Re: GNAT: Performance of String functions

[mheaney@ni.net (Matthew Heaney)]

In article <33D11806.6C59@online.no>, tarjei@online.no wrote:

>VAX Pascal passes strings by descriptors. The string pointed to by
>descriptor is of type counted string. A counted string has a maximum
>size, an actual size and the string itself.

That's all an Ada bounded string is: a maximum size (implicit in the size
of the array used to implement Bounded_String), and another component to
record the logical length of the string.

So I'm confused by your comment, because it sounds like they're very much
similar.

>This means that comparing Ada bounded strings and VAX Pascal strings
>compares two different ways of doing things.

This appears not the be the case; perhaps you can elaborate.

>Each time you change the
>size of an Ada bounded string there can be serious overhead.

What size do you mean: the physical length or the logical length.  I assume
logical length, because the physical size can't be changed.

But where's the "serious overhead" when changing the (logical) length of a
bounded string?  For example, if I have this (a hypothetical implementation
- I don't have GNAT source handy):

type Bounded_String is
   record
      Items : String (1 .. Max);
      Length : Natural := 0;
   end record;

and I do an append:

procedure Append (Bounded : in out Bounded_String; Item : in String) is
   Length : constant Natural := Bounded.Length + Item'Length;
   Items : String renames Bounded.Items;
begin
   if Length <= Max then

     Items (Bounded.Length + 1 .. Length) := Item;
     Bounded.Length := Length;

elsif ...

end Append;

Where's the overhead, exactly?

>The apparent overhead of bounded strings is an strong argument for
>providing a standard counted string package in Ada.

I think that that package is already included in Ada: it's called
Ada.Strings.Bounded.

Are you sure you aren't thinking of Unbounded_String?  That abstraction
definately is very likely to have greater overhead than any form of bounded
string.

--------------------------------------------------------------------
Matthew Heaney
Software Development Consultant
<mailto:matthew_heaney@acm.org>
(818) 985-1271