fix(oci): falcon-rate shipped as a core module — export cabi_realloc, fail loud, assert the header - #323
Merged
Merged
Conversation
…+ fail loud + assert the header jess pulled falcon-rate:1.129.0 and found it is a raw CORE MODULE (0061736d01000000), not a component (0061736d0d000100) — `meld fuse` rejects it outright, so the consumption path stopped at step one. Confirmed independently by pulling the published blob. Three compounding bugs, mine at the root: 1. ROOT CAUSE — my no_std conversion (#314) dropped the `cabi_realloc` export. wit-bindgen-rt provides it only `#[cfg(not(target_env = "p2"))]` — on wasm32-wasip2 it expects std to supply the symbol. no_std has no std, so `wasm-component-ld` failed: "module does not export a function named `cabi_realloc`". (The local wasip1 build DID componentize, which is why this passed my earlier check — the release target is wasip2.) Now exported from the no_std path, backed by the same single-threaded allocator. 2. SILENT FAILURE — build-components.sh ended the build in `|| true`, so that error was printed and ignored; the script fell through to a stale pre-componentization artifact and bundled it. Build failures now stop the bundle. 3. NO VALIDATION — nothing checked the payload. The OCI config mediaType says "component" regardless of the bytes, so metadata cannot catch this. Now asserts the 8-byte header on every component before it enters the bundle; verified the assert REJECTS the published 1.129.0 artifact and ACCEPTS both good components. Verified after the fix: wasip2 build produces a component (4581 bytes, was a 3749-byte core module), still ZERO wasi imports, and the through-wasm closed-loop proof is unchanged — converges 0.193 s, |err| 0.0059 rad/s. Also corrects the record: the "15x smaller" claim compared a core module against components. Componentized it is ~4.6 KB vs 53-60 KB — still a large win (~12x), just not the number I quoted. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01HvusAXYbHLyv3uTzfBcMbG
avrabe
enabled auto-merge (squash)
July 29, 2026 16:35
This was referenced Jul 29, 2026
Merged
avrabe
added a commit
that referenced
this pull request
Jul 30, 2026
#325) FV-FALCON-OCI-001 claimed distribution was "validated end-to-end" on v1.127/ v1.128. That evidence was NOT sufficient, and jess proved it in the field: falcon-rate:1.129.0 shipped as a raw core module. My check confirmed the OCI config mediaType — which reads "component" regardless of the payload — so it verified the WRAPPER, not the BYTES. Record it honestly rather than quietly re-verifying: the gap, its root cause (no_std dropped `cabi_realloc`, wit-bindgen-rt only provides it when std is linked; the build failure was then swallowed by `|| true`), and the four strengthenings that close it (#323, #324) — cabi_realloc exported, build failures abort, an 8-byte HEADER assert on every component (verified to REJECT the published 1.129.0 artifact), and by-name artifact selection. This is the release-planning loop working as intended: a field regression re-opens the requirement's verification rather than being patched silently. Also defers SWREQ-FALCON-MATHF32-P06 (Cody-Waite argument reduction) from v1.130 to v1.131 — an explicit, logged scope move, not a silent slip. It is the multi-week proof of the four and is not ready; v1.130 is a fix release that republishes falcon-rate as a real component. rivet validate: PASS. Claude-Session: https://claude.ai/code/session_01HvusAXYbHLyv3uTzfBcMbG Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
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.
jess caught a real bug in v1.129.0.
falcon-rate:1.129.0was published as a raw core module (0061736d01000000), not a component (0061736d0d000100) —meld fuserejects it outright, so their consumption path stopped at step one. I pulled the published blob and confirmed it independently.Three compounding bugs, mine at the root:
1. Root cause — my
no_stdconversion droppedcabi_reallocwit-bindgen-rtprovidescabi_realloconly#[cfg(not(target_env = "p2"))]— on wasm32-wasip2 it expects std to supply the symbol.no_stdhas no std, sowasm-component-ldfailed:Why I missed it: my local verification used wasip1, where wit-bindgen-rt does provide it — so it componentized fine. The release build targets wasip2. Now exported explicitly from the
no_stdpath, backed by the same single-threaded allocator.2. Silent failure — the build swallowed the error
build-components.shended that invocation in|| true: the error was printed and ignored, and the script fell through to a stale pre-componentization artifact. Build failures now stop the bundle.3. No validation — nothing checked the payload
The OCI
config.mediaTypesays "component" regardless of the actual bytes, so metadata cannot catch this (jess's point). Now asserts the 8-byte header on every component before it enters the bundle.The assert is verified against the real artifacts:
Verified after the fix
wasi:*importsCorrecting my own claim
I said "15× smaller" — that compared a core module against components. Componentized it's ~4.6 KB vs 53–60 KB: still a large win (~12×), but not the number I quoted. jess caught that too.
Refs: jess#167, #314 (the no_std conversion that introduced this).
🤖 Generated with Claude Code