cleanups
[epoint] / document / document.go
index c550619..0ac99b6 100644 (file)
@@ -1,26 +1,47 @@
-package document
-
-// An epoint document is a clear signed utf-8 text of key-value pairs
-// according to OpenPGP RFC 4880. The body contains a content-type
-// MIME header so it can be used in OpenPGP/MIME RFC 3156 emails with
-// the signature detached. The format of the key-value pairs are
-// similar to MIME header fields.
+// Package document implements epoint document parsing and creation.
+//
+// An epoint document is an OpenPGP (RFC 4880) clear signed
+// utf-8 text of key-value pairs.
+// The body contains a content-type MIME header so the document
+// can be used in OpenPGP/MIME (RFC 3156) emails.
+// The format of the key-value pairs are similar to MIME header
+// fields: keys and values are separated by ": ", repeated keys
+// are not allowed, long values can be split before a space.
 //
 // Example:
 //
 // -----BEGIN PGP SIGNED MESSAGE-----
 // Hash: SHA1
 //
-// Content-Type: text/plain.epoint.cert; charset=utf-8
+// Content-Type: text/plain.epoint.type; charset=utf-8
 //
 // Key: Value1
 // Another-Key: Value2
+// Last-Key: Long
+//  value that spans
+//  multiple lines
 // -----BEGIN PGP SIGNATURE-----
+//
 // pgp signature
 // -----END PGP SIGNATURE-----
+package document
+
+// TODO: error wrapper (so reporting to user or creating bounce cert is simple)
+// TODO: optional fields: exact semantics ("" vs "-" vs nil)
+// TODO: trailing space handling in ParseFields
+// TODO: fields of notice (last notice, serial, failure notice,..)
+// TODO: limits and cert type specific input validation
+// TODO: fix Cert mess
+// TODO: nonce is id, id is even number of hex digits
+// TODO: denom, issuer from key (key representation: armor?)
 
 import (
        "bytes"
+       "crypto"
+       "crypto/openpgp"
+       "crypto/openpgp/armor"
+       "crypto/openpgp/packet"
+       "crypto/sha1"
        "encoding/hex"
        "fmt"
        "reflect"
@@ -29,462 +50,683 @@ import (
        "time"
 )
 
-const ClearSignedHeader = "-----BEGIN PGP SIGNED MESSAGE-----\n"
+// limits
+const (
+       MaxFields             = 20
+       MaxLineLength         = 160  // 1 sha512 + 1 key (without \n)
+       MaxValueLength        = 1300 // 20 sha256 space separated (without \n)
+       MaxNonceLength        = 20
+       MaxDenominationLength = 100
+)
+
+const ClearSignedHeader = "-----BEGIN PGP SIGNED MESSAGE-----"
 
-// (non-standard) MIME subtype for epoint documents, see RFC 2045 and RFC 2046
+// MIME type for epoint documents, see RFC 4288
 var ContentType = map[string]string{
-       "cert":  "text/plain.epoint.cert; charset=utf-8",
-       "draft": "text/plain.epoint.draft; charset=utf-8",
+       "Draft":      "text/vnd.epoint.draft; charset=utf-8",
+       "Notice":     "text/vnd.epoint.notice; charset=utf-8",
+       "DebitCert":  "text/vnd.epoint.debit; charset=utf-8",
+       "CreditCert": "text/vnd.epoint.credit; charset=utf-8",
+       "BounceCert": "text/vnd.epoint.bounce; charset=utf-8",
 }
 
 // OpenPGP signed cleartext document representation
-type ClearSigned struct {
+type Signed struct {
+       // Sign and CleanSigned sets Hash for FormatSigned
+       // TODO: CreationDate
        Hash string
-       // Signed text (no dash escape, no trailing space)
+       // Signed text (no dash escape, no trailing space, \n new lines)
        Body []byte
        // Armored detached text signature of the Body
-       ArmoredSignature []byte
+       Signature []byte
 }
 
