github.com/provpn/mobile@v0.0.0-20210315122651-28c475f89f6c/internal/binres/binres.go

// Copyright 2015 The Go Authors.  All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.

//go:generate go run genarsc.go
//go:generate stringer -output binres_string.go -type ResType,DataType

// Package binres implements encoding and decoding of android binary resources.
//
// Binary resource structs support unmarshalling the binary output of aapt.
// Implementations of marshalling for each struct must produce the exact input
// sent to unmarshalling. This allows tests to validate each struct representation
// of the binary format as follows:
//
//  * unmarshal the output of aapt
//  * marshal the struct representation
//  * perform byte-to-byte comparison with aapt output per chunk header and body
//
// This process should strive to make structs idiomatic to make parsing xml text
// into structs trivial.
//
// Once the struct representation is validated, tests for parsing xml text
// into structs can become self-referential as the following holds true:
//
//  * the unmarshalled input of aapt output is the only valid target
//  * the unmarshalled input of xml text may be compared to the unmarshalled
//    input of aapt output to identify errors, e.g. text-trims, wrong flags, etc
//
// This provides validation, byte-for-byte, for producing binary xml resources.
//
// It should be made clear that unmarshalling binary resources is currently only
// in scope for proving that the BinaryMarshaler works correctly. Any other use
// is currently out of scope.
//
// A simple view of binary xml document structure:
//
//  XML
//    Pool
//    Map
//    Namespace
//    [...node]
//
// Additional resources:
// https://android.googlesource.com/platform/frameworks/base/+/master/libs/androidfw/include/androidfw/ResourceTypes.h
// https://justanapplication.wordpress.com/2011/09/13/ (a series of articles, increment date)
package binres

import (
	"encoding"
	"encoding/binary"
	"encoding/xml"
	"fmt"
	"io"
	"sort"
	"strconv"
	"strings"
	"unicode"
)

func errWrongType(have ResType, want ...ResType) error {
	return fmt.Errorf("wrong resource type %s, want one of %v", have, want)
}

type ResType uint16

func (t ResType) IsSupported() bool {
	return t != ResNull
}

// explicitly defined for clarity and resolvability with apt source
const (
	ResNull       ResType = 0x0000
	ResStringPool ResType = 0x0001
	ResTable      ResType = 0x0002
	ResXML        ResType = 0x0003

	ResXMLStartNamespace ResType = 0x0100
	ResXMLEndNamespace   ResType = 0x0101
	ResXMLStartElement   ResType = 0x0102
	ResXMLEndElement     ResType = 0x0103
	ResXMLCharData       ResType = 0x0104

	ResXMLResourceMap ResType = 0x0180

	ResTablePackage  ResType = 0x0200
	ResTableType     ResType = 0x0201
	ResTableTypeSpec ResType = 0x0202
	ResTableLibrary  ResType = 0x0203
)

var (
	btou16 = binary.LittleEndian.Uint16
	btou32 = binary.LittleEndian.Uint32
	putu16 = binary.LittleEndian.PutUint16
	putu32 = binary.LittleEndian.PutUint32
)

// unmarshaler wraps BinaryUnmarshaler to provide byte size of decoded chunks.
type unmarshaler interface {
	encoding.BinaryUnmarshaler

	// size returns the byte size unmarshalled after a call to
	// UnmarshalBinary, or otherwise zero.
	size() int
}

// chunkHeader appears at the front of every data chunk in a resource.
type chunkHeader struct {
	// Type of data that follows this header.
	typ ResType

	// Advance slice index by this value to find its associated data, if any.
	headerByteSize uint16

	// This is the header size plus the size of any data associated with the chunk.
	// Advance slice index by this value to completely skip its contents, including
	// any child chunks. If this value is the same as headerByteSize, there is
	// no data associated with the chunk.
	byteSize uint32
}

