name: 7.7.4.js
+ ECMA Section: 7.7.4 String Literals
+
+ Description: A string literal is zero or more characters enclosed in
+ single or double quotes. Each character may be
+ represented by an escape sequence.
+
+
+ Author: christine@netscape.com
+ Date: 16 september 1997
+*/
+
+var SECTION = "7.7.4";
+var VERSION = "ECMA_1";
+startTest();
+var TITLE = "String Literals";
+
+writeHeaderToLog( SECTION + " "+ TITLE);
+
+// StringLiteral:: "" and ''
+
+new TestCase( SECTION, "\"\"", "", "" );
+new TestCase( SECTION, "\'\'", "", '' );
+
+// DoubleStringCharacters:: DoubleStringCharacter :: EscapeSequence :: CharacterEscapeSequence
+new TestCase( SECTION, "\\\"", String.fromCharCode(0x0022), "\"" );
+new TestCase( SECTION, "\\\'", String.fromCharCode(0x0027), "\'" );
+new TestCase( SECTION, "\\", String.fromCharCode(0x005C), "\\" );
+new TestCase( SECTION, "\\b", String.fromCharCode(0x0008), "\b" );
+new TestCase( SECTION, "\\f", String.fromCharCode(0x000C), "\f" );
+new TestCase( SECTION, "\\n", String.fromCharCode(0x000A), "\n" );
+new TestCase( SECTION, "\\r", String.fromCharCode(0x000D), "\r" );
+new TestCase( SECTION, "\\t", String.fromCharCode(0x0009), "\t" );
+new TestCase( SECTION, "\\v", String.fromCharCode(0x000B), "\v" );
+
+// DoubleStringCharacters:DoubleStringCharacter::EscapeSequence::OctalEscapeSequence
+
+new TestCase( SECTION, "\\00", String.fromCharCode(0x0000), "\00" );
+new TestCase( SECTION, "\\01", String.fromCharCode(0x0001), "\01" );
+new TestCase( SECTION, "\\02", String.fromCharCode(0x0002), "\02" );
+new TestCase( SECTION, "\\03", String.fromCharCode(0x0003), "\03" );
+new TestCase( SECTION, "\\04", String.fromCharCode(0x0004), "\04" );
+new TestCase( SECTION, "\\05", String.fromCharCode(0x0005), "\05" );
+new TestCase( SECTION, "\\06", String.fromCharCode(0x0006), "\06" );
+new TestCase( SECTION, "\\07", String.fromCharCode(0x0007), "\07" );
+
+new TestCase( SECTION, "\\010", String.fromCharCode(0x0008), "\010" );
+new TestCase( SECTION, "\\011", String.fromCharCode(0x0009), "\011" );
+new TestCase( SECTION, "\\012", String.fromCharCode(0x000A), "\012" );
+new TestCase( SECTION, "\\013", String.fromCharCode(0x000B), "\013" );
+new TestCase( SECTION, "\\014", String.fromCharCode(0x000C), "\014" );
+new TestCase( SECTION, "\\015", String.fromCharCode(0x000D), "\015" );
+new TestCase( SECTION, "\\016", String.fromCharCode(0x000E), "\016" );
+new TestCase( SECTION, "\\017", String.fromCharCode(0x000F), "\017" );
+new TestCase( SECTION, "\\020", String.fromCharCode(0x0010), "\020" );
+new TestCase( SECTION, "\\042", String.fromCharCode(0x0022), "\042" );
+
+new TestCase( SECTION, "\\0", String.fromCharCode(0x0000), "\0" );
+new TestCase( SECTION, "\\1", String.fromCharCode(0x0001), "\1" );
+new TestCase( SECTION, "\\2", String.fromCharCode(0x0002), "\2" );
+new TestCase( SECTION, "\\3", String.fromCharCode(0x0003), "\3" );
+new TestCase( SECTION, "\\4", String.fromCharCode(0x0004), "\4" );
+new TestCase( SECTION, "\\5", String.fromCharCode(0x0005), "\5" );
+new TestCase( SECTION, "\\6", String.fromCharCode(0x0006), "\6" );
+new TestCase( SECTION, "\\7", String.fromCharCode(0x0007), "\7" );
+
+new TestCase( SECTION, "\\10", String.fromCharCode(0x0008), "\10" );
+new TestCase( SECTION, "\\11", String.fromCharCode(0x0009), "\11" );
+new TestCase( SECTION, "\\12", String.fromCharCode(0x000A), "\12" );
+new TestCase( SECTION, "\\13", String.fromCharCode(0x000B), "\13" );
