Version 1.3 - 05-Mar-2002 - by Juergen Helbing

Revisions

version release date author
v1.3 05-Mar-2002 Juergen Helbing
v1.2 28-Feb-2002 Juergen Helbing
v1.1 17-Feb-2002 Steve Blinch
v1.0 31-Jul-2001 Juergen Helbing

Introduction

This document describes a mechanism for encoding arbitrary binary information for transmission by electronic mail and newsgroups. Unlike similar encoding schemes, yEncode takes advantage of the entire 8-bit character set, rendering output only 1-2% larger than the original binary source.

Motivation

Existing mechanisms for transmission of binary information by electronic mail and newsgroups make use of only 7-bit ASCII text. The resulting encoded data are up to 40% larger than the original binary information.

yEncode intends to reduce the additional overhead of existing encoding schemes by taking advantage of the full 8-bit character set, which has become widely used and acceptable in Internet newsgroups. Special consideration is provided for specific reserved ASCII control characters to avoid interference with existing message transfer protocols.

The overhead of yEncoded binary data can be as little as 1-2%.

Encoding Principle

The encoding process represents each octet of input data with a single corresponding encoded output character. The ASCII value of each output character is derived by the following simple formula:

O = (I+42) % 256

That is, the output value is equal to the ASCII value of each input character plus 42, all modulo 256. This reduces overhead by reducing the number of NULL characters (ASCII 00) that would otherwise have had needed to be escaped, since many binaries contain a disproportionately large number of NULLs).

Under special circumstances, a single escape character (ASCII 3Dh, "=") is used to indicate that the following output character is "critical", and requires special handling.

Critical characters include the following:

These characters should always be escaped. Additionally, technique used to encode critical characters (described in the next section) provides for any character to be escaped; yDecoder implementations should be capable of decoding any character following an escape sequence.

The probability of occurance of these 4 characters in binary input data is approximately 0.4%. On average, escape sequences cause approximately 1.6% overhead when only these 4 characters are escaped.

The carriage return/linefeed overhead for every line depends on the developer-defined line length. Header and trailer lines are relatively small, and cause negligible impact on output size.

Careful writers of encoders will encode TAB (09h) SPACES (20h) if they would appear in the first or last column of a line. Implementors who write directly to a TCP stream will care about the doubling of dots in the first column - or also encode a DOT in the first column.

Encoding Technique

A typical encoding process might look something like this:

To facilitate transmission via existing standard protocols (most notably NNTP), carriage return/linefeed pairs should be written to the output stream after every n characters, where n is the desired line length. Typical values for n are 128 and 256. See additional experience information

If a critical character appears in the nth position of a line, both the escape character and the encoded critical character must be written to the same line, before the carriage return/linefeed. In this event, the actual number of characters in the line is equal to n+1. Effectively, this means that a line cannot end with an escape character, and that a line with n+1 characters must end with an encoded critical character.

Headers and Trailers

Similar to other binary encoding mechanisms, yEncode makes use of special keyword lines to mark the beginning and end of encoded data blocks. These blocks may be embedded in any standard 8-bit ASCII text file. yDecoder implementations must ignore any text outside the header/trailer blocks.

All keyword lines must begin with an escape character ('='), followed by an ASCII 79h ('y'). This '=y' combination uniquely identifies a line as a keyword line, since 'y' is not a valid encoded critical character. Header and trailer keyword lines always begin with an escape character, followed by a keyword indicating the line type, followed by any keywords appropriate for that particular line type. A typical header line should look similar to this:

=ybegin line=128 size=123456 name=mybinary.dat

Future versions of yEnc (if any) might use a different keyword than =ybegin. Perhaps "=ybegin2". Decoders should scan for "=ybegin " - with a SPACE behind =ybegin.

If the parameters "line=" "size=" "name=" are not present then the =ybegin might be part of a text-message with a discussion about yEnc. In such cases the decoder should assume that there is no binary.

