Epic: Linux Delivery, Distro Proof, and Remaining Parity Gaps

[Jesssullivan]

Linux Delivery, Distro Proof, And Remaining Parity Gaps

This issue is the current umbrella for the Linux program in Jesssullivan/cmux.

The old "Linux MVP roadmap" framing is stale. cmux-linux is already broad and real; the remaining work is now mostly validation, packaging proof, and a smaller set of explicit parity gaps.

Current reality as of 2026-04-21

• cmux-linux exists in-tree and is materially implemented
• hosted Linux CI is already green across the container/build matrix
• Linux release packaging exists, including signed artifacts and multi-arch release builds
• release-gated distro proof now executes real .sandboxed VM runs for Ubuntu 24.04, Debian 12, and Fedora 42
• the branch now also wires a dedicated Rocky 10 terminal-first RPM and distro-rocky10 validation path
• the biggest remaining gaps are first green CI evidence for the new Fedora/Rocky lanes, direct QA promotion, and a few explicit capability gaps

Active lanes

1. Distro proof and packaging confidence

• #209 Fedora 42 fresh-install VM proof and first green CI evidence
• #187 Rocky 10 terminal-first lane proof and retirement of the Rocky 9 proxy
• keep Ubuntu 24.04 and Debian 12 release-gated validation green

2. Explicit parity gaps

• #206 complete the Linux WebAuthn / FIDO2 bridge
• #216 expand Linux tests_v2 socket coverage beyond the current stable baseline

3. Remote/fleet follow-up

• #201 Tailnet-direct cmuxd-remote listener mode
• valid future lane, but not a distro-shipping blocker

4. Naming RFC

• #76 stays explicitly non-blocking until the distro-proof lane is in a better state

What is no longer accurate

Do not use this issue as evidence that Linux is still at the architecture or research stage. That is no longer true.

The blocker is not "can Linux cmux exist?" The blocker is "can the fork prove the right capability level on the right distros without overstating what is finished?"

Exit criteria for this umbrella

• Fedora 42 has recorded green fresh-install proof
• Rocky 10 has recorded green fresh-install proof and Rocky 9 is retired as the standing proxy
• Linux WebAuthn bridge is either implemented or deliberately deferred with rationale
• Linux socket/API coverage is meaningfully broader than the current baseline allowlist
• the public tracker matches repo reality and no longer describes Linux as an early MVP concept

References

• docs/program-status.md
• docs/distro-testing-readiness-plan.md
• docs/linux-parity-matrix.md
• docs/linux-validation-checklist.md
• docs/linear-qa-shard-punchlist.md

[Jesssullivan]

M5a Dependency Graph & Execution Order

Phase 0 (Week 1) — Foundation [parallel]
├── #85 Nix flake dev environment (GTK4 deps, VM apps, QEMU testing)
└── #84 Linux CI matrix (ubuntu/fedora/arch containers)

Phase 1 (Weeks 2-4) — Architecture [sequential]
└── #77 cmux_gtk apprt variant (minimal: single terminal window renders)

Phase 2 (Weeks 4-6) — UI Shell [parallel after #77]
├── #78 CmuxWindow + AdwTabView tabs
├── #79 SplitTree integration
└── #81 Socket command handler (Zig port) [can start parallel]

Phase 3 (Weeks 5-7) — Features [parallel after #77]
├── #80 Workspace model + sidebar (depends on #78)
├── #82 Config parser (Zig port)
└── #83 Session persistence (depends on #82)

Critical Path

#85 → #77 → #78 → #80 (Nix flake → apprt → tabs → workspaces)

Parallel Tracks

• CI track: Set up Linux CI matrix #84 + Nix flake for Linux dev environment #85 (week 1)
• UI track: Implement CmuxWindow with AdwTabView tabs #78 + Implement split pane management via SplitTree #79 (weeks 4-6, after Build cmux-linux GTK4 application linking to libghostty #77)
• Backend track: Port socket command handler to Zig #81 + Port configuration parser to Zig #82 + Port session persistence to Zig #83 (weeks 5-7, after Build cmux-linux GTK4 application linking to libghostty #77)

[Jesssullivan]

M5b Browser Panel sprint complete — all issues closed:

• Port WebAuthn JS bridge to WebKitGTK #87 WebAuthn JS bridge (PR M5b Sprint 2: WebAuthn JS bridge for FIDO2/YubiKey (#87) #125)
• Port cookie management to WebKitGTK #88 Cookie management (PR M5b Sprint 3: Cookie persistence (#88) #128)
• Port devtools + find-in-page to WebKitGTK #89 DevTools + find-in-page (PR M5b Sprint 4+5: DevTools, find-in-page, media permissions (#89, #90) #127)
• Port media capture permissions to WebKitGTK #90 Media capture permissions (PR M5b Sprint 4+5: DevTools, find-in-page, media permissions (#89, #90) #127)

All PRs merged to main as of 2026-03-30.

