- Add build/register for a delegate to record each archive in the tag's
release with radicle-artifact, from their own machine
- Build the explorer bundle with config/default.json in build/register,
and register and seed it as a collection in the same release
- Record the CI build that fetch leaves in artifacts/; make build drop it
- Add build/attest for other delegates to attest a matching build
- Document registration, attestation and how to verify a download
- Add build/ci to build the archives of an annotated release tag
- Run it on the large_linux_arm Buildkite queue
- Keep the archives as artifacts and show checksums in an annotation
- Pass --load to docker, as a remote builder keeps the image in its cache
- Add build/fetch to download the archives of the CI release job
- Pick the job by its step key, so a failed test elsewhere does not hide it
- Add build/compare to check a build against another
- Document the CI build, the download and the comparison
- Pin the rust and alpine images by digest
- Check that the rust image matches rust-toolchain
- Install pinned zig 0.17.0 and cargo-zigbuild 0.23.4
- Install cargo-zigbuild before the source copy, to cache it
- Build on linux/arm64 on every host, with docker or podman
- Use 8-character commit IDs and read RADICLE_HTTPD_VERSION
- Name the tagged commit in the announcement, not a build ID
- Link macOS binaries with -dead_strip_dylibs and -rtlib=none
- Move signing out of build into build/sign, which checks checksums
- Fix Dockerfile lint warnings
- Rename RADICLE_VERSION to RADICLE_HTTPD_VERSION, apart from
heartwood's RADICLE_VERSION
- Add Unreleased sections to the changelog and API changes log, so
patches add their highlights as they land
- Replace the release patch with a signed tag that one delegate
pushes
- Stop pushing the tag from upload: the delegate who makes the tag
pushes it
- Remove build/changelog
- Add build/prepare to rename the Unreleased sections, in POSIX sh
and awk
- Ask a major release to set the target date to drop the previous
major
- Fix the signing key setup: use the bare key that build/tag
compares against
httpd:
- Serve the API at /api, not only under /v1
- Keep serving it at /api/v1, without a redirect, until the first
major release after explorer and the Open Graph worker switch
to /api
- Remove apiVersion: released explorers do not read it
- Add `httpd.releases` to `/info`, set by the artifacts feature
explorer:
- Remove the API version check, its config and the unused
compare-versions dependency
- Keep the node version in a plain parse error message
- Add `minimumSupportedVersion`, set to 0.28.0, and warn on nodes
whose major is older, treating 0.y as the major before 1.0
- Do not warn on builds without a release number
- Show a warning on the node page of an outdated node
- Mark an outdated node in the seed picker on every page
- Name the node version and the floor in parse errors
- Drop the `/node` fallback for httpd without `/info`, which
predates the floor
- Check the status, not `ResponseError`, so the Open Graph worker
can import compat without the fetcher
- Name the httpd release each shim serves, not a date
- Drop the mime shim for httpd 0.18.1 and the unused `lib/file.ts`
- Read the node from `/info` in repo views
- Pass the node and its features to repo views as one `node` param
- Show the releases tab and pages when the node supports releases,
and the counter only when there is a count
- Read a release count as release support on older httpd
- Test the flag and the fallback
docs:
- Restructure docs/api-changes.md: the supported versions, one
table of removals, fixed fields per change, and a link to ADR
0001 for the rules
- Say in ADR 0001 that each compat comment names what the other
side does
- Say in ADR 0001 when a capability needs a flag in /info
tests:
- Run e2e tests on httpd 0.30.0, above the floor, instead of 0.26.0
- Run the release tests on it too, so they cover the compat shims
- Skip only those that need release counts per scope on it
- Test a node without release support through a mocked `/info`
flag, instead of an httpd that predates the floor
Repos need not carry a `xyz.radicle.project` payload; the default
branch can come from `xyz.radicle.crefs` alone. Such repos got no
counts and no default branch, and failed schema validation in
explorer, which broke the parse of a whole listing.
