diff options
Diffstat (limited to 'node_modules/bl/test')
-rw-r--r-- | node_modules/bl/test/indexOf.js | 463 | ||||
-rw-r--r-- | node_modules/bl/test/test.js | 782 |
2 files changed, 0 insertions, 1245 deletions
diff --git a/node_modules/bl/test/indexOf.js b/node_modules/bl/test/indexOf.js deleted file mode 100644 index 68435e1..0000000 --- a/node_modules/bl/test/indexOf.js +++ /dev/null @@ -1,463 +0,0 @@ -'use strict' - -var tape = require('tape') - , BufferList = require('../') - , Buffer = require('safe-buffer').Buffer - -tape('indexOf single byte needle', t => { - const bl = new BufferList(['abcdefg', 'abcdefg', '12345']) - t.equal(bl.indexOf('e'), 4) - t.equal(bl.indexOf('e', 5), 11) - t.equal(bl.indexOf('e', 12), -1) - t.equal(bl.indexOf('5'), 18) - t.end() -}) - -tape('indexOf multiple byte needle', t => { - const bl = new BufferList(['abcdefg', 'abcdefg']) - t.equal(bl.indexOf('ef'), 4) - t.equal(bl.indexOf('ef', 5), 11) - t.end() -}) - -tape('indexOf multiple byte needles across buffer boundaries', t => { - const bl = new BufferList(['abcdefg', 'abcdefg']) - t.equal(bl.indexOf('fgabc'), 5) - t.end() -}) - -tape('indexOf takes a buffer list search', t => { - const bl = new BufferList(['abcdefg', 'abcdefg']) - const search = new BufferList('fgabc') - t.equal(bl.indexOf(search), 5) - t.end() -}) - -tape('indexOf a zero byte needle', t => { - const b = new BufferList('abcdef') - const buf_empty = Buffer.from('') - t.equal(b.indexOf(''), 0) - t.equal(b.indexOf('', 1), 1) - t.equal(b.indexOf('', b.length + 1), b.length) - t.equal(b.indexOf('', Infinity), b.length) - t.equal(b.indexOf(buf_empty), 0) - t.equal(b.indexOf(buf_empty, 1), 1) - t.equal(b.indexOf(buf_empty, b.length + 1), b.length) - t.equal(b.indexOf(buf_empty, Infinity), b.length) - t.end() -}) - -tape('indexOf buffers smaller and larger than the needle', t => { - const bl = new BufferList(['abcdefg', 'a', 'bcdefg', 'a', 'bcfgab']) - t.equal(bl.indexOf('fgabc'), 5) - t.equal(bl.indexOf('fgabc', 6), 12) - t.equal(bl.indexOf('fgabc', 13), -1) - t.end() -}) - -// only present in node 6+ -;(process.version.substr(1).split('.')[0] >= 6) && tape('indexOf latin1 and binary encoding', t => { - const b = new BufferList('abcdef') - - // test latin1 encoding - t.equal( - new BufferList(Buffer.from(b.toString('latin1'), 'latin1')) - .indexOf('d', 0, 'latin1'), - 3 - ) - t.equal( - new BufferList(Buffer.from(b.toString('latin1'), 'latin1')) - .indexOf(Buffer.from('d', 'latin1'), 0, 'latin1'), - 3 - ) - t.equal( - new BufferList(Buffer.from('aa\u00e8aa', 'latin1')) - .indexOf('\u00e8', 'latin1'), - 2 - ) - t.equal( - new BufferList(Buffer.from('\u00e8', 'latin1')) - .indexOf('\u00e8', 'latin1'), - 0 - ) - t.equal( - new BufferList(Buffer.from('\u00e8', 'latin1')) - .indexOf(Buffer.from('\u00e8', 'latin1'), 'latin1'), - 0 - ) - - // test binary encoding - t.equal( - new BufferList(Buffer.from(b.toString('binary'), 'binary')) - .indexOf('d', 0, 'binary'), - 3 - ) - t.equal( - new BufferList(Buffer.from(b.toString('binary'), 'binary')) - .indexOf(Buffer.from('d', 'binary'), 0, 'binary'), - 3 - ) - t.equal( - new BufferList(Buffer.from('aa\u00e8aa', 'binary')) - .indexOf('\u00e8', 'binary'), - 2 - ) - t.equal( - new BufferList(Buffer.from('\u00e8', 'binary')) - .indexOf('\u00e8', 'binary'), - 0 - ) - t.equal( - new BufferList(Buffer.from('\u00e8', 'binary')) - .indexOf(Buffer.from('\u00e8', 'binary'), 'binary'), - 0 - ) - t.end() -}) - -tape('indexOf the entire nodejs10 buffer test suite', t => { - const b = new BufferList('abcdef') - const buf_a = Buffer.from('a') - const buf_bc = Buffer.from('bc') - const buf_f = Buffer.from('f') - const buf_z = Buffer.from('z') - - const stringComparison = 'abcdef' - - t.equal(b.indexOf('a'), 0) - t.equal(b.indexOf('a', 1), -1) - t.equal(b.indexOf('a', -1), -1) - t.equal(b.indexOf('a', -4), -1) - t.equal(b.indexOf('a', -b.length), 0) - t.equal(b.indexOf('a', NaN), 0) - t.equal(b.indexOf('a', -Infinity), 0) - t.equal(b.indexOf('a', Infinity), -1) - t.equal(b.indexOf('bc'), 1) - t.equal(b.indexOf('bc', 2), -1) - t.equal(b.indexOf('bc', -1), -1) - t.equal(b.indexOf('bc', -3), -1) - t.equal(b.indexOf('bc', -5), 1) - t.equal(b.indexOf('bc', NaN), 1) - t.equal(b.indexOf('bc', -Infinity), 1) - t.equal(b.indexOf('bc', Infinity), -1) - t.equal(b.indexOf('f'), b.length - 1) - t.equal(b.indexOf('z'), -1) - // empty