Embarcadero DelphiStrings

String types

Delphi has the following string types (in order of popularity):

TypeMaximum lengthMinimum sizeDescription
string2GB16 bytesA managed string. An alias for AnsiString through Delphi 2007, and an alias for UnicodeString as of Delphi 2009.
UnicodeString2GB16 bytesA managed string in UTF-16 format.
AnsiString2GB16 bytesA managed string in pre-Unicode ANSI format. As of Delphi 2009, it carries an explicit code-page indicator.
UTF8String2GB16 bytesA managed string in UTF-8 format, implemented as an AnsiString with a UTF-8 code page.
ShortString255 chars2 bytesA legacy, fixed-length, unmanaged string with very little overhead
WideString2GB4 bytesIntended for COM interop, a managed string in UTF-16 format. Equivalent to the Windows BSTR type.

UnicodeString and AnsiString are reference counted and copy-on-write (COW).
ShortString and WideString are not reference counted and do not have COW semantics.



  S1, S2: string;
  S1 := 'Foo';
  S2 := ToLower(S1); // Convert the string to lower-case
  S1 := ToUpper(S2); // Convert the string to upper-case



  C1, C2: Char;
  C1 := 'F';
  C2 := ToLower(C1); // Convert the char to lower-case
  C1 := ToUpper(C2); // Convert the char to upper-case

The uses clause should be System.Character if version is XE2 or above.

UPPER and lower case


  S1, S2: string;
  S1 := 'Foo';
  S2 := LowerCase(S1); // S2 := 'foo';
  S1 := UpperCase(S2); // S1 := 'FOO';


Assigning string to different string types and how the runtime environment behaves regarding them. Memory allocation, reference counting, indexed access to chars and compiler errors described briefly where applicable.

  SS5: string[5]; {a shortstring of 5 chars + 1 length byte, no trailing `0`}
  WS: Widestring; {managed pointer, with a bit of compiler support}
  AS: ansistring; {ansistring with the default codepage of the system}
  US: unicodestring; {default string type}
  U8: UTF8string;//same as AnsiString(65001)
  A1251: ansistring(1251); {ansistring with codepage 1251: Cryllic set}
  RB: RawbyteString; {ansistring with codepage 0: no conversion set}
  SS5:= 'test'; {S[0] = Length(SS254) = 4, S[1] = 't'...S[5] = undefined}
  SS5:= 'test1'; {S[0] = 5, S[5] = '1', S[6] is out of bounds}
  SS5:= 'test12'; {compile time error}
  WS:= 'test'; {WS now points to a constant unicodestring hard compiled into the data segment}
  US:= 'test'+IntToStr(1); {New unicode string is created with reference count = 1}
  WS:= US; {SysAllocateStr with datacopied to dest, US refcount = 1 !}
  AS:= US; {the UTF16 in US is converted to "extended" ascii taking into account the codepage in AS possibly losing data in the process}  
  U8:= US; {safe copy of US to U8, all data is converted from UTF16 into UTF8}
  RB:= US; {RB = 'test1'#0 i.e. conversion into RawByteString uses system default codepage}
  A1251:= RB; {no conversion takes place, only reference copied. Ref count incremented }

Reference counting

Counting references on strings is thread-safe. Locks and exception handlers are used to safeguard the process. Consider the following code, with comments indicating where the compiler inserts code at compile time to manage reference counts:

procedure PassWithNoModifier(S: string);
// prologue: Increase reference count of S (if non-negative),
//           and enter a try-finally block
  // Create a new string to hold the contents of S and 'X'. Assign the new string to S,
  // thereby reducing the reference count of the string S originally pointed to and
  // brining the reference count of the new string to 1.
  // The string that S originally referred to is not modified.
  S := S + 'X';
// epilogue: Enter the `finally` section and decrease the reference count of S, which is
//           now the new string. That count will be zero, so the new string will be freed.
procedure PassWithConst(const S: string);
  TempStr: string;
// prologue: Clear TempStr and enter a try-finally block. No modification of the reference
//           count of string referred to by S.
  // Compile-time error: S is const.
  S := S + 'X';
  // Create a new string to hold the contents of S and 'X'. TempStr gets a reference count
  // of 1, and reference count of S remains unchanged.
  TempStr := S + 'X';
// epilogue: Enter the `finally` section and decrease the reference count of TempStr,
//           freeing TempStr because its reference count will be zero.

As shown above, introducing temporary local string to hold the modifications to a parameter involves the same overhead as making modifications directly to that parameter. Declaring a string const only avoids reference counting when the string parameter is truly read-only. However, to avoid leaking implementation details outside a function, it is advisable to always use one of const, var, or out on string parameter.


String types like UnicodeString, AnsiString, WideString and UTF8String are stored in a memory using their respective encoding (see String Types for more details). Assigning one type of string into another may result in a conversion. Type string is designed to be encoding independent - you should never use its internal representation.

The class Sysutils.TEncoding provides method GetBytes for converting string to TBytes (array of bytes) and GetString for converting TBytes to string. The class Sysutils.TEncoding also provides many predefined encodings as class properties.

One way how to deal with encodings is to use only string type in your application and use TEncoding every time you need to use specific encoding - typically in I/O operations, DLL calls, etc...

procedure EncodingExample;
var hello,response:string;
  hello := 'Hello World!Привет мир!';
  dataout := SysUtils.TEncoding.UTF8.GetBytes(hello); //Conversion to UTF8
  datain := SomeIOFunction(dataout); //This function expects input as TBytes in UTF8 and returns output as UTF8 encoded TBytes.
  response := SysUtils.TEncoding.UTF8.GetString(datain); //Convertsion from UTF8

  //In case you need to send text via pointer and length using specific encoding (used mostly for DLL calls)
  dataout := SysUtils.TEncoding.GetEncoding('ISO-8859-2').GetBytes(hello); //Conversion to ISO 8859-2
  //The same is for cases when you get text via pointer and length
  expectedLength := DLLCallToGetDataLength();
  response := Sysutils.TEncoding.GetEncoding(1250).getString(datain);

   //TStringStream and TStringList can use encoding for I/O operations
   stringList.text := hello;
   stringStream := TStringStream(hello,SysUtils.TEncoding.Unicode);