git.lukeshu.com/go/lowmemjson@v0.3.9-0.20230723050957-72f6d13f6fb2/reencode.go

// Copyright (C) 2022-2023  Luke Shumaker <lukeshu@lukeshu.com>
//
// SPDX-License-Identifier: GPL-2.0-or-later

package lowmemjson

import (
	"fmt"
	"io"
	"unicode/utf8"

	"git.lukeshu.com/go/lowmemjson/internal/fastio"
	"git.lukeshu.com/go/lowmemjson/internal/jsonparse"
)

// A ReEncoderConfig controls how a ReEncoder should behave.
type ReEncoderConfig struct {
	// A JSON document is specified to be a single JSON element;
	// but it is often desirable to handle streams of multiple
	// JSON elements.
	AllowMultipleValues bool

	// Whether to minify the JSON.
	//
	// Trims all whitespace, except that it emits a newline
	// between two *number* top-level values (or puts a newline
	// after all top-level values if ForceTrailingNewlines).
	//
	// Trims superflous 0s from numbers.
	Compact bool

	// CompactIfUnder causes the *ReEncoder to behave as if
	// Compact=true for individual elements if doing so would
	// cause that element to be under this number of bytes.
	//
	// Has no affect if Compact is true or Indent is empty.
	//
	// This has O(2^min(CompactIfUnder, depth)) time overhead, so
	// set with caution.
	CompactIfUnder int

	// String to use to indent; ignored if Compact is true.
	//
	// Newlines are emitted *between* top-level values; a newline is
	// not emitted after the *last* top-level value (unless
	// ForceTrailingNewlines is on).
	Indent string

	// String to put before indents.
	Prefix string

	// Whether to emit a newline after each top-level value.  See
	// the comments on Compact and Indent for discussion of how
	// this is different than the usual behavior.
	ForceTrailingNewlines bool

	// CompactFloats causes the *ReEncoder to trim unnecessary '0'
	// digits from floating-point number values.
	CompactFloats bool

	// A JSON document is specified to be a sequence of Unicode
	// codepoints; InvalidUTF8 controls how the *ReEncoder behaves
	// when it encounters invalid UTF-8 bytes in a JSON string
	// (i.e. the string is not representable as a sequence of
	// Unicode codepoints, and thus the document is invalid JSON).
	InvalidUTF8 InvalidUTF8Mode

	// Returns whether a given character in a string should be
	// backslash-escaped.  The bool argument is whether it was
	// \u-escaped in the input.  This does not affect characters
	// that must or must-not be escaped to be valid JSON.
	//
	// If not set, then EscapeDefault is used.
	BackslashEscape BackslashEscaper
}

// NewReEncoder returns a new ReEncoder instance.
//
// A ReEncoder tends to make many small writes; if Out.Write
// calls are syscalls, then you may want to wrap Out in a
// bufio.Writer.
func NewReEncoder(out io.Writer, cfg ReEncoderConfig) *ReEncoder {
	var module reEncoderModule

	// Basic
	module = &reEncodeWrite{
		out: fastio.NewAllWriter(out),
	}

	// Whitespace
	if cfg.ForceTrailingNewlines {
		module = &reEncodeForceNL{
			out: module,
		}
	}
	switch {
	case cfg.Compact:
		module = &reEncodeCompactWS{
			out: module,
		}
	case cfg.Indent != "":
		if cfg.CompactIfUnder > 0 {
			module = &reEncodeCompactWSIfUnder{
				out:              module,
				CompactWSIfUnder: cfg.CompactIfUnder,
			}
		}
		module = &reEncodeIndent{
			out:    module,
			Indent: cfg.Indent,
			Prefix: cfg.Prefix,
		}
	}

	// Numbers
	if cfg.CompactFloats {
		module = &reEncodeCompactNum{
			out: module,
		}
	}