-type Field struct {
-       Key   string
-       Value string
+// parsed epoint document
+type Document struct {
+       Type   string
+       Fields map[string]string
+       Order  []string
+}
+
+var fieldtype = map[string]string{
+       "Amount":             "int",
+       "Authorized-By":      "id",
+       "Balance":            "int",
+       "Beneficiary":        "id",
+       "Date":               "date",
+       "Debit-Cert":         "id",
+       "Denomination":       "text",
+       "Difference":         "int",
+       "Draft":              "id",
+       "Drawer":             "id",
+       "Expiry-Date":        "date",
+       "Holder":             "id",
+       "Issuer":             "id",
+       "Last-Cert":          "id",
+       "Last-Credit-Serial": "int",
+       "Last-Debit-Serial":  "int",
+       "Maturity-Date":      "date",
+       "Nonce":              "id",
+       "Notes":              "text",
+       "References":         "ids",
+       "Serial":             "int",
+}
+
+var fieldname = map[string]string{
+       "AuthorizedBy":     "Authorized-By",
+       "DebitCert":        "Debit-Cert",
+       "ExpiryDate":       "Expiry-Date",
+       "LastCert":         "Last-Cert",
+       "LastCreditSerial": "Last-Credit-Serial",
+       "LastDebitSerial":  "Last-Debit-Serial",
+       "MaturityDate":     "Maturity-Date",
 }
 
-// Draft document represents an obligation transfer order
 type Draft struct {
-       Drawer       string `marshal:"id"` // ID of the payer (signer of the document)
-       Beneficiary  string `marshal:"id"` // ID of the payee
-       Amount       int64  // amount transfered
+       Drawer       string
+       Beneficiary  string
+       Amount       int64
        Denomination string
-       IssueDate    int64 `marshal:"date" key:"Issue-Date"`
-       // Draft is bounced before this date
-       MaturityDate int64  `marshal:"date" key:"Maturity-Date"`
-       Notes        string // Arbitrary text notes of the drawer
-       Nonce        string // unique number
-       Server       string `marshal:"id"` // ID of the server
-       Drawee       string `marshal:"id"` // ID of the obligation issuer
-       // useful if more strict date of issue information is needed
-       //References []string
-}
-
-// Obligation certificate after a transfer
-// References previous certificate (if any)
-// and the transfer related other documents
+       Issuer       string
+       AuthorizedBy string
+       MaturityDate *int64 // optional
+       ExpiryDate   *int64 // optional
+       Nonce        string
+       Notes        *string // optional
+}
+
+type Notice struct {
+       Date         int64
+       AuthorizedBy string
+       Notes        *string  // optional
+       References   []string // may be empty (startup notice)
+}
+
 type Cert struct {
-       Holder           string `marshal:"id"` // ID of the creditor
-       Serial           uint32 // serial number, number of certs of the holder
-       Date             int64  `marshal:"date"` // date of issue
-       Balance          int64  // current obligation value
+       IsDebit          bool
+       Holder           string
+       Serial           int64
+       Balance          int64
        Denomination     string
-       Server           string   `marshal:"id"`                 // ID of the server
-       Issuer           string   `marshal:"id"`                 // ID of the obligation issuer (drawee?)
-       LastDebitSerial  uint32   `key:"Last-Debit-Serial"`      // serial of the last draft cert or 0
-       LastCreditSerial uint32   `key:"Last-Credit-Serial"`     // serial of the last credit cert or 0
-       LastCert         string   `marshal:"id" key:"Last-Cert"` // ID of the previous cert if any
-       Difference       int64    // difference from previous balance
-       Draft            string   `marshal:"id"`                   // draft ID related to the transfer
-       Drawer           string   `marshal:"id"`                   // ID of the drawer in the transaction
-       DrawerSerial     uint32   `key:"Drawer-Serial"`            // serial of the drawer's related debit cert
-       DrawerCert       string   `marshal:"id" key:"Drawer-Cert"` // ID of the drawer's related debit cert
-       Notes            string   // Arbitrary text notes of the server (signer)
-       References       []string `marshal:"idlist"` // cert IDs for timestamping the system
-}
-
-func DecodeClearSigned(s []byte) (c *ClearSigned, err error) {
-       hash, body, sig := split(s)
-       if len(sig) == 0 {
-               err = fmt.Errorf("DecodeClearSigned could parse the signed document")
-               return
+       Issuer           string
+       Date             int64
+       Difference       int64
+       Draft            string
+       Beneficiary      *string // only in debit cert
+       Drawer           *string // only in credit cert
+       DebitCert        *string // only in credit cert
+       AuthorizedBy     string
+       Notes            *string // optional
+       LastDebitSerial  int64   // 0 if none
+       LastCreditSerial int64   // 0 if none
+       LastCert         *string // nil if serial == 1
+       References       []string
+}
+
+type DebitCert struct {
+       Holder           string
+       Serial           int64
+       Balance          int64
+       Denomination     string
+       Issuer           string
+       Date             int64
+       Difference       int64
+       Draft            string
+       Beneficiary      string
+       AuthorizedBy     string
+       Notes            *string // optional
+       LastDebitSerial  int64   // 0 if none
+       LastCreditSerial int64   // 0 if none
+       LastCert         *string // nil if serial == 1
+       References       []string
+}
+
+type CreditCert struct {
+       Holder           string
+       Serial           int64
+       Balance          int64
+       Denomination     string
+       Issuer           string
+       Date             int64
+       Difference       int64
+       Draft            string
+       Drawer           string
+       DebitCert        string
+       AuthorizedBy     string
+       Notes            *string // optional
+       LastDebitSerial  int64   // 0 if none
+       LastCreditSerial int64   // 0 if none
+       LastCert         *string // ? if serial == 1
+       References       []string
+}
+
+type BounceCert struct {
+       Drawer       string
+       Draft        string
+       LastCert     *string // optional
+       Balance      int64   // 0 if none
+       Date         int64
+       AuthorizedBy string
+       Notes        *string // optional
+       References   []string
+}
+
+func ToCert(v interface{}) (cert *Cert, err error) {
+       cert = new(Cert)
+       switch x := v.(type) {
+       case *DebitCert:
+               cert.IsDebit = true
+               cert.Beneficiary = &x.Beneficiary
+
+               cert.Holder = x.Holder
+               cert.Serial = x.Serial
+               cert.Balance = x.Balance
+               cert.Denomination = x.Denomination
+               cert.Issuer = x.Issuer
+               cert.Date = x.Date
+               cert.Difference = x.Difference
+               cert.Draft = x.Draft
+               cert.AuthorizedBy = x.AuthorizedBy
+               cert.Notes = x.Notes
+               cert.LastDebitSerial = x.LastDebitSerial
+               cert.LastCreditSerial = x.LastCreditSerial
+               cert.LastCert = x.LastCert
+               cert.References = x.References
+
+       case *CreditCert:
+               cert.IsDebit = false
+               cert.Drawer = &x.Drawer
+               cert.DebitCert = &x.DebitCert
+
+               cert.Holder = x.Holder
+               cert.Serial = x.Serial
+               cert.Balance = x.Balance
+               cert.Denomination = x.Denomination
+               cert.Issuer = x.Issuer
+               cert.Date = x.Date
+               cert.Difference = x.Difference
+               cert.Draft = x.Draft
+               cert.AuthorizedBy = x.AuthorizedBy
+               cert.Notes = x.Notes
+               cert.LastDebitSerial = x.LastDebitSerial
+               cert.LastCreditSerial = x.LastCreditSerial
+               cert.LastCert = x.LastCert
+               cert.References = x.References
+       default:
+               err = fmt.Errorf("ToCert: only debit or credit document can be converted to cert")
        }
-       c = &ClearSigned{string(hash), trimspace(dashunesc(body)), sig}
        return
 }
 
