Re: GNAT: Performance of String functions

[Samuel Mize <smize@link.com>]

Tarjei Jensen wrote:

> I'd be delighted if it is possible to write a portable bounded
> string package
> that does not copy blindly on assignment.

Here's what I understand you want:

1 Each string has its own individual maximum length, and a
  currently-used length.

2 Each string allocates ONLY its maximum length, not a max
  length for all strings of this subtype.

3 The user can assign A to B so long as the "used" length of
  A is less than the maximum length of B.

4 Normal assignment is available.

5 Unused bytes are not copied on assignment.

You can build a type to do what you want (except you have
to use a "copy" procedure), but Bounded_Type doesn't quite do
what you want.

You can get all but 4 (normal assignment) with a limited variant
record with a discriminant (max length).  To assign the value,
you provide a "copy" routine.  Don't provide a default value
for the discriminant, so the user will have to explicitly give
a max size for each variable of this type.  (If you provide a
default value, the implementation may allocate the subtype's
max size for each record.)

Since Ada's "user-defined" assignment is actually user code
adjusting and finalizing values, you can't get 4 (unused bytes
not copied) with normal assignment.  (You might get it by
using a variant record, but then the compiler can allocate
the subtype's max size for each record.)

Sam Mize