You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Production export: one package with several roots, sharing one runtime #387
Pinned at 387735155da3; origin/main and origin/release/v0.5 (ed549c24f) read. A feature request. No cost was measured beyond the sizes below.
Today
§24.7: every root of a lexlean/production-request/1 names its own crate. Two roots with one package name are refused, as tried at the pin:
error[LLC0201]: package `shared` is exported twice for target `rust-std`
Manifest.exports in calculus/rust/package.rs is a list, so a manifest can already state several exports. The request cannot ask for that.
The committed helpdesk export on release/v0.5 has three roots and three crates, each with its own src/lib.rs (47,093, 63,759 and 40,458 bytes), no Cargo dependencies, and the same runtime digest in all three provenance.json files. I did not measure how much of each lib.rs is runtime.
Ask
A request form that exports several roots from one crate, with one copy of the runtime and of shared functions. A large application model with many roots would otherwise link many runtime copies and many crates.
Questions
Is one crate per root deliberate, for example because each certificate concerns one root and the program lowered for it (§17.17)? If so, a shared-runtime crate that the per-root crates depend on may fit better than a shared root crate.
How do certificates B and E bind a crate holding several roots?
I’m taking a bounded design-only pass on this unclaimed request: compare one crate with several roots against a shared-runtime dependency, and state the required certificate B/E, replay, provenance and package bindings against the current accepted v0.5 contract. The existing per-root refusal is deliberate contract behavior, not being treated as a bug.
This reserves only the proposal/review work. No request/schema/compiler change, measurement, generated product or identity wave is started, and no performance or binary-size benefit is assumed. #405’s forward integration and existing packager work retain their owners. I’ll bring back a concrete proposed boundary and remaining acceptance decisions before implementation.
Design-only recommendation, independently reviewed against accepted v0.5 2b0395ef (SPEC §§17.17/24.7 and eleven exact source/schema blobs): prefer investigating a single package with isolated per-root modules and an internal shared runtime before an external runtime dependency. This is an unaccepted design target, not an implementation commitment or existing proof capability.
The present B theorem quantifies over every fuel/function for one lowered program and its rendered crate; E uses that same crate and its recorded entry. Manifest.exports supporting several entries of one program does not prove several independently lowered roots can share a crate. A flat merge, identifier relocation or runtime extraction cannot be justified by matching hashes. The proposed route retains each original program, A certificate, entry, ADT/closure identities, effects and policy, then requires either qualified B/E for the actual final context or a kernel-checked embedding/context theorem. Proving the actual public wrapper is an additional proposed obligation; current E does not certify an arbitrary wrapper.
Required design decisions before R3 implementation:
Define closed namespace/runtime resolution and prove noninterference, typed calls, boundary-validator behavior, ownership/overflow semantics and exact axiom policy for every root in the final context.
Define explicit versioned grouped requests, root/export naming and target eligibility; preserve existing request /1 behavior. Bind root→program→entry→A/B/E and context/runtime to one exact package in provenance/result/preservation records, with deterministic full-package replay and offline publication. Replay remains bindings verification, with its existing consistent-forgery limitation.
Choose the proof shape and version/identity domains. A shared-runtime dependency additionally expands dependency resolution, linked-program, license/source/lock and offline replay obligations; it is not merely moving text into another crate.
This bounded route intentionally leaves arbitrary shared-function deduplication unsatisfied. #387 remains open for that requirement; acceptance of a narrower research or implementation milestone does not authorize closure. No speedup, binary-size saving, hard budget or current theorem sufficiency is claimed. If the route needs unresolved calculus group-union semantics, it stops under the existing R25 admission/STOP; #431/TC03 is not a feasibility shortcut.
After a design disposition, the implementation proposal would need the normal R3 sequence, native A/B/context/E and exact-policy checks, semantic plants (cross-wired exports, bypassed validator, aliased identities/runtime), exact product/replay evidence and complete final-head gates. None was run or admitted by this design pass. Claude’s #405 forward lane, existing packager work and reserved #316 remain separate. Full local proposal SHA256: 094a25856afd603463cbd401bdd525197cec49425fc915ae3b28da7f0961dbc8; independent verdict is clear for proposal only.
Design-only proof-shape followup to the proposal above, independently reviewed against exact v0.5 2b0395ef:
The proposed isolated-module/shared-runtime route needs a modeled package context before a final contextual theorem can be stated. RustSyntax has unqualified numeric function/dispatch identities and a flat Crate { profile, items }, without a module constructor. findFunction/invoke search one item list. Distinct public export strings therefore do not isolate two roots' local f0, recursive calls, closure dispatch or type identities.
A concrete design recommendation is to retain existing isolated A/B/E as premises and introduce a closed outer context with a root-to-local-crate table, explicit export-to-(root, entry) bindings, qualified invocation, and one pinned runtime realization. The new embedding relation must preserve each root's observations in that context; wrapper correspondence then connects the actual public export to its original entry, including boundary-validator acceptance/refusal. This is new modeled meaning, not a textual wrapper or hash/prefix check. Runtime operations currently use runItem(crate.profile, ...), so shared runtime bytes still need explicit renderer/resolution correspondence. Any administrative evaluation steps also need a stated fuel relation rather than an assumed equal-fuel simulation.
