Updated 2014-12-28 14:09:26 by aspect

A subcommand of string

Synopsis edit

string is class ?-strict? ?-failindex varname? string

Official Documentation

Dealing with Integers (what's an entier?) edit

To test if a string is an integer, one would use string is integer, right? Not so fast!

A careful perusal of the manual will show that you might want one of several things:

  • string is integer: for an integer whose absolute value fits in 32 bits (not platform dependent!), written in one of the forms Tcl can parse
  • string is entier: an integer (of any size), written in one of the forms Tcl can parse
  • string is digit: a string that consists of only digits

Often, Tcl code shouldn't need to care about the size of an integer, so string is entier is probably what you want. See entier for some historical etymology. For more background on the limits of string is integer through history, see 32-bit integer overflow.

Some examples:
% string is integer -1234
% string is integer 0xdeadbeef   ;# hex is fine too
% string is integer 007          ;# so is octal ...
% string is integer 09           ;# but this is not!
% string is integer ""
% string is integer -strict ""
% string is integer 12345678999  ;# too big for 32 bits!
% string is entier 12345678999

Note that string is integer checks that the unsigned value fits in 32 bits:
% set x [expr 2**32-1]
% string is integer $x
% string is integer -$x
% incr x
% string is integer $x
% string is integer -$x

If you find yourself tripping over decimal numbers written with leading 0, the solution is probably to scan them as entiers:
% scan 012345678999 %d x
% set x

Discussion edit

 Heritage of [string is ...]

A comparison with the ctype(3) man page (e.g., [1]) shows much agreement with the string is classes. A not-too-daring guess is that C has contributed also this piece of Tcl heritage.

 Unicode and [string is digit ...]

Note that string is digit recognises Unicode digits that are not valid in expr expressions string is digit string will return 1 if string is composed of "Any Unicode digit character." Then it goes on to say that "this includes characters outside of the [0-9] range." What other characters, aside from [0-9], are members of the digit character class?

RS As Tcl is great for introspection, a few lines of code give the answer:
 proc udigits max {
   set res {}
   for {set i 0} {$i<=$max} {incr i} {
      if [string is digit [format %c $i]] {
         append res "\\u[format %04x $i] "
   set res
 % udigits 65535

\u0030 \u0031 \u0032 \u0033 \u0034 \u0035 \u0036 \u0037 \u0038 \u0039 \u0660 \u0661 \u0662 \u0663 \u0664 \u0665 \u0666 \u0667 \u0668 \u0669 \u06f0 \u06f1 \u06f2 \u06f3 \u06f4 \u06f5 \u06f6 \u06f7 \u06f8 \u06f9 \u0966 \u0967 \u0968 \u0969 \u096a \u096b \u096c \u096d \u096e \u096f \u09e6 \u09e7 \u09e8 \u09e9 \u09ea \u09eb \u09ec \u09ed \u09ee \u09ef \u0a66 \u0a67 \u0a68 \u0a69 \u0a6a \u0a6b \u0a6c \u0a6d \u0a6e \u0a6f \u0ae6 \u0ae7 \u0ae8 \u0ae9 \u0aea \u0aeb \u0aec \u0aed \u0aee \u0aef \u0b66 \u0b67 \u0b68 \u0b69 \u0b6a \u0b6b \u0b6c \u0b6d \u0b6e \u0b6f \u0be7 \u0be8 \u0be9 \u0bea \u0beb \u0bec \u0bed \u0bee \u0bef \u0c66 \u0c67 \u0c68 \u0c69 \u0c6a \u0c6b \u0c6c \u0c6d \u0c6e \u0c6f \u0ce6 \u0ce7 \u0ce8 \u0ce9 \u0cea \u0ceb \u0cec \u0ced \u0cee \u0cef \u0d66 \u0d67 \u0d68 \u0d69 \u0d6a \u0d6b \u0d6c \u0d6d \u0d6e \u0d6f \u0e50 \u0e51 \u0e52 \u0e53 \u0e54 \u0e55 \u0e56 \u0e57 \u0e58 \u0e59 \u0ed0 \u0ed1 \u0ed2 \u0ed3 \u0ed4 \u0ed5 \u0ed6 \u0ed7 \u0ed8 \u0ed9 \u0f20 \u0f21 \u0f22 \u0f23 \u0f24 \u0f25 \u0f26 \u0f27 \u0f28 \u0f29 \u1040 \u1041 \u1042 \u1043 \u1044 \u1045 \u1046 \u1047 \u1048 \u1049 \u1369 \u136a \u136b \u136c \u136d \u136e \u136f \u1370 \u1371 \u17e0 \u17e1 \u17e2 \u17e3 \u17e4 \u17e5 \u17e6 \u17e7 \u17e8 \u17e9 \u1810 \u1811 \u1812 \u1813 \u1814 \u1815 \u1816 \u1817 \u1818 \u1819 \uff10 \uff11 \uff12 \uff13 \uff14 \uff15 \uff16 \uff17 \uff18 \uff19