[Jesssullivan]

Status refresh as of April 13, 2026.

This epic no longer reads accurately as a greenfield Linux architecture note. Current repo state is a Linux delivery and validation program:

• cmux-linux is materially implemented in-tree
• branch sid/linux-program-hygiene now has green Linux CI lanes for Ubuntu 24.04, Fedora 42, Debian 12 baseline, Rocky 10 terminal-first, and Arch Linux on run 24360116285
• the main remaining work is parity classification, distro validation, packaging proof, and tracker hygiene

Current checked-in execution docs:

• docs/program-status.md
• docs/component-portfolio.md
• docs/linux-program-plan.md
• docs/linux-parity-matrix.md
• docs/linux-validation-checklist.md

Critical-path interpretation:

1. validate Ubuntu 24.04 and Fedora 42 as the broad-feature distros
2. validate Debian 12 as the package/runtime baseline and record browser status explicitly
3. validate Rocky 10 as terminal-first
4. keep naming and upstream-prep explicitly non-blocking

[Jesssullivan]

This epic is now better treated as the Linux program umbrella than as a
greenfield architecture note.

Current state:

• #205 is merged
• hosted Linux CI is green across Ubuntu 24.04, Fedora 42, Debian 12 baseline,
  Rocky 10, Arch, and the vendor-library lane
• fresh-install KVM package validation is real and release-gating
• the remaining blocker is proof and parity truth, not initial Linux bring-up

New dedicated parity issues are now split out:

• #206 Linux WebAuthn bridge completion
• #207 Linux socket/control-plane parity
• #208 Linux restore/headless runtime support

Current execution docs:

• docs/program-review-2026-04-14.md
• docs/linux-work-week-2026-04-14.md
• docs/linear-qa-shard-punchlist.md

Critical-path interpretation:

1. direct Tier A QA on Ubuntu 24.04 and Fedora 42
2. explicit Debian 12 browser status
3. explicit Rocky 10 terminal-first proof
4. Fedora 42 / Rocky 10 fresh-install VM strategy beyond the current
   nix-vm-test ceiling

[Jesssullivan]

Program update — 2026-04-25

Near-term Linux work should stay on testing, packaging, and QA proof rather than opening new architecture lanes.

Current cadence proposal:

1. re-pin nix-vm-test to upstream now that numtide/nix-vm-test#172 has merged
2. keep Fedora 42 and Rocky 10 VM proof visible on #209 and #187
3. finish signed-package install docs and the Flathub/AUR/COPR handoff runbooks
4. recruit QA with explicit tiers: Ubuntu/Fedora broad-feature, Debian baseline, Rocky terminal-first, plus early Arch/Mint/NixOS reports
5. stage Ghostty/Bonsplit/Manaflow upstreamables only as owned handoff packets for Jess to submit manually

Docs updated in docs/program-status.md, docs/linux-packaging-cd-plan.md, docs/distro-testing-readiness-plan.md, docs/linear-qa-shard-punchlist.md, and docs/upstream-candidate-ledger.md.

[Jesssullivan]

2026-04-25 reconciliation note:

Current owned-surface state:

• Root main is clean at 6eb26eb7; no open Jesssullivan/cmux PRs.
• Broad packaging/toolchain WIP is parked on codex/park-linux-packaging-wip at 208f2638; no PR opened yet.
• Active proof issues remain ci(distro): establish Fedora 42 fresh-install VM proof #209, ci(distro): establish Rocky 10 fresh-install proof and retire the Rocky 9 proxy #187, tests_v2: expand Linux socket-test coverage beyond the current stable baseline #216, and linux: complete WebAuthn bridge handling #206.
• audit: fork landscape and human-gated upstream preparation #199 is closed and should stay archival; future upstream work should go through owned handoff docs/issues first, with Jess handling non-owned submissions manually.
• Linear now has TIN-621 for the active Zig vendor productionization parent-pin worktree.

The next Linux delivery decision is not naming. It is whether to turn the parked packaging WIP into a reviewed PR now, or first land the smaller Zig vendor pin PR.