-func EncodeClearSigned(c *ClearSigned) (data []byte, err error) {
-       s := ClearSignedHeader
-       if c.Hash != "" {
-               s += "Hash: " + c.Hash + "\n"
+// sha1 sum of the (cleaned) document body as uppercase hex string
+func Id(c *Signed) string {
+       h := sha1.New()
+       h.Write(c.Body)
+       return fmt.Sprintf("%040X", h.Sum())
+}
+
+// parse an epoint document without checking the signature and format details
+func Parse(s []byte) (iv interface{}, c *Signed, err error) {
+       c, err = ParseSigned(s)
+       if err != nil {
+               return
        }
-       // TODO: check if space was trimmed from body before signature
-       s += "\n"
-       s += string(dashesc(c.Body))
-       s += "\n"
-       s += string(c.ArmoredSignature)
-       data = []byte(s)
+       doc, err := ParseDocument(c.Body)
+       if err != nil {
+               return
+       }
+       iv, err = ParseStruct(doc)
        return
 }
 
-func ParseFields(s []byte) (fields []Field, rest []byte, err error) {
-       rest = s
-       for len(rest) > 0 {
-               var line []byte
-               line, rest = getLine(rest)
-               // empty line after the parsed fields (consumed)
-               if len(line) == 0 {
-                       break
-               }
-               i := bytes.Index(line, []byte(": "))
-               if i < 0 {
-                       err = fmt.Errorf("ParseFields: missing ': '\n")
-                       break
-               }
-               fields = append(fields, Field{string(line[:i]), string(line[i+2:])})
+// format and sign an epoint document
+func Format(iv interface{}, key *openpgp.Entity) (s []byte, c *Signed, err error) {
+       doc, err := FormatStruct(iv)
+       if err != nil {
+               return
        }
+       body, err := FormatDocument(doc)
+       if err != nil {
+               return
+       }
+       c, err = Sign(body, key)
+       if err != nil {
+               return
+       }
+       s, err = FormatSigned(c)
        return
 }
 