search tests - t.equal(b.indexOf(buf_a), 0) - t.equal(b.indexOf(buf_a, 1), -1) - t.equal(b.indexOf(buf_a, -1), -1) - t.equal(b.indexOf(buf_a, -4), -1) - t.equal(b.indexOf(buf_a, -b.length), 0) - t.equal(b.indexOf(buf_a, NaN), 0) - t.equal(b.indexOf(buf_a, -Infinity), 0) - t.equal(b.indexOf(buf_a, Infinity), -1) - t.equal(b.indexOf(buf_bc), 1) - t.equal(b.indexOf(buf_bc, 2), -1) - t.equal(b.indexOf(buf_bc, -1), -1) - t.equal(b.indexOf(buf_bc, -3), -1) - t.equal(b.indexOf(buf_bc, -5), 1) - t.equal(b.indexOf(buf_bc, NaN), 1) - t.equal(b.indexOf(buf_bc, -Infinity), 1) - t.equal(b.indexOf(buf_bc, Infinity), -1) - t.equal(b.indexOf(buf_f), b.length - 1) - t.equal(b.indexOf(buf_z), -1) - t.equal(b.indexOf(0x61), 0) - t.equal(b.indexOf(0x61, 1), -1) - t.equal(b.indexOf(0x61, -1), -1) - t.equal(b.indexOf(0x61, -4), -1) - t.equal(b.indexOf(0x61, -b.length), 0) - t.equal(b.indexOf(0x61, NaN), 0) - t.equal(b.indexOf(0x61, -Infinity), 0) - t.equal(b.indexOf(0x61, Infinity), -1) - t.equal(b.indexOf(0x0), -1) - - // test offsets - t.equal(b.indexOf('d', 2), 3) - t.equal(b.indexOf('f', 5), 5) - t.equal(b.indexOf('f', -1), 5) - t.equal(b.indexOf('f', 6), -1) - - t.equal(b.indexOf(Buffer.from('d'), 2), 3) - t.equal(b.indexOf(Buffer.from('f'), 5), 5) - t.equal(b.indexOf(Buffer.from('f'), -1), 5) - t.equal(b.indexOf(Buffer.from('f'), 6), -1) - - t.equal(Buffer.from('ff').indexOf(Buffer.from('f'), 1, 'ucs2'), -1) - - // test invalid and uppercase encoding - t.equal(b.indexOf('b', 'utf8'), 1) - t.equal(b.indexOf('b', 'UTF8'), 1) - t.equal(b.indexOf('62', 'HEX'), 1) - t.throws(() => b.indexOf('bad', 'enc'), TypeError) - - // test hex encoding - t.equal( - Buffer.from(b.toString('hex'), 'hex') - .indexOf('64', 0, 'hex'), - 3 - ) - t.equal( - Buffer.from(b.toString('hex'), 'hex') - .indexOf(Buffer.from('64', 'hex'), 0, 'hex'), - 3 - ) - - // test base64 encoding - t.equal( - Buffer.from(b.toString('base64'), 'base64') - .indexOf('ZA==', 0, 'base64'), - 3 - ) - t.equal( - Buffer.from(b.toString('base64'), 'base64') - .indexOf(Buffer.from('ZA==', 'base64'), 0, 'base64'), - 3 - ) - - // test ascii encoding - t.equal( - Buffer.from(b.toString('ascii'), 'ascii') - .indexOf('d', 0, 'ascii'), - 3 - ) - t.equal( - Buffer.from(b.toString('ascii'), 'ascii') - .indexOf(Buffer.from('d', 'ascii'), 0, 'ascii'), - 3 - ) - - // test optional offset with passed encoding - t.equal(Buffer.from('aaaa0').indexOf('30', 'hex'), 4) - t.equal(Buffer.from('aaaa00a').indexOf('3030', 'hex'), 4) - - { - // test usc2 encoding - const twoByteString = Buffer.from('\u039a\u0391\u03a3\u03a3\u0395', 'ucs2') - - t.equal(8, twoByteString.indexOf('\u0395', 4, 'ucs2')) - t.equal(6, twoByteString.indexOf('\u03a3', -4, 'ucs2')) - t.equal(4, twoByteString.indexOf('\u03a3', -6, 'ucs2')) - t.equal(4, twoByteString.indexOf( - Buffer.from('\u03a3', 'ucs2'), -6, 'ucs2')) - t.equal(-1, twoByteString.indexOf('\u03a3', -2, 'ucs2')) - } - - const mixedByteStringUcs2 = - Buffer.from('\u039a\u0391abc\u03a3\u03a3\u0395', 'ucs2') - t.equal(6, mixedByteStringUcs2.indexOf('bc', 0, 'ucs2')) - t.equal(10, mixedByteStringUcs2.indexOf('\u03a3', 0, 'ucs2')) - t.equal(-1, mixedByteStringUcs2.indexOf('\u0396', 0, 'ucs2')) - - t.equal( - 6, mixedByteStringUcs2.indexOf(Buffer.from('bc', 'ucs2'), 0, 'ucs2')) - t.equal( - 10, mixedByteStringUcs2.indexOf(Buffer.from('\u03a3', 'ucs2'), 0, 'ucs2')) - t.equal( - -1, mixedByteStringUcs2.indexOf(Buffer.from('\u0396', 'ucs2'), 0, 'ucs2')) - - { - const twoByteString = Buffer.from('\u039a\u0391\u03a3\u03a3\u0395', 'ucs2') - - // Test single char pattern - t.equal(0, twoByteString.indexOf('\u039a', 0, 'ucs2')) - let index = twoByteString.indexOf('\u0391', 0, 'ucs2') - t.equal(2, index, `Alpha - at index ${index}`) - index = twoByteString.indexOf('\u03a3', 0, 'ucs2') - t.equal(4, index, `First Sigma - at index ${index}`) - index = twoByteString.indexOf('\u03a3', 6, 'ucs2') - t.equal(6, index, `Second Sigma - at index ${index}`) - index = twoByteString.indexOf('\u0395', 0, 'ucs2') - t.equal(8, index, `Epsilon - at index ${index}`) - index = twoByteString.indexOf('\u0392', 0, 'ucs2') - t.equal(-1, index, `Not beta - at index ${index}`) - - // Test multi-char pattern - index = twoByteString.indexOf('\u039a\u0391', 0, 'ucs2') - t.equal(0, index, `Lambda Alpha - at index ${index}`) - index = twoByteString.indexOf('\u0391\u03a3', 