Roc Zulip Daily Summary — 2026-03-05

  • Realm: https://roc.zulipchat.com
  • Window: 2026-03-05 12:00 AM to 2026-03-06 12:00 AM (America/Los_Angeles)
  • Messages: 24
  • LLM: vertex_open_model model=zai-org/glm-5-maas topics=3 ok=3/3

Top threads

List concatMany and insert

Summary

The discussion explores the potential value of implementing persistent data structures in Roc, specifically considering Roc’s support for mutation of singly-held data, before shifting to details about the standard Dict type and the future Hash ability.

Key points

  • Persistent data structures (like HAMT) in userspace could still be useful in Roc for specific use cases, such as history tracking or snapshotting, where structural sharing is beneficial.
  • Steve Howell shared a past experience learning data structures in Elm, noting that userspace implementations can be highly educational for the community.
  • The standard Roc Dict currently uses an indexmap (a flat hash map that remembers insertion order) and will retain this behavior, likely moving to a Zig implementation.
  • The Hash ability will be implementable in userspace and will support compiler magic for auto-derivation, though manual implementation will be possible.

Open questions

  • None explicitly open; the participant asking about Dict availability received confirmation that it will be ported with its current behavior.

Action items

Key links

Datastar

Summary

Steve Howell argues that modern tech stacks and AI assistance allow for replicating core functionality of mature systems like Zulip in a fraction of the original development time, suggesting a shift toward custom clients consuming server data via technologies like Datastar.

Key points

  • While a full Zulip rewrite in hours is unrealistic, replicating its core value (permissions, message distribution, topics) could be done in about a month using better modern tools.
  • Zulip’s current performance challenges at scale stem from managing long-polling HTTP connections in Python, which could potentially be alleviated by Datastar’s SSE model.
  • The conversation anticipates a paradigm shift where users run personalized, AI-generated custom clients to interact with standardized server data.

Open questions

  • None identified.

Action items

compile time evaluation

Summary

Luke Boswell proposes refactoring the interpreter to baseline off the MIR or LIR in the future, which would allow support for any target Zig supports.

Key points

  • The refactoring would occur when MIR (and potentially LIR) land.
  • The proposal suggests moving the baseline from the current state to MIR or possibly LIR.
  • This change would enable support for “exotic” targets supported by Zig.

Open questions

  • Should the interpreter be based on the MIR or the LIR?

Action items