Here they are "literally":

0 1 2 3 4 5 6 7 8 9 ٠ ١ ٢ ٣ ٤ ٥ ٦ ٧ ٨ ٩ ۰ ۱ ۲ ۳ ۴ ۵ ۶ ۷ ۸ ۹ ० १ २ ३ ४ ५ ६ ७ ८ ९ ০ ১ ২ ৩ ৪ ৫ ৬ ৭ ৮ ৯ ੦ ੧ ੨ ੩ ੪ ੫ ੬ ੭ ੮ ੯ ૦ ૧ ૨ ૩ ૪ ૫ ૬ ૭ ૮ ૯ ୦ ୧ ୨ ୩ ୪ ୫ ୬ ୭ ୮ ୯ ௧ ௨ ௩ ௪ ௫ ௬ ௭ ௮ ௯ ౦ ౧ ౨ ౩ ౪ ౫ ౬ ౭ ౮ ౯ ೦ ೧ ೨ ೩ ೪ ೫ ೬ ೭ ೮ ೯ ൦ ൧ ൨ ൩ ൪ ൫ ൬ ൭ ൮ ൯ ๐ ๑ ๒ ๓ ๔ ๕ ๖ ๗ ๘ ๙ ໐ ໑ ໒ ໓ ໔ ໕ ໖ ໗ ໘ ໙ ༠ ༡ ༢ ༣ ༤ ༥ ༦ ༧ ༨ ༩ ၀ ၁ ၂ ၃ ၄ ၅ ၆ ၇ ၈ ၉ ፩ ፪ ፫ ፬ ፭ ፮ ፯ ፰ ፱ ០ ១ ២ ៣ ៤ ៥ ៦ ៧ ៨ ៩ ᠐ ᠑ ᠒ ᠓ ᠔ ᠕ ᠖ ᠗ ᠘ ᠙ 0 1 2 3 4 5 6 7 8 9

For instance, \u0660-\u0669 are "Indo-Arabic" digits as used in Arab countries. \uFF10-\uFF19 are "fullwidth" variants of 0-9, etc.

Lars H: Quite a lot, as you can see; there are plenty of digit sets in Unicode. The authorative source on the subject should be the Unicode Character Database (see [2] for format and links), but whether Tcl really uses that is another matter. A comparison of the above with the UCD shows that [string is digit] returns 1 for most (but not quite all!) of the characters from the class Nd (decimal digits), so that is probably what it is supposed to test.

escargo 14 Mar 2006 - I had my own peek starting at http://www.unicode.org/charts/symbols.html where there are links to four PDF files under Numbers and Digits: ASCII Digits, Fullwidth ASCII Digits, Number Forms, and Super and Subscripts. All of these might legitimately contain digits.

NEM 14 Mar 2006: Interesting. Note though that most of these "digits" are not valid for expressions fed to expr. So, if you are using [string is digit] to validate arguments to expressions then you probably have a bug.

escargo - If there are digits that are not valid when fed to expr, is that a problem in the implementation of expr? It almost makes me think that there ought to be a string normalize that changes Unicode digits of different stripes down to the [0-9] range expected by expr.

It's like if the character looks to humans like a digit but it is not one to expr then the problem is really in expr. (The solution might not be in expr though.)

