xeiaso.net / ircd.rs
Proof of concept mechanical port of ircnet/ircd to Rust as part of a bit about C being insecure for network services
0
Fork 0

Configure Feed

Select the types of activity you want to include in your feed.

ircd.rs / docs / claude-memory / leveva-linelen-2048-graphemes.md
2.6 kB

name: leveva-linelen-2048-graphemes description: "leveva supports 2048-grapheme (not 512-byte) non-tag lines, advertised as LINELEN=2048; where the cap lives + the framer ceiling formula (P11 slice 121)" metadata: node_type: memory type: project originSessionId: 5cc0cd61-d1bb-40b1-8777-3238c363db78#

Maintainer directive (2026-06-12): leveva MUST support IRC lines whose non-tag portion can be up to 2048 graphemes (UAX #29 clusters, not bytes — a 2048-char multi-byte line of accented Latin / CJK / emoji is fully supported), and advertise LINELEN=2048 in ISUPPORT.

P11 slice 121 implemented it. Key facts (so you don't re-derive or break them):

  • The cap is a count of grapheme clusters, measured with unicode-segmentation. Constant leveva::message::LINELEN = 2048. Enforcement point = Message::to_wire (output): it truncates the assembled non-tag body at the start of grapheme index LINELEN (body.grapheme_indices(true).nth(LINELEN)...), never splitting a cluster. The old 510-byte (MAX_CONTENT_LEN) cut is gone from to_wire.
  • MAX_MESSAGE_LEN (512) / MAX_CONTENT_LEN (510) still exist but now mean "the RFC floor / the chunker budget" — command::ison, command::monitor, and s2s::burst still pack to MAX_CONTENT_LEN, deliberately staying RFC-sized so a linked peer never sees an over-512 line (no S2S protocol drift). Don't "modernize" those to LINELEN.
  • Framer ceiling: session::MAX_INBOUND_LINE = MAX_TAG_DATA_LEN + MAX_BODY_BYTES where MAX_BODY_BYTES = LINELEN * 64 (= 139263, was 8703 = 8191+512). 64 bytes/grapheme is the DoS backstop allowance (covers ZWJ emoji families ≈25 B, flags 8 B); a line past this still gets 417 ERR_INPUTTOOLONG + resync (the slice-118 recvQ guard, just moved). If you change LINELEN, this and golden_recvq's blob size (must exceed 139263) move with it.
  • Divergence: LINELEN is advertised as 2048; the internal limit is 2048 graphemes. Since grapheme count ≤ byte count, a client respecting LINELEN=2048 bytes is always within budget — a safe lower bound. leveva-native, no oracle differential (C oracle truncates at 512 B, advertises no LINELEN).

Tested by message/isupport/session units + leveva/tests/linelen_proptest.rs (headline: to_wire body == full body truncated to LINELEN graphemes). Recording discipline: [[leveva-slice-recording]]. Related: [[leveva-utf8-only-parse-boundary]] (the other client-frame message-boundary hardening), [[leveva-s2s-port-based-routing]].