// size implements unmarshaler.
func (hdr chunkHeader) size() int { return int(hdr.byteSize) }

func (hdr *chunkHeader) UnmarshalBinary(bin []byte) error {
	hdr.typ = ResType(btou16(bin))
	if !hdr.typ.IsSupported() {
		return fmt.Errorf("%s not supported", hdr.typ)
	}
	hdr.headerByteSize = btou16(bin[2:])
	hdr.byteSize = btou32(bin[4:])
	if len(bin) < int(hdr.byteSize) {
		return fmt.Errorf("too few bytes to unmarshal chunk body, have %v, need at-least %v", len(bin), hdr.byteSize)
	}
	return nil
}

func (hdr chunkHeader) MarshalBinary() ([]byte, error) {
	if !hdr.typ.IsSupported() {
		return nil, fmt.Errorf("%s not supported", hdr.typ)
	}

	bin := make([]byte, 8)
	putu16(bin, uint16(hdr.typ))
	putu16(bin[2:], hdr.headerByteSize)
	putu32(bin[4:], hdr.byteSize)
	return bin, nil
}

type XML struct {
	chunkHeader

	Pool *Pool
	Map  *Map

	Namespace *Namespace
	Children  []*Element

	// tmp field used when unmarshalling binary
	stack []*Element
}

// RawValueByName returns the original raw string value of first matching element attribute, or error if not exists.
// Given <manifest package="VAL" ...> then RawValueByName("manifest", xml.Name{Local: "package"}) returns "VAL".
func (bx *XML) RawValueByName(elname string, attrname xml.Name) (string, error) {
	elref, err := bx.Pool.RefByName(elname)
	if err != nil {
		return "", err
	}
	nref, err := bx.Pool.RefByName(attrname.Local)
	if err != nil {
		return "", err
	}
	nsref := PoolRef(NoEntry)
	if attrname.Space != "" {
		nsref, err = bx.Pool.RefByName(attrname.Space)
		if err != nil {
			return "", err
		}
	}

	for el := range bx.iterElements() {
		if el.Name == elref {
			for _, attr := range el.attrs {
				// TODO enforce TypedValue DataString constraint?
				if nsref == attr.NS && nref == attr.Name {
					return bx.Pool.strings[int(attr.RawValue)], nil
				}
			}
		}
	}
	return "", fmt.Errorf("no matching element %q for attribute %+v found", elname, attrname)
}

const (
	androidSchema = "http://schemas.android.com/apk/res/android"
	toolsSchema   = "http://schemas.android.com/tools"
)

// skipSynthesize is set true for tests to avoid synthesis of additional nodes and attributes.
var skipSynthesize bool

// UnmarshalXML decodes an AndroidManifest.xml document returning type XML
// containing decoded resources.
func UnmarshalXML(r io.Reader, withIcon bool) (*XML, error) {
	tbl, err := OpenTable()
	if err != nil {
		return nil, err
	}

	lr := &lineReader{r: r}
	dec := xml.NewDecoder(lr)
	bx := new(XML)

	// temporary pool to resolve real poolref later
	pool := new(Pool)

	type ltoken struct {
		xml.Token
		line int
	}
	var q []ltoken

	for {
		line := lr.line(dec.InputOffset())
		tkn, err := dec.Token()
		if err != nil {
			if err == io.EOF {
				break
			}
			return nil, err
		}
		tkn = xml.CopyToken(tkn)

		switch tkn := tkn.(type) {
		case xml.StartElement:
			switch tkn.Name.Local {
			default:
				q = append(q, ltoken{tkn, line})
			case "uses-sdk":
				return nil, fmt.Errorf("manual declaration of uses-sdk in AndroidManifest.xml not supported")
			case "manifest":
				// synthesize additional attributes and nodes for use during encode.
				tkn.Attr = append(tkn.Attr,
					xml.Attr{
						Name: xml.Name{
							Space: "",
							Local: "platformBuildVersionCode",
						},
						Value: "15",
					},
					xml.Attr{
						Name: xml.Name{
							Space: "",
							Local: "platformBuildVersionName",
						},
						Value: "4.0.4-1406430",
					})

				q = append(q, ltoken{tkn, line})

				if !skipSynthesize {
					s := xml.StartElement{
						Name: xml.Name{
							Space: "",
							Local: "uses-sdk",
						},
						Attr: []xml.Attr{
							xml.Attr{
								Name: xml.Name{
									Space: androidSchema,
									Local: "minSdkVersion",
								},
								Value: fmt.Sprintf("%v", MinSDK),
							},
						},
					}
					e := xml.EndElement{Name: xml.Name{Local: "uses-sdk"}}

