Repository navigation
chore(deps): upgrade to Holochain 0.7.1 - #130
Conversation
Pin holonix to main-0.7 and the workspace to hdk =0.7.0 / hdi =0.8.0 / holochain =0.7.0. The 0.7 action model replaces the per-variant action structs with ActionHeader + ActionData, so the integrity zomes move to TypedAction: - FlatOp variants renamed: StoreEntry -> CreateEntry, StoreRecord -> CreateRecord. - EntryCreationAction is gone; narrowing a fetched Action now goes through TypedAction<EntryCreationData>::try_from. - OpRecord::DeleteEntry no longer carries original_action_hash; read action.deletes_address. Same for UpdateEntry -> action.original_action_address. - action.author is now the accessor action.author(); Create/Update parameter types become TypedAction<CreateData> / TypedAction<UpdateData>. - Coordinator post_commit handlers match on action.hashed.content.data (ActionData) instead of the removed Action::* variants. Validation logic is unchanged: every check reads the same value from its new location.
- @holochain/client ^0.21.0 (root, ui, tests); @holochain/hc-spin ^0.700.0. - scripts/launch-happ.mjs: 0.7 removed the WebRTC/tx5 transport, so `hc sandbox create network` now takes `quic` instead of `webrtc <signalUrl>`. - Version references across CLAUDE.md, README.md and documentation/ updated to HDK 0.7 / HDI 0.8 / holochain 0.7 / client 0.21. - Tryorama has no 0.7-compatible release; tests/DEPRECATED.md and the testing stack table now say the legacy suite cannot run against a 0.7 conductor.
@holochain/client 0.21 narrows AppWebsocket.client to the AppClientTransport interface (request/on only), so the e2e harness feature-detects close() instead of calling it on the concrete WsClient. shared-types peerDependency moves to ^0.21.0. svelte-check: 0 errors, 0 warnings.
- await_consistency_20_s is gone. 0.7 has await_consistency(cells) with a hard 60s timeout and await_consistency_s(timeout, cells) as the configurable form; existing 20s and 10s waits are preserved via the latter rather than silently widened to 60s. - SweetConductorBatch no longer derefs mutably or exposes iter_mut, but does implement IndexMut, so the NDO cell setup indexes the batch directly. All three suites (nondominium, group, lobby) compile against holochain =0.7.0 with test_utils.
The vendored hREA DNA ships as a role in this hApp and runs in the same 0.7 conductor, so a 0.6-built wasm will not load. Upstream h-REA has no main-0.7 branch, so the Sensorica fork carries the port on chore/holochain-0.7. Note the flake's shellHook runs `git submodule update --init vendor/hrea` on every `nix develop`; without this pointer commit that reset the submodule back to the 0.6 commit.
The flake shellHook runs `git submodule update --init vendor/hrea` on every `nix develop`, which silently discards an uncommitted branch checkout inside the submodule. Hit during the 0.7 port.
Author self-check on the highest-risk item (not an independent review)Item 2 in "What is still missing" asks whether the replaced Where 0.6 populated each field, from
Narrowing failure branches. Every site that swapped a hand-written
No branch flipped One genuinely new branch, worth a reviewer's eye. In hREA's Still unreviewed by anyone but me: the |
…can write (#138) rsync -a preserves the read-only mode of nix store paths, so after the first materialization .claude/ was mode 555 and the agent-skills hook could not mkdir .claude/skills/holochain. Every nix shell entry and every CI job logged 'rsync: mkdir failed: Permission denied (13)' three times. Non-fatal but noisy, and a red herring when diagnosing real e2e failures: in PR #130's failing run the error appears immediately before the launcher timeout, though the two are unrelated. .cursor/ already had this exact chmod; .claude/ was missing it.
The input was unpinned and the lock held ab02c4f from 2026-04-06, which predates every tagged release of the upstream repository. Two upstream breaking changes had to be absorbed to move it forward: - The repository was renamed to holochain-agent-skills (plural). The input attribute stays singular so `nix flake update holochain-agent-skill` and the references in CONTRIBUTING.md and pai/README.md keep working. - The skill moved from the repository root to skills/holochain/. Fetching the root used to drag book.toml, SUMMARY.md, CHANGELOG.md, docs/ and .github/ into the installed skill; the src now names the subpath. v1.0.0-rc.1 targets Holochain 0.7 only, while dev is still on hdk 0.6 / holonix main-0.6. The 0.6-compatible release is v0.2.0. Pinning the candidate here is deliberate: it lands the skill ahead of the 0.7 upgrade in #130 rather than after it.
Commit the @valueflows/vf-graphql-holochain 0.700.0-rc.0 tarball built from h-REA/hREA tag happ-0.5.0-beta.1 (79c8d4fb) under vendor/, beside the hrea submodule, with its sha256 and a README recording source commit, rebuild steps and how to drop it once npm serves 0.700. The UI installs it through a file: dependency; bun.lock regenerated by bun install. The adapter brings its own @holochain/client ^0.21, installed nested; the UI keeps ^0.20 until the Holochain 0.7 layer. Closes #153
… stream Building inside a full hREA checkout fails under npm 11 (EUNSUPPORTEDPROTOCOL on the workspace link: dependency), so the recipe now extracts the module with git archive. A rebuild reproduces the tar stream, not the gzip layer, so the README records the tar stream sha256 and points at h-REA/hREA#421 for the pack script.