+new TestCase( SECTION, "\\14", String.fromCharCode(0x000C), "\14" );
+new TestCase( SECTION, "\\15", String.fromCharCode(0x000D), "\15" );
+new TestCase( SECTION, "\\16", String.fromCharCode(0x000E), "\16" );
+new TestCase( SECTION, "\\17", String.fromCharCode(0x000F), "\17" );
+new TestCase( SECTION, "\\20", String.fromCharCode(0x0010), "\20" );
+new TestCase( SECTION, "\\42", String.fromCharCode(0x0022), "\42" );
+
+new TestCase( SECTION, "\\000", String.fromCharCode(0), "\000" );
+new TestCase( SECTION, "\\111", String.fromCharCode(73), "\111" );
+new TestCase( SECTION, "\\222", String.fromCharCode(146), "\222" );
+new TestCase( SECTION, "\\333", String.fromCharCode(219), "\333" );
+
+// following line commented out as it causes a compile time error
+// new TestCase( SECTION, "\\444", "444", "\444" );
+
+// DoubleStringCharacters:DoubleStringCharacter::EscapeSequence::HexEscapeSequence
+new TestCase( SECTION, "\\xF0", String.fromCharCode(240), "\xF0" );
+new TestCase( SECTION, "\\xE1", String.fromCharCode(225), "\xE1" );
+new TestCase( SECTION, "\\xD2", String.fromCharCode(210), "\xD2" );
+new TestCase( SECTION, "\\xC3", String.fromCharCode(195), "\xC3" );
+new TestCase( SECTION, "\\xB4", String.fromCharCode(180), "\xB4" );
+new TestCase( SECTION, "\\xA5", String.fromCharCode(165), "\xA5" );
+new TestCase( SECTION, "\\x96", String.fromCharCode(150), "\x96" );
+new TestCase( SECTION, "\\x87", String.fromCharCode(135), "\x87" );
+new TestCase( SECTION, "\\x78", String.fromCharCode(120), "\x78" );
+new TestCase( SECTION, "\\x69", String.fromCharCode(105), "\x69" );
+new TestCase( SECTION, "\\x5A", String.fromCharCode(90), "\x5A" );
+new TestCase( SECTION, "\\x4B", String.fromCharCode(75), "\x4B" );
+new TestCase( SECTION, "\\x3C", String.fromCharCode(60), "\x3C" );
+new TestCase( SECTION, "\\x2D", String.fromCharCode(45), "\x2D" );
+new TestCase( SECTION, "\\x1E", String.fromCharCode(30), "\x1E" );
+new TestCase( SECTION, "\\x0F", String.fromCharCode(15), "\x0F" );
+
+// string literals only take up to two hext digits. therefore, the third character in this string
+// should be interpreted as a StringCharacter and not part of the HextEscapeSequence
+
+new TestCase( SECTION, "\\xF0F", String.fromCharCode(240)+"F", "\xF0F" );
+new TestCase( SECTION, "\\xE1E", String.fromCharCode(225)+"E", "\xE1E" );
+new TestCase( SECTION, "\\xD2D", String.fromCharCode(210)+"D", "\xD2D" );
+new TestCase( SECTION, "\\xC3C", String.fromCharCode(195)+"C", "\xC3C" );
+new TestCase( SECTION, "\\xB4B", String.fromCharCode(180)+"B", "\xB4B" );
+new TestCase( SECTION, "\\xA5A", String.fromCharCode(165)+"A", "\xA5A" );
+new TestCase( SECTION, "\\x969", String.fromCharCode(150)+"9", "\x969" );
+new TestCase( SECTION, "\\x878", String.fromCharCode(135)+"8", "\x878" );
+new TestCase( SECTION, "\\x787", String.fromCharCode(120)+"7", "\x787" );
+new TestCase( SECTION, "\\x696", String.fromCharCode(105)+"6", "\x696" );
+new TestCase( SECTION, "\\x5A5", String.fromCharCode(90)+"5", "\x5A5" );
+new TestCase( SECTION, "\\x4B4", String.fromCharCode(75)+"4", "\x4B4" );
+new TestCase( SECTION, "\\x3C3", String.fromCharCode(60)+"3", "\x3C3" );
+new TestCase( SECTION, "\\x2D2", String.fromCharCode(45)+"2", "\x2D2" );
+new TestCase( SECTION, "\\x1E1", String.fromCharCode(30)+"1", "\x1E1" );
+new TestCase( SECTION, "\\x0F0", String.fromCharCode(15)+"0", "\x0F0" );
+
+// DoubleStringCharacter::EscapeSequence::CharacterEscapeSequence::\ NonEscapeCharacter