					q = append(q, ltoken{s, line}, ltoken{e, line})
				}
			case "application":
				if !skipSynthesize {
					for _, attr := range tkn.Attr {
						if attr.Name.Space == androidSchema && attr.Name.Local == "icon" {
							return nil, fmt.Errorf("manual declaration of android:icon in AndroidManifest.xml not supported")
						}
					}
					if withIcon {
						tkn.Attr = append(tkn.Attr,
							xml.Attr{
								Name: xml.Name{
									Space: androidSchema,
									Local: "icon",
								},
								Value: "@mipmap/icon",
							})
					}
				}
				q = append(q, ltoken{tkn, line})
			}
		default:
			q = append(q, ltoken{tkn, line})
		}
	}

	for _, ltkn := range q {
		tkn, line := ltkn.Token, ltkn.line

		switch tkn := tkn.(type) {
		case xml.StartElement:
			el := &Element{
				NodeHeader: NodeHeader{
					LineNumber: uint32(line),
					Comment:    0xFFFFFFFF,
				},
				NS:   NoEntry,
				Name: pool.ref(tkn.Name.Local),
			}
			if len(bx.stack) == 0 {
				bx.Children = append(bx.Children, el)
			} else {
				n := len(bx.stack)
				var p *Element
				p, bx.stack = bx.stack[n-1], bx.stack[:n-1]
				p.Children = append(p.Children, el)
				bx.stack = append(bx.stack, p)
			}
			bx.stack = append(bx.stack, el)

			for _, attr := range tkn.Attr {
				if (attr.Name.Space == "xmlns" && attr.Name.Local == "tools") || attr.Name.Space == toolsSchema {
					continue // TODO can tbl be queried for schemas to determine validity instead?
				}

				if attr.Name.Space == "xmlns" && attr.Name.Local == "android" {
					if bx.Namespace != nil {
						return nil, fmt.Errorf("multiple declarations of xmlns:android encountered")
					}
					bx.Namespace = &Namespace{
						NodeHeader: NodeHeader{
							LineNumber: uint32(line),
							Comment:    NoEntry,
						},
						prefix: 0,
						uri:    0,
					}
					continue
				}

				nattr := &Attribute{
					NS:       pool.ref(attr.Name.Space),
					Name:     pool.ref(attr.Name.Local),
					RawValue: NoEntry,
				}
				el.attrs = append(el.attrs, nattr)

				if attr.Name.Space == "" {
					nattr.NS = NoEntry
					// TODO it's unclear how to query these
					switch attr.Name.Local {
					case "platformBuildVersionCode":
						nattr.TypedValue.Type = DataIntDec
						i, err := strconv.Atoi(attr.Value)
						if err != nil {
							return nil, err
						}
						nattr.TypedValue.Value = uint32(i)
					default: // "package", "platformBuildVersionName", and any invalid
						nattr.RawValue = pool.ref(attr.Value)
						nattr.TypedValue.Type = DataString
					}
				} else {
					// get type spec and value data type
					ref, err := tbl.RefByName("attr/" + attr.Name.Local)
					if err != nil {
						return nil, err
					}
					nt, err := ref.Resolve(tbl)
					if err != nil {
						return nil, err
					}
					if len(nt.values) == 0 {
						panic("encountered empty values slice")
					}

					if len(nt.values) == 1 {
						val := nt.values[0]
						if val.data.Type != DataIntDec {
							panic("TODO only know how to handle DataIntDec type here")
						}

