The current version provides implementations of smart pointers, directed graphs, sets, maps, B-trees, stacks, tables, string editing, unbounded arrays, expression analyzers, lock-free data structures, synchronization primitives (events, race condition free pulse events, arrays of events, reentrant mutexes, deadlock-free arrays of mutexes), arbitrary precision arithmetic, pseudo-random non-repeating numbers, symmetric encoding and decoding, IEEE 754 representations support, streams, persistent storage, multiple connections server/client designing tools and protocols implementations.
Changes to the previous version:
- The package Unbounded_Unsigneds implementing arbitrary precision unsigned arithmetic was added;
- The package Unbounded_Integers implementing arbitrary precision integer arithmetic was added;
- The package Unbounded_Unsigneds Primes implementing operations with prime numbers was added;
- The package Unbounded_Unsigneds.Montgomery implementing Montgomery domain operations was added;
- The package Unbounded_Unsigneds.Barrett implementing Barrett reduction was added;
- The package Strings_Edit.Unbounded_Unsigned_Edit string editing for arbitrary precision unsigned numbers was added;
- The package Strings_Edit.Unbounded_Integer_Edit string editing for arbitrary precision integer numbers was added;
- The package Unbounded_Unsigneds.Parallel implementing parallel arbitrary precision algorithms was added;
- The package Job_Servers was added implementing servers of jobs backed by a task pool;
- The number protocol added to Python bindings;
- Reply_Text and Reply_HTML in GNAT.Sockets.Connection_State_Machine.HTTP_Server modified to call Send_Body passing the Get parameter rather than skipping it when Get = False;
- UUID v6 and v7 generation was added to Universally_Unique_Identifiers.
The key points of the arbitrary precision arithmetic packages design and functionality:
- Advanced memory management preventing excessive copying;
- The number internal representation vector is shared between objects if possible;
- No limit on the number size, except for the storage pool size;
- In-place versions of operations (e.g. for addition, subtraction) further reduce need of copying;
- Lazy memory deallocation strategy, the memory is kept between variable updates;
- Swapping variables;
- Long to short operations;
- Squaring;
- Square root, square root with remainder;
- Multiplicative inverse;
- 2’s complement;
- Bit representation access, slicing, truncation;
- Full division with remainder, remainder only division;
- Karatsuba multiplication and squaring;
- Specialized operations involving powers of two and words;
- Exponentiation under modulo;
- Fibonacci number under modulo;
- Montgomery domain multiplication, squaring, exponentiation under modulo and primality tests of the domain modulus;
- Barrett reduction, multiplication, exponentiation;
- Primality tests: Miller-Rabin, Fibonacci, Lucas-Lehmer, strong Lucas;
- Parallel algorithms for very large numbers;
- String editing and formatting packages for the numbers.
Performance notes. In order to get optimal performance -O2 switch need to be used. It does 3x performance boost. 64-bit (with 128-bit integer) outperform 32-bit by many multiplies. See GPR variables: Simple components for Ada E.g.
gprbuild -P components-tests.gpr -XTarget_OS=Linux -Xarch=aarch64 -XDevelopment=Release