The integration strategy said hREA integration bypasses GraphQL entirely, while this branch vendors the GraphQL adapter into the UI. State the bound: the adapter serves read-only queries, and every write stays zome-level so governance evaluation and PPR issuance cannot be skipped.
Point vendor/hrea at the upstream repository over https and pin it to 79c8d4fb, the commit tagged happ-0.5.0-beta.1. The branch line is dropped: the pin is a release tag, not a branch to follow. Refs #154
The README, the vendor README, the implementation status and the integration strategy still named the Sensorica fork on main-0.6. They now name the upstream release tag and its commit, and the integration strategy records that the pin needs the Holochain 0.7 toolchain. Refs #154
Point vendor/hrea at the upstream repository over https and pin it to 79c8d4fb, the commit tagged happ-0.5.0-beta.1. The branch line is dropped: the pin is a release tag, not a branch to follow. Refs #154
The README, the vendor README, the implementation status and the integration strategy still named the Sensorica fork on main-0.6. They now name the upstream release tag and its commit, and the integration strategy records that the pin needs the Holochain 0.7 toolchain. Refs #154
…t API hdk 0.7.1 replaced ZomeCallCapGrant and the CapAccess grant variants with CapGrant::new_zome_call_grant and GrantConstraint. Both grant sites keep the same tag, secret, assignees and function list, so what a grant allows is unchanged.
Workspace pins hdk =0.7.1, hdi =0.8.1 and holochain =0.7.1. Holonix main-0.7 is updated to the 0.7.1 revision, and the JS side moves to @holochain/client ^0.21.1 and @holochain/hc-spin ^0.700.1. Docs that named the exact pins follow. Refs #130
Staying a draft, one blocker (full text in the body, section "Waiting on Soushi"). Blocker. The hREA DNA at the Independent review of the validation diff, verdict: no blocker for the port. Reviewer was a fresh-context agent given the diff and five questions. It read hdi 0.7.0 and 0.8.1, holochain_integrity_types 0.6.0 and 0.7.1 and holochain 0.7.1 sources, and ran nothing.
Also found on the way. Local results, inside |
vendor/hrea now points at Sensorica/hREA c4c44adb: the release commit 79c8d4fb with only the hdk =0.7.1 / hdi =0.8.1 pins and the lockfile changed. A DNA built on hdk 0.7.0 cannot be granted a signing capability by a Holochain 0.7.1 conductor. Retired once upstream releases the bump. Refs #130
Review APPROVE at head 15ad9a5, merging under Soushi's explicit
|
Summary
This moves Nondominium from Holochain 0.6.1 to Holochain 0.7.1 (
hdk 0.7.1,hdi 0.8.1,holochain 0.7.1, Holonixmain-0.7at the 0.7.1 revision,@holochain/client0.21.1,@holochain/hc-spin0.700.1). It also moves the hREA submodule off the 0.6 fork:vendor/hreanow points atSensorica/hREAcommitc4c44adb, which is the h-REA/hREA releasehapp-0.5.0-beta.1plus one commit that changes only thehdk/hdipins to 0.7.1 / 0.8.1. That pin bump is proposed upstream as h-REA/hREA#423, and the fork pin is retired once upstream releases it. This PR absorbs what was #169, because a Holochain 0.7 hREA DNA cannot build on the 0.6 shell alone.What changes for people. Every DNA hash changes, so a 0.7 network is a new network and existing conductor data must be cleared. Nothing is migrated: the project has no production users, and the break costs nothing now. What the validation rules accept is meant to be unchanged, and an independent review found no port bug (verdict in the folds below). The only new behaviour is stricter validation inside hREA's own DNA, which comes with the release.
Why the fork pin. A Holochain 0.7.1 conductor cannot grant a signing capability to a DNA built on
hdk 0.7.0, because 0.7.1 changed the capability grant entry that DNA's validation callback decodes. At the release as published,bun run e2efails on its first test; with the two pins bumped it passes 23 of 23.Changes
Toolchain. Workspace pins
hdi =0.8.1,hdk =0.7.1,holochain =0.7.1withtest_utils.flake.lockmovesholonix(main-0.7) toa6cdfbd3, which locksholochain-0.7.1and kitsune2 0.5.2. JS side:@holochain/client^0.21.1,@holochain/hc-spin^0.700.1; the vendored hREA adapter (@valueflows/vf-graphql-holochain0.700.0-rc.0, from #155) depends on^0.21.0, which 0.21.1 satisfies.bun.lockwas regenerated bybun install, not edited.0.7.1 capability grants.