0, 'ucs2') - t.equal(2, index, `Alpha Sigma - at index ${index}`) - index = twoByteString.indexOf('\u03a3\u03a3', 0, 'ucs2') - t.equal(4, index, `Sigma Sigma - at index ${index}`) - index = twoByteString.indexOf('\u03a3\u0395', 0, 'ucs2') - t.equal(6, index, `Sigma Epsilon - at index ${index}`) - } - - const mixedByteStringUtf8 = Buffer.from('\u039a\u0391abc\u03a3\u03a3\u0395') - t.equal(5, mixedByteStringUtf8.indexOf('bc')) - t.equal(5, mixedByteStringUtf8.indexOf('bc', 5)) - t.equal(5, mixedByteStringUtf8.indexOf('bc', -8)) - t.equal(7, mixedByteStringUtf8.indexOf('\u03a3')) - t.equal(-1, mixedByteStringUtf8.indexOf('\u0396')) - - - // Test complex string indexOf algorithms. Only trigger for long strings. - // Long string that isn't a simple repeat of a shorter string. - let longString = 'A' - for (let i = 66; i < 76; i++) { // from 'B' to 'K' - longString = longString + String.fromCharCode(i) + longString - } - - const longBufferString = Buffer.from(longString) - - // pattern of 15 chars, repeated every 16 chars in long - let pattern = 'ABACABADABACABA' - for (let i = 0; i < longBufferString.length - pattern.length; i += 7) { - const index = longBufferString.indexOf(pattern, i) - t.equal((i + 15) & ~0xf, index, - `Long ABACABA...-string at index ${i}`) - } - - let index = longBufferString.indexOf('AJABACA') - t.equal(510, index, `Long AJABACA, First J - at index ${index}`) - index = longBufferString.indexOf('AJABACA', 511) - t.equal(1534, index, `Long AJABACA, Second J - at index ${index}`) - - pattern = 'JABACABADABACABA' - index = longBufferString.indexOf(pattern) - t.equal(511, index, `Long JABACABA..., First J - at index ${index}`) - index = longBufferString.indexOf(pattern, 512) - t.equal( - 1535, index, `Long JABACABA..., Second J - at index ${index}`) - - // Search for a non-ASCII string in a pure ASCII string. - const asciiString = Buffer.from( - 'arglebargleglopglyfarglebargleglopglyfarglebargleglopglyf') - t.equal(-1, asciiString.indexOf('\x2061')) - t.equal(3, asciiString.indexOf('leb', 0)) - - // Search in string containing many non-ASCII chars. - const allCodePoints = [] - for (let i = 0; i < 65536; i++) allCodePoints[i] = i - const allCharsString = String.fromCharCode.apply(String, allCodePoints) - const allCharsBufferUtf8 = Buffer.from(allCharsString) - const allCharsBufferUcs2 = Buffer.from(allCharsString, 'ucs2') - - // Search for string long enough to trigger complex search with ASCII pattern - // and UC16 subject. - t.equal(-1, allCharsBufferUtf8.indexOf('notfound')) - t.equal(-1, allCharsBufferUcs2.indexOf('notfound')) - - // Needle is longer than haystack, but only because it's encoded as UTF-16 - t.equal(Buffer.from('aaaa').indexOf('a'.repeat(4), 'ucs2'), -1) - - t.equal(Buffer.from('aaaa').indexOf('a'.repeat(4), 'utf8'), 0) - t.equal(Buffer.from('aaaa').indexOf('你好', 'ucs2'), -1) - - // Haystack has odd length, but the needle is UCS2. - t.equal(Buffer.from('aaaaa').indexOf('b', 'ucs2'), -1) - - { - // Find substrings in Utf8. - const lengths = [1, 3, 15]; // Single char, simple and complex. - const indices = [0x5, 0x60, 0x400, 0x680, 0x7ee, 0xFF02, 0x16610, 0x2f77b] - for (let lengthIndex = 0; lengthIndex < lengths.length; lengthIndex++) { - for (let i = 0; i < indices.length; i++) { - const index = indices[i] - let length = lengths[lengthIndex] - - if (index + length > 0x7F) { - length = 2 * length - } - - if (index + length > 0x7FF) { - length = 3 * length - } - - if (index + length > 0xFFFF) { - length = 4 * length - } - - const patternBufferUtf8 = allCharsBufferUtf8.slice(index, index + length) - t.equal(index, allCharsBufferUtf8.indexOf(patternBufferUtf8)) - - const patternStringUtf8 = patternBufferUtf8.toString() - t.equal(index, allCharsBufferUtf8.indexOf(patternStringUtf8)) - } - } - } - - { - // Find substrings in Usc2. - const lengths = [2, 4, 16]; // Single char, simple and complex. - const indices = [0x5, 0x65, 0x105, 0x205, 0x285, 0x2005, 0x2085, 0xfff0] - for (let lengthIndex = 0; lengthIndex < lengths.length; lengthIndex++) { - for (let i = 0; i < indices.length; i++) { - const index = indices[i] * 2 - const length = lengths[lengthIndex] - - const patternBufferUcs2 = - allCharsBufferUcs2.slice(index, index + length) - t.equal( - index, allCharsBufferUcs2.indexOf(patternBufferUcs2, 0, 'ucs2')) - - const patternStringUcs2 = patternBufferUcs2.toString('ucs2') - t.equal( - index, allCharsBufferUcs2.indexOf(patternStringUcs2, 0, 'ucs2')) - } - } - } - - [ - () => {}, - {}, - [] - ].forEach(val => { - debugger - t.throws(() => b.indexOf(val), TypeError, `"${JSON.stringify(val)}" should throw`) - }) - - // Test weird offset arguments. - // The following offsets coerce to NaN or 0, searching the whole Buffer - t.equal(b.indexOf('b', undefined), 1) - t.equal(b.indexOf('b', {}), 1) - t.equal(b.indexOf('b', 0), 1) - t.equal(b.indexOf('b', null), 1) - t.equal(b.indexOf('b', []), 1) - - // The following offset coerces to 2, in other words +[2] === 2 - t.equal(b.indexOf('b', [2]), -1) - - // Behavior should match String.indexOf() - t.equal( - b.indexOf('b', undefined), - stringComparison.indexOf('b', undefined)) - t.equal( - b.indexOf('b', {}), - stringComparison.indexOf('b', {})) - t.equal( - b.indexOf('b', 0), - stringComparison.indexOf('b', 0)) - t.equal( - b.indexOf('b', null), - stringComparison.indexOf('b', null)) - t.equal( - b.indexOf('b', []), - stringComparison.indexOf('b', [])) - t.equal( - b.indexOf('b', [2]), - stringComparison.indexOf('b', [2])) - - // test truncation of Number arguments to uint8 - { - const buf = Buffer.from('this is a test') - t.equal(buf.indexOf(0x6973), 3) - t.equal(buf.indexOf(0x697320), 4) - t.equal(buf.indexOf(0x69732069), 2) - t.equal(buf.indexOf(0x697374657374), 0) - t.equal(buf.indexOf(0x69737374), 0) - t.equal(buf.indexOf(0x69737465), 11) - t.equal(buf.indexOf(0x69737465), 11) - t.equal(buf.indexOf(-140), 0) - t.equal(buf.indexOf(-152), 1) - t.equal(buf.indexOf(0xff), -1) - t.equal(buf.indexOf(0xffff), -1) - } - - // Test that Uint8Array arguments are okay. - { - const needle = new Uint8Array([ 0x66, 0x6f, 0x6f ]) - const haystack = new BufferList(Buffer.from('a foo b foo')) - t.equal(haystack.indexOf(needle), 2) - } - t.end() -}) diff --git a/node_modules/bl/test/test.js b/node_modules/bl/test/test.js deleted file mode 100644 index 42fcad4..0000000 --- a/node_modules/bl/test/test.js +++ /dev/null @@ -1,782 +0,0 @@ -'use strict' - -var tape = require('tape') - , crypto = require('crypto') - , fs = require('fs') - , hash = require('hash_file') - , BufferList = require('../') - , Buffer = require('safe-buffer').Buffer - - , encodings = - ('hex utf8 utf-8 ascii binary base64' - + (process.browser ? '' : ' ucs2 ucs-2 utf16le utf-16le')).split(' ') - -// run the indexOf tests -require('./indexOf') - -tape('single bytes from single buffer', function (t) { - var bl = new BufferList() - bl.append(Buffer.from('abcd')) - - t.equal(bl.length, 4) - t.equal(bl.get(-1), undefined) - t.equal(bl.get(0), 97) - t.equal(bl.get(1), 98) - t.equal(bl.get(2), 99) - t.equal(bl.get(3), 100) - t.equal(bl.get(4), undefined) - - t.end() -}) - -tape('single bytes from multiple buffers', function (t) { - var bl = new BufferList() - bl.append(Buffer.from('abcd')) - bl.append(Buffer.from('efg')) - bl.append(Buffer.from('hi')) - bl.append(Buffer.from('j')) - - t.equal(bl.length, 10) - - t.equal(bl.get(0), 97) - t.equal(bl.get(1), 98) - t.equal(bl.get(2), 99) - t.equal(bl.get(3), 100) - t.equal(bl.get(4), 101) - t.equal(bl.get(5), 102) - t.equal(bl.get(6), 103) - t.equal(bl.get(7), 104) - t.equal(bl.get(8), 105) - t.equal(bl.get(9), 106) - t.end() -}) - -tape('multi bytes from single buffer', function (t) { - var bl = new BufferList() - bl.append(Buffer.from('abcd')) - - t.equal(bl.length, 4) - - t.equal(bl.slice(0, 4).toString('ascii'), 'abcd') - t.equal(bl.slice(0, 3).toString('ascii'), 'abc') - t.equal(bl.slice(1, 4).toString('ascii'), 'bcd') - t.equal(bl.slice(-4, -1).toString('ascii'), 'abc') - - t.end() -}) - -tape('multi bytes from single buffer (negative indexes)', function (t) { - var bl = new BufferList() - bl.append(Buffer.from('buffer')) - - t.equal(bl.length, 6) - - t.equal(bl.slice(-6, -1).toString('ascii'), 'buffe') - t.equal(bl.slice(-6, -2).toString('ascii'), 'buff') - t.equal(bl.slice(-5, -2).toString('ascii'), 'uff') - - t.end() -}) - -tape('multiple bytes from multiple buffers', function (t) { - var bl = new BufferList() - - bl.append(Buffer.from('abcd')) - bl.append(Buffer.from('efg')) - bl.append(Buffer.from('hi')) - bl.append(Buffer.from('j')) - - t.equal(bl.length, 10) - - t.equal(bl.slice(0, 10).toString('ascii'), 'abcdefghij') - t.equal(bl.slice(3, 