threadweave

The canonical JWZ email message-threading algorithm for Python — pure stdlib, zero runtime dependencies.

threadweave turns a flat list of email messages into conversation trees using Jamie Zawinski's threading algorithm, built on top of RFC 5322 §3.6.4 identification-field parsing (Message-ID, References, In-Reply-To). It is the algorithm every serious mail client uses to group a mailbox into threads.

It exists because the obviously-wrong approaches — grouping by subject alone, or naively chaining In-Reply-To — mis-thread real mail: subjects collide across unrelated conversations, references arrive out of order, roots go missing, and malformed headers form reference loops. JWZ handles all of these; threadweave implements it faithfully and loop-safely.

What it does

• Builds the JWZ id-table of containers and links References chains without creating loops and without overriding a message's good existing parent.
• Recovers missing roots as empty placeholder containers, then prunes empty containers correctly (nuking childless empties and splice-promoting the children of empty ones, with the special root-level single-child handling).
• Optionally groups the root set by base subject (Re:/Fwd:/Fw: stripped) — a heuristic, off by default.
• Terminates on adversarial input: self-references and mutual reference cycles never loop or crash.

Install

pip install threadweave

No runtime dependencies — only the standard library (re, hashlib, dataclasses, typing).

Quickstart

from threadweave import Message, thread_messages

threads = thread_messages([
    Message(message_id="a", subject="Deploy plan"),
    Message(message_id="b", in_reply_to="a", references=["a"], subject="Re: Deploy plan"),
    Message(message_id="c", references=["a", "b"], subject="Re: Deploy plan"),
])

assert len(threads) == 1            # one conversation
root = threads[0]                   # a Container
print(root.message.message_id)      # "a"
for node in root.iter_descendants():
    print(node.message.message_id)  # "b", then "c"

Group unrelated-but-same-subject roots when references are missing:

threads = thread_messages(messages, group_by_subject=True)

API

Container.iter_descendants and the internal linking are loop-safe: traversal visits each node at most once, so a cyclic reference graph can never hang.

One source, multi use (OSMU)

The RFC 5322 header primitives in threadweave/headers.py are extracted behaviour-preserving from a production control plane (naruon), where they normalize Message-ID/References headers for canonical email threading. The JWZ assembly here is a fresh canonical implementation built on top of those primitives — one source, usable both as a standalone dependency and as a git submodule.

Research grounding

See docs/research: Zawinski's threading algorithm (https://www.jwz.org/doc/threading.html) and RFC 5322 §3.6.4 (Identification Fields). Base-subject grouping is documented there as a heuristic fallback.

License

Apache-2.0. See LICENSE.