hdk 0.7.1removedZomeCallCapGrantand theCapAccess::Assigned/Transferablegrant variants. The two grant sites inzome_person(grant_private_data_access,create_transferable_private_data_access) now build aCapGrant::new_zome_call_grantwithGrantConstraint::Assigned/Transferable. Tag, secret, assignees and function list are the same, so what a grant allows is unchanged.Integrity zomes, the v2 action model. 0.7 removes the per-variant action structs; an
Actionis anActionHeaderplus anActionData, and flat-op variants carry aTypedAction<D>. Acrosszome_person,zome_resource,zome_gouvernance,zome_group_integrityandzome_lobby_integrity:StoreEntry/StoreRecordbecomeCreateEntry/CreateRecord;RegisterCreateLink/RegisterDeleteLinkfold intoFlatOp::Link(OpLink::*);EntryCreationActionbecomesTypedAction<EntryCreationData>; the hash fields that sat on the flat-op variant are read from the action (deletes_address,original_action_address,link_add_address);action.authorbecomesaction.author().Coordinator zomes.
post_commitsignal dispatch matches onActionData::*.hREA. The submodule is
Sensorica/hREAc4c44adb(.gitmodulesurl updated,vendor/README.mdsays what it is). It is the upstream releasehapp-0.5.0-beta.1(79c8d4fb) plus the pin bump. The release ships real validation (required fields, quantity and date rules, an action whitelist), so the earlier "everyvalidate_*is a TODO" description no longer holds; the Sweettest and e2e runs below were made against it.Dev launcher and tests.
scripts/launch-happ.mjspassesquic(0.7 drops WebRTC).await_consistency_20_sis gone; the existing waits go throughawait_consistency_s(timeout, cells)with the same timings. Tryorama has no 0.7-compatible release, so the deprecatedtests/suite cannot connect andtests/DEPRECATED.mdsays so.Contributor trap. The flake
shellHookrunsgit submodule update --init vendor/hreaon everynix develop, which discards an uncommitted checkout inside the submodule.CONTRIBUTING.mddocuments it.Independent review of the validation diff
Reviewed by a fresh-context reviewer given the diff and the questions, not this PR's conclusions. It read the real
hdi 0.7.0andhdi 0.8.1sources,holochain_integrity_types0.6.0 and 0.7.1, andholochain 0.7.1. It did not compile or run anything.original_*_hashfields are still reachable fromaction. No op type that was validated now falls into a catch-all returning Valid; catch-all arms sit at the same places before and after.FlatOp::Link(OpLink::*)arms: equivalent. Base, target, tag and link type come from the same underlyingCreateLink, and the delete arms read them fromoriginal_action, which is built from it.prev_action: None => Validin hREA's CreateAgent arm: a behaviour change, unreachable. In 0.6prev_actionwas non-optional, so the branch only satisfies the newOptiontype. If it were reached it would accept a CreateAgent with no predecessor without checking the membrane proof. It cannot be reached: sys validation (check_prev_action,holochain 0.7.1 sys_validate.rs:153-172) rejects any non-genesis action with no predecessor before app validation runs. Not traced: whether sys validation always completes before the app callback for this op. Optional hardening, in hREA upstream: return Invalid onNoneso it fails closed.deletes_address,original_action_address,link_add_address: equivalent at every call site in both diffs. Same hash type (ActionHash) and meaning. Thetry_into()narrowing keeps each call site's old reaction (person, resource and hREA return Invalid; gouvernance and lobby return Valid).rea_agent.rsnow requires a non-blanknameand anagent_typeofPersonorOrganization). Nondominium'shrea_bridge.rsalways sendsPerson. Not verified by the reviewer: whether a blank name can reach the bridge, and the 18 other hREA integrity files (only skimmed). The Sweettest and e2e runs exercise the bridge and passed.Blocking findings: none for the port.
Decisions
await_consistency_skeeps the timings; retuning them is separate.How to test
Clear old conductor data first (
hc sandbox clean, and remove/tmp/ndo-e2e). 0.7 cannot read 0.6 databases and the failure is a confusing startup error.What was verified locally, inside
nix developbun run build:happworkdir/nondominium.happ, hREA built from the fork pincargo test -p nondominium_sweettest -- --test-threads 6#[ignore]ondev)cargo test -p group_sweettest -- --test-threads 6cargo test -p lobby_sweettest -- --test-threads 6bun run --filter ui checkLifecycleTransitionModal.svelte,state_referenced_locally)bun run e2e(23 specs, real conductors)hc dna hashof the rebuilthrea.dnauhC0kqRm...)Documentation
.rules(read asCLAUDE.md,AGENTS.md,GEMINI.md),README.md,CHANGELOG.md,documentation/IMPLEMENTATION_STATUS.md,ARCHITECTURE_COMPONENTS.md,DOCUMENTATION_INDEX.md,Testing_Infrastructure.md,specifications/ui_architecture.md: exact pins and client version.documentation/zomes/architecture_overview.md,documentation/hREA/integration-strategy.md,tests/DEPRECATED.md,CONTRIBUTING.md: earlier commits of this PR (0.7 notes, submodule trap, Tryorama).Related
Closes #134
Closes #154
Pre-open review
dev, no conflict (git rev-list HEAD..origin/devis 0). chore(deps): move the hREA submodule to happ-0.5.0-beta.1 #169 folded in. Pass.