go-parsekit/reader/reader.go

// Package reader provides a buffered Reader that wraps around an io.Reader.
//
// Functionally, it provides an input buffer in the form of a sliding window.
// Let's say we've got the following input coming up in the io.Reader that is
// wrapped by the Reader:
//
//     |H|e|l|l|o|,| |w|o|r|l|d|!|  <-- runes
//      0           6           12  <-- rune offset
//
// The Reader can now be used to retrieve runes from the input, based on their
// offset, using RuneAt(offset). Normally these runes will be retrieved in
// sequence, but that is not a requirement. Let's say we retrieve the rune with
// offset 6 from the input (the 'w'), then the Reader buffer be filled with runes
// from the io.Reader until there are enough runes available to return the rune
// for offset 6:
//
//     |H|e|l|l|o| |w|
//      0           6
//
// Using RuneAt, you can retrieve arbitrary runes. If you request one that is
// in the Reader buffer, then the buffered rune is returned. If you request one
// that is not in the buffer, then the buffer will be expanded.
//
// To make this into a sliding window, the Reader provides the method
// Flush(numberOfRunes). This method will drop the provided number of runes from
// the Reader buffer. So when we'd do a Flush(3) on the example buffer from above,
// then the Reader buffer would become:
//
//     |l|o| |w|
//      0     3
//
// Note that the offset for the first rune 'l' in the buffer is now 0.
// You can consider the input to be changed in a similar way:
//
//     |l|o|,| |w|o|r|l|d|!|
//      0           6     9
//
// So after a flush, the first upcoming rune after the flushed runes
// will always have index 0.
package reader

import (
	"bufio"
	"fmt"
	"io"
	"strings"
	"unicode/utf8"
)

// Reader wraps around a bufio.Reader and provides an additional layer of
// buffering that allows us to read the same runes over and over again.
// This is useful for implementing a parser that must be able to do lookahead
// on the input, returning to the original input position after finishing
// that lookahead).
//
// To minimze memory use, it is also possible to flush the read buffer when there is
// no more need to go back to previously read runes.
//
// The parserkit.reader.Reader is used internally by parsekit.TokenAPI.
type Reader struct {
	bufio        *bufio.Reader // Used for ReadRune()
	buffer       []rune        // Input buffer, holding runes that were read from input
	bufferOffset int           // The offset of the buffer, relative to the start of the input
	bufferLen    int           // Input size, the number of runes in the buffer
	err          error         // A read error, if one occurred
	errOffset    int           // The offset in the buffer at which the read error was encountered
}

// New initializes a new reader struct, wrapped around the provided input.
//
// The input can be any one of the following types:
// - string
// - type implementing io.Reader
// - bufio.Reader
func New(input interface{}) *Reader {
	return &Reader{
		bufio:  makeBufioReader(input),
		buffer: []rune{},
	}
}

func makeBufioReader(input interface{}) *bufio.Reader {
	switch input := input.(type) {
	case bufio.Reader:
		return &input
	case *bufio.Reader:
		return input
	case io.Reader:
		return bufio.NewReader(input)
	case string:
		return bufio.NewReader(strings.NewReader(input))
	default:
		panic(fmt.Sprintf("parsekit.reader.New(): no support for input of type %T", input))
	}
}

// RuneAt reads the rune at the provided rune offset.
//
// This offset is relative to the current starting position of the buffer in
// the reader. When starting reading, offset 0 will point at the start of the
// input. After flushing, offset 0 will point at the input up to where
// the flush was done.
//
// The error return value will be nil when reading was successful.
// When an invalid rune is encountered on the input, the error will be nil,
// but the rune will be utf8.RuneError
//
// When reading failed, the rune will be utf8.RuneError and the error will
// be not nil. One special read fail is actually a normal situation: end
// of file reached. In that case, the returned error wille be io.EOF.
func (r *Reader) RuneAt(offset int) (rune, error) {
	// Re-issue a previously seen read error.
	if r.err != nil && offset >= r.errOffset {
		return utf8.RuneError, r.err
	}

	// Rune at provided offset is not yet available in the input buffer.
	// Read runes until we have enough runes to satisfy the offset.
	for r.bufferLen <= offset {
		readRune, _, err := r.bufio.ReadRune()

		// Handle errors.
		if err != nil {
			r.err = err
			r.errOffset = r.bufferLen
			return utf8.RuneError, err
		}

		// Skip BOM.
		if readRune == '\uFEFF' && r.bufferOffset == 0 {
			r.bufferOffset++
			continue
		}

		r.buffer = append(r.buffer, readRune)
		r.bufferLen++
	}
	return r.buffer[offset], nil
}

// Flush deletes the provided number of runes from the start of the
// reader buffer. After flushing the buffer, offset 0 as used by RuneAt()
// will point to the rune that comes after the flushed runes.
// So what this basically does is turn the Reader into a sliding window.
func (r *Reader) Flush(numberOfRunes int) {
	if numberOfRunes > r.bufferLen {
		panic(fmt.Sprintf(
			"parsekit.Input.Reader.Flush(): number of runes to flush (%d) "+
				"exceeds size of the buffer (%d)", numberOfRunes, r.bufferLen))
	}
	r.bufferOffset += numberOfRunes
	r.bufferLen -= numberOfRunes
	r.buffer = r.buffer[numberOfRunes:]
	if r.err != nil {
		r.errOffset -= numberOfRunes
	}
}