	// Strings
	escaper := cfg.BackslashEscape
	if escaper == nil {
		escaper = EscapeDefault
	}
	module = &reEncodeString{
		out:             module,
		BackslashEscape: escaper,
	}

	return &ReEncoder{
		out:                 module,
		esc:                 escaper,
		utf:                 cfg.InvalidUTF8,
		allowMultipleValues: cfg.AllowMultipleValues,
	}
}

// A ReEncoder takes a stream of JSON elements (by way of implementing
// io.Writer, io.StringWriter, io.ByteWriter, and WriteRune), and
// re-encodes the JSON, writing it to the .Out member.
//
// This is useful for prettifying, minifying, sanitizing, and/or
// validating JSON.
//
// The memory use of a ReEncoder is O(CompactIfUnder+depth).
type ReEncoder struct {
	out                 reEncoderModule
	esc                 BackslashEscaper
	utf                 InvalidUTF8Mode
	allowMultipleValues bool

	// state: .Write's/.WriteString's/.WriteRune's utf8-decoding buffer
	buf    [utf8.UTFMax]byte
	bufLen int

	// state: contract between the public API and .handleRune
	err      error
	par      jsonparse.Parser
	inputPos int64

	// state: .pushWriteBarrier and .popWriteBarrier
	barriers []barrier

	// state: .handleRuneType
	uhex [3]byte // "\uABCD"-encoded characters in strings
}

type barrier struct {
	inputPos  int64
	stackSize int
}

type reEncoderModule interface {
	HandleRune(c rune, t jsonparse.RuneType, escape BackslashEscapeMode, stackSize int) error
	PopWriteBarrier()
}

// public API //////////////////////////////////////////////////////////////////

var (
	_ fastio.AllWriter = (*ReEncoder)(nil)
	_ io.Closer        = (*ReEncoder)(nil)
)

func (enc *ReEncoder) getRuneFromBytes(str []byte, pos int) (c rune, size int, full, isRune bool) {
	var tmp []byte
	if pos < enc.bufLen {
		var buf [utf8.UTFMax]byte
		n := copy(buf[:], enc.buf[pos:enc.bufLen])
		n += copy(buf[n:], str)
		tmp = buf[:n]
	} else {
		tmp = str[pos-enc.bufLen:]
	}
	c, size = utf8.DecodeRune(tmp)
	switch {
	case c == utf8.RuneError && size <= 1 && !utf8.FullRune(tmp):
		return c, size, false, true
	case c == utf8.RuneError && size == 1 && enc.utf != InvalidUTF8Replace:
		return rune(tmp[0]), 1, true, false
	default:
		return c, size, true, true
	}
}

func (enc *ReEncoder) getRuneFromString(str string, pos int) (c rune, size int, full, isRune bool) {
	if pos < enc.bufLen {
		var buf [utf8.UTFMax]byte
		var tmp []byte
		n := copy(buf[:], enc.buf[pos:enc.bufLen])
		n += copy(buf[n:], str)
		tmp = buf[:n]
		c, size = utf8.DecodeRune(tmp)
		switch {
		case c == utf8.RuneError && size <= 1 && !utf8.FullRune(tmp):
			return c, size, false, true
		case c == utf8.RuneError && size == 1 && enc.utf != InvalidUTF8Replace:
			return rune(tmp[0]), 1, true, false
		default:
			return c, size, true, true
		}
	} else {
		tmp := str[pos-enc.bufLen:]
		c, size := utf8.DecodeRuneInString(tmp)
		switch {
		case c == utf8.RuneError && size <= 1 && !utf8.FullRuneInString(tmp):
			return c, size, false, true
		case c == utf8.RuneError && size == 1 && enc.utf != InvalidUTF8Replace:
			return rune(tmp[0]), 1, true, false
		default:
			return c, size, true, true
		}
	}
}