-func ParseBody(s []byte) (doctype string, fields []Field, err error) {
-       // parse content type header first
-       fs, s, err := ParseFields(s)
+// verify an epoint document, return the cleaned version as well
+func Verify(c *Signed, key openpgp.KeyRing) (err error) {
+       msg := bytes.NewBuffer(c.Body)
+       sig := bytes.NewBuffer(c.Signature)
+       // TODO: verify signature
+       _, _ = msg, sig
+       //      _, err = openpgp.CheckArmoredDetachedSignature(key, msg, sig)
+       return
+}
+
+// sign body with given secret key
+func Sign(body []byte, key *openpgp.Entity) (c *Signed, err error) {
+       c = new(Signed)
+       c.Hash = "SHA256"
+       c.Body = body
+       w := new(bytes.Buffer)
+       err = openpgp.ArmoredDetachSignText(w, key, bytes.NewBuffer(c.Body))
        if err != nil {
                return
        }
-       if len(fs) != 1 || fs[0].Key != "Content-Type" {
-               return "", nil, fmt.Errorf("ParseBody: single Content-Type header was expected\n")
-       }
-       ctype := fs[0].Value
-       for k, v := range ContentType {
-               if ctype == v {
-                       doctype = k
-                       break
+       // close armored document with a \n
+       _, _ = w.Write([]byte{'\n'})
+       c.Signature = w.Bytes()
+       return
+}
+
+// split a clear signed document into body and armored signature
+func ParseSigned(s []byte) (c *Signed, err error) {
+       // look for clear signed header
+       for !bytes.HasPrefix(s, []byte(ClearSignedHeader)) {
+               _, s = getLine(s)
+               if len(s) == 0 {
+                       err = fmt.Errorf("ParseSigned: clear signed header is missing")
+                       return
                }
        }
-       if doctype == "" {
-               return "", nil, fmt.Errorf("ParseBody: unknown Content-Type: %s", ctype)
+       s = s[len(ClearSignedHeader):]
+       // end of line after the header
+       empty, s := getLine(s)
+       if len(empty) != 0 {
+               err = fmt.Errorf("ParseSigned: bad clear signed header")
+               return
+       }
+       // skip all hash headers, section 7.
+       for bytes.HasPrefix(s, []byte("Hash: ")) {
+               _, s = getLine(s)
        }
-       fields, s, err = ParseFields(s)
-       if err == nil && len(s) > 0 {
-               err = fmt.Errorf("ParseBody: extra data after fields: %q", s)
+       // skip empty line
+       empty, s = getLine(s)
+       if len(empty) != 0 {
+               err = fmt.Errorf("ParseSigned: expected an empty line after armor headers")
+               return
        }
+       lines := [][]byte{}
+       for !bytes.HasPrefix(s, []byte("-----BEGIN")) {
+               var line []byte
+               line, s = getLine(s)
+               // dash unescape, section 7.1.
+               if bytes.HasPrefix(line, []byte("- ")) {
+                       line = line[2:]
+               }
+               // empty values are not supported: "Key: \n"
+               lines = append(lines, bytes.TrimRight(line, " \t"))
+       }
+       c = new(Signed)
+       // last line is not closed by \n
+       c.Body = bytes.Join(lines, []byte("\n"))
+       // signature is just the rest of the input data
+       c.Signature = s
        return
 }
 
