Buffer(缓冲) API

buffer库提供了便于使用的IO流。本库提供的流更像是io库提供的，由io.open返回的流，不像filesystem.open返回的原始流，后者支持的方法没有前者那么多。下文的实例方法板块列出的方法可被用于从io.open函数获取的文件句柄。因此就算你没打算构建自定义带缓冲IO流，此API文档也会大有裨益。

此外，该API还允许你创建带缓冲IO流。你需要提供后端的流式读取和流式写入，buffer库提供数据缓冲与格式化。一般而言，用户无需构建自己的带缓冲输入输出流。以备参考，io库就使用了带缓冲IO流（包含文件io和控制台io）。

下列方法从buffer库自身调用。

• buffer.new([mode: string], stream: table)

用读写模式(mode)来修饰传入的流(stream)，以创建一个带缓冲IO流。模式(mode)可以为只读（r或nil）、读写（rw）或仅写入（w）。请查看stream对象所需的流实例方法有关内容。

The following methods can only be called on instances created by buffer.new (note file handles returned by io.open are also buffered streams, created with buffer.new). These methods are instance methods, requiring instance call notation :. In order to help differentiate these instance methods from static methods (e.g. buffer.new), b: will be used to prefix the method names.

• b:flush()

  If any data is buffered it is immediately written to the stream and released.

• b:close()

  Flushes the buffer and closes the wrapped stream.

• b:setvbuf([mode: string], [size: number]) mode, size

  Sets the buffering mode and size and returns the result mode and size. The amount of data buffered is specified by size which defaults to [512, 8192] bytes, depending on available system memory. mode and size can be nil, in which case the previous values are used for either. size is also used in read(n) calls to the stream.
  Modes only affect write, which include:

  • “no” writes are immediately pushed to the stream.
  • “full” writes are buffered up to size bytes. This is the default mode.
  • “line” writes are buffered until newlines are found or size is reached, whichever comes first.
• b:write([values...])

  Writes each value to the stream, first buffering based on the mode and buffer size (see setvbuf). Note that to write to a file, you have to open it for write.

snippet.lua

local file = io.open("/tmp/foo.txt", "w")
file:write("abc", "def", "\n")
file:close()
-- foo.txt now has "abcdef\n"

• b:lines([line_formats...]) string array

  Returns a function iterator which reads from the stream until it reaches nil. On each read, the line_formats list of args as passed to stream:read(...). The overwhelmingly typical use is to not define line_formats, i.e. passing no args to lines(). The default behavior (i..e without line_formats) is to read a “line” at a time from the stream.

snippet.lua

local file = io.open("/tmp/foobar.txt")
for line in file:lines() do
  process_next_line(line)
end
file:close()

• b:read([formats...]) string...

  A fairly advanced reader that support various formats. First of all, if called with no format, i.e an empty param list, it reads the next line from the stream, which is equivalent to read("*l")
  Each format is read from the stream and all returned in a multiple return value list of the results. Note all format strings are prefixed with * and also note that only the first char of the string names of the formats matters, the rest is ignored. These are the supported formats:

  • a number value, e.g. 10

    Read n bytes (in binary mode) or chars (in text mode) from the stream; result is returned as a string. See io.open for more details about how to open files in different modes.

    local chars = b:read(10)

  • “*n” or “*number”

    Read the next series of bytes from the stream that can be interpreted as a number. Note that reading numbers is also affected by the open mode, binary or text. See io.open for more details about how to open files in different modes..

    local number = b:read("*n")

  • “*l” or “*line”

    Read the next line from the stream, chopping off the line ending marker (which may be \n, \r, or \r\n)

    local line = b:read("*l")

  • “*L” or “*Line”

    Read the next line from the stream, like “*line”, but preserves the line ending marker as part of the result

    local whole_line = b:read("*L")

  • “*a” or “*all”

    Reads all remaining data from the stream until nil. There would be no point in having formats following this.

    local the_whole_file = b:read("*a")

• b:getTimeout() number

  Returns the current timeout (in seconds) set on the buffered stream. math.huge is the default timeout. Read setTimeout for more information about the effects of a buffered stream timeout.

• b:setTimeout(timeout)

  Sets the time in seconds a buffered stream will try to limit a read operation. Note that this timeout cannot be strictly adhered to. A read operation that completes within a single readChunk (an internal method that invokes the actual read on the stream) does not check the timeout limit. Timeout is only checked between stream reads within a single buffered read (an example follows). Thus, if a read requires multiple chunk reads, and the time between the start of the first read before the start of the last read is greater than or equal to the timeout, then the buffered stream will error. Again note that a timeout is default math.huge.

snippet.lua

local file = buffer.new("r", { read = function() os.sleep(5) return "a" end })
file:setvbuf("full", 1) -- set buffer size to 1 char
file:setTimeout(1) -- set buffer timeout to 1 second
-- this will time out before trying to read the 2nd char
local a, b = file:read(1, 1) -- read 1 char, then read 1 char again

• b:seek([whence:string], [offset:number])

  Moves the stream position by offset bytes from whence, both optional params. whence defaults to “cur”, and offset defaults to 0. Valid whence values:

  • “cur” from the current position.
  • “set” from the start of the stream.
  • “end” from the end of the stream. Returns the result of the seek operation on the stream (which may fail).

The following methods are expected to be implemented on the buffered streams passed to buffer.new.

• close() ok, reason

  Close handles, release resources, disconnect – and return success

• write(arg: string) ok, reason

  Write arg as bytes, assume a string of plain unformatted chars. Return falsey and reason on failure.

• read(n: number) ok, reason

  Return n bytes, and not n unicode-aware chars. Assume your data is binary data and let the buffer library manage the mode and the unicode string packaging (if applicable). Note that this is exactly how the filesystem library operates.The caller assumes there is more data to read until nil is returned. A empty string or a string shorter than n chars long is a valid return, but the caller may assume there is more data to request until nil is returned.

• seek([whence: string], [offset: number]) [offset from start] or falsey, reason

  Refer to b:seek() for details. In short, move the stream position to offset from whence, and return the offset from the start of the stream of the position after the seek operation. Note that seek("cur", 0) is a valid request, typical of the caller wanting to determine the current position of the stream. Your stream is not required to support seek, in such case (or in any case of failure) you should return nil, and the reason (as a string) for the failure.