PROPOSAL: Underscores in Numbers (Version 2)
Derek Foster
vapor1 at teleport.com
Wed May 20 21:06:03 PDT 2009
Replies inline.
-----Original Message-----
>From: Joe Darcy <Joe.Darcy at Sun.COM>
>Sent: May 7, 2009 9:59 PM
>To: Derek Foster <vapor1 at teleport.com>
>Cc: coin-dev at openjdk.java.net
>Subject: Re: PROPOSAL: Underscores in Numbers (Version 2)
>
>Hello.
>
>A few comments.
>
>This is well-localized change and judicious use of the feature would
>improve the readability of some numerical constants.
>
>However, I generally agree with Bruce's earlier comments that phone
>numbers, credit card numbers, social security numbers, etc. are not a
>good motivating example since they are really strings whose digits are
>(mostly) constrained to be numeric as opposed to actual integral or
>real-valued data. (Using floating-point types to store dollars and
>cents is also ill-advised because fractions like 0.12 generally cannot
>be stored exactly as binary fractions.)
Hi, Joe.
I agree with this comment -- my major intended use of the feature would be for separating digits in large hex and/or binary (I hope) numbers, and secondarily for making large decimal numbers more readable by introducing order-of-magnitude marks. (Every three digits, for Americans). I mostly included the credit card numbers, etc. just to illustrate that the feature could potentially be used for numbers in other regular formats, if desired.
>For more substantive feedback, I believe the grammar changes still need
>some work.
I'll take another look at the proposal and revise the grammar to handle your concerns. Thanks for pointing out these issues.
Derek
> First, my reading of the grammar is that it is ambiguous
>regarding underscores, there are multiple ways for "0__1" to be parsed
>and the grammar as written does not ban "5_" (NonZeroDigit followed by
>Undersccores_opt). How about treating underscore as an extra digit in
>some cases, something like:
>
>Decimal Numeral
> 0
> NonZeroDigit DigitsAndUnderscores_opt
>
>DigitsAndUnderscores
> DigitOrUnderscore_opt Digit
>
>DigitOrUnderscore
> 0 ... 9 _
>
>This grammar:
>
>* forces a nonzero leading digit
>* forces a non-underscore final character
>* allows any number of intermixed digits and underscores in between
>
>-Joe
>
>On 04/30/09 09:01 PM, Derek Foster wrote:
>> Per feedback on the Project Coin mailing list, the following proposal has been revised to disallow placing underscores at the beginning or end of a sequence of digits within a number. I have also revised the value used for the "alsoMaxLong" example (as it was previously illegal), as well as the other examples showing where underscores can/cannot be placed.
>>
>> AUTHOR(S): Derek Foster
>>
>> OVERVIEW
>>
>> In Java, currently, numbers of various types currently are expressed in their pure form, as a long string of digits possibly interspersed with other punctuation (periods, an exponent specifier, etc.) needed to separate distinct sections of the number. While this is easy for a compiler to process, it is often difficult for a human being to visually parse.
>>
>> The ability of a human to visually separate separate items tops out somewhere near "seven plus or minus two" items. Research done by telephone companies suggests that for many practical purposes, the longest string of numbers an average human can successfully hold in memory at a time is around three or four. Also, it is difficult for the human eye to find common positions in numbers that have no break in their visual structure.
>>
>> As a result, most numbers that humans deal with in day-to-day life have separators included in them to break the long sequence of digits into manageable chunks that are easy to deal with as separate entities. This includes items such as (apologies to non-USA readers...):
>>
>> Phone numbers: 555-555-1212
>> Credit card numbers: 1234-5678-9012-3456
>> Social security numbers: 999-99-9999
>> Monetary amounts: $12,345,132.12
>>
>> and a wide variety of other types of numbers.
>>
>> However, Java provides no way to add these kinds of visual separators into a number. Java expects the number to be essentially an unbroken string of digits.
>>
>> This proposal suggests that Java follow the lead of the Ruby programming language in allowing the underscore character to be inserted into numbers in most positions, for readability purposes.
>>
>>
>> FEATURE SUMMARY:
>>
>> Java numeric literals will allow underscores to be placed in (nearly) arbitrary positions within the number, at the programmer's discretion, for readability purposes. These underscores shall be ignored by the compiler for the purposes of code generation.