This narrows the prerequisite for the proposed module route; it is not an impossibility claim for all multi-root packaging. Flat renaming/relocation was not investigated or admitted. Proof feasibility, exact schema/identity decisions and shared-function deduplication remain open; no implementation, Lean run, producer or gate was started. The existing R25 admission/STOP and #431/TC03 STOP remain intact.
Source-only review SHA256 babd9067950247605ff697e08aa5bff94af0a6fa8088e10d4ce8c62be98128ef; corrected exact-blob citation receipt 0ee4257818583816a5f32f639cd5c92d8ae8eadd99b8d05c86f7005017c99677. #387 stays open; this is a recommendation for design disposition, not acceptance.
Activity
afflom commentedon Nov 10, 2026
I’m taking a bounded design-only pass on this unclaimed request: compare one crate with several roots against a shared-runtime dependency, and state the required certificate B/E, replay, provenance and package bindings against the current accepted v0.5 contract. The existing per-root refusal is deliberate contract behavior, not being treated as a bug.
This reserves only the proposal/review work. No request/schema/compiler change, measurement, generated product or identity wave is started, and no performance or binary-size benefit is assumed. #405’s forward integration and existing packager work retain their owners. I’ll bring back a concrete proposed boundary and remaining acceptance decisions before implementation.
afflom commentedon Nov 10, 2026
Design-only recommendation, independently reviewed against accepted v0.5
2b0395ef(SPEC §§17.17/24.7 and eleven exact source/schema blobs): prefer investigating a single package with isolated per-root modules and an internal shared runtime before an external runtime dependency. This is an unaccepted design target, not an implementation commitment or existing proof capability.The present B theorem quantifies over every fuel/function for one lowered program and its rendered crate; E uses that same crate and its recorded entry.
Manifest.exportssupporting several entries of one program does not prove several independently lowered roots can share a crate. A flat merge, identifier relocation or runtime extraction cannot be justified by matching hashes. The proposed route retains each original program, A certificate, entry, ADT/closure identities, effects and policy, then requires either qualified B/E for the actual final context or a kernel-checked embedding/context theorem. Proving the actual public wrapper is an additional proposed obligation; current E does not certify an arbitrary wrapper.Required design decisions before R3 implementation:
/1behavior. Bind root→program→entry→A/B/E and context/runtime to one exact package in provenance/result/preservation records, with deterministic full-package replay and offline publication. Replay remains bindings verification, with its existing consistent-forgery limitation.This bounded route intentionally leaves arbitrary shared-function deduplication unsatisfied. #387 remains open for that requirement; acceptance of a narrower research or implementation milestone does not authorize closure. No speedup, binary-size saving, hard budget or current theorem sufficiency is claimed. If the route needs unresolved calculus group-union semantics, it stops under the existing R25 admission/STOP; #431/TC03 is not a feasibility shortcut.
After a design disposition, the implementation proposal would need the normal R3 sequence, native A/B/context/E and exact-policy checks, semantic plants (cross-wired exports, bypassed validator, aliased identities/runtime), exact product/replay evidence and complete final-head gates. None was run or admitted by this design pass. Claude’s #405 forward lane, existing packager work and reserved #316 remain separate. Full local proposal SHA256:
094a25856afd603463cbd401bdd525197cec49425fc915ae3b28da7f0961dbc8; independent verdict is clear for proposal only.afflom commentedon Nov 11, 2026
Design-only proof-shape followup to the proposal above, independently reviewed against exact v0.5
2b0395ef:The proposed isolated-module/shared-runtime route needs a modeled package context before a final contextual theorem can be stated. RustSyntax has unqualified numeric function/dispatch identities and a flat
Crate { profile, items }, without a module constructor. findFunction/invoke search one item list. Distinct public export strings therefore do not isolate two roots' localf0, recursive calls, closure dispatch or type identities.A concrete design recommendation is to retain existing isolated A/B/E as premises and introduce a closed outer context with a root-to-local-crate table, explicit export-to-(root, entry) bindings, qualified invocation, and one pinned runtime realization. The new embedding relation must preserve each root's observations in that context; wrapper correspondence then connects the actual public export to its original entry, including boundary-validator acceptance/refusal. This is new modeled meaning, not a textual wrapper or hash/prefix check. Runtime operations currently use
runItem(crate.profile, ...), so shared runtime bytes still need explicit renderer/resolution correspondence. Any administrative evaluation steps also need a stated fuel relation rather than an assumed equal-fuel simulation.This narrows the prerequisite for the proposed module route; it is not an impossibility claim for all multi-root packaging. Flat renaming/relocation was not investigated or admitted. Proof feasibility, exact schema/identity decisions and shared-function deduplication remain open; no implementation, Lean run, producer or gate was started. The existing R25 admission/STOP and #431/TC03 STOP remain intact.
Source-only review SHA256
babd9067950247605ff697e08aa5bff94af0a6fa8088e10d4ce8c62be98128ef; corrected exact-blob citation receipt0ee4257818583816a5f32f639cd5c92d8ae8eadd99b8d05c86f7005017c99677. #387 stays open; this is a recommendation for design disposition, not acceptance.