						t := DataType(val.data.Value)
						switch t {
						case DataString, DataAttribute, DataType(0x3e):
							// TODO identify 0x3e, in bootstrap.xml this is the native lib name
							nattr.RawValue = pool.ref(attr.Value)
							nattr.TypedValue.Type = DataString
							nattr.TypedValue.Value = uint32(nattr.RawValue)
						case DataIntBool, DataType(0x08):
							nattr.TypedValue.Type = DataIntBool
							switch attr.Value {
							case "true":
								nattr.TypedValue.Value = 0xFFFFFFFF
							case "false":
								nattr.TypedValue.Value = 0
							default:
								return nil, fmt.Errorf("invalid bool value %q", attr.Value)
							}
						case DataIntDec, DataFloat, DataFraction:
							// TODO DataFraction needs it's own case statement. minSdkVersion identifies as DataFraction
							// but has accepted input in the past such as android:minSdkVersion="L"
							// Other use-cases for DataFraction are currently unknown as applicable to manifest generation
							// but this provides minimum support for writing out minSdkVersion="15" correctly.
							nattr.TypedValue.Type = DataIntDec
							i, err := strconv.Atoi(attr.Value)
							if err != nil {
								return nil, err
							}
							nattr.TypedValue.Value = uint32(i)
						case DataReference:
							nattr.TypedValue.Type = DataReference
							dref, err := tbl.RefByName(attr.Value)
							if err != nil {
								if strings.HasPrefix(attr.Value, "@mipmap") {
									// firstDrawableId is a TableRef matching first entry of mipmap spec initialized by NewMipmapTable.
									// 7f is default package, 02 is mipmap spec, 0000 is first entry; e.g. R.drawable.icon
									// TODO resource table should generate ids as required.
									const firstDrawableId = 0x7f020000
									nattr.TypedValue.Value = firstDrawableId
									continue
								}
								return nil, err
							}
							nattr.TypedValue.Value = uint32(dref)
						default:
							return nil, fmt.Errorf("unhandled data type %0#2x: %s", uint8(t), t)
						}
					} else {
						// 0x01000000 is an unknown ref that doesn't point to anything, typically
						// located at the start of entry value lists, peek at last value to determine type.
						t := nt.values[len(nt.values)-1].data.Type
						switch t {
						case DataIntDec:
							for _, val := range nt.values {
								if val.name == 0x01000000 {
									continue
								}
								nr, err := val.name.Resolve(tbl)
								if err != nil {
									return nil, err
								}
								if attr.Value == nr.key.Resolve(tbl.pkgs[0].keyPool) { // TODO hard-coded pkg ref
									nattr.TypedValue = *val.data
									break
								}
							}
						case DataIntHex:
							nattr.TypedValue.Type = t
							for _, x := range strings.Split(attr.Value, "|") {
								for _, val := range nt.values {
									if val.name == 0x01000000 {
										continue
									}
									nr, err := val.name.Resolve(tbl)
									if err != nil {
										return nil, err
									}
									if x == nr.key.Resolve(tbl.pkgs[0].keyPool) { // TODO hard-coded pkg ref
										nattr.TypedValue.Value |= val.data.Value
										break
									}
								}
							}
						default:
							return nil, fmt.Errorf("unhandled data type for configuration %0#2x: %s", uint8(t), t)
						}
					}
				}
			}
		case xml.CharData:
			if s := poolTrim(string(tkn)); s != "" {
				cdt := &CharData{
					NodeHeader: NodeHeader{
						LineNumber: uint32(line),
						Comment:    NoEntry,
					},
					RawData: pool.ref(s),
				}
				el := bx.stack[len(bx.stack)-1]
				if el.head == nil {
					el.head = cdt
				} else if el.tail == nil {
					el.tail = cdt
				} else {
					return nil, fmt.Errorf("element head and tail already contain chardata")
				}
			}
		case xml.EndElement:
			if tkn.Name.Local == "manifest" {
				bx.Namespace.end = &Namespace{
					NodeHeader: NodeHeader{
						LineNumber: uint32(line),
						Comment:    NoEntry,
					},
					prefix: 0,
					uri:    0,
				}
			}
			n := len(bx.stack)
			var el *Element
			el, bx.stack = bx.stack[n-1], bx.stack[:n-1]
			if el.end != nil {
				return nil, fmt.Errorf("element end already exists")
			}
			el.end = &ElementEnd{
				NodeHeader: NodeHeader{
					LineNumber: uint32(line),
					Comment:    NoEntry,
				},
				NS:   el.NS,
				Name: el.Name,
			}
		case xml.Comment, xml.ProcInst:
			// discard
		default:
			panic(fmt.Errorf("unhandled token type: %T %+v", tkn, tkn))
		}
	}

