Streams

A stream represents a stream of bytes, either being read from some source, or written to some destination. Storm provides different streams for different sources and destinations. These all present the same interface, which means that programs typically do not need to be aware of the exact source or destination of the data.

All streams in Storm present a synchronous interface. Utilizing the cheap user level threads in Storm, it is possible to create nonblocking calls by simply spawning a new user level thread (spawn in Basic Storm). Using multiple user level threads in this fashion makes Storm issue nonblocking IO requests to the operating system, so that other user level threads are not blocked by long-running IO requests (this is not the case for disk access on Linux currently due to the POSIX API, but it will be fixed eventually).

There are three types of streams in Storm: output streams, input streams, and random access input streams.

Output Streams

Output streams accept data and writes it to some destination. An output stream is implemented as subclasses to the OStream class. Output streams are not buffered in general, so it is usually a good idea to write data in chunks when possible. The class core.io.BufferedOStream can be used to wrap an output to ensure data is buffered.

The OStream class has the following members:

void write(core.io.Buffer buf)
Write all data from the start of the buffer and up to, but not including the filled marker in the buffer. Blocks until all data is written.
void write(core.io.Buffer buf, core.Nat start)
Write all data from start up to, but not including the filled marker in the buffer. Blocks until all data is written.
void flush()
Flush any buffered data to the destination. The exact behavior of this operation depends on the stream that is used. For example, file- and network streams are generally unbuffered by default, so it is not necessary to flush them. There are, however, streams that buffer the input for efficiency (e.g. core.io.BufferedOStream). These use flush() to allow manually flushing the buffer. Calls to flush are generally passed along chains of output streams. For example, calling flush on a BufferedOStream will flush the BufferedOStream and call flush on whichever stream the BufferedOStream is writing its data to.
void close()
Closes the stream and frees any associated resources. As with flush, this is generally propagated in the case of chained streams.

The destructor of the OStream calls close as a last resort. Since output streams are garbage collected, it might be a while before the destructor is called. As such, manually closing streams is generally preferred.

Input Streams

An input stream reads data from some source. Input streams are generally not buffered, so it is usually a good idea to read data in chunks. This is done by many consumers of data from input streams, such as the text IO system.

The IStream class has the following members:

core.Bool more()
Returns false if the end of the stream has been reached, and true otherwise. In some cases it is not possible to determine the end of stream ahead of time. In such cases, more might return true even if read will later report that it managed to read zero bytes. Whenever read succeeds with zero bytes, more will start returning false afterwards.
core.io.Buffer read(core.io.Buffer to)
Read data from the stream and attempt to fill the bytes from filled to count in the buffer to with bytes from the stream. Returns a buffer with the newly acquired data, and a filled member that has been updated to reflect the number of read bytes. Generally, the returned buffer refers to the same contents as to to avoid copying potentially large amounts of data. The implementation is, however, free to re-allocate the buffer as needed. In particular, this is always the case when the system needs to execute code on another thread, and thereby copy the buffers. Because of this, users of the read function shall always use the returned buffer and not rely on the buffer passed as to being updated.

The function blocks until at least one byte has been successfully read from the stream, or if the end of stream has been reached. This means that if filled has not been updated from its original location, then it is guaranteed that the end of the stream has been reached. Note, however, that the read function does not guarantee that the entire buffer has been filled. This often happens in the case of networking, for example. If the remote end has only sent one byte, a read operation of 10 bytes will typically only result in 1 out of 10 bytes being returned, since filling the buffer would potentially block foreveer (for example, the remote end might wait for a response).
core.io.Buffer read(core.Nat maxBytes)
Like read(Buffer), but allocates a new buffer rather than reusing an existing one.
core.io.Buffer fill(core.io.Buffer to)
Like read, but repeats the read operation until the buffer is full, or until the end of the stream has been reached. This means that if the buffer is not full after read has been called, then the end of stream has always been reached. Do, however, note that this behavior may not always be desirable when working with pipes or networking.
core.io.Buffer fill(core.Nat bytes)
Like fill(Buffer), but allocates a new buffer rather than reusing an existing one.
core.io.Buffer peek(core.io.Buffer to)
The exact number of bytes that can be safely acquired through a peek operation depends on the stream in use.
core.io.Buffer peek(core.Nat maxBytes)
Like peek(Buffer), but allocates a new buffer rather than reusing an existing one.
void close()
Closes the stream and frees any associated resources. As with flush, this is generally propagated in the case of chained streams.

The destructor of the OStream calls close as a last resort. Since output streams are garbage collected, it might be a while before the destructor is called. As such, manually closing streams is generally preferred.
core.io.RIStream randomAccess()
Create a random access stream based on this stream. For streams that are random access, this function simply returns the same stream, but casted to a random access stream. For streams that do not natively support random access, it instead a core.io.LazyMemIStream that buffers data to allow seeking from this point onwards. Regardless of which case was used, the original stream does not need to be closed.

Random Access Input Streams

Random access input stream are derived from core.io.RIStream. The class RIStream inherit from IStream, which means that random-access input streams are usable as regular input streams. Random streams provide the following members in addition to regular streams:

core.Word length()
Get the length of the input data, in bytes.
core.Word tell()
Get current position.
void seek(core.Word to)
Seek relative the start.