The docker image build compiled secp256k1 with -march=native and failed.
waku.nimble's getNimParams reads the NIM_PARAMS env var, not NIMFLAGS, so
the -d:disableMarchNative passed via NIMFLAGS on the make line never
reached `nim c` and config.nims applied -march=native. #3916 added
NIM_PARAMS to ci.yml but missed container-image.yml; set it here too.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Stripping nim from the lock breaks nimble 0.22.3: 41 packages list nim in
their dependencies, so removing the nim entry makes resolution fail with
"key not found: nim". nimble already has the right knob — --useSystemNim
("Use system nim and ignore nim from the lock") — which skips the locked
nim (no download, no checksum) while still verifying every other locked
dependency, and uses the nim the CI already installed.
This is the proper fix for the Windows checksum mismatch: nim's tree
checks out differently per platform, so its locked checksum is unreliable
there. Verified locally with nimble 0.22.3 (exit 0, system nim used, lock
left untouched).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Restores a fix that was dropped in e5b46e5e and re-exposed when cda01971
reintroduced nim/nimble into nimble.lock: the nim package never matches
its Linux-computed checksum on Windows because nim's own .gitattributes
forces line endings on parts of its tree, so the SHA1 differs regardless
of core.autocrlf. autocrlf alone therefore cannot fix the nim mismatch.
Remove nim/nimble from the lock before `nimble setup` (they are provided
by the toolchain installed in CI, not real project deps). Added to the
Windows workflow, which never had it, and restored in ci.yml's setup
steps since the cache-key bump now forces those to run on a fresh tree.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The Linux nimbledeps cache held a corrupt lsquic/nimblemeta.json (invalid
JSON), so restoring it made `make wakunode2` fail parsing it while a fresh
lsquic download in the same run was fine. The key is derived from
nimble.lock, which is unchanged, so the poison persisted across runs.
Add a v2 salt to the key to force a clean rebuild of the cache.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The earlier autocrlf fix ran in a pwsh step, which only configures Git
for Windows. nimble clones its dependency packages from the MSYS2 shell,
whose git reads a separate global config, so autocrlf stayed enabled
there. That kept altering dependency source trees, breaking their SHA1
match against nimble.lock (e.g. the nim package), so nimble re-downloaded
and hung. Repeat the setting in an MSYS2-shell step so nimble's own git
clones produce Linux-identical trees.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Windows Git defaults to core.autocrlf=true. When nimble clones dependency
packages into nimbledeps/, the LF→CRLF conversion changes the SHA1 of the
source tree relative to the Linux-computed checksums in nimble.lock.
nimble then treats every cached package as invalid and re-downloads on each
invocation; in PR #3919 the re-download of bearssl_pkey_decoder during
`nimble wakunode2` hung until the 6h job timeout.
Set core.autocrlf=false / core.eol=lf globally before checkout so every
package nimble clones keeps LF endings and matches the Linux SHA1s. This
also obsoletes the targeted sed patch for the nim checksum, which is
removed.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(node-info): real Version + new Commit in Nix/lgpm builds
getNodeInfo Version returned "n/a" on Nix-built libs (and lgpm
releases built from the flake) because nix/default.nix never passed
-d:git_version. A flake sandbox has no .git, so git describe is
impossible; derive the semver from waku.nimble's version field plus
the flake short commit, and expose the full commit SHA via a new
Commit node info id.
- waku_state_info: add Commit to NodeInfoId + dispatch git_commit
- waku_node: add git_commit {.strdefine.} (default "n/a")
- node start logs ("Starting Waku node" / "Running nwaku node") now
print commit = git_commit alongside version
- Makefile: inject -d:git_commit (full SHA), mirrors docker label
- nix/default.nix: accept gitVersion/gitCommit, pass as nim defines
- flake.nix: gitVersion = <nimble version>-g<shortRev>, gitCommit = rev
- CI version-check (PR only): ancestor-aware `git describe --tags
--abbrev=0` vs PR HEAD, base-version compare, so waku.nimble is kept
current early and a new tag never breaks in-flight PRs
- release-assets.yml: gate build/upload on a verify-version job
asserting tag base == waku.nimble (RC tags allowed), so a mismatched
tag publishes no artifacts
- docs: prepare_release.md explains the bump-before-tag requirement
Refs: status-im/infra-logos#4Closes: logos-messaging/logos-delivery#3884
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* docs: simplify
* chore: update version in waku.nimble
* fix(node-info): remove Commit node info field
Drop the newly added Commit (full SHA) node info id and its
git_commit compile-time define plumbing across Makefile, flake.nix
and nix/default.nix; revert the start/run log lines to version only.
The PR now solely fixes the getNodeInfo Version regression.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* chore(nix): align git_version format closer to Makefile
Adds the `v` prefix and uses a 6-char SHA so Nix-built nodes report
e.g. `v0.38.1-g52e980`, matching the shape of `git describe --abbrev=6
--always --tags` aside from the unreachable commit-count segment (tag
metadata isn't exposed through the flake input protocol).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* allow create libwaku pkg
* fix Makefile create library extension libwaku
* make sure libwaku is built as part of assets
* Makefile: avoid rm libwaku before building it
* properly format debian pkg in gh release workflow
* waku.nimble set dylib extension correctly
* properly pass lib name and ext to waku.nimble
* Bump commit for vendor wakurlnv2contract
* Update RLN registration proc for contract updates
* add option to runAnvil for state dump or load with optional contract deployment on setup
* Code clean up
* Upodate rln relay tests to use cached anvil state
* Minor updates to utils and new test for anvil state dump
* stopAnvil needs to wait for graceful shutdown
* configure runAnvil to use load state in other tests
* reduce ci timeout
* Allow for RunAnvil load state file to be compressed
* Fix linting
* Change return type of sendMintCall to Futre[void]
* Update naming of ci path for interop tests
* properly pass userMessageLimit to OnchainGroupManager
* waku.nimble 2.2.4 Nim compiler
* rm stew/shims/net import
* change ValidIpAddress.init with parseIpAddress
* fix serialize for zerokit
* group_manager: separate if statements
* protocol_types: add encode UInt32 with zeros up to 32 bytes
* windows build: skip libunwind build and rm libunwind.a inlcusion step
* bump nph to overcome the compilation issues with 2.2.x
* bump nim-libp2p to v1.10.1
* avoid compile TRACE level to reduce libwaku size
* waku_rln_relay/constants.nim: avoid adding constant seq that is used in tests only
* ci:yml USE_LIBBACKTRACE=0 to force -d:debug when running tests
* undo apt install libpcre (not circuit-relay related.)
* nat.nim: protect against possible exceptions when calling getExternalIP
* new external CLI argument, isRelayClient
* waku factory change to mount circuit hop proto by default
* waku_node: move autonat_service to a separate module