NEM My gut reaction is the same as yours: that expr should be enhanced to recognize these alternative numeral characters as digits. This seems to be sensible, and would be backwards compatible as the characters are apparently not allowed at all by expr currently. The expr(n) manpage could also do with clarification about what counts as an integer or float -- at present, it just says "decimal" (or octal/hexidecimal), as far as I can see.

slebetman 15 Mar 2006: Well, if you are not using a number larger than 18446744073709551615.99, then you can use [string is double $number] as the test instead of [string is digit $number]. Currently under Microsoft Windows on 32-bit Pentium [string is double] returns 0 for 18446744073709551616 and larger (found out by trial and error). Alternatively, if you want to check for integers and you're not using a number larger than 4294967295 then you can use [string is integer $number]:
    (bin) 59 % string is digit \u0669
    (bin) 60 % string is double \u0669

 Octal and [string is integer ...]

string is integer has a complication because a leading zero implies octal, but 09 etc are not valid octal numbers

NLast - 2010-01-05

string is integer shows unexpected behaviour as does string is double:
1 % string is integer 09
2 % string is integer 9

Why is 09 not an integer?

AMG: 'Cuz it's not, and neither is 08. The leading zero indicates that it is an octal number, but 8 and 9 are not valid octal digits. By the same token, 12ABC is not a valid integer, even though it looks like hexadecimal.
% expr {09 + 1}
missing operator at _@_
in expression "0_@_9 + 1";
looks like invalid octal number

Valid formats for integers:
      0 - zero is always valid
  56789 - lack of leading zeros - decimal
0xABCDE - leading 0x or 0X      - hexadecimal
 034567 - leading 0             - octal
0o34567 - leading 0o or 0O      - octal (Tcl 8.5+)
0b01010 - leading 0b or 0B      - binary (Tcl 8.5+)

If you want to force a number with leading zeros to be interpreted as decimal, use the %d scan format code.
scan 09 %d   ;# returns 9
scan 09 %i   ;# returns 0, since the 9 is ignored as invalid
scan 0123 %d ;# returns 123
scan 0123 %i ;# returns 83, which is 1*8² + 2*8¹ + 3*8⁰ = 64 + 16 + 3

NLast - 2010-01-06 15:25:38

Thank you AMG for sharing your opinion. However it doesn't sound logical for me. The main reason is absence of class "octal" (or "hexadecimal"): we only have class "integer". You've just written: '"Valid formats for integers: ... blah-blah ... 034567 - leading 0 - octal"' meaning 034567 is digital (namely octal). That's what you've stated, not me. And that's right.

I just expected string is to test if the string is compatible with expr. And now I see I have to use catch which is just not very nice.

slebetman: Actually, for what you want to do (test if string is compatible with expr), string is IS doing what you want. You should note that in expr 09 is an invalid number because 9 is an invalid octal digit. And string is is telling you that.

The 0nn format for octals was inherited from C. Yes it sucks. Yes it's prone to bugs - especially when you're trying to calculate time. And yes there was a proposal to change it and yes the majority agreed that it should go away but the Tcl Core Team highly values backwards compatibility so for now it's staying until we're sure removing the syntax doesn't break anything.

But it is not an unexpected behavior of [string is]. It's the behavior of Tcl itself in how numbers are interpreted. And besides C and Tcl, 09 is also an invalid octal number in: Java, Perl, Ruby, Javascript, Python.. in fact most mainstream programming languages from the 60s until today (I blame it on C). C# stands out for not supporting the 0nn notation.

AMG: So... why don't we have [string is decimal], [string is octal], [string is hexadecimal], and [string is binary]? One question that comes to mind is whether or not they will all accept "0". Is "0" an octal string? It starts with a leading zero. Is "0" a hexadecimal string? It doesn't start with "0x". Is "00" valid decimal? It has an extra leading zero, signifying octal. And so on. Probably we don't need to bloat the core with these new commands, since they can easily be implemented using regexp, enabling the programmer to decide on a case-by-case basis whether "0" is valid octal, etc.

 EIAS humor

AMG: Here's an April 1 TIP for you: [string is string]. :^) It'll return true (1) whenever given one argument, no matter what that argument is, since everything is a string! A -strict option would cause it to return false (0) if given empty string. (update) On second thought, -strict would have no effect, since an empty string is still a string!!

aspect: while I can't invoke TIP #323 here, I think [string is string] ought to treturn true with no argument. Not sure about [string is string -strict] though!