Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(1189)

Unified Diff: pkg/dev_compiler/tool/input_sdk/lib/core/double.dart

Issue 2698353003: unfork DDC's copy of most SDK libraries (Closed)
Patch Set: revert core_patch Created 3 years, 10 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View side-by-side diff with in-line comments
Download patch
Index: pkg/dev_compiler/tool/input_sdk/lib/core/double.dart
diff --git a/pkg/dev_compiler/tool/input_sdk/lib/core/double.dart b/pkg/dev_compiler/tool/input_sdk/lib/core/double.dart
deleted file mode 100644
index 87a8b51073d599f14d4b76f5a464cd8cc0075b0a..0000000000000000000000000000000000000000
--- a/pkg/dev_compiler/tool/input_sdk/lib/core/double.dart
+++ /dev/null
@@ -1,206 +0,0 @@
-// Copyright (c) 2012, the Dart project authors. Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-part of dart.core;
-
-// TODO: Convert this abstract class into a concrete class double
-// that uses the patch class functionality to account for the
-// different platform implementations.
-
-/**
- * A double-precision floating point number.
- *
- * Representation of Dart doubles containing double specific constants
- * and operations and specializations of operations inherited from
- * [num]. Dart doubles are 64-bit floating-point numbers as specified in the
- * IEEE 754 standard.
- *
- * The [double] type is contagious. Operations on [double]s return
- * [double] results.
- *
- * It is a compile-time error for a class to attempt to extend or implement
- * double.
- */
-abstract class double extends num {
- static const double NAN = 0.0 / 0.0;
- static const double INFINITY = 1.0 / 0.0;
- static const double NEGATIVE_INFINITY = -INFINITY;
- static const double MIN_POSITIVE = 5e-324;
- static const double MAX_FINITE = 1.7976931348623157e+308;
-
- double remainder(num other);
-
- /** Addition operator. */
- double operator +(num other);
-
- /** Subtraction operator. */
- double operator -(num other);
-
- /** Multiplication operator. */
- double operator *(num other);
-
- double operator %(num other);
-
- /** Division operator. */
- double operator /(num other);
-
- /**
- * Truncating division operator.
- *
- * The result of the truncating division `a ~/ b` is equivalent to
- * `(a / b).truncate()`.
- */
- int operator ~/(num other);
-
- /** Negate operator. */
- double operator -();
-
- /** Returns the absolute value of this [double]. */
- double abs();
-
- /**
- * Returns the sign of the double's numerical value.
- *
- * Returns -1.0 if the value is less than zero,
- * +1.0 if the value is greater than zero,
- * and the value itself if it is -0.0, 0.0 or NaN.
- */
- double get sign;
-
- /**
- * Returns the integer closest to `this`.
- *
- * Rounds away from zero when there is no closest integer:
- * `(3.5).round() == 4` and `(-3.5).round() == -4`.
- *
- * If `this` is not finite (`NaN` or infinity), throws an [UnsupportedError].
- */
- int round();
-
- /**
- * Returns the greatest integer no greater than `this`.
- *
- * If `this` is not finite (`NaN` or infinity), throws an [UnsupportedError].
- */
- int floor();
-
- /**
- * Returns the least integer no smaller than `this`.
- *
- * If `this` is not finite (`NaN` or infinity), throws an [UnsupportedError].
- */
- int ceil();
-
- /**
- * Returns the integer obtained by discarding any fractional
- * digits from `this`.
- *
- * If `this` is not finite (`NaN` or infinity), throws an [UnsupportedError].
- */
- int truncate();
-
- /**
- * Returns the integer double value closest to `this`.
- *
- * Rounds away from zero when there is no closest integer:
- * `(3.5).roundToDouble() == 4` and `(-3.5).roundToDouble() == -4`.
- *
- * If this is already an integer valued double, including `-0.0`, or it is not
- * a finite value, the value is returned unmodified.
- *
- * For the purpose of rounding, `-0.0` is considered to be below `0.0`,
- * and `-0.0` is therefore considered closer to negative numbers than `0.0`.
- * This means that for a value, `d` in the range `-0.5 < d < 0.0`,
- * the result is `-0.0`.
- */
- double roundToDouble();
-
- /**
- * Returns the greatest integer double value no greater than `this`.
- *
- * If this is already an integer valued double, including `-0.0`, or it is not
- * a finite value, the value is returned unmodified.
- *
- * For the purpose of rounding, `-0.0` is considered to be below `0.0`.
- * A number `d` in the range `0.0 < d < 1.0` will return `0.0`.
- */
- double floorToDouble();
-
- /**
- * Returns the least integer double value no smaller than `this`.
- *
- * If this is already an integer valued double, including `-0.0`, or it is not
- * a finite value, the value is returned unmodified.
- *
- * For the purpose of rounding, `-0.0` is considered to be below `0.0`.
- * A number `d` in the range `-1.0 < d < 0.0` will return `-0.0`.
- */
- double ceilToDouble();
-
- /**
- * Returns the integer double value obtained by discarding any fractional
- * digits from `this`.
- *
- * If this is already an integer valued double, including `-0.0`, or it is not
- * a finite value, the value is returned unmodified.
- *
- * For the purpose of rounding, `-0.0` is considered to be below `0.0`.
- * A number `d` in the range `-1.0 < d < 0.0` will return `-0.0`, and
- * in the range `0.0 < d < 1.0` it will return 0.0.
- */
- double truncateToDouble();
-
- /**
- * Provide a representation of this [double] value.
- *
- * The representation is a number literal such that the closest double value
- * to the representation's mathematical value is this [double].
- *
- * Returns "NaN" for the Not-a-Number value.
- * Returns "Infinity" and "-Infinity" for positive and negative Infinity.
- * Returns "-0.0" for negative zero.
- *
- * For all doubles, `d`, converting to a string and parsing the string back
- * gives the same value again: `d == double.parse(d.toString())` (except when
- * `d` is NaN).
- */
- String toString();
-
- /**
- * Parse [source] as an double literal and return its value.
- *
- * Accepts an optional sign (`+` or `-`) followed by either the characters
- * "Infinity", the characters "NaN" or a floating-point representation.
- * A floating-point representation is composed of a mantissa and an optional
- * exponent part. The mantissa is either a decimal point (`.`) followed by a
- * sequence of (decimal) digits, or a sequence of digits
- * optionally followed by a decimal point and optionally more digits. The
- * (optional) exponent part consists of the character "e" or "E", an optional
- * sign, and one or more digits.
- *
- * Leading and trailing whitespace is ignored.
- *
- * If the [source] is not a valid double literal, the [onError]
- * is called with the [source] as argument, and its return value is
- * used instead. If no `onError` is provided, a [FormatException]
- * is thrown instead.
- *
- * The [onError] function is only invoked if [source] is a [String] with an
- * invalid format. It is not invoked if the [source] is invalid for some
- * other reason, for example by being `null`.
- *
- * Examples of accepted strings:
- *
- * "3.14"
- * " 3.14 \xA0"
- * "0."
- * ".0"
- * "-1.e3"
- * "1234E+7"
- * "+.12e-9"
- * "-NaN"
- */
- external static double parse(String source,
- [double onError(String source)]);
-}
« no previous file with comments | « pkg/dev_compiler/tool/input_sdk/lib/core/date_time.dart ('k') | pkg/dev_compiler/tool/input_sdk/lib/core/duration.dart » ('j') | no next file with comments »

Powered by Google App Engine
This is Rietveld 408576698