>>
>>
>> MAJOR ADVANTAGE:
>>
>> Programmers won't have to visually parse long strings of digits (a task humans are quite poor at). The internal digit-oriented structure of many numbers can be made more clear.
>>
>>
>> MAJOR BENEFIT:
>>
>> Increased readability of code.
>>
>>
>> MAJOR DISADVANTAGE:
>>
>> The number parsers in the Java compiler would have to be adjusted to parse and ignore the underscores. This is a small amount of effort, but nonzero. There might also be some small performance impact.
>>
>> If someone were to use this feature inappropriately, it could result in difficult to read code.
>>
>>
>> ALTERNATIVES:
>>
>> Do without separators in numbers, or use some other character for them.
>>
>>
>> EXAMPLES
>>
>>
>> SIMPLE EXAMPLE: Show the simplest possible program utilizing the new feature.
>>
>> int phoneNumber = 555_555_1212;
>>
>>
>> ADVANCED EXAMPLE:
>>
>> long creditCardNumber = 1234_5678_9012_3456L;
>> long socialSecurityNumbers = 999_99_9999L;
>> float monetaryAmount = 12_345_132.12;
>> long hexBytes = 0xFF_EC_DE_5E;
>> long hexWords = 0xFFEC_DE5E;
>> long maxLong = 0x7fff_ffff_ffff_ffffL;
>> long alsoMaxLong = 9_223_372_036_854_775_807L;
>> double whyWouldYouEverDoThis = 0x1_.ffff_ffff_ffff_fp10_23;
>>
>> (Additionally, if binary literals are ever added to the Java language, the following might also be possible...
>> byte nybbles = 0b0010_0101;
>> long bytes = 0b11010010_01101001_10010100_10010010;
>> int weirdBitfields = 0b000_10_101;
>> )
>>
>> Note that according to this proposal, underscores can only be placed between digits. They cannot be placed by themselves in positions where a string of digits would normally be expected:
>>
>> int x1 = _52; // This is an identifier, not a numeric literal.
>> int x2 = 5_2; // OK. (Decimal literal)
>> int x2 = 52_; // Illegal. (Underscores must always be between digits)
>> int x3 = 5_______2; // OK. (Decimal literal.)
>>
>> int x4 = 0_x52; // Illegal. Can't put underscores in the "0x" radix prefix.
>> int x5 = 0x_52; // Illegal. (Underscores must always be between digits)
>> int x6 = 0x5_2; // OK. (Hexadecimal literal)
>> int x6 = 0x52_; // Illegal. (Underscores must always be between digits)
>> int x6 = 0x_; // Illegal. (Not valid with the underscore removed)
>>
>> int x7 = 0_52; // OK. (Octal literal)
>> int x7 = 05_2; // OK. (Octal literal)
>> int x8 = 052_; // Illegal. (Underscores must always be between digits)
>>
>>
>> DETAILS
>>
>> SPECIFICATION:
>>
>> DECIMALS:
>>
>> Section 3.10.1 ("Integer Literals") of the Java Language Specification 3rd edition shall be modified like so:
>>
>> Underscores:
>> _
>> _ Underscores
>>
>> DecimalNumeral:
>> 0
>> NonZeroDigit Underscores_opt DigitsAndUnderscores_opt
>>
>> DigitsAndUnderscores:
>> Digit
>> DigitsAndUnderscores Underscores_opt Digit
>>
>> Digit:
>> 0
>> NonZeroDigit
>>
>> NonZeroDigit: one of
>> 1 2 3 4 5 6 7 8 9
>>
>> HexNumeral:
>> 0 x HexDigitsAndUnderscores
>> 0 X HexDigitsAndUnderscores
>>
>> HexDigitsAndUnderscores:
>> HexDigit
>> HexDigit Underscores_opt HexDigitsAndUnderscores
>>
>> HexDigit: one of
>> 0 1 2 3 4 5 6 7 8 9 a b c d e f A B C D E F
>>
>> OctalNumeral:
>> 0 OctalDigitsAndUnderscores
>>
>> OctalDigitsAndUnderscores:
>> OctalDigit
>> OctalDigit Underscores_opt OctalDigitsAndUnderscores
>>
>> OctalDigit: one of
>> 0 1 2 3 4 5 6 7
>>
>> Section 3.10.2 ("Floating-Point Literals") would be modified as follows:
>>
>>
>> FloatingPointLiteral:
>> DecimalFloatingPointLiteral
>> HexadecimalFloatingPointLiteral
>>
>> DecimalFloatingPointLiteral:
>> DigitsAndUnderscores_opt . DigitsAndUnderscores_opt ExponentPart_opt FloatTypeSuffix_opt