httpd:
- Add top-level `cobs` with issue, patch and bucketed release counts
- Add top-level `defaultBranch`, qualified like the crefs payload
- Omit a count the node cannot determine instead of reporting zero
- Drop `head`: the tip is `refs[defaultBranch]`
- Keep counts in the project payload for older clients until a
major release
- Keep the project payload when a COB read fails
- Keep `meta.releases` a single count: delegate + other
- Log and skip canonical ref rules and refs that fail to read,
so one bad ref doesn't drop the default branch tip
explorer:
- Model `cobs` and `defaultBranch`, adapting older nodes to one shape
- Resolve the default branch from the crefs HEAD on older nodes
- Collect the shims for older httpd in http-client/lib/compat.ts,
each naming the httpd release that the client can drop it with
- Add `defaultBranchTip`, replacing `head`, with a fallback to the
legacy `meta.head`
- Make the project payload and its legacy meta optional
- Fall back to the RID for the repo name, and show a placeholder
for a missing name in repo cards and the header
- Hide a counter instead of reading a missing count as zero
- Fall back to a full-page check for "More" without counts
- Add a `repo.sourceUnavailable` route for repos without a
default branch or tip
- Browse explicit revisions and offer the branch selector without
a default branch
- List these repos in repo grids and on the landing page
- Say the node is older or built without releases
- Scope release pages in the client, so views never call the shims
- Read OG worker repos through the same compat shims
- Share the repo name and default branch helpers with the OG worker
CI:
- Add check:open-graph, a wrangler deploy dry run
- Run it in GitHub CI, Buildkite and check:ci
- Install explorer's dependencies first: the worker
resolves compat.ts imports, such as zod, from there
Tests:
- Test that a build without releases omits the count
- Test that one failing rule keeps the others
- Test the repo shape across httpd versions and the release shims
- Test `hasMorePages` and `repoName`
- Test repos without a project payload end to end
- Upgrade radicle-artifact to 0.23
- Use radicle-artifact Filters, Trust and Scope for releases
- Count only releases a user can browse: delegate + other
- Count releases only for a single repo, not in repo lists
- Add the scope=trusted|untrusted query
- Keep allAuthors for older clients until a major release
- Replace the All filter with Delegates and Others in the UI
- Open lists and releases on Delegates, also when empty
- Show the Others filter whenever it has content
- Name the scope in the empty artifacts message
- Filter by creator on older nodes, which ignore scope
- Page releases on the raw response length
- Warn on a release by a non-delegate
- Warn on a delegate's release while it shows others' artifacts
- Count only delegate attestations in a badge
- Warn about browser links added by non-delegates
- Say browser downloads are not checked against the CID
- Move the scope and warning logic to `src/lib/releases.ts`
- Test artifact scopes, scope routes, opening on Delegates and the
warnings
- Add ADR 0001: one SemVer version for httpd, explorer
supports every release from a minimum version, a 3-month
guideline to drop a replaced major, expand and contract,
and one API path at /api
- Treat 0.y as the major before 1.0
- Track API changes in docs/api-changes.md
- List API changes in httpd release notes
- Take the version from the release tag, not Cargo.toml
- Pass the version to the changelog script
- Link the compatibility rules from the README
- Point to the compat module from the backwards compatibility rules
A rescan can take tens of minutes on permissive seeds and blocked the
event loop meanwhile. radicle-node drops a subscriber once 8192 events
are queued, so every rescan disconnected us and the events emitted
until the resubscribe were lost.
Rescans and batch reindexes now run alongside the event loop, and the
subscription reconnects on its own task. A repo that an event adds
during a rescan only gets back into the seeded cache when its batch
runs, so it can be missing for up to RADICLE_SEARCH_BATCH_SECS.
Co-authored-by: Daniel Norman <daniel@norman.life>
Makes explorer and desktop sibling projects, as proposed in ADR 1.
This commit only moves files; paths in CI, scripts and docs are fixed
in the following commits.
Daniel Norman