Repository navigation
chore: bump the rustc nightly to 2026-11-12, with a canary on the newest - #729
Conversation
Quest-Claim: 2157d1a20ef95679
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… and Allocator stable Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…s, and IoSlice::advance to 1.102's std Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…fault, Fuse::advance_back_by, and EscapeDebug's Debug Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… cover the unfused paths Skip takes no step when advanced by none while it still skips. std::vec::from_elem is now a stable deprecated wrapper over from_elem_global, so both map. The iters fixture gains an unfused iterator through Skip and Fuse, and panics covers unimplemented!() with no message. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…earnings Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
There was a problem hiding this comment.
ℹ️ No critical issues. There are two rough edges in the new canary job; the port itself looks right.
Reviewed changes
I reviewed the whole bump. That covers the rustc port across driver/, ffi/, go/, and ts/, the runtime mirrors of std 1.102, the new fixture cases, the canary job, and the rescheduled quest. I spot-checked the regenerated expected/ code snapshots rather than reading them line by line: the differential runs against native are what prove their behavior.
- Mechanical rustc port:
GenericArgsRefbecomesGenericArgs,caller_bounds().all_clauses(tcx).iter()replaces the iterator,layout.alignbecomes a plainAlign, anddoc_str_and_fragment_kindreads doc comments. rust2-lint readsAttribute::Tool(AttrItem)throughtcx.tool_attrsandMetaItemKind::list_from_tokens. - Derived
Debug: newfmtDebugFields/fmtDebugEnum(ts) anddebugFields/debugEnum(go) handle the space-joined names and the byte offset ofdebug_c_like_enum_write_str. The caller-site needles are retargeted. - std behavior mirrors: the PR maps
vec::from_elem_global, the*_default_msgpanics, andstringFromChars.str::EscapeDefault's specializedExtend/FromIterator/ToStringget their own path. It also coversFuse::advance_back_by,Skip::advance_by(0),EscapeDebug's literal char,IoSlice::advance, and theReadymessages. New cases cover each of these (iters::unfused, capacity op 24 and ops 49–52, panics case 36). - ffi reject fixture:
unsafe impl Pattern, and theallocator_apifeature is dropped now thatAllocatoris stable. - Canary:
sh/canary.shupdates rust-overlay and pointsts/rust-toolchain.tomlat the newest nightly that has every listed component. The newcontinue-on-errorjob runs every module's check and test on it, andreportdoes not wait on it. - Quest:
quest/m0/nightly-bump.mdis rewritten for its next run on 2026-12-24 and carries this run's findings forward.
I did not run the suite locally, because the 2026-11-12 nightly isn't in this environment. This review relies on the just gate / just ci results in the PR description.
claude-opus-5-5 | 𝕏
… when its toolchain step fails Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… not its IntoIter The capacity caller now drives every strings op, 48 to 53, which it skipped. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…-2026-11-12 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
There was a problem hiding this comment.
✅ No new issues found. Both canary findings from the last review are fixed, and the new
Extendchange matches std.
Reviewed changes
This run covers the commits since 7b55fb2: the canary fixes, a fix to how String's Extend specializes, and a merge of main.
- Canary toolchain profile:
sh/canary.shnow selects the nightly witht.${file.profile or "default"}. That is the same profilefromRustupToolchainFilebuilds for the dev shell, so the canary can no longer pick a nightly the shell can't build. - Canary step gating: the Toolchain step gets
id: toolchain, and every module step requiressteps.toolchain.outcome == 'success'. A failed toolchain selection no longer runs the modules on the pinned nightly. String'sExtendspecialization:string_extendnow takes(source, iter)and checks forEscapeDefaulton the source type. At3ef0c22cd, std'sSpecExtendCharsdispatches onIitself, whileFromCharIterSpec(FromIterator<char>) dispatches onI::IntoIter. Socollectcorrectly keeps passing(from, from). Capacity op 53 tests a wrapper whoseinto_iterreturns anEscapeDefault.- Merged main: rust-overlay stays at
8bbfc858, the PR's newer rev that providesnightly-2026-11-12.
