Is this a delphi string bug ?

-
Is this a delphi string bug ?
declaring a type as short string (subtypes) cause it's variable to be non unicode :

type
  SmallString = String[20];
var
  SmStr: SmallString;
  Str: String;
begin
  Str := 'UnicodeString';
  SmStr := Str; // SmStr is no longer Unicode !
  SmStr := SmStr;
end;

here is the sequence of allocated bytes (variable Str):img1.
https://lh5.googleusercontent.com/-4lTgu3CbG_Q/VETvaOtUVrI/AAAAAAAAAUE/Hj34jzJmv8g/s219-p/unicode.PNG

and img2 for the sequence of allocated bytes (variable SmStr)
https://lh4.googleusercontent.com/-uKcgkITdwBw/VETv79YasLI/AAAAAAAAAUM/SKVqHl1dpn8/s219-p/nounicode.PNG

Tested using the latest delphi version (XE7).
Any explanation ?

Comments

  1. David MillingtonOctober 20, 2014 at 4:40 AM

    The documentation says, "For a short string or AnsiString, S[i] is of type AnsiChar."  ( http://docwiki.embarcadero.com/RADStudio/XE7/en/String_Types_(Delphi) )

    So I'd say it's not a bug that it loses Unicode and becomes single-byte - unless there's no compiler warning about it.

    Interestingly the documentation also says a short string is statically allocated at 256 bytes, even if the size dynamically changes (because it can only vary between 0-255 bytes long.) I didn't know that; I always thought if I declared a string[32] it was up to 32 characters long and took only 32/33 bytes of memory.

    What are you using it for? Unless it's a backward-compatible record written to disk or something similar, an ordinary string is probably better. Otherwise, some variation of using an array of (Unicode) characters that you convert to/from a string?

    ReplyDelete
  2. William WongOctober 20, 2014 at 4:42 AM

    Isn't short string a Pascal string from the good old days? It is essentially an ANSI string with length stored at the zero indexed byte.

    ReplyDelete
  3. Dalija PrasnikarOctober 20, 2014 at 4:44 AM

    David Millington ShortString is equivalent of saying string[255] and therefore its allocated size will be 256 bytes. If you declare string[10], its size will be 11 bytes.

    ReplyDelete
  4. David MillingtonOctober 20, 2014 at 4:46 AM

    Dalija Prasnikar So it is (I say while furiously reading the docs. http://docwiki.embarcadero.com/Libraries/XE7/en/System.ShortString )

    The more you learn... thanks for the correction :) I always though a declaration like "string[32]" was a "short string". I have never actually used them since, I suppose, Delphi 1, which is too far away for memory.

    ReplyDelete
  5. Bruce McGeeOctober 20, 2014 at 5:31 AM

    There is a compiler warning about an implicit cast and potential data loss.

    ReplyDelete
  6. Uwe RaabeOctober 20, 2014 at 5:50 AM

    David Millington There is a small difference between the "type ShortString" and a "short-string type".

    ReplyDelete
  7. Mahdi SafsafiOctober 20, 2014 at 9:42 AM

    David Millington "I always thought if I declared a string[32] it was up to 32 characters long" 
    That's true : declaring a string[32] will end with 33 bytes of memory .

    BTW , the compiler will not allow you to write something like that :
    type
      SmallStr = AnsiString[32];

    ReplyDelete
  8. Mahdi SafsafiOctober 20, 2014 at 9:51 AM

    Dalija Prasnikar The issue isn't from the allocated size .
    it's about the lose of unicode .

    ReplyDelete
  9. Dalija PrasnikarOctober 20, 2014 at 10:30 AM

    Mahdi Safsafi That is because short strings are legacy type from pre-Unicode versions of Delphi. Actually, their roots go all way back to Turbo Pascal. They store 8-bit characters in Ansi encoding and are not suited for storing Unicode characters.

    ReplyDelete
  10. Alexey PetushkovOctober 20, 2014 at 12:50 PM

    I feeling myself old reading this question :(

    ReplyDelete
  11. Moz LeOctober 20, 2014 at 2:03 PM

    Dalija Prasnikar disturbingly, you can store unicode in them very easily because they're known-length strings and can have nulls in them. Delphi will work perfectly well if you put UTF-8 strings in provided you set the length to the byte-count of the string not the char-count. UTF-16 or wide-char can be done but it's harder, and UTF-32 is kinda pointless but again can be done.

    Not that I would ever do such a horrid thing. Not me.

    ReplyDelete
  12. David MillingtonOctober 20, 2014 at 2:22 PM

    Wouldn't a statically sized array of Unicode characters be the best solution here, if you cannot use a normal string for some reason? Known size (good for streaming / saving, records, etc), can be easily converted to/from string, no data loss.

    ReplyDelete
  13. Moz LeOctober 20, 2014 at 6:47 PM

    David Millington define "normal string"... the default has changed twice in the time I've been using Delphi.

    ReplyDelete
  14. David MillingtonOctober 21, 2014 at 3:25 AM

    Moz Le "Normal string" is whatever you get when you write "string".

    Unicode since 2009, non-Unicode before that. (Going way back 20 years, I think D1 had a simpler implementation similar or identical to shortstrings, and I don't even remember Turbo Pascal's strings. But the only change since D1->D2 in 1996 is Unicode, and that is a good one.)

    ReplyDelete
  15. Moz LeOctober 21, 2014 at 1:26 PM

    David Millington huge strings were introduced with Delphi 2 and didn't become the default for a while.

    ReplyDelete

Post a Comment