	// pools appear to be sorted as follows:
	// * attribute names prefixed with android:
	// * "android", [schema-url], [empty-string]
	// * for each node:
	//   * attribute names with no prefix
	//   * node name
	//   * attribute value if data type of name is DataString, DataAttribute, or 0x3e (an unknown)
	bx.Pool = new(Pool)

	var arecurse func(*Element)
	arecurse = func(el *Element) {
		for _, attr := range el.attrs {
			if attr.NS == NoEntry {
				continue
			}
			if attr.NS.Resolve(pool) == androidSchema {
				bx.Pool.strings = append(bx.Pool.strings, attr.Name.Resolve(pool))
			}
		}
		for _, child := range el.Children {
			arecurse(child)
		}
	}
	for _, el := range bx.Children {
		arecurse(el)
	}

	// TODO encoding/xml does not enforce namespace prefix and manifest encoding in aapt
	// appears to ignore all other prefixes. Inserting this manually is not strictly correct
	// for the general case, but the effort to do otherwise currently offers nothing.
	bx.Pool.strings = append(bx.Pool.strings, "android", androidSchema)

	// there always appears to be an empty string located after schema, even if one is
	// not present in manifest.
	bx.Pool.strings = append(bx.Pool.strings, "")

	var brecurse func(*Element)
	brecurse = func(el *Element) {
		for _, attr := range el.attrs {
			if attr.NS == NoEntry {
				bx.Pool.strings = append(bx.Pool.strings, attr.Name.Resolve(pool))
			}
		}

		bx.Pool.strings = append(bx.Pool.strings, el.Name.Resolve(pool))

		for _, attr := range el.attrs {
			if attr.RawValue != NoEntry {
				bx.Pool.strings = append(bx.Pool.strings, attr.RawValue.Resolve(pool))
			} else if attr.NS == NoEntry {
				bx.Pool.strings = append(bx.Pool.strings, fmt.Sprintf("%+v", attr.TypedValue.Value))
			}
		}

		if el.head != nil {
			bx.Pool.strings = append(bx.Pool.strings, el.head.RawData.Resolve(pool))
		}
		if el.tail != nil {
			bx.Pool.strings = append(bx.Pool.strings, el.tail.RawData.Resolve(pool))
		}

		for _, child := range el.Children {
			brecurse(child)
		}
	}
	for _, el := range bx.Children {
		brecurse(el)
	}

	// do not eliminate duplicates until the entire slice has been composed.
	// consider <activity android:label="label" .../>
	// all attribute names come first followed by values; in such a case, the value "label"
	// would be a reference to the same "android:label" in the string pool which will occur
	// within the beginning of the pool where other attr names are located.
	bx.Pool.strings = asSet(bx.Pool.strings)

	// TODO consider cases of multiple declarations of the same attr name that should return error
	// before ever reaching this point.
	bx.Map = new(Map)
	for _, s := range bx.Pool.strings {
		ref, err := tbl.RefByName("attr/" + s)
		if err != nil {
			break // break after first non-ref as all strings after are also non-refs.
		}
		bx.Map.rs = append(bx.Map.rs, ref)
	}

	// resolve tmp pool refs to final pool refs
	// TODO drop this in favor of sort directly on Table
	var resolve func(el *Element)
	resolve = func(el *Element) {
		if el.NS != NoEntry {
			el.NS = bx.Pool.ref(el.NS.Resolve(pool))
			el.end.NS = el.NS
		}
		el.Name = bx.Pool.ref(el.Name.Resolve(pool))
		el.end.Name = el.Name
		for _, attr := range el.attrs {
			if attr.NS != NoEntry {
				attr.NS = bx.Pool.ref(attr.NS.Resolve(pool))
			}
			attr.Name = bx.Pool.ref(attr.Name.Resolve(pool))
			if attr.RawValue != NoEntry {
				attr.RawValue = bx.Pool.ref(attr.RawValue.Resolve(pool))
				if attr.TypedValue.Type == DataString {
					attr.TypedValue.Value = uint32(attr.RawValue)
				}
			}
		}
		for _, child := range el.Children {
			resolve(child)
		}
	}
	for _, el := range bx.Children {
		resolve(el)
	}

	var asort func(*Element)
	asort = func(el *Element) {
		sort.Sort(byType(el.attrs))
		sort.Sort(byNamespace(el.attrs))
		sort.Sort(byName(el.attrs))
		for _, child := range el.Children {
			asort(child)
		}
	}
	for _, el := range bx.Children {
		asort(el)
	}