10).toString('ascii'), 'defghij') - t.equal(bl.slice(3, 6).toString('ascii'), 'def') - t.equal(bl.slice(3, 8).toString('ascii'), 'defgh') - t.equal(bl.slice(5, 10).toString('ascii'), 'fghij') - t.equal(bl.slice(-7, -4).toString('ascii'), 'def') - - t.end() -}) - -tape('multiple bytes from multiple buffer lists', function (t) { - var bl = new BufferList() - - bl.append(new BufferList([ Buffer.from('abcd'), Buffer.from('efg') ])) - bl.append(new BufferList([ Buffer.from('hi'), Buffer.from('j') ])) - - t.equal(bl.length, 10) - - t.equal(bl.slice(0, 10).toString('ascii'), 'abcdefghij') - - t.equal(bl.slice(3, 10).toString('ascii'), 'defghij') - t.equal(bl.slice(3, 6).toString('ascii'), 'def') - t.equal(bl.slice(3, 8).toString('ascii'), 'defgh') - t.equal(bl.slice(5, 10).toString('ascii'), 'fghij') - - t.end() -}) - -// same data as previous test, just using nested constructors -tape('multiple bytes from crazy nested buffer lists', function (t) { - var bl = new BufferList() - - bl.append(new BufferList([ - new BufferList([ - new BufferList(Buffer.from('abc')) - , Buffer.from('d') - , new BufferList(Buffer.from('efg')) - ]) - , new BufferList([ Buffer.from('hi') ]) - , new BufferList(Buffer.from('j')) - ])) - - t.equal(bl.length, 10) - - t.equal(bl.slice(0, 10).toString('ascii'), 'abcdefghij') - - t.equal(bl.slice(3, 10).toString('ascii'), 'defghij') - t.equal(bl.slice(3, 6).toString('ascii'), 'def') - t.equal(bl.slice(3, 8).toString('ascii'), 'defgh') - t.equal(bl.slice(5, 10).toString('ascii'), 'fghij') - - t.end() -}) - -tape('append accepts arrays of Buffers', function (t) { - var bl = new BufferList() - bl.append(Buffer.from('abc')) - bl.append([ Buffer.from('def') ]) - bl.append([ Buffer.from('ghi'), Buffer.from('jkl') ]) - bl.append([ Buffer.from('mnop'), Buffer.from('qrstu'), Buffer.from('vwxyz') ]) - t.equal(bl.length, 26) - t.equal(bl.slice().toString('ascii'), 'abcdefghijklmnopqrstuvwxyz') - t.end() -}) - -tape('append accepts arrays of BufferLists', function (t) { - var bl = new BufferList() - bl.append(Buffer.from('abc')) - bl.append([ new BufferList('def') ]) - bl.append(new BufferList([ Buffer.from('ghi'), new BufferList('jkl') ])) - bl.append([ Buffer.from('mnop'), new BufferList([ Buffer.from('qrstu'), Buffer.from('vwxyz') ]) ]) - t.equal(bl.length, 26) - t.equal(bl.slice().toString('ascii'), 'abcdefghijklmnopqrstuvwxyz') - t.end() -}) - -tape('append chainable', function (t) { - var bl = new BufferList() - t.ok(bl.append(Buffer.from('abcd')) === bl) - t.ok(bl.append([ Buffer.from('abcd') ]) === bl) - t.ok(bl.append(new BufferList(Buffer.from('abcd'))) === bl) - t.ok(bl.append([ new BufferList(Buffer.from('abcd')) ]) === bl) - t.end() -}) - -tape('append chainable (test results)', function (t) { - var bl = new BufferList('abc') - .append([ new BufferList('def') ]) - .append(new BufferList([ Buffer.from('ghi'), new BufferList('jkl') ])) - .append([ Buffer.from('mnop'), new BufferList([ Buffer.from('qrstu'), Buffer.from('vwxyz') ]) ]) - - t.equal(bl.length, 26) - t.equal(bl.slice().toString('ascii'), 'abcdefghijklmnopqrstuvwxyz') - t.end() -}) - -tape('consuming from multiple buffers', function (t) { - var bl = new BufferList() - - bl.append(Buffer.from('abcd')) - bl.append(Buffer.from('efg')) - bl.append(Buffer.from('hi')) - bl.append(Buffer.from('j')) - - t.equal(bl.length, 10) - - t.equal(bl.slice(0, 10).toString('ascii'), 'abcdefghij') - - bl.consume(3) - t.equal(bl.length, 7) - t.equal(bl.slice(0, 7).toString('ascii'), 'defghij') - - bl.consume(2) - t.equal(bl.length, 5) - t.equal(bl.slice(0, 5).toString('ascii'), 'fghij') - - bl.consume(1) - t.equal(bl.length, 4) - t.equal(bl.slice(0, 4).toString('ascii'), 'ghij') - - bl.consume(1) - t.equal(bl.length, 3) - t.equal(bl.slice(0, 3).toString('ascii'), 'hij') - - bl.consume(2) - t.equal(bl.length, 1) - t.equal(bl.slice(0, 1).toString('ascii'), 'j') - - t.end() -}) - -tape('complete consumption', function (t) { - var bl = new BufferList() - - bl.append(Buffer.from('a')) - bl.append(Buffer.from('b')) - - bl.consume(2) - - t.equal(bl.length, 0) - t.equal(bl._bufs.length, 0) - - t.end() -}) - -tape('test readUInt8 / readInt8', function (t) { - var buf1 = Buffer.alloc(1) - , buf2 = Buffer.alloc(3) - , buf3 = Buffer.alloc(3) - , bl = new BufferList() - - buf2[1] = 0x3 - buf2[2] = 0x4 - buf3[0] = 0x23 - buf3[1] = 0x42 - - bl.append(buf1) - bl.append(buf2) - bl.append(buf3) - - t.equal(bl.readUInt8(2), 0x3) - t.equal(bl.readInt8(2), 0x3) - t.equal(bl.readUInt8(3), 0x4) - t.equal(bl.readInt8(3), 0x4) - t.equal(bl.readUInt8(4), 0x23) - t.equal(bl.readInt8(4), 0x23) - t.equal(bl.readUInt8(5), 0x42) - t.equal(bl.readInt8(5), 0x42) - t.end() -}) - -tape('test readUInt16LE / readUInt16BE / readInt16LE / readInt16BE', function (t) { - var buf1 = Buffer.alloc(1) - , buf2 = Buffer.alloc(3) - , buf3 = Buffer.alloc(3) - , bl = new BufferList() - - buf2[1] = 0x3 - buf2[2] = 0x4 - buf3[0] = 0x23 - buf3[1] = 0x42 - - bl.append(buf1) - bl.append(buf2) - bl.append(buf3) - - t.equal(bl.readUInt16BE(2), 0x0304) - t.equal(bl.readUInt16LE(2), 0x0403) - t.equal(bl.readInt16BE(2), 0x0304) - t.equal(bl.readInt16LE(2), 0x0403) - t.equal(bl.readUInt16BE(3), 0x0423) - t.equal(bl.readUInt16LE(3), 0x2304) - t.equal(bl.readInt16BE(3), 0x0423) - t.equal(bl.readInt16LE(3), 0x2304) - t.equal(bl.readUInt16BE(4), 0x2342) - t.equal(bl.readUInt16LE(4), 0x4223) - t.equal(bl.readInt16BE(4), 0x2342) - t.equal(bl.readInt16LE(4), 0x4223) - t.end() -}) - -tape('test readUInt32LE / readUInt32BE / readInt32LE / readInt32BE', function (t) { - var buf1 = Buffer.alloc(1) - , buf2 = Buffer.alloc(3) - , buf3 = Buffer.alloc(3) - , bl = new BufferList() - - buf2[1] = 0x3 - buf2[2] = 0x4 - buf3[0] = 0x23 - buf3[1] = 0x42 - - bl.append(buf1) - bl.append(buf2) - bl.append(buf3) - - t.equal(bl.readUInt32BE(2), 0x03042342) - t.equal(bl.readUInt32LE(2), 0x42230403) - t.equal(bl.readInt32BE(2), 0x03042342) - t.equal(bl.readInt32LE(2), 0x42230403) - t.end() -}) - -tape('test readUIntLE / readUIntBE / readIntLE / readIntBE', function (t) { - var buf1 = Buffer.alloc(1) - , buf2 = Buffer.alloc(3) - , buf3 = Buffer.alloc(3) - , bl = new BufferList() - - buf2[0] = 0x2 - buf2[1] = 0x3 - buf2[2] = 0x4 - buf3[0] = 0x23 - buf3[1] = 0x42 - buf3[2] = 0x61 - - bl.append(buf1) - bl.append(buf2) - bl.append(buf3) - - t.equal(bl.readUIntBE(1, 1), 0x02) - t.equal(bl.readUIntBE(1, 2), 0x0203) - t.equal(bl.readUIntBE(1, 3), 0x020304) - t.equal(bl.readUIntBE(1, 4), 0x02030423) - t.equal(bl.readUIntBE(1, 5), 0x0203042342) - t.equal(bl.readUIntBE(1, 6), 0x020304234261) - t.equal(bl.readUIntLE(1, 1), 0x02) - t.equal(bl.readUIntLE(1, 2), 0x0302) - t.equal(bl.readUIntLE(1, 3), 0x040302) - t.equal(bl.readUIntLE(1, 4), 0x23040302) - t.equal(bl.readUIntLE(1, 5), 0x4223040302) - t.equal(bl.readUIntLE(1, 6), 0x614223040302) - t.equal(bl.readIntBE(1, 1), 0x02) - t.equal(bl.readIntBE(1, 2), 0x0203) - t.equal(bl.readIntBE(1, 3), 0x020304) - t.equal(bl.readIntBE(1, 4), 0x02030423) - t.equal(bl.readIntBE(1, 5), 0x0203042342) - t.equal(bl.readIntBE(1, 6), 0x020304234261) - t.equal(bl.readIntLE(1, 1), 0x02) - t.equal(bl.readIntLE(1, 2), 0x0302) - t.equal(bl.readIntLE(1, 3), 0x040302) - t.equal(bl.readIntLE(1, 4), 0x23040302) - t.equal(bl.readIntLE(1, 5), 0x4223040302) - t.equal(bl.readIntLE(1, 6), 0x614223040302) - t.end() -}) - -tape('test readFloatLE / readFloatBE', function (t) { - var buf1 = Buffer.alloc(1) - , buf2 = Buffer.alloc(3) - , buf3 = Buffer.alloc(3) - , bl = new BufferList() - - buf2[1] = 0x00 - buf2[2] = 0x00 - buf3[0] = 0x80 - buf3[1] = 0x3f - - bl.append(buf1) - bl.append(buf2) - bl.append(buf3) - - t.equal(bl.readFloatLE(2), 0x01) - t.end() -}) - -tape('test readDoubleLE / readDoubleBE', function (t) { - var buf1 = Buffer.alloc(1) - , buf2 = Buffer.alloc(3) - , buf3 = Buffer.alloc(10) - , bl = new BufferList() - - buf2[1] = 0x55 - buf2[2] = 0x55 - buf3[0] = 0x55 - buf3[1] = 0x55 - buf3[2] = 0x55 - buf3[3] = 0x55 - buf3[4] = 0xd5 - buf3[5] = 0x3f - - bl.append(buf1) - bl.append(buf2) - bl.append(buf3) - - t.equal(bl.readDoubleLE(2), 0.3333333333333333) - t.end() -}) - -tape('test toString', function (t) { - var bl = new BufferList() - - bl.append(Buffer.from('abcd')) - bl.append(Buffer.from('efg')) - bl.append(Buffer.from('hi')) - bl.append(Buffer.from('j')) - - t.equal(bl.toString('ascii', 0, 10), 'abcdefghij') - t.equal(bl.toString('ascii', 3, 10), 'defghij') - t.equal(bl.toString('ascii', 3, 6), 'def') - t.equal(bl.toString('ascii', 3, 8), 'defgh') - t.equal(bl.toString('ascii', 5, 10), 'fghij') - - t.end() -}) - -tape('test toString encoding', function (t) { - var bl = new BufferList() - , b = Buffer.from('abcdefghij\xff\x00') - - bl.append(Buffer.from('abcd')) - bl.append(Buffer.from('efg')) - bl.append(Buffer.from('hi')) - bl.append(Buffer.from('j')) - bl.append(Buffer.from('\xff\x00')) - - encodings.forEach(function (enc) { - t.equal(bl.toString(enc), b.toString(enc), enc) - }) - - t.end() -}) - -tape('uninitialized memory', function (t) { - const secret = crypto.randomBytes(256) - for (let i = 0; i < 1e6; i++) { - const clone = Buffer.from(secret) - const bl = new BufferList() - bl.append(Buffer.from('a')) - bl.consume(-1024) - const buf = bl.slice(1) - if (buf.indexOf(clone) !