-func parse(s []byte, doctype string) (v interface{}, err error) {
-       t, fields, err := ParseBody(s)
+// clean up, check and reencode signature
+// used on drafts before calculating the signed document hash
+func CleanSigned(c *Signed) (err error) {
+       b, err := armor.Decode(bytes.NewBuffer(c.Signature))
        if err != nil {
                return
        }
-       if doctype != t && doctype != "" {
-               err = fmt.Errorf("parse: expected doctype %s; got %s", doctype, t)
+       if b.Type != openpgp.SignatureType {
+               err = fmt.Errorf("CleanSigned: invalid armored signature type")
                return
        }
-       switch t {
-       case "draft":
-               v = new(Draft)
-       case "cert":
-               v = new(Cert)
-       default:
-               err = fmt.Errorf("parse: unkown doc type: %s", t)
+       p, err := packet.Read(b.Body)
+       if err != nil {
                return
        }
-       err = unmarshal(fields, v)
-       return
-}
-
-// TODO: limit errors
-func render(v interface{}) ([]byte, error) {
-       doctype := ""
-       switch v.(type) {
-       case *Draft:
-               doctype = "draft"
-       case *Cert:
-               doctype = "cert"
+       sig, ok := p.(*packet.Signature)
+       if !ok {
+               err = fmt.Errorf("CleanSigned: invalid signature packet")
+               return
+       }
+       // section 5.2.3
+       if sig.SigType != packet.SigTypeText {
+               err = fmt.Errorf("CleanSigned: expected text signature")
+               return
+       }
+       switch sig.Hash {
+       case crypto.SHA1:
+               c.Hash = "SHA1"
+       case crypto.SHA256:
+               c.Hash = "SHA256"
        default:
-               panic("reder: unknown type")
+               err = fmt.Errorf("CleanSigned: expected SHA1 or SHA256 signature hash")
+               return
        }
-       s := "Content-Type: " + ContentType[doctype] + "\n\n"
-       fields := marshal(v)
-       for _, f := range fields {
-               s += f.Key + ": " + f.Value + "\n"
+       // TODO: check CreationTime and other subpackets
+       if sig.SigLifetimeSecs != nil && *sig.SigLifetimeSecs != 0 {
+               err = fmt.Errorf("CleanSigned: signature must not expire")
+               return
        }
-       return []byte(s), nil
-}
-
-func ParseDraft(s []byte) (draft *Draft, err error) {
-       v, err := parse(s, "draft")
+       out := new(bytes.Buffer)
+       w, err := armor.Encode(out, openpgp.SignatureType, nil)
+       if err != nil {
+               return
+       }
+       err = sig.Serialize(w)
+       if err != nil {
+               return
+       }
+       err = w.Close()
        if err != nil {
                return
        }
-       draft = v.(*Draft)
+       c.Signature = out.Bytes()
        return
 }
 
-func RenderDraft(draft *Draft) ([]byte, error) {
-       return render(draft)
+// create clear signed document
+func FormatSigned(c *Signed) (data []byte, err error) {
+       s := ClearSignedHeader + "\n"
+       if c.Hash != "" {
+               s += "Hash: " + c.Hash + "\n"
+       }
+       s += "\n"
+       s += string(c.Body)
+       s += "\n"
+       s += string(c.Signature)
+       data = []byte(s)
+       return
 }
 
-func ParseCert(s []byte) (cert *Cert, err error) {
-       v, err := parse(s, "cert")
+// parse type and fields of a document body
+func ParseDocument(body []byte) (doc *Document, err error) {
+       // parse content type header first
+       fields, s, err := ParseFields(body)
        if err != nil {
-               return nil, err
+               return
+       }
+       ctype, ok := fields["Content-Type"]
+       if len(fields) != 1 || !ok {
+               return nil, fmt.Errorf("ParseBody: expected a single Content-Type header field")
+       }
+       doc = new(Document)
+       for k, v := range ContentType {
+               if ctype == v {
+                       doc.Type = k
+                       break
+               }
+       }
+       if doc.Type == "" {
+               return nil, fmt.Errorf("ParseBody: unknown Content-Type: %s", ctype)
+       }
+       // TODO: doc.Order
+       doc.Fields, s, err = ParseFields(s)
+       if err == nil && len(s) > 0 {
+               err = fmt.Errorf("ParseBody: extra data after fields: %q", s)
        }
-       cert = v.(*Cert)
        return
 }
 
-func RenderCert(cert *Cert) ([]byte, error) {
-       return render(cert)
-}
-
-func formatId(s string) string {
-       return fmt.Sprintf("%040X", s)
-}
-
-func parseId(s string) (string, error) {
-       dst := make([]byte, 20)
-       if len(s) != 40 {
-               return "", fmt.Errorf("parseId: expected 40 characters; got %d", len(s))
+// create document body
+func FormatDocument(doc *Document) (body []byte, err error) {
+       ctype, ok := ContentType[doc.Type]
+       if !ok {
+               err = fmt.Errorf("FormatDocument: unknown document type: %s", doc.Type)
+               return
        }
-       _, err := hex.Decode(dst, []byte(s))
-       return string(dst), err
-}
-
-func quoteValue(s string) string {
-       return s
-}
-
-func unquoteValue(s string) (string, error) {
-       return s, nil
-}
-
-func formatDate(i int64) string {
-       return time.SecondsToUTC(i).Format(time.RFC3339)
-}
-
-func parseDate(s string) (int64, error) {
-       t, err := time.Parse(time.RFC3339, s)
-       if err != nil {
-               return 0, err
+       s := "Content-Type: " + ctype + "\n\n"
+       for _, k := range doc.Order {
+               s += k + ": " + doc.Fields[k] + "\n"
        }
-       return t.Seconds(), nil
+       return []byte(s), nil
 }
 