// Write implements io.Writer; it does what you'd expect.
//
// It is worth noting that Write returns the number of bytes consumed
// from p, not number of bytes written to the output stream.  This
// distinction that most io.Writer implementations don't need to make,
// but *ReEncoder does because it transforms the data written to it,
// and the number of bytes written may be wildly different than the
// number of bytes handled.
//
//nolint:dupl // Yes, this is mostly a duplicate of .WriteString().
func (enc *ReEncoder) Write(str []byte) (int, error) {
	if len(str) == 0 {
		return 0, nil
	}
	origBufLen := enc.bufLen
	var n int
	for {
		c, size, full, isRune := enc.getRuneFromBytes(str, n)
		if !full {
			if n < enc.bufLen {
				l := copy(enc.buf[:], enc.buf[n:enc.bufLen])
				l += copy(enc.buf[l:], str)
				enc.bufLen = l
			} else {
				enc.bufLen = copy(enc.buf[:], str[n-enc.bufLen:])
			}
			return len(str), nil
		}
		if enc.utf == InvalidUTF8Error && !isRune {
			return n - origBufLen, &ReEncodeSyntaxError{
				Offset: enc.inputPos,
				Err:    fmt.Errorf("invalid UTF-8: %#02x", c),
			}
		}
		enc.handleRune(c, size, isRune)
		if enc.err != nil {
			return n - origBufLen, enc.err
		}
		n += size
	}
}

// WriteString implements io.StringWriter; it does what you'd expect,
// but see the notes on the Write method.
//
//nolint:dupl // Yes, this is mostly a duplicate of .Write().
func (enc *ReEncoder) WriteString(str string) (int, error) {
	if len(str) == 0 {
		return 0, nil
	}
	origBufLen := enc.bufLen
	var n int
	for {
		c, size, full, isRune := enc.getRuneFromString(str, n)
		if !full {
			if n < enc.bufLen {
				l := copy(enc.buf[:], enc.buf[n:enc.bufLen])
				l += copy(enc.buf[l:], str)
				enc.bufLen = l
			} else {
				enc.bufLen = copy(enc.buf[:], str[n-enc.bufLen:])
			}
			return len(str), nil
		}
		if enc.utf == InvalidUTF8Error && !isRune {
			return n - origBufLen, &ReEncodeSyntaxError{
				Offset: enc.inputPos,
				Err:    fmt.Errorf("invalid UTF-8: %#02x", c),
			}
		}
		enc.handleRune(c, size, isRune)
		if enc.err != nil {
			return n - origBufLen, enc.err
		}
		n += size
	}
}

// WriteByte implements io.ByteWriter; it does what you'd expect.
func (enc *ReEncoder) WriteByte(b byte) error {
	return fastio.WriteByte(enc, b)
}

// WriteRune does what you'd expect.
func (enc *ReEncoder) WriteRune(c rune) (n int, err error) {
	return fastio.WriteRune(enc, c)
}

// Close implements io.Closer; it does what you'd expect, mostly.
//
// The *ReEncoder may continue to be written to with new JSON values
// if enc.AllowMultipleValues is set.
func (enc *ReEncoder) Close() error {
	if enc.bufLen > 0 {
		if enc.utf == InvalidUTF8Error {
			return &ReEncodeSyntaxError{
				Offset: enc.inputPos,
				Err:    fmt.Errorf("truncated UTF-8: %q", enc.buf[:enc.bufLen]),
			}
		}
		for i := 0; i < enc.bufLen; i++ {
			if enc.utf == InvalidUTF8Replace {
				enc.handleRune(utf8.RuneError, 1, true)
			} else {
				enc.handleRune(rune(enc.buf[i]), 1, false)
			}
			if enc.err != nil {
				return enc.err
			}
		}
	}
	if _, err := enc.par.HandleEOF(); err != nil {
		enc.err = &ReEncodeSyntaxError{
			Err:    err,
			Offset: enc.inputPos,
		}
		return enc.err
	}
	if len(enc.barriers) == 0 {
		if err := enc.handleRuneType(0, jsonparse.RuneTypeEOF, enc.stackSize(), true); err != nil {
			enc.err = &ReEncodeWriteError{
				Err:    err,
				Offset: enc.inputPos,
			}
			return enc.err
		}
		if enc.allowMultipleValues {
			enc.par.Reset()
		}
	}
	return nil
}