Header lines must always begin with the "ybegin" keyword, and contain the typical line length, the size of the original unencoded binary (in bytes), and the name of the original binary file.

The filename must always be the last item on the header line. This ensures that all characters and character sequences may be included in the filename without interfering with other keywords. Although quotes (ASCII 22h, '"') are technically permitted, they are not recommended for use in filenames.

Leading and trailing spaces will be cut by decoders!
See additional experience information.
Implementors of decoders should be careful about the filename. It can contain non-US-ASCII-characters (80h-FFh), control-characters (01h..1Fh), and characters which conflict with the current platform: > / \ < | > : ? * @
It can be a very long parameter (up to 256 characters).

A typical trailer line should look similar to this:

=yend size=123456

Trailer lines must always begin with the "yend" keyword, and must contain the size of the original unencoded binary (in bytes).

The size of the original binary must be repeated in the trailer for redundancy checking. yDecoder implementations should compare the header size value with both the trailer size value and the actual size of the resulting decoded binary. If any of these three values differ then the attachment is corrupt, and a warning must be issued; the resulting decoded binary must be discarded. See additional experience information

Verifying Integrity

yEncoded documents may also include a 32-bit Cyclic Redundancy Check (CRC) value, to assist in verifying the integrity of the encoded binary data. A CRC32 value, if present, should be included as a "crc32" keyword in the trailer line. Such a trailer line might look similar to this:

=yend size=123456 crc32=abcdef12

It should be noted that CRC32 values are not mandatory, but should, if possible, be processed if present.

See additional experience information

Sample yEncoded File Part

The following is an excerpt from an actual yEncoded file block:

=ybegin line=128 size=111401 name=al_larsonbw030_ball.jpg
)_)=J*:tpsp*++++V+V**)_*m*0./0/.00/011024:44334>896:A>
....
....
....
....
R̴R̴R̴R̴R̴R̴R̴R̴R̴R̴R̴R̴R̴R̴R̴R̴R̴R̴R̴R̴R̴R̴R̴R̴R̴R̴R̴
´R̴R̴R̴R̴R̴R̴R̴R̴R̴R̴R̴R̴Rͩ)_
=yend size=111401

Complete yEncoded file samples are also available at www.yenc.org.

Multi-part Encoded Binaries

It is frequently desirable to split large binary files into multiple parts for transmissio n over the Internet. Such binaries are often rendered unusable by missing parts and/or data corruption.

To address these problems, yEncode defines an additional keyword line, "ypart", and several additional keywords to handle multipart binaries. Each individual file part begins with a standard "ybegin" header line, but an additional keyword, "part", is added to specify the part number and identify the file as a multipart binary.

When the "part" keyword is included in a header line, the following line must be a "ypart" keyword line which specifies information about the part. The "ypart" keyword line requires a "begin" and "end" keyword; these specify the starting and ending points, in bytes, of the block in the original file.

The file part must end with a slightly modified "ypart" trailer line. An additional keyword, "part", is added to specify the part number. This part number must match the part number found in the header line.

An additional keyword "total" should be also added. This total number must match the total number of parts found in the header line. First implementation of yEnc do NOT include this parameter.

The trailer line must also contain a "pcrc32" keyword representing the CRC32 of the preceeding encoded part. As always, it is also desirable (but not required) to include a "crc32" keyword representing the CRC32 of the entire encoded binary.

Unlike single-part yEncoded documents, the "size" keyword in the trailer lines of multipart encoded binaries must represent the size of the file part, not the size of the entire file. To verify integrity, a decoder implementation must recompute the expected part size from the "begin" and "end" keyword values in the "ypart" line. If the expected part size differs from the part size specified in the "yend" line, the file is corrupt.