	for i, s := range bx.Pool.strings {
		switch s {
		case androidSchema:
			bx.Namespace.uri = PoolRef(i)
			bx.Namespace.end.uri = PoolRef(i)
		case "android":
			bx.Namespace.prefix = PoolRef(i)
			bx.Namespace.end.prefix = PoolRef(i)
		}
	}

	return bx, nil
}

// UnmarshalBinary decodes all resource chunks in buf returning any error encountered.
func (bx *XML) UnmarshalBinary(buf []byte) error {
	if err := (&bx.chunkHeader).UnmarshalBinary(buf); err != nil {
		return err
	}
	buf = buf[8:]
	for len(buf) > 0 {
		k, err := bx.unmarshalBinaryKind(buf)
		if err != nil {
			return err
		}
		buf = buf[k.size():]
	}
	return nil
}

// unmarshalBinaryKind decodes and stores the first resource chunk of bin.
// It returns the unmarshaler interface and any error encountered.
// If k.size() < len(bin), subsequent chunks can be decoded at bin[k.size():].
func (bx *XML) unmarshalBinaryKind(bin []byte) (k unmarshaler, err error) {
	k, err = bx.kind(ResType(btou16(bin)))
	if err != nil {
		return nil, err
	}
	if err = k.UnmarshalBinary(bin); err != nil {
		return nil, err
	}
	return k, nil
}

func (bx *XML) kind(t ResType) (unmarshaler, error) {
	switch t {
	case ResStringPool:
		if bx.Pool != nil {
			return nil, fmt.Errorf("pool already exists")
		}
		bx.Pool = new(Pool)
		return bx.Pool, nil
	case ResXMLResourceMap:
		if bx.Map != nil {
			return nil, fmt.Errorf("resource map already exists")
		}
		bx.Map = new(Map)
		return bx.Map, nil
	case ResXMLStartNamespace:
		if bx.Namespace != nil {
			return nil, fmt.Errorf("namespace start already exists")
		}
		bx.Namespace = new(Namespace)
		return bx.Namespace, nil
	case ResXMLEndNamespace:
		if bx.Namespace.end != nil {
			return nil, fmt.Errorf("namespace end already exists")
		}
		bx.Namespace.end = new(Namespace)
		return bx.Namespace.end, nil
	case ResXMLStartElement:
		el := new(Element)
		if len(bx.stack) == 0 {
			bx.Children = append(bx.Children, el)
		} else {
			n := len(bx.stack)
			var p *Element
			p, bx.stack = bx.stack[n-1], bx.stack[:n-1]
			p.Children = append(p.Children, el)
			bx.stack = append(bx.stack, p)
		}
		bx.stack = append(bx.stack, el)
		return el, nil
	case ResXMLEndElement:
		n := len(bx.stack)
		var el *Element
		el, bx.stack = bx.stack[n-1], bx.stack[:n-1]
		if el.end != nil {
			return nil, fmt.Errorf("element end already exists")
		}
		el.end = new(ElementEnd)
		return el.end, nil
	case ResXMLCharData: // TODO assure correctness
		cdt := new(CharData)
		el := bx.stack[len(bx.stack)-1]
		if el.head == nil {
			el.head = cdt
		} else if el.tail == nil {
			el.tail = cdt
		} else {
			return nil, fmt.Errorf("element head and tail already contain chardata")
		}
		return cdt, nil
	default:
		return nil, fmt.Errorf("unexpected type %s", t)
	}
}

func (bx *XML) MarshalBinary() ([]byte, error) {
	bx.typ = ResXML
	bx.headerByteSize = 8

	var (
		bin, b []byte
		err    error
	)
	b, err = bx.chunkHeader.MarshalBinary()
	if err != nil {
		return nil, err
	}
	bin = append(bin, b...)

	b, err = bx.Pool.MarshalBinary()
	if err != nil {
		return nil, err
	}
	bin = append(bin, b...)

	b, err = bx.Map.MarshalBinary()
	if err != nil {
		return nil, err
	}
	bin = append(bin, b...)

