mox/message/writer.go
Mechiel Lukkien 361bc2b516
when accepting an incoming message, turn any bare newlines (without carriage return) into crlf
because that is what most of the code expects. we could work around having bare
lf, but it would complicate too much code.

currently, a message with bare lf is accepted (in smtpserver delivery,
imapserver append, etc), but when an imap session would try to fetch parsed
parts, that would fail because and even cause a imapserver panic (closing the
connection).

in message imports we would already convert bare lf to crlf (because it is
expected those messages are all lf-only-ending).

we store messages with crlf-ending instead of lf-ending so the imapserver has
all correct information at hand (line counts, byte counts).

found by using emclient with mox. it adds a message to the inbox that can have
mixed crlf and bare lf line endings in a few header fields (in some
localization, emclient authors explained how that happened, thanks!).  we can
now convert those lines and read those messages over imap. emclient already
switched to all-crlf line endings in newer (development) versions.
2023-11-21 13:19:54 +01:00

100 lines
2 KiB
Go

package message
import (
"io"
)
// Writer is a write-through helper, collecting properties about the written
// message and replacing bare \n line endings with \r\n.
type Writer struct {
writer io.Writer
HaveBody bool // Body is optional. ../rfc/5322:343
Has8bit bool // Whether a byte with the high/8bit has been read. So whether this is 8BITMIME instead of 7BIT.
Size int64
tail [3]byte // For detecting header/body-separating crlf.
// todo: should be parsing headers here, as we go
}
func NewWriter(w io.Writer) *Writer {
// Pretend we already saw \r\n, for handling empty header.
return &Writer{writer: w, tail: [3]byte{0, '\r', '\n'}}
}
// Write implements io.Writer.
func (w *Writer) Write(buf []byte) (int, error) {
origtail := w.tail
if !w.HaveBody && len(buf) > 0 {
get := func(i int) byte {
if i < 0 {
return w.tail[3+i]
}
return buf[i]
}
for i, b := range buf {
if b == '\n' && (get(i-1) == '\n' || get(i-1) == '\r' && get(i-2) == '\n') {
w.HaveBody = true
break
}
}
n := len(buf)
if n > 3 {
n = 3
}
copy(w.tail[:], w.tail[n:])
copy(w.tail[3-n:], buf[len(buf)-n:])
}
if !w.Has8bit {
for _, b := range buf {
if b&0x80 != 0 {
w.Has8bit = true
break
}
}
}
wrote := 0
o := 0
Top:
for o < len(buf) {
for i := o; i < len(buf); i++ {
if buf[i] == '\n' && (i > 0 && buf[i-1] != '\r' || i == 0 && origtail[2] != '\r') {
// Write buffer leading up to missing \r.
if i > o+1 {
n, err := w.writer.Write(buf[o:i])
if n > 0 {
wrote += n
w.Size += int64(n)
}
if err != nil {
return wrote, err
}
}
n, err := w.writer.Write([]byte{'\r', '\n'})
if n == 2 {
wrote += 1 // For only the newline.
w.Size += int64(2)
}
if err != nil {
return wrote, err
}
o = i + 1
continue Top
}
}
n, err := w.writer.Write(buf[o:])
if n > 0 {
wrote += n
w.Size += int64(n)
}
if err != nil {
return wrote, err
}
break
}
return wrote, nil
}