A sample multipart encoded binary might look similar to this:

=ybegin part=1 line=128 size=500000 name=mybinary.dat
....
....
....
=yend size=100000 part=1 pcrc32=abcdef12

=ybegin part=1 total=10 line=128 size=500000 name=mybinary.dat
=ypart begin=1 end=100000
....
....
....
=yend size=100000 part=1 pcrc32=abcdef12

=ybegin part=5 line=128 size=500000 name=mybinary.dat
=ypart begin=400001 end=500000
....
....
....
=yend size=100000 part=10 pcrc32=12a45c78 crc32=abcdef12

It should be noted that if a decoder does not implement multipart support, or fails to detect a multipart encoded binary, then it will not successfully decode the individual file parts because the "size" keyword in the "ybegin" line will differ from the "size" keyword in the "yend" line.

Multipart binaries are usually quite sensitive to corruption. Transferring hundreds of megabytes in vain, simply because a corrupt part cannot be identified is a significant waste of bandwidth.

Using the "begin" and "end" keywords, yEncode allows decoders to identify the position of an individual part in a larger file, which allows parts to be combined from several different sources regardless of the part size. This feature is unique to yEncode, and is very easy to include in an encoder implementation.

Subject Line Conventions

Standard single-part yEncoded binaries require no special conventions for the subject line. It is recommended, however, that yEncoded binaries be specifically identified as such, until the yEncode encoding format becomes more widely implemented. The suggested format for subject lines for single-part binaries is:

[Comment1] "filename" 12345 yEnc bytes [Comment2]

[Comment1] and [Comment2] are optional. The filename should always be enclosed in quotes; this allows for easy detection, even when the filename includes spaces or other special characters. The word "yEnc" should be placed in between the file size and the word "bytes".

see additional experience information. Placing the word "yEnc" between filename+bytes or bytes+comment2 is acceptable.

Multi-part archives should always be identified as such. As with single-part binaries, they should also be identified as yEncoded until yEncoding becomes more mainstream.

The (strongly) recommended format for subject lines for multi-part binaries is:

[Comment1] "filename" yEnc (partnum/numparts) [size] [Comment2]

Again, [Comment1] and [Comment2] are optional. The [size] value is also optional here. The filename must be included, in quotes. The keyword "yEnc" is mandatory, and must appear between the filename and the size (or Comment2, if size is omitted). Future revisions of the draft may specify additional information may be inserted between the "yEnc" keyword and the opening parenthesis of the part number.

see additional experience information. Placing the word "yEnc" between (#/#)+size or size+comment2 is acceptable.

Handling of corrupt messages

Decoders should use error-detection whenever possible. The user should be notified about corrupt messages. If warnings are disabled then it is strongly recommended to store binaries with an error-text in the filename. Examples:

picture(size-error).jpg
homemovie(crc32-error).avi
document(line-error).rtf
longmusic(missing-parts).mp3

It is acceptable to store also corrupt binaries (some might be even partially usable). But it is _not_ acceptable to hide detected errors from the user entirely.

yEnc has the design target to _detect_ corruption. Advanced newsreaders might fetch corrupt messages even from other sources.

Protection and Copyright

The yEncode encoding method is released into the public domain. Everyone is permitted to copy it, to use it, and to implement it.

Neither this document nor the yEncode encoding method may be patented, protected, or restricted in any way. Everyone should benefit from it, and its predecessors.

This document may be freely distributed, as long as credit remains with the original author(s). Do not claim that it's your own work!

Public domain example software is also available at www.yenc.org.

Credits

This document has been created based on my [Juergen Helbing] own personal experience, and help and input from a few Usenet activists. Thanks to: Jeremy Nixon, Curt Welch, Ed, Andrew, Stuart, JBerg, Marco d'Itri, The Meowbot, Jan Ingvoldstat, The UseFor taskforce. All others: please remind me!

Draft revised (02/17/02) by Steve Blinch
Draft extended (02/28/02) by Juergen Helbing

Conclusion

This is an informal proposal, not an RFC. Your input is greatly appreciated. The author is just a poor programmer - with a few years of binary experience.

Thanks for reading.

Juergen Helbing (yenc@infostar.de)

Changes from 1.1 -> 1.2

Changes from 1.2 -> 1.3