>> . DigitsAndUnderscores ExponentPartopt FloatTypeSuffix_opt
>> DigitsAndUnderscores_opt ExponentPart FloatTypeSuffix_opt
>> DigitsAndUnderscores_opt ExponentPart_opt FloatTypeSuffix
>>
>> ExponentPart:
>> ExponentIndicator SignedInteger
>>
>> ExponentIndicator: one of
>> e E
>>
>> SignedInteger:
>> Sign_opt DigitsAndUnderscores
>>
>> Sign: one of
>> + -
>>
>> FloatTypeSuffix: one of
>> f F d D
>>
>> HexadecimalFloatingPointLiteral:
>> HexSignificand BinaryExponent FloatTypeSuffix_opt
>>
>> HexSignificand:
>> HexNumeral
>> HexNumeral .
>> 0x HexDigitsAndUnderscores_opt . HexDigitsAndUnderscores
>> 0X HexDigitsAndUnderscores_opt . HexDigitsAndUnderscores
>>
>> BinaryExponent:
>> BinaryExponentIndicator SignedInteger
>>
>> BinaryExponentIndicator:one of
>> p P
>>
>>
>> COMPILATION:
>>
>> Numbers containing underscores are to be parsed and evaluated by the compiler exactly as if the underscores were not present. The above grammar ensures that removing underscores will not result in an unparseable number.
>>
>> A simple strategy for achieving this is that once a number has been parsed by the compiler lexer and determined to be syntactically valid according to the above grammar, then if the number contains any underscores, then all underscores in it may be removed (by something as simple as numberAsString.replaceAll("_","")) before passing the number on to the code that would normally have parsed the number prior to this proposal.
>>
>> More performant approaches are certainly possible.
>>
>>
>> TESTING: How can the feature be tested?
>>
>> A variety of literals may be generated, of the cross product of each of the following radicies:
>>
>> hex, decimal, octal
>>
>> with each of the following types:
>>
>> byte, char, short, int, long, float, double
>>
>> such that for each possible numeric field in the result, that one or more underscores are inserted at the beginning, in the middle, and at the end of the digits.
>>
>> Note that the above grammar is specifically designed to disallow any underscores from appearing which are not either preceded by or followed by a digit.
>>
>>
>> LIBRARY SUPPORT:
>>
>> Methods such as Integer.decode(String) and Long.decode(String) should probably be updated to ignore underscores in their inputs, since these methods attempt to parse according to Java conventions.
>>
>> I suggest that methods such as Integer.parseInt(), Float.parseFloat(), etc. should probably NOT be updated to ignore underscores, since these methods deal with numbers in their pure form, and are more focused and much more widely used. To alter them to ignore underscores would introduce ambiguity in (and have a performance impact on) various parsing code that uses them.
>>
>>
>> REFLECTIVE APIS:
>>
>> No changes to reflective APIs are needed.
>>
>>
>> OTHER CHANGES:
>>
>> No other changes are needed.
>>
>>
>> MIGRATION:
>>
>> Underscores can be inserted into numbers within an existing code base as desired for readability.
>>
>>
>> COMPATIBILITY
>>
>> BREAKING CHANGES:
>>
>> Since use of underscores within numbers was previously a syntax error, this should not break any existing programs.
>>
>>
>> EXISTING PROGRAMS:
>>
>> This feature does not affect the format of class files. It is purely a notational convenience. Hence, interaction with existing class files would not be affected.
>>
>>
>> REFERENCES
>>
>> EXISTING BUGS:
>>
>> A search of the Bug Database did not find any bug ID's related to this proposal.
>>
>> URL FOR PROTOTYPE (optional):
>>
>> None.
>>
>>
>>
>>
>
More information about the coin-dev
mailing list