+new TestCase( SECTION, "\\a", "a", "\a" );
+new TestCase( SECTION, "\\c", "c", "\c" );
+new TestCase( SECTION, "\\d", "d", "\d" );
+new TestCase( SECTION, "\\e", "e", "\e" );
+new TestCase( SECTION, "\\g", "g", "\g" );
+new TestCase( SECTION, "\\h", "h", "\h" );
+new TestCase( SECTION, "\\i", "i", "\i" );
+new TestCase( SECTION, "\\j", "j", "\j" );
+new TestCase( SECTION, "\\k", "k", "\k" );
+new TestCase( SECTION, "\\l", "l", "\l" );
+new TestCase( SECTION, "\\m", "m", "\m" );
+new TestCase( SECTION, "\\o", "o", "\o" );
+new TestCase( SECTION, "\\p", "p", "\p" );
+new TestCase( SECTION, "\\q", "q", "\q" );
+new TestCase( SECTION, "\\s", "s", "\s" );
+new TestCase( SECTION, "\\w", "w", "\w" );
+new TestCase( SECTION, "\\y", "y", "\y" );
+new TestCase( SECTION, "\\z", "z", "\z" );
+new TestCase( SECTION, "\\9", "9", "\9" );
+
+new TestCase( SECTION, "\\A", "A", "\A" );
+new TestCase( SECTION, "\\B", "B", "\B" );
+new TestCase( SECTION, "\\C", "C", "\C" );
+new TestCase( SECTION, "\\D", "D", "\D" );
+new TestCase( SECTION, "\\E", "E", "\E" );
+new TestCase( SECTION, "\\F", "F", "\F" );
+new TestCase( SECTION, "\\G", "G", "\G" );
+new TestCase( SECTION, "\\H", "H", "\H" );
+new TestCase( SECTION, "\\I", "I", "\I" );
+new TestCase( SECTION, "\\J", "J", "\J" );
+new TestCase( SECTION, "\\K", "K", "\K" );
+new TestCase( SECTION, "\\L", "L", "\L" );
+new TestCase( SECTION, "\\M", "M", "\M" );
+new TestCase( SECTION, "\\N", "N", "\N" );
+new TestCase( SECTION, "\\O", "O", "\O" );
+new TestCase( SECTION, "\\P", "P", "\P" );
+new TestCase( SECTION, "\\Q", "Q", "\Q" );
+new TestCase( SECTION, "\\R", "R", "\R" );
+new TestCase( SECTION, "\\S", "S", "\S" );
+new TestCase( SECTION, "\\T", "T", "\T" );
+new TestCase( SECTION, "\\U", "U", "\U" );
+new TestCase( SECTION, "\\V", "V", "\V" );
+new TestCase( SECTION, "\\W", "W", "\W" );
+new TestCase( SECTION, "\\X", "X", "\X" );
+new TestCase( SECTION, "\\Y", "Y", "\Y" );
+new TestCase( SECTION, "\\Z", "Z", "\Z" );
+
+// DoubleStringCharacter::EscapeSequence::UnicodeEscapeSequence
+
+new TestCase( SECTION, "\\u0020", " ", "\u0020" );
+new TestCase( SECTION, "\\u0021", "!", "\u0021" );
+new TestCase( SECTION, "\\u0022", "\"", "\u0022" );
+new TestCase( SECTION, "\\u0023", "#", "\u0023" );
+new TestCase( SECTION, "\\u0024", "$", "\u0024" );
+new TestCase( SECTION, "\\u0025", "%", "\u0025" );
+new TestCase( SECTION, "\\u0026", "&", "\u0026" );
+new TestCase( SECTION, "\\u0027", "'", "\u0027" );
+new TestCase( SECTION, "\\u0028", "(", "\u0028" );
+new TestCase( SECTION, "\\u0029", ")", "\u0029" );
+new TestCase( SECTION, "\\u002A", "*", "\u002A" );
+new TestCase( SECTION, "\\u002B", "+", "\u002B" );
+new TestCase( SECTION, "\\u002C", ",", "\u002C" );
+new TestCase( SECTION, "\\u002D", "-", "\u002D" );
+new TestCase( SECTION, "\\u002E", ".", "\u002E" );
+new TestCase( SECTION, "\\u002F", "/", "\u002F" );
+new TestCase( SECTION, "\\u0030", "0", "\u0030" );
+new TestCase( SECTION, "\\u0031", "1", "\u0031" );
+new TestCase( SECTION, "\\u0032", "2", "\u0032" );
+new TestCase( SECTION, "\\u0033", "3", "\u0033" );
+new TestCase( SECTION, "\\u0034", "4", "\u0034" );
+new TestCase( SECTION, "\\u0035", "5", "\u0035" );
+new TestCase( SECTION, "\\u0036", "6", "\u0036" );
+new TestCase( SECTION, "\\u0037", "7", "\u0037" );
+new TestCase( SECTION, "\\u0038", "8", "\u0038" );
+new TestCase( SECTION, "\\u0039", "9", "\u0039" );
+
+test();