tanya/README.md

7.8 KiB

Tanya

Build status Build status codecov Dub version Dub downloads License

Tanya is a general purpose library for D programming language.

Its aim is to simplify the manual memory management in D and to provide a guarantee with @nogc attribute that there are no hidden allocations on the Garbage Collector heap. Everything in the library is usable in @nogc code. Tanya provides data structures and utilities to facilitate painless systems programming in D.

Overview

Tanya consists of the following packages and (top-level) modules:

  • algorithm: Collection of generic algorithms.
  • async: Event loop (epoll, kqueue and IOCP).
  • container: Queue, Array, Singly and doubly linked lists, Buffers, UTF-8 string, Hash table.
  • conv: This module provides functions for converting between different types.
  • encoding: This package provides tools to work with text encodings.
  • exception: Common exceptions and errors.
  • format: Formatting and conversion functions.
  • hash: Hash algorithms.
  • math: Arbitrary precision integer and a set of functions.
  • memory: Tools for manual memory management (allocators, smart pointers).
  • meta: Template metaprogramming. This package contains utilities to acquire type information at compile-time, to transform from one type to another. It has also different algorithms for iterating, searching and modifying template arguments.
  • net: URL-Parsing, network programming.
  • network: Socket implementation. network is currently under rework. After finishing the new socket implementation will land in the net package and network will be deprecated.
  • os: Platform-independent interfaces to operating system functionality.
  • range: Generic functions and templates for D ranges.
  • test: Test suite for unittest-blocks.
  • typecons: Templates that allow to build new types based on the available ones.

NogcD

To achieve programming without the Garbage Collection tanya uses a subset of D: NogcD.

Allocators

Memory management is done with allocators. Instead of using new to create an instance of a class, an allocator is used:

import tanya.memory;

class A
{
    this(int arg)
    {
    }
}

A a = defaultAllocator.make!A(5);

defaultAllocator.dispose(a);

As you can see, the user is responsible for deallocation, therefore dispose is called at the end.

If you want to change the defaultAllocator to something different, you probably want to do it at the program's beginning. Or you can invoke another allocator directly. It is important to ensure that the object is destroyed using the same allocator that was used to allocate it.

What if I get an allocated object from some function? The generic rule is: If you haven't requested the memory yourself (with make), you don't need to free it.

tanya.memory.smartref contains smart pointers, helpers that can take care of a proper deallocation in some cases for you.

Exceptions

Since exceptions are normal classes in D, they are allocated and dellocated the same as described above, but:

  1. The caller is always responsible for destroying a caught exception.
  2. Exceptions are always allocated and should be always allocated with the defaultAllocator.
import tanya.memory;

void functionThatThrows()
{
    throw defaultAlocator.make!Exception("An error occurred");
}

try
{
    functionThatThrows()
}
catch (Exception e)
{
    defaultAllocator.dispose(e);
}

Built-in array operations and containers

Arrays are commonly used in programming. D's built-in arrays often rely on the GC. It is inconvenient to change their size, reserve memory for future use and so on. Containers can help here. The following example demonstrates how tanya.container.array.Array can be used instead of int[].

import tanya.container.array;

Array!int arr;

// Reserve memory if I know that my container will contain approximately 15
// elements.
arr.reserve(15);

arr.insertBack(5); // Add one element.
arr.length = 10; // New elements are initialized to 0.

// Iterate over the first five elements.
foreach (el; arr[0 .. 5])
{
}

int i = arr[7]; // Access 8th element.

There are more containers in the tanya.container package.

Immutability

Immutability doesn't play nice with manual memory management since the allocated storage should be initialized (mutated) and then released (mutated). immutable is used only for non-local immutable declarations (that are evaluated at compile time), static immutable data, strings (immutable(char)[], immutable(wchar)[] and immutable(dchar)[]).

Unsupported features

The following features depend on GC and aren't supported:

  • lazy parameters (allocate a closure which is evaluated when then the parameter is used)

  • synchronized blocks

Development

Supported compilers

DMD GCC
2.080.0 master
2.079.1
2.078.3
2.077.1

Current status

Following modules are under development:

Feature Branch Build status
Hash table bitvector bitvector bitvector

Release management

Tanya is still under active development and it isn't possible to provide great backwards-compatibility at this stage. This won't change until 1.0.0. Almost every release contains new features or API changes alongside bug fixes. Thus:

  • Patch releases add new functionality and bug fixes in a backwards-compatible manner

  • Minor releases contain API breakages

  • Major release number is always the same: 0.x.x

Deprecated functionality is where possible marked as such before getting removed. It is left in the library for one release: If 0.8.1 deprecates some feature, it is removed in the next release: 0.9.0.

Further characteristics

  • Tanya is a native D library

  • Tanya is cross-platform. The development happens on a 64-bit Linux, but it is being tested on Windows and FreeBSD as well

  • Tanya favours generic algorithms therefore there is no auto-decoding. Char arrays are handled as any other array type

  • The library isn't thread-safe yet

  • Complex numbers (cfloat, cdouble, creal, ifloat, idouble, ireal) aren't supported

Feedback

Any feedback about your experience with tanya would be greatly appreciated. Feel free to contact me.