-func marshal(iv interface{}) []Field {
-       v := reflect.ValueOf(iv)
-       if v.Kind() != reflect.Ptr || v.IsNil() || v.Elem().Kind() != reflect.Struct {
-               panic("unmarshal: input is not a pointer to struct")
+// parse doc fields into a struct according to the document type
+func ParseStruct(doc *Document) (iv interface{}, err error) {
+       switch doc.Type {
+       case "Draft":
+               iv = new(Draft)
+       case "Notice":
+               iv = new(Notice)
+       case "DebitCert":
+               iv = new(DebitCert)
+       case "CreditCert":
+               iv = new(CreditCert)
+       case "BounceCert":
+               iv = new(BounceCert)
+       default:
+               err = fmt.Errorf("ParseStruct: unkown doc type: %s", doc.Type)
+               return
        }
-       v = v.Elem()
+       seen := make(map[string]bool)
+       v := reflect.ValueOf(iv).Elem()
        t := v.Type()
        n := v.NumField()
-       fields := []Field{}
+       nokey := 0
        for i := 0; i < n; i++ {
                ft := t.Field(i)
                fv := v.Field(i)
-               m := ft.Tag.Get("marshal")
-               k := ft.Tag.Get("key")
-               if k == "" {
-                       k = ft.Name
+               key := fieldname[ft.Name]
+               if key == "" {
+                       key = ft.Name
                }
-               val := ""
-               switch fv.Kind() {
-               case reflect.String:
-                       switch m {
-                       case "id":
-                               val = formatId(fv.String())
-                       case "":
-                               val = quoteValue(fv.String())
-                       default:
-                               panic("bad string field tag")
+               seen[key] = true
+               s, ok := doc.Fields[key]
+               if !ok {
+                       if fv.Kind() == reflect.Ptr {
+                               // missing optional key: leave the pointer as nil
+                               nokey++
+                               continue
                        }
-               case reflect.Int, reflect.Int32, reflect.Int64:
-                       switch m {
-                       case "date":
-                               val = formatDate(fv.Int())
-                       case "":
-                               val = strconv.Itoa64(fv.Int())
-                       default:
-                               panic("bad int field tag")
-                       }
-               case reflect.Uint, reflect.Uint32, reflect.Uint64:
-                       switch m {
-                       case "":
-                               val = strconv.Uitoa64(fv.Uint())
-                       default:
-                               panic("bad uint field tag")
+                       return nil, fmt.Errorf("ParseStruct: field %s of %s is missing\n", key, t.Name())
+               }
+               if fv.Kind() == reflect.Ptr {
+                       if s == "" || s == "-" {
+                               // TODO
+                               // empty optional key: same as missing
+                               continue
                        }
-               case reflect.Slice:
-                       switch m {
-                       case "idlist":
-                               if fv.Type().Elem().Kind() != reflect.String {
-                                       panic("only string slice is supported")
-                               }
-                               k := fv.Len()
-                               for j := 0; j < k; j++ {
-                                       if j > 0 {
-                                               val += " "
-                                       }
-                                       val += formatId(fv.Index(j).String())
+                       fv.Set(reflect.New(fv.Type().Elem()))
+                       fv = fv.Elem()
+               }
+               switch fieldtype[key] {
+               case "id":
+                       var val string
+                       val, err = parseId(s)
+                       fv.SetString(val)
+               case "text":
+                       var val string
+                       val, err = parseString(s)
+                       fv.SetString(val)
+               case "int":
+                       var val int64
+                       val, err = strconv.Atoi64(s)
+                       fv.SetInt(val)
+               case "date":
+                       var val int64
+                       val, err = parseDate(s)
+                       fv.SetInt(val)
+               case "ids":
+                       // TODO: empty slice?
+                       ids := strings.Split(s, " ")
+                       val := make([]string, len(ids))
+                       for j, id := range ids {
+                               val[j], err = parseId(id)
+                               if err != nil {
+                                       return
                                }
-                       default:
-                               panic("bad slice field tag")
                        }
+                       fv.Set(reflect.ValueOf(val))
                default:
-                       panic("bad field type")
+                       panic("bad field type " + key + " " + fieldtype[key])
+               }
+               if err != nil {
+                       return
+               }
+       }
+       if len(doc.Fields)+nokey != n {
+               for k := range doc.Fields {
+                       if !seen[k] {
+                               err = fmt.Errorf("ParseStruct: unknown field %s in %s", k, t.Name())
+                               return
+                       }
                }
-               fields = append(fields, Field{k, val})
        }
-       return fields
+       return
 }
 