I still haven't run the toolchain locally, because nightly-2026-11-12 isn't installed in this environment.
claude-opus-5-5 | 𝕏

Bumps the nightly every driver links from
nightly-2026-10-15tonightly-2026-11-12(rustc 1.102.0,3ef0c22cd), withflake.lock's rust-overlay bumped to name it. Portsdriver/,ffi/,go/, andts/to the new compiler and std, and adds a canary job tonightly.ymlthat runs every module on the newest nightly, so the next bump's breaks show up before that bump starts. Implements quest/m0/nightly-bump.md and schedules its next run for 2026-12-24.Summary
rustc internals
GenericArgsRefis gone.GenericArgsis now aCopyvalue with no rangeIndex(useas_slice/as_tcx_slice); derefs applied withcargo fix --broken-code.ParamEnv::caller_bounds()is now an accessor:.all_clauses(tcx).iter().LayoutData::alignis a plainAlign.Attribute::Tool(AttrItem), so rust2-lint reads its budget throughtcx.tool_attrsand parses the arguments withMetaItemKind::list_from_tokens.AttributeExt::doc_str_and_fragment_kind. Both///and#[doc = ".."]still becomeDocComment, so the result is the same.Expansion and std changes. Only the differential tests caught these.
Debug:debug_struct_field{N}_finishanddebug_struct_fields_finishnow take the type's and fields' names joined by spaces. A name with no space is a tuple struct's.debug_c_like_enum_write_str(names, byte_offset).fmtDebugFields/fmtDebugEnum(ts) anddebugFields/debugEnum(go). The go codec fixture's thiserror enum reachedslice::Splitthrough std's new path, so go refused it until this mapping existed.vec![x; n]now callsvec::from_elem_global.panic!(),unreachable!(),unimplemented!(), andtodo!()with no message now callcore::panicking::*_default_msg.String: FromIterator<char>starts fromwith_capacity(size_hint().0), which changes capacities. Covers the generic collect (newstringFromChars) and ts'sfrom_utf16_lossy.str::EscapeDefaulthas specializedExtend/FromIterator/ToStringthat reserve by its flat map's parts, plus ten bytes before each\u{..}. Mirrored exactly (stringExtendEscapeDefault); the capacity fixture gains four cases (ops 49–52) for it.Fuseforwardsadvance_back_by, so aTakeunder aChainnow trims its source from the back.Skip::advance_by(0)takes no step while it still skips, so an iterator that is not fused is not seen ending early. The iters fixture gainsiters::unfused, which runs a non-fused iterator throughSkipandFuse.std::vec::from_elemis now a stable, deprecated wrapper overfrom_elem_global, so ts maps both, and capacity's op 24 calls it.EscapeDebug's Display/Debug drops its literal char once it has been yielded.IoSlice::advancenow panicsmid > lenfromsplit_at.Readyrewords both of its panics.Patternis anunsafe traitandAllocatoris stable as of 1.100, so the ffi reject fixture is updated for both.Canary. A new
canaryjob ("Newest nightly") innightly.yml:sh/canary.shupdates rust-overlay, then pointsts/rust-toolchain.tomlat the newest nightly that ships every component and target the file lists (selectLatestNightlyWith).continue-on-error, andreportdoes not wait on it, so it fails no merge and opens no issue.Review fixes (this round).
sh/canary.shselects with the profile the dev shell builds: the file'sprofile, elsedefault, asfromRustupToolchainFiledoes. Before, it checkedminimal, so it could pick a nightly the shell then failed to build.String: Extendnow specializes on the argument's own type, as std does. Before, the translator checked theIntoItertype, so a wrapper whoseinto_iteris astr::EscapeDefaultgot std's reservation forEscapeDefaultitself (Codex found this). capacity op 53 covers it. The capacity caller now drives ops 48–53: it stopped at 47, so ops 48–52 never ran. Without the fix, the differential run shows 22 differences.flake.lockkeeps main's other inputs and this branch's rust-overlay revision (2026-11-12).Quest.
quest/m0/nightly-bump.mdis rewritten for its next run: scheduled 2026-12-24, with this run's findings carried forward.Code snapshots (128
expected/files) move for two reasons:breaklabels, temp numbering.Debugcalls above.Every fixture's differential run against native Rust passes, so behavior did not move.
go/ port cost, for weighing whether to keep it: 41 lines added and 35 removed over 7 files, all derived
Debugor mechanical renames. ts/ took 380 added and 302 removed over 36.Public API: none. Wire: none.