== -1) { - t.fail(`Match (at ${i})`) - break - } - } - t.end() -}) - -!process.browser && tape('test stream', function (t) { - var random = crypto.randomBytes(65534) - , rndhash = hash(random, 'md5') - , md5sum = crypto.createHash('md5') - , bl = new BufferList(function (err, buf) { - t.ok(Buffer.isBuffer(buf)) - t.ok(err === null) - t.equal(rndhash, hash(bl.slice(), 'md5')) - t.equal(rndhash, hash(buf, 'md5')) - - bl.pipe(fs.createWriteStream('/tmp/bl_test_rnd_out.dat')) - .on('close', function () { - var s = fs.createReadStream('/tmp/bl_test_rnd_out.dat') - s.on('data', md5sum.update.bind(md5sum)) - s.on('end', function() { - t.equal(rndhash, md5sum.digest('hex'), 'woohoo! correct hash!') - t.end() - }) - }) - - }) - - fs.writeFileSync('/tmp/bl_test_rnd.dat', random) - fs.createReadStream('/tmp/bl_test_rnd.dat').pipe(bl) -}) - -tape('instantiation with Buffer', function (t) { - var buf = crypto.randomBytes(1024) - , buf2 = crypto.randomBytes(1024) - , b = BufferList(buf) - - t.equal(buf.toString('hex'), b.slice().toString('hex'), 'same buffer') - b = BufferList([ buf, buf2 ]) - t.equal(b.slice().toString('hex'), Buffer.concat([ buf, buf2 ]).toString('hex'), 'same buffer') - t.end() -}) - -tape('test String appendage', function (t) { - var bl = new BufferList() - , b = Buffer.from('abcdefghij\xff\x00') - - bl.append('abcd') - bl.append('efg') - bl.append('hi') - bl.append('j') - bl.append('\xff\x00') - - encodings.forEach(function (enc) { - t.equal(bl.toString(enc), b.toString(enc)) - }) - - t.end() -}) - -tape('test Number appendage', function (t) { - var bl = new BufferList() - , b = Buffer.from('1234567890') - - bl.append(1234) - bl.append(567) - bl.append(89) - bl.append(0) - - encodings.forEach(function (enc) { - t.equal(bl.toString(enc), b.toString(enc)) - }) - - t.end() -}) - -tape('write nothing, should get empty buffer', function (t) { - t.plan(3) - BufferList(function (err, data) { - t.notOk(err, 'no error') - t.ok(Buffer.isBuffer(data), 'got a buffer') - t.equal(0, data.length, 'got a zero-length buffer') - t.end() - }).end() -}) - -tape('unicode string', function (t) { - t.plan(2) - var inp1 = '\u2600' - , inp2 = '\u2603' - , exp = inp1 + ' and ' + inp2 - , bl = BufferList() - bl.write(inp1) - bl.write(' and ') - bl.write(inp2) - t.equal(exp, bl.toString()) - t.equal(Buffer.from(exp).toString('hex'), bl.toString('hex')) -}) - -tape('should emit finish', function (t) { - var source = BufferList() - , dest = BufferList() - - source.write('hello') - source.pipe(dest) - - dest.on('finish', function () { - t.equal(dest.toString('utf8'), 'hello') - t.end() - }) -}) - -tape('basic copy', function (t) { - var buf = crypto.randomBytes(1024) - , buf2 = Buffer.alloc(1024) - , b = BufferList(buf) - - b.copy(buf2) - t.equal(b.slice().toString('hex'), buf2.toString('hex'), 'same buffer') - t.end() -}) - -tape('copy after many appends', function (t) { - var buf = crypto.randomBytes(512) - , buf2 = Buffer.alloc(1024) - , b = BufferList(buf) - - b.append(buf) - b.copy(buf2) - t.equal(b.slice().toString('hex'), buf2.toString('hex'), 'same buffer') - t.end() -}) - -tape('copy at a precise position', function (t) { - var buf = crypto.randomBytes(1004) - , buf2 = Buffer.alloc(1024) - , b = BufferList(buf) - - b.copy(buf2, 20) - t.equal(b.slice().toString('hex'), buf2.slice(20).toString('hex'), 'same buffer') - t.end() -}) - -tape('copy starting from a precise location', function (t) { - var buf = crypto.randomBytes(10) - , buf2 = Buffer.alloc(5) - , b = BufferList(buf) - - b.copy(buf2, 0, 5) - t.equal(b.slice(5).toString('hex'), buf2.toString('hex'), 'same buffer') - t.end() -}) - -tape('copy in an interval', function (t) { - var rnd = crypto.randomBytes(10) - , b = BufferList(rnd) // put the random bytes there - , actual = Buffer.alloc(3) - , expected = Buffer.alloc(3) - - rnd.copy(expected, 0, 5, 8) - b.copy(actual, 0, 5, 8) - - t.equal(actual.toString('hex'), expected.toString('hex'), 'same buffer') - t.end() -}) - -tape('copy an interval between two buffers', function (t) { - var buf = crypto.randomBytes(10) - , buf2 = Buffer.alloc(10) - , b = BufferList(buf) - - b.append(buf) - b.copy(buf2, 0, 5, 15) - - t.equal(b.slice(5, 15).toString('hex'), buf2.toString('hex'), 'same buffer') - t.end() -}) - -tape('shallow slice across buffer boundaries', function (t) { - var bl = new BufferList(['First', 'Second', 'Third']) - - t.equal(bl.shallowSlice(3, 13).toString(), 'stSecondTh') - t.end() -}) - -tape('shallow slice within single buffer', function (t) { - t.plan(2) - var bl = new BufferList(['First', 'Second', 'Third']) - - t.equal(bl.shallowSlice(5, 10).toString(), 'Secon') - t.equal(bl.shallowSlice(7, 10).toString(), 'con') - t.end() -}) - -tape('shallow slice single buffer', function (t) { - t.plan(3) - var bl = new BufferList(['First', 'Second', 'Third']) - - t.equal(bl.shallowSlice(0, 5).toString(), 'First') - t.equal(bl.shallowSlice(5, 11).toString(), 'Second') - t.equal(bl.shallowSlice(11, 16).toString(), 'Third') -}) - -tape('shallow slice with negative or omitted indices', function (t) { - t.plan(4) - var bl = new BufferList(['First', 'Second', 'Third']) - - t.equal(bl.shallowSlice().toString(), 'FirstSecondThird') - t.equal(bl.shallowSlice(5).toString(), 'SecondThird') - t.equal(bl.shallowSlice(5, -3).toString(), 'SecondTh') - t.equal(bl.shallowSlice(-8).toString(), 'ondThird') -}) - -tape('shallow slice does not make a copy', function (t) { - t.plan(1) - var buffers = [Buffer.from('First'), Buffer.from('Second'), Buffer.from('Third')] - var bl = (new BufferList(buffers)).shallowSlice(5, -3) - - buffers[1].fill('h') - buffers[2].fill('h') - - t.equal(bl.toString(), 'hhhhhhhh') -}) - -tape('shallow slice with 0 length', function (t) { - t.plan(1) - var buffers = [Buffer.from('First'), Buffer.from('Second'), Buffer.from('Third')] - var bl = (new BufferList(buffers)).shallowSlice(0, 0) - t.equal(bl.length, 0) -}) - -tape('shallow slice with 0 length from middle', function (t) { - t.plan(1) - var buffers = [Buffer.from('First'), Buffer.from('Second'), Buffer.from('Third')] - var bl = (new BufferList(buffers)).shallowSlice(10, 10) - t.equal(bl.length, 0) -}) - -tape('duplicate', function (t) { - t.plan(2) - - var bl = new BufferList('abcdefghij\xff\x00') - , dup = bl.duplicate() - - t.equal(bl.prototype, dup.prototype) - t.equal(bl.toString('hex'), dup.toString('hex')) -}) - -tape('destroy no pipe', function (t) { - t.plan(2) - - var bl = new BufferList('alsdkfja;lsdkfja;lsdk') - bl.destroy() - - t.equal(bl._bufs.length, 0) - t.equal(bl.length, 0) -}) - -!process.browser && tape('destroy with pipe before read end', function (t) { - t.plan(2) - - var bl = new BufferList() - fs.createReadStream(__dirname + '/test.js') - .pipe(bl) - - bl.destroy() - - t.equal(bl._bufs.length, 0) - t.equal(bl.length, 0) - -}) - -!process.browser && tape('destroy with pipe before read end with race', function (t) { - t.plan(2) - - var bl = new BufferList() - fs.createReadStream(__dirname + '/test.js') - .pipe(bl) - - setTimeout(function () { - bl.destroy() - setTimeout(function () { - t.equal(bl._bufs.length, 0) - t.equal(bl.length, 0) - }, 500) - }, 500) -}) - -!process.browser && tape('destroy with pipe after read end', function (t) { - t.plan(2) - - var bl = new BufferList() - fs.createReadStream(__dirname + '/test.js') - .on('end', onEnd) - .pipe(bl) - - function onEnd () { - bl.destroy() - - t.equal(bl._bufs.length, 0) - t.equal(bl.length, 0) - } -}) - -!process.browser && tape('destroy with pipe while writing to a destination', function (t) { - t.plan(4) - - var bl = new BufferList() - , ds = new BufferList() - - fs.createReadStream(__dirname + '/test.js') - .on('end', onEnd) - .pipe(bl) - - function onEnd () { - bl.pipe(ds) - - setTimeout(function () { - bl.destroy() - - t.equals(bl._bufs.length, 0) - t.equals(bl.length, 0) - - ds.destroy() - - t.equals(bl._bufs.length, 0) - t.equals(bl.length, 0) - - }, 100) - } -}) - -!process.browser && tape('handle error', function (t) { - t.plan(2) - fs.createReadStream('/does/not/exist').pipe(BufferList(function (err, data) { - t.ok(err instanceof Error, 'has error') - t.notOk(data, 'no data') - })) -}) |