-func unmarshal(fields []Field, iv interface{}) error {
+// turn a struct into a document
+func FormatStruct(iv interface{}) (doc *Document, err error) {
        v := reflect.ValueOf(iv)
        if v.Kind() != reflect.Ptr || v.IsNil() || v.Elem().Kind() != reflect.Struct {
-               panic("unmarshal: input is not a pointer to struct")
+               panic("input is not a pointer to struct")
        }
        v = v.Elem()
        t := v.Type()
        n := v.NumField()
-       if len(fields) != n {
-               return fmt.Errorf("unmarshal: %s has %d fields, got %d\n", t.Name(), n, len(fields))
-       }
+       doc = new(Document)
+       doc.Type = t.Name()
+       doc.Fields = make(map[string]string)
        for i := 0; i < n; i++ {
                ft := t.Field(i)
                fv := v.Field(i)
-               m := ft.Tag.Get("marshal")
-               k := ft.Tag.Get("key")
-               if k == "" {
-                       k = ft.Name
+               key := fieldname[ft.Name]
+               if key == "" {
+                       key = ft.Name
                }
-               if fields[i].Key != k {
-                       return fmt.Errorf("unmarshal: field %d of %s (%s) is missing\n", i, t.Name(), k)
-               }
-               s := fields[i].Value
-               var err error
-               switch fv.Kind() {
-               case reflect.String:
-                       var val string
-                       switch m {
-                       case "id":
-                               val, err = parseId(s)
-                       case "":
-                               val, err = unquoteValue(s)
-                       default:
-                               panic("bad string field tag")
-                       }
-                       fv.SetString(val)
-               case reflect.Int, reflect.Int32, reflect.Int64:
-                       var val int64
-                       switch m {
-                       case "date":
-                               val, err = parseDate(s)
-                       case "":
-                               val, err = strconv.Atoi64(s)
-                       default:
-                               panic("bad int field tag")
-                       }
-                       fv.SetInt(val)
-               case reflect.Uint, reflect.Uint32, reflect.Uint64:
-                       var val uint64
-                       switch m {
-                       case "":
-                               val, err = strconv.Atoui64(s)
-                       default:
-                               panic("bad uint field tag")
+               val := ""
+               if fv.Kind() == reflect.Ptr {
+                       if fv.IsNil() {
+                               // keep empty optional fields but mark them
+                               val = "-"
+                               goto setval
                        }
-                       fv.SetUint(val)
-               case reflect.Slice:
-                       var val []string
-                       switch m {
-                       case "idlist":
-                               if fv.Type().Elem().Kind() != reflect.String {
-                                       panic("only string slice is supported")
-                               }
-                               ids := strings.Split(s, " ")
-                               val = make([]string, len(ids))
-                               for j := range val {
-                                       val[j], err = parseId(ids[j])
-                                       if err != nil {
-                                               return err
-                                       }
+                       fv = fv.Elem()
+               }
+               switch fieldtype[key] {
+               case "id":
+                       val = formatId(fv.String())
+               case "text":
+                       val = formatString(fv.String())
+               case "int":
+                       val = strconv.Itoa64(fv.Int())
+               case "date":
+                       val = formatDate(fv.Int())
+               case "ids":
+                       k := fv.Len()
+                       for j := 0; j < k; j++ {
+                               if j > 0 {
+                                       val += "\n "
                                }
-                       default:
-                               panic("bad slice field tag")
+                               val += formatId(fv.Index(j).String())
                        }
-                       fv.Set(reflect.ValueOf(val))
                default:
-                       panic("bad field type")
-               }
-               if err != nil {
-                       return err
+                       panic("bad field type " + key + " " + fieldtype[key])
                }
+       setval:
+               doc.Fields[key] = val
+               doc.Order = append(doc.Order, key)
        }
-       return nil
+       return
 }
 