// isRune=false indicates that 'c' is a raw byte from invalid UTF-8.
func (enc *ReEncoder) handleRune(c rune, size int, isRune bool) {
	t, err := enc.par.HandleRune(c, isRune)
	if err != nil {
		enc.err = &ReEncodeSyntaxError{
			Err:    err,
			Offset: enc.inputPos,
		}
		return
	}
	if err := enc.handleRuneType(c, t, enc.stackSize(), isRune); err != nil {
		enc.err = &ReEncodeWriteError{
			Err:    err,
			Offset: enc.inputPos,
		}
		return
	}
	if t == jsonparse.RuneTypeEOF {
		if len(enc.barriers) == 0 {
			panic(fmt.Errorf("should not happen: EOF for rune %q without write barriers", c))
		}
		enc.err = &ReEncodeSyntaxError{
			Err:    fmt.Errorf("invalid character %q after top-level value", c),
			Offset: enc.inputPos,
		}
		return
	}

	enc.inputPos += int64(size)
}

// semi-public API /////////////////////////////////////////////////////////////

func (enc *ReEncoder) pushWriteBarrier() {
	enc.barriers = append(enc.barriers, barrier{
		inputPos:  enc.inputPos,
		stackSize: enc.stackSize(),
	})
	enc.par.PushWriteBarrier()
	enc.inputPos = 0
}

func (enc *ReEncoder) popWriteBarrier() {
	enc.par.PopBarrier()
	enc.inputPos += enc.barriers[len(enc.barriers)-1].inputPos
	enc.barriers = enc.barriers[:len(enc.barriers)-1]
	enc.out.PopWriteBarrier()
}

// internal ////////////////////////////////////////////////////////////////////

func (enc *ReEncoder) stackSize() int {
	sz := enc.par.StackSize()
	if len(enc.barriers) > 0 {
		sz += enc.barriers[len(enc.barriers)-1].stackSize
	}
	return sz
}

func (enc *ReEncoder) handleRuneType(c rune, t jsonparse.RuneType, stackSize int, isRune bool) error {
	switch t {
	case jsonparse.RuneTypeStringEsc, jsonparse.RuneTypeStringEscU:
		return nil
	case jsonparse.RuneTypeStringEsc1:
		switch c {
		case '"', '\\', '/':
			// self
		case 'b':
			c = '\b'
		case 'f':
			c = '\f'
		case 'n':
			c = '\n'
		case 'r':
			c = '\r'
		case 't':
			c = '\t'
		default:
			panic(fmt.Errorf("should not happen: rune %q is not a RuneTypeStringEsc1", c))
		}
		return enc.out.HandleRune(c, jsonparse.RuneTypeStringChar, BackslashEscapeShort, stackSize)
	case jsonparse.RuneTypeStringEscUA:
		enc.uhex[0] = byte(c)
		return nil
	case jsonparse.RuneTypeStringEscUB:
		enc.uhex[1] = byte(c)
		return nil
	case jsonparse.RuneTypeStringEscUC:
		enc.uhex[2] = byte(c)
		return nil
	case jsonparse.RuneTypeStringEscUD:
		mode := hexToMode(enc.uhex[0], enc.uhex[1], enc.uhex[2], byte(c))
		c = hexToRune(enc.uhex[0], enc.uhex[1], enc.uhex[2], byte(c))
		return enc.out.HandleRune(c, jsonparse.RuneTypeStringChar, mode, stackSize)
	case jsonparse.RuneTypeError:
		panic(fmt.Errorf("should not happen: handleRune called with %#v", t))
	default:
		if t > jsonparse.RuneTypeEOF {
			panic(fmt.Errorf("should not happen: handleRune called with %#v", t))
		}
		esc := BackslashEscapeNone
		if !isRune {
			esc = BackslashEscapeRawByte
		}
		return enc.out.HandleRune(c, t, esc, stackSize)
	}
}