plan(v0.39): the rowan fork has a measured expiry, not a permanent one (REQ-350) - #932
Merged
Conversation
…e (REQ-350) The maintainer closed our rust-analyzer/rowan#211 on 2026-09-07 with "We're about to rewrite rowan anyway and the new code should pass Miri (and is also a change of pretty much all code)". That turns the fork from a permanent maintenance burden into one with a defined expiry, and changes the posture from maintaining to watching. TWO OF OUR OWN NOTES WERE FACTUALLY WRONG and are corrected in Cargo.toml. There is no "Future Rowan" GSoC project: GSoC 2026's rust-analyzer project was "Migrating rust-analyzer assists to SyntaxEditor", the stated PREREQUISITE, closed completed 2026-06-20 as rust-analyzer#18285. The rewrite issue itself, rust-analyzer#15710, has been open and untouched since 2024-09-02. The rewrite is landing as incremental breaking changes on rowan master — #213 tree-top, #217 mutable-API removal, 0.17.0 on 2026-08-02, #219 trivia, #220 open — not on a branch and with no announced date. MEASURED ON OUR OWN SURFACE rather than on rowan's internals. An isolated worktree pinned to crates.io rowan 0.17.0 instead of the fork: cargo check -p rivet-core exit 0 compiles clean, no code changes SB_EXIT=1 rowan-0.17.0/src/arc.rs:264 retag for SharedReadOnly, tag absent from borrow stack TB_EXIT=1 rowan-0.17.0/src/cursor.rs:136 deallocation forbidden Both reached through yaml_cst::parse -> parse_root -> parse_block_mapping, so this is a real consumer failing on ordinary construction and traversal rather than a synthetic exercise of rowan's own tests. The SB site is the same header-fattening cast reported upstream as #108 in 2021; the TB site is #192's. Two consequences. The fork earns its keep — dropping it today turns the Miri gate red under BOTH aliasing models, not merely the stricter one. And the rationale #211 was closed under, that rowan should pass Tree Borrows, does not hold for our usage on the current release. The clean compile matters separately: rivet touches only the immutable core and none of the mutable API #217 removed, so the fork is NOT what keeps this project on 0.16.x. CONTEXT FROM AN ECOSYSTEM SURVEY: of 30 rowan consumers, zero carry a fork or patch and zero execute rowan under Miri. rowan's own CI has no Miri job; rust-analyzer's covers only the intern crate, which has no rowan dependency. So this gate is a standard we hold rather than shared pain we are fixing — which is defensible for a compliance tool, and worth stating rather than assuming. cstree, a declared fork of rowan with the same unsafe core, does run Miri under Stacked Borrows across three platforms and passes. A SECOND ROWAN PROBLEM, raised cross-repo as pulseengine/spar#446: rivet's default-on aadl feature pulls spar, and the lock carries two rowans — rivet on the forked 0.16.2, five spar crates on unforked 0.16.1. Our Miri gate covers yaml_cst and sexpr only, so spar's rowan is never interpreted. That issue is framed as a coordination request, not a bug report, because we have NOT tested whether spar's usage reaches the failing paths. Caveats recorded on the artifact: toolchain is miri on nightly 2026-04-19, roughly five months stale, and Tree Borrows is experimental and moves. Re-run before citing upstream. Confirmed with fmt 0, clippy 1.97.0 --all-targets -D warnings 0, cargo test --workspace 0 (2368 passed), cargo test -p rivet-cli --test cli_commands 0 (213 passed), rivet validate 0, rivet docs check 0, yamllint 0, diagnose_test.sh 0, cargo metadata 0. 319 artifacts, no duplicate ids. Refs: FEAT-001 Trace: skip
Codecov Report✅ All modified and coverable lines are covered by tests. 📢 Thoughts on this report? Let us know! |
📐 Rivet artifact delta
Graphgraph LR
REQ_350["REQ-350"]:::added
classDef added fill:#d4edda,stroke:#28a745,color:#155724
classDef removed fill:#f8d7da,stroke:#dc3545,color:#721c24
classDef modified fill:#fff3cd,stroke:#ffc107,color:#856404
classDef overflow fill:#e2e3e5,stroke:#6c757d,color:#495057,stroke-dasharray: 3 3
Added
Posted by |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The rowan maintainer closed our #211 on 2026-09-07:
That turns the fork from a permanent burden into one with a defined expiry, and the posture from maintaining to watching.
Two of our own notes were wrong
Corrected in
Cargo.toml. There is no "Future Rowan" GSoC project — GSoC 2026's rust-analyzer project was "Migrating rust-analyzer assists to SyntaxEditor", the prerequisite, closed 2026-06-20 (#18285). The rewrite issue (#15710) has been open and untouched since 2024-09-02. The rewrite is landing as incremental breaking changes onmaster, not on a branch, with no announced date.Measured on our own surface
Isolated worktree pinned to crates.io rowan 0.17.0 instead of the fork:
Both reached via
yaml_cst::parse→parse_root→parse_block_mapping— a real consumer on ordinary construct-and-traverse, not rowan's own tests. SB site is the header-fattening cast of #108 (2021); TB site is #192's.The fork earns its keep — dropping it turns the gate red under both models. And the rationale #211 was closed under ("rowan should pass Tree Borrows") does not hold for our usage on the current release.
The clean compile matters separately: the fork is not what keeps us on 0.16.x. rivet touches only the immutable core, untouched by #217's mutable-API removal.
Ecosystem context
Of 30 rowan consumers: zero carry a fork, zero execute rowan under Miri. rowan's own CI has no Miri job; rust-analyzer's covers only
intern(no rowan). So this gate is a standard we hold, not shared pain we're fixing — defensible for a compliance tool, worth stating rather than assuming. cstree, a declared rowan fork with the same unsafe core, does run Miri under SB on three platforms and passes.A second rowan problem
Raised as pulseengine/spar#446: rivet's default-on
aadlfeature pulls spar, and the lock carries two rowans — rivet forked 0.16.2, five spar crates unforked 0.16.1. Our gate coversyaml_cst/sexpronly, so spar's rowan is never interpreted. Framed as coordination, not a bug report — we have not tested whether spar's usage reaches the failing paths, and said so there.Caveats on the artifact: nightly
2026-04-19(~5 months stale), TB experimental. Re-run before citing upstream.fmt/clippy -D warnings(1.97.0)cargo test --workspacecargo test -p rivet-cli --test cli_commandsvalidate/docs check/yamllint/diagnose_test.sh/cargo metadataRefs: FEAT-001