-func getLine(data []byte) (line, rest []byte) {
-       i := bytes.Index(data, []byte{'\n'})
-       j := i + 1
-       if i < 0 {
-               i = len(data)
-               j = i
-       } else if i > 0 && data[i-1] == '\r' {
-               i--
+func ParseFields(s []byte) (fields map[string]string, rest []byte, err error) {
+       rest = s
+       fields = make(map[string]string)
+       key := ""
+       // \n is optional after the last field and an extra \n is allowed as well
+       for len(rest) > 0 {
+               var line []byte
+               line, rest = getLine(rest)
+               // empty line after the last field is consumed
+               if len(line) == 0 {
+                       break
+               }
+               if line[0] == ' ' && key != "" {
+                       // "Key: v1\n v2\n" is equivalent to "Key: v1 v2\n"
+                       fields[key] += string(line)
+                       continue
+               }
+               if line[0] < 'A' || line[0] > 'Z' {
+                       err = fmt.Errorf("ParseFields: field name must start with an upper-case ascii letter")
+                       return
+               }
+               i := bytes.IndexByte(line, ':')
+               if i < 0 {
+                       err = fmt.Errorf("ParseFields: missing ':'")
+                       return
+               }
+               key = string(line[:i])
+               if _, ok := fields[key]; ok {
+                       err = fmt.Errorf("ParseFields: repeated fields are not allowed")
+                       return
+               }
+               fields[key] = string(line[i+1:])
        }
-       return data[:i], data[j:]
+       for key, v := range fields {
+               // either a single space follows ':' or the value is empty
+               // good: "Key:\n", "Key:\n value\n", "Key: value\n", "Key: v1\n v2\n"
+               // bad: "Key:value\n", "Key: \nvalue\n"
+               // bad but not checked here: "Key: \n", "Key: value \n", "Key:\n \n value\n"
+               if len(v) == 0 {
+                       continue
+               }
+               if v[0] != ' ' {
+                       err = fmt.Errorf("ParseFields: ':' is not followed by ' '")
+                       return
+               }
+               fields[key] = v[1:]
+       }
+       return
 }
 
-func trimspace(s []byte) []byte {
-       a := bytes.Split(s, []byte("\n"))
-       for i := range a {
-               a[i] = bytes.TrimRight(a[i], " \t\r")
-       }
-       return bytes.Join(a, []byte("\n"))
+// TODO: limit errors
+
+func parseId(s string) (string, error) {
+       // check if hex decodable
+       // TODO: length check
+       dst := make([]byte, len(s)/2)
+       _, err := hex.Decode(dst, []byte(s))
+       return s, err
 }
 
-func dashesc(s []byte) []byte {
-       r := bytes.Replace(s, []byte("\n-"), []byte("\n- -"), -1)
-       if len(r) > 0 && r[0] == '-' {
-               r = append([]byte("- "), r...)
-       }
-       return r
+func formatId(s string) string {
+       return s
 }
 
-func dashunesc(s []byte) []byte {
-       r := bytes.Replace(s, []byte("\n- "), []byte("\n"), -1)
-       if len(r) >= 2 && r[0] == '-' && r[1] == ' ' {
-               r = r[2:]
+func parseString(s string) (string, error) {
+       if len(s) > MaxValueLength {
+               return "", fmt.Errorf("parseString: length limit is exceeded")
        }
-       return r
+       return s, nil
 }
 
-// RFC 4880 is unclear about multiple Hash header semantics, section 7. says
-//  "One or more "Hash" Armor Headers"
-// then
-//  "If more than one message digest is used in the signature, the "Hash"
-//   armor header contains a comma-delimited list of used message digests."
-// in section 6.2.
-//  "there is no limit to the length of Armor Headers.  Care should
-//   be taken that the Armor Headers are short enough to survive
-//   transport.  One way to do this is to repeat an Armor Header key
-//   multiple times with different values for each so that no one line is
-//   overly long."
-// we accept a single Hash header with a list of hash algorithms for now
-// but use the one specified by the signature
+func formatString(s string) string {
+       return s
+}
 
-func split(s []byte) (hash, body, sig []byte) {
-       if !bytes.HasPrefix(s, []byte(ClearSignedHeader)) {
-               return
-       }
-       s = s[len(ClearSignedHeader):]
-       // only allow a single Hash: header
-       if bytes.HasPrefix(s, []byte("Hash: ")) {
-               s = s[len("Hash: "):]
-               hash, s = getLine(s)
-       }
-       // skip empty line
-       empty, s := getLine(s)
-       if len(empty) != 0 {
-               return
+func parseDate(s string) (int64, error) {
+       // TODO: fractional seconds?
+       t, err := time.Parse(time.RFC3339, s)
+       if err != nil {
+               return 0, err
        }
-       i := bytes.Index(s, []byte("\n-----BEGIN"))
+       return t.Seconds(), nil
+}
+
+func formatDate(i int64) string {
+       return time.SecondsToUTC(i).Format(time.RFC3339)
+}
+
+func getLine(data []byte) (line, rest []byte) {
+       i := bytes.IndexByte(data, '\n')
+       j := i + 1
        if i < 0 {
-               return
-       }
-       body, sig = s[:i], s[i+1:]
-       if i > 0 && body[i-1] == '\r' {
-               body = body[:i-1]
+               i = len(data)
+               j = i
+       } else if i > 0 && data[i-1] == '\r' {
+               i--
        }
-       return
+       return data[:i], data[j:]
 }