Can someone confirm what range checking FreePascal does (if enabled) when indexing a string. I imagine shortstrings and ansistrings are handled differently, but cannot find this in the documentation.

It looks like the following:

String Type Range Checked for s[i]
----------- ----------------------
s: ShortString 0 <= i <= 255
s: AnsiString 1 <= i <= Length(s)

I am surprised that for ShortString, the range is not also 1 <= i <= Length(s).

IIRC Turbo Pascal only allowed you to access s[0] when range checking was off, and similarly for s[i] with i > Length(s).

Thanks,

Tom

No. See transcript:

-- 8< -- the program --
{$R+}
var
    s1 : String;
    s2 : String[40];
begin
   s1 := '';
   s2 := '';
   WriteLn (Length (s1));
   WriteLn (Length (s2));
   WriteLn (Ord (s1[0]));
   WriteLn (Ord (s2[0]));
   WriteLn (Ord (s1[255]));
   WriteLn (Ord (s2[255]));
end.
-- 8< --

results in:

-- 8< --
Borland Pascal Version 7.0 Copyright (c) 1983,92 Borland International TEST.PAS(18): Error 76: Constant out of range.

   WriteLn (Ord (s2[255]));
                       ^

-- 8< --

Commenting out the offending last line of code:

-- 8< --
Borland Pascal Version 7.0 Copyright (c) 1983,92 Borland International TEST.PAS(19)
19 lines, 2432 bytes code, 968 bytes data.
-- 8< --

Running the program:

-- 8< --
.\TEST.EXE
0
0
0
0
0
-- 8< --

No errors, although the index exceeds the actual length of the string.

Vinzent.

That looks rather convincing.

Borland Pascal Version 7.0 Copyright (c) 1983,92 Borland International

However, IIRC there were some differences between Borland Pascal and Turbo Pascal (as it was named before they marketed it as Borland Pascal).

But my memory may be confusing this with UCSD Pascal (Oh, those days :-).

If find it more important that the rules are clear, than that a specific behavior is provided.

Nevertheless, this means that AnsiStrings and ShortStrings are slightly more different than suggested in the documentation. AnsiString provide more protection with range checking (also implying that a program that works will with range checking and ShortStrings, might fail when using AnsiStrings, even though those are more flexible).

In what way can this information be incorporated into the documentation? Where should it go? Can I generate a patch for that?

Thanks,

Tom

IIRC the differences were in how big a program you could write. Accessing the length byte in a string did not change.

[digs out an old copy of Turbo.com -- TurboPascal version 3. Amazingly, it will run under WinXP.]

var
   x : String[50];
   i : Integer;
Begin
   x := 'Hello, World';
   for i := 0 to length(x) do
     write(ord(x[i]),' ');
   writeln;
   x[0] := #5;
   writeln(x);
end.

Outputs this:
12 72 101 108 108 111 44 32 87 111 114 108 100
Hello

so read and write access to element 0

Yes, there were. But regarding string handling, nothing changed, at least not between TP5.5 and BP7. And I know for sure that accessing element 0 was always allowed, because it could be taken as shortcut for reading or even changing the length.

Nevertheless, this means that AnsiStrings and ShortStrings are slightly more different than suggested in the documentation.

Well, if you put it this way, yes. With ShortStrings you rather get a maximum size (where indexing is allowed regardless of the actual length, although the behaviour could be considered more or less undefined - just like accessing uninitialized memory), whilst AnsiString's behaviour doesn't distinguish between length and (maximum) size.

Vinzent.