Validation
just gatepasses on the new nightly (lint, quest, sh, driver, ffi, go, ts; ts 92 fixtures).capacity: 3 cases;chars: 27;iters: 1;shapes;futures: 24). Each now matches native.just ffi censusandjust ts censuspass. The ts censuses for moq-net, model, and kio match main's latest nightly artifacts exactly (309 / 41 / 87 refusals). ffi's census matches or exceeds the 2026-11-11 run.just ffi moqgets as far as the Swift step, which fails on theNoUseUnstructuredThrowingTaskerror that quest/m0/nightly-red.md owns. Not touched here.sh/canary.shrun locally: from a toolchain file pinned to 2026-10-15 it rewrites the channel tonightly-2026-11-12, the newest the updated overlay names with every component.just cipasses.just cipasses exceptapi_swiftonce: its check "Rust gives up on an async closure" (within(0), inffi/tests/api.swift, which this PR does not touch) races the drain under load. Run alone, it passed 3 of 3 times. That includes ts's 92 fixtures and the newiters::unfused, capacity op 24, and panics case 36 against native.Choices made without asking
Code snapshots regenerated. The quest said "a fixture whose expected output moves is a finding to explain in the PR, not to regenerate". I read that as differential output, which did not change. Code snapshots necessarily move with rustc's MIR shape, so I regenerated them after the differential runs passed and explained the causes above. The rewritten quest states it that way.
The canary runs each module's checks and tests, not only builds. Most of this run's cost was std behavior that only the differential tests caught, so a build-only canary would have missed it.
The canary updates the rust-overlay input before choosing, so it tests the newest nightly rather than the newest one the pinned lock names.
ts runtime keeps
fmtDebugStructFields/fmtDebugTupleFieldsfor its own impls. They drop the length assertion that mirrored std's removedassert_eq!.This round's review ran Codex alone. opencode is still out of quota. Three rounds: one finding (above), then 10/10 with no findings on efd1d29 and again on the merge head f380be9.
The previous round's review ran Codex alone, in a new session. opencode is out of quota. Its round-1 findings (Fuse's clearing branch untested; no
unimplemented!()fixture) are both fixed in 131e926:iters::unfusedcases 1 and 2, andpanicscase 36. Codex's round-1 findings (Skip's zero advance;vec::from_elem) are fixed in the same commit.quest reviewwould not resume Codex while opencode's findings were unsettled, so--newstarted a fresh Codex review of the whole change.Newly stable std APIs are not mapped here. A std-diff audit found that
String::into_chars, slicesplit_once/rsplit_once, andRc/Arc::is_uniquebecame stable. No translator supported them before, and an unsupported one is still a compile-time refusal, so nothing regresses. Mapping them is coverage work, not part of this bump.Follow-ups
None.
Caused by
No single PR: an upstream compiler and std change.
Models
Size: M
Implementer: Claude, claude-opus-5-5, unknown effort
Implementer: Claude Code, claude-opus-5-5, low effort (iterate, S)
Review
Rewritten by
quest reviewafter each round.Head: f380be9
Base: origin/main at cf1f0f9
Reviewer: Codex, gpt-6.1-sol, high effort
Session:
01a1d624-0dff-7280-9672-18892eae8729, round 3Score: 10/10,
The merge preserves the previously reviewed branch changes and rust-overlay pin, with no new issues found by inspection; builds and tests were not run.Findings: none
Summary by CodeRabbit
New Features
Bug Fixes
Maintenance
Summary by CodeRabbit
New Features
Bug Fixes
Improvements