Repository navigation
chore: bump the moq pin past moq-tokio's fallback runtime - #196
Conversation
Quest-Claim: ab2ec9cf05b1176f
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
# Conflicts: # ffi/moq/moq.c
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
kixelated
left a comment
There was a problem hiding this comment.
Automated review by review (OpenAI)
Reviewed commit: 4747602
No actionable bugs found. Pinning the upstream merge and adapting the export exclusions/tests is a focused way to bring the fallback runtime to C. The census guard is limited to unresolved type parameters in polymorphic mode, leaving concrete trait-object lowering intact. The added C calls directly cover runtime-free construction; they do not establish an end-to-end session, and the removed catalog-publication coverage remains an explicitly documented generic-export limitation.
Verification: static review of the pin, exclusions, binding-test changes and surrounding coercion lowering. I did not run builds, census, tests or a C session, and did not independently audit every upstream change in the pin bump.
# Conflicts: # ffi/moq/ffi.toml
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
kixelated
left a comment
There was a problem hiding this comment.
Automated review by review (OpenAI)
Reviewed commit: e1d1d80
No new actionable bugs found in the update since 4747602. Most changes import main's foreign-type support, inference gate and TS reference counting; the pin/coercion fix itself is unchanged. Reconciling the export configuration with those capabilities is the right direction: accept::Health is now exposed while its unsupported new/failed methods remain excluded, rather than suppressing the whole type. The moved QUIC exclusion and existing generated-name disambiguation are retained.
Verification: static comparison against the previous review and the updated base, including exclusion lists and binding-test integration. I did not run builds, census, tests or a live C session; runtime-free construction still does not demonstrate an end-to-end session.
# Conflicts: # ts/fixtures/reject/census.md # ts/fixtures/reject/src/lib.rs
#196's pin drops binary_snapshot; the JSON track takes its place in moq.c, moq.cpp, and moq.kt. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…e-check the bench (#229) #196's moq pin dropped Ietf DRAFT_16 and DRAFT_17, so the hand-written route passed undefined as both forms' versions, which moq reads as leading ones: its QUIC varint decode of 2 bytes and wider failed the output check. Use Lite DRAFT_06 for QUIC and DRAFT_07 for leading ones, typed as moq's StreamVersion, and type-check bench/ against moq at the pin in just check. Co-authored-by: Luke Curley <432854+kixelated@users.noreply.github.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Problem
moq_rev, whichtsandffiboth build against, predates moq-tokio's fallback runtime (moq-dev/moq#7123), so C still can't reach moq-tokio outside a tokio runtime. At moq'smain,ffi/moq/ffi.tomlfails with stale and missing excludes andjust ts censusICEs.Approach
moq_revmoves to30f25c04, #7123's merge commit.&dyn Traitfrom a pointee that is still a type parameter (moq's<S as Source>::Item) normalized it twice and ICEd.object_ofnow leaves that coercion undefined in the census, since only an instance decides the class. The new case is in therejectfixture.ffi.tomlexcludes track moq's API, after feat(ffi): export Url, SocketAddr, paths, and Vec<T> across the C ABI #190: moq-tokio's QUIC driver moved underquic::driver::{connection, endpoint}, kio gainedBridge, and names moq removed are dropped. Every remaining exclude is something the export refuses.Path::newnow rejects an empty segment, so the tests assert that/cam/fails withEmptySegment.binary_snapshot. Its replacementjson_snapshot<T: Serialize>is generic and was already excluded, so the catalog block now covers a genericis_claimedinstance andmutateinstead.moq_origin_spawn()andmoq_connect_config_init()outside any runtime. At the old pin the first aborted and the second failed with "no async runtime".m1/moq-pinand addsm1/ffi-export/warnings(see Follow-ups).Verified:
just ffi moq,just ts moq,just ts census,just check,just test.Census, same translators, old pin -> new pin:
The model and moq-net growth is moq's move to
hashlink's linked hash map, whoseNonNullcalls the translator refuses (+244unsupported callin model), plus more foreign types (+66) and casts (+31).moq-mux and hang shrink as moq deleted API (moq#7046). moq-tokio's 3 newly refused types (
quic::driver::endpoint::{Error, Socket},unix::Error) are objects that no export returns. Against this PR's first head, before #190's foreign types and #205's slices, moq-tokio went from 56/96 types and 213/293 methods and fields to 65/96 and 279/325.Impact
moq_path_too_longbecomes themoq_path_invalidenum, themoq_binary_*snapshot names go away, andmoq_origin_spawn/moq_connect_config_initwork outside a runtime.Alternatives
main: it isn't onmain's history, and would skip the API the next bump has to absorb anyway.Follow-ups
m1/ffi-export/warnings: a unit-returning callback (catalog::Producer::mutate) emitslet result = ...; (), which warnsunused variable3 times in moq's glue.json_snapshot<T: Serialize>exports; this belongs with the instance-keys quest.kiousesAtomicUsize::fetch_update, deprecated on the ffi nightly; this is moq's own code.🤖 Generated with Claude Code
(Written by Claude Opus 5.5)