License:
BSD style: see license.txt
Version:
Initial release: October 2004
Outback release: December 2006
author:
Kris
Allocators to use in conjunction with the Reader class. These are
intended to manage array allocation for a variety of Reader.get()
methods
- class
HeapCopy
: tango.io.protocol.model.IProtocol.IAllocator;
- Simple allocator, copying into the heap for each array requested:
this is the default behaviour for Reader instances
- this(IProtocol protocol);
- final IProtocol
protocol
();
- final void
reset
();
- final void[]
allocate
(void[] delegate(void* dst, uint bytes, Type type) reader, uint bytes, Type type);
- class
HeapSlice
: tango.io.protocol.model.IProtocol.IAllocator;
- Allocate from within a private heap space. This supports reading
data as 'records', reusing the same chunk of memory for each record
loaded. The ctor takes an argument defining the initial allocation
made, and this will be increased as necessary to accomodate larger
records. Use the reset() method to indicate end of record (reuse
memory for subsequent requests), or set the autoreset flag to reuse
upon each array request.
- this(IProtocol protocol, uint width = 4096, bool autoreset = false);
- final IProtocol
protocol
();
- final void
reset
();
- Reset content length to zero
- final void[]
allocate
(void[] delegate(void* dst, uint bytes, Type type) reader, uint bytes, Type type);
- No allocation: copy into a reserved arena.
With HeapSlice, it is normal to
allocate
space large
enough to contain, say, a record of data. The reserved
space will grow to accomodate larger records. A reset()
call should be made between each record read, to ensure
the space is being reused.
- class
BufferSlice
: tango.io.protocol.model.IProtocol.IAllocator;
- Alias directly from the buffer instead of allocating from the heap.
This avoids both heap activity and copying, but requires some care
in terms of usage. See methods allocate() for details
- this(IProtocol protocol);
- final IProtocol
protocol
();
- final void
reset
();
- Move all unconsumed data to the front of the buffer, freeing
up space for more
- final void[]
allocate
(void[] delegate(void* dst, uint bytes, Type type) reader, uint bytes, Type type);
- No alloc or copy: alias directly from buffer. While this is
very efficient (no heap activity) it should be used only in
scenarios where content is known to fit within a buffer, and
there is no conversion of said content e.g. take care when
using with EndianProtocol since it will convert within the
buffer, potentially confusing additional buffer clients.
With BufferSlice, it is considered normal to create a Buffer
large enough to contain, say, a file and subsequently slice
all strings/arrays directly from this buffer. Smaller Buffers
can be used in a record-oriented manner similar to HeapSlice:
invoke reset() before each record is processed to ensure here
is sufficient space available in the buffer to house a complete
record. GrowBuffer could be used in the latter case, to ensure
the largest record width is always accomodated.
A good use of this is in handling of network traffic, where
incoming data is often transient and of a known extent. For
another potential use, consider the quantity of distinct text
arrays generated by an XML parser -- would be convenient to
slice all of them from a single allocation instead
|
|