imf-codec

Work in progress, do not use in production This is currently only a decoder (parser), encoding is not supported.

Goals

Correctness: do no deviate from the RFC, support edge and obsolete cases
Straightforward/maintainable: implement the RFCs as close as possible, minimizing the amount of clever tricks and optimizations
Multiple syntax: Write the parser so it's easy to alternate between the strict and obsolete/compatible syntax
Never fail: Provide as many fallbacks as possible

Non goals

Parsing optimization (greedy parser, etc.) as it would require to significantly deviate from the RFC ABNF syntax (would consider this case if we could prove that the transformation we make are equivalent)
Pipelining/streaming/buffering as the parser can arbitrarily backtrack + our result contains reference to the whole buffer, imf-codec must keep the whole buffer in memory. Avoiding the sequential approach would certainly speed-up a little bit the parsing, but it's too much work to implement currently.
Zerocopy. It might be implementable in the future, but to quickly bootstrap this project, I avoided it for now.

Missing / known bugs

Current known limitations/bugs:

Resent Header Fields are not implemented
Return-Path/Received headers might be hard to use as their order is important, and it's currently lost in the final datastructure.
Datetime parsing of invalid date might return None instead of falling back to the bad_body field
Comments are dropped

Design

Based on nom, a parser combinator lib in Rust. multipass parser

extract header block: &[u8] (find \r\n\r\n OR \n\n OR \r\r OR \r\n)
decode/convert it with chardet + encoding_rs to support latin-1: Cow<&str>
extract header lines iter::&str (requires only to search for FWS + obs_CRLF)
extract header names iter::Name::From(&str)
extract header body iter::Body::From(Vec)
extract header section Section

recovery

based on multipass, equivalent to sentinel / synchronization tokens

Testing strategy

imf-codec aims to be as much tested as possible against reald

Unit testing: parser combinator independently (done)

Selected full emails (expected)

Existing datasets

Enron 500k - Took 20 minutes to parse ~517k emails and check that RFC5322 headers (From, To, Cc, etc.) are correctly parsed. From this list, we had to exclude ~50 emails on which the From/To/Cc fields were simply completely wrong, but while some fields failed to parse, the parser did not crash and parsed the other fields of the email correctly.

Planned: jpbush, my inbox, etc.

Fuzzing (expected)

Across reference IMAP servers (dovevot, cyrus) (expected)

Development status

Early development. Not ready. Do not use it in production or any software at all.

Todo:

test over the enron dataset
convert to multipass parser
implement mime part 3 (encoded headers)
implement mime part 1 (new headers)
review part 2 (media types) and part 4 (registration procedure) but might be out of scope
implement some targeted testing as part of mime part 5
implement fuzzing through cargo fuzz
test over other datasets (jpbush, ml, my inbox)
backport to aerogramme
fix warnings, put examples, document the public API a little bit

Targeted RFC

🚩	#	Name
🟩	822	ARPA INTERNET TEXT MESSAGES
🟩	2822	Internet Message Format (2001)
🟩	5322	Internet Message Format (2008)
🔴	2045	↳ Multipurpose Internet Mail Extensions (MIME) Part One: Format of Internet Message Bodies
🔴	2046	↳ Multipurpose Internet Mail Extensions (MIME) Part Two: Media Types
🔴	2047	↳ MIME (Multipurpose Internet Mail Extensions) Part Three: Message Header Extensions for Non-ASCII Text
🔴	2048	↳ Multipurpose Internet Mail Extensions (MIME) Part Four: Registration Procedures
🔴	2049	↳ Multipurpose Internet Mail Extensions (MIME) Part Five: Conformance Criteria and Examples
🟩	6532	Internationalized Email Headers
🔴	9228	Delivered-To Email Header Field

Alternatives

stalwartlab/mail_parser

4.1 KiB Raw Blame History