	b, err = bx.Namespace.MarshalBinary()
	if err != nil {
		return nil, err
	}
	bin = append(bin, b...)

	for _, child := range bx.Children {
		if err := marshalRecurse(child, &bin); err != nil {
			return nil, err
		}
	}

	b, err = bx.Namespace.end.MarshalBinary()
	if err != nil {
		return nil, err
	}
	bin = append(bin, b...)

	putu32(bin[4:], uint32(len(bin)))
	return bin, nil
}

func marshalRecurse(el *Element, bin *[]byte) error {
	b, err := el.MarshalBinary()
	if err != nil {
		return err
	}
	*bin = append(*bin, b...)

	if el.head != nil {
		b, err := el.head.MarshalBinary()
		if err != nil {
			return err
		}
		*bin = append(*bin, b...)
	}

	for _, child := range el.Children {
		if err := marshalRecurse(child, bin); err != nil {
			return err
		}
	}

	b, err = el.end.MarshalBinary()
	if err != nil {
		return err
	}
	*bin = append(*bin, b...)

	return nil
}

func (bx *XML) iterElements() <-chan *Element {
	ch := make(chan *Element, 1)
	go func() {
		for _, el := range bx.Children {
			iterElementsRecurse(el, ch)
		}
		close(ch)
	}()
	return ch
}

func iterElementsRecurse(el *Element, ch chan *Element) {
	ch <- el
	for _, e := range el.Children {
		iterElementsRecurse(e, ch)
	}
}

// asSet returns a set from a slice of strings.
func asSet(xs []string) []string {
	m := make(map[string]bool)
	fo := xs[:0]
	for _, x := range xs {
		if !m[x] {
			m[x] = true
			fo = append(fo, x)
		}
	}
	return fo
}

// poolTrim trims all but immediately surrounding space.
// \n\t\tfoobar\n\t\t becomes \tfoobar\n
func poolTrim(s string) string {
	var start, end int
	for i, r := range s {
		if !unicode.IsSpace(r) {
			if i != 0 {
				start = i - 1 // preserve preceding space
			}
			break
		}
	}

	for i := len(s) - 1; i >= 0; i-- {
		r := rune(s[i])
		if !unicode.IsSpace(r) {
			if i != len(s)-1 {
				end = i + 2
			}
			break
		}
	}

	if start == 0 && end == 0 {
		return "" // every char was a space
	}

	return s[start:end]
}

// byNamespace sorts attributes based on string pool position of namespace.
// Given that "android" always preceeds "" in the pool, this results in the
// correct ordering of attributes.
type byNamespace []*Attribute

func (a byNamespace) Len() int { return len(a) }
func (a byNamespace) Less(i, j int) bool {
	return a[i].NS < a[j].NS
}
func (a byNamespace) Swap(i, j int) { a[i], a[j] = a[j], a[i] }

// byType sorts attributes by the uint8 value of the type.
type byType []*Attribute

func (a byType) Len() int { return len(a) }
func (a byType) Less(i, j int) bool {
	return a[i].TypedValue.Type < a[j].TypedValue.Type
}
func (a byType) Swap(i, j int) { a[i], a[j] = a[j], a[i] }

// byName sorts attributes that have matching types based on string pool position of name.
type byName []*Attribute

func (a byName) Len() int { return len(a) }
func (a byName) Less(i, j int) bool {
	return (a[i].TypedValue.Type == DataString || a[i].TypedValue.Type == DataIntDec) &&
		(a[j].TypedValue.Type == DataString || a[j].TypedValue.Type == DataIntDec) &&
		a[i].Name < a[j].Name
}
func (a byName) Swap(i, j int) { a[i], a[j] = a[j], a[i] }

type lineReader struct {
	off   int64
	lines []int64
	r     io.Reader
}

func (r *lineReader) Read(p []byte) (n int, err error) {
	n, err = r.r.Read(p)
	for i := 0; i < n; i++ {
		if p[i] == '\n' {
			r.lines = append(r.lines, r.off+int64(i))
		}
	}
	r.off += int64(n)
	return n, err
}

func (r *lineReader) line(pos int64) int {
	return sort.Search(len(r.lines), func(i int) bool {
		return pos < r.lines[i]
	}) + 1
}