Repository navigation
CI modernization: current runners/actions, macOS ARM (supersedes #25) - #27
Open
Wid-L-Hacker wants to merge 42 commits into
Open
Wid-L-Hacker wants to merge 42 commits into
Wid-L-Hacker wants to merge 42 commits into
Conversation
…cannot write .github/workflows/ - to be renamed into place)
…can be renamed into place
rusqlite's load_extension gained a generic entry-point parameter (bound: Name), so passing a bare None no longer infers (E0283 at lib.rs:128 with rusqlite 0.40.2). Annotate None::<&str>; behavior unchanged (default entry point). Verified locally: cargo test --no-default-features passes in full against rusqlite 0.40.2 (2 integration tests + doc-test, 2026-10-17). The JDBC pom targeted Java 1.7, which modern JDKs refuse (javac 12+ dropped source/target 7). Bump maven.compiler.source/target to 1.8, the modern minimum. Local Maven verification is blocked by the sandbox JVM/proxy issue (runbook); CI will verify.
Drop libcurl4-openssl-dev from APT_DEPS (conflicts with libhts-dev's libcurl4-gnutls-dev on Ubuntu 24.04; apt exit 100, conflict verified against noble package metadata) and drop the invalid pip --system flag from the lint/test/linux-x86-64-so installs (pip: no such option, exit 2; verified with Ubuntu 24.04's pip 24.0). Staged at the root because the connector cannot write workflow paths; takes effect when renamed over .github/workflows/build.yml.
…n be renamed into place
CMake 4 removed compatibility with projects declaring cmake_minimum_required() below 3.5; the pinned SQLiteCpp fetched via FetchContent declares 3.1, so configure fails outright under CMake 4 (e.g. the macos-15 runner image, CMake 4.4.4) in a matter of seconds. Setting CMAKE_POLICY_VERSION_MINIMUM=3.5 before FetchContent raises the policy-version floor for the fetched subprojects. Verified locally: CMake 4.4.4 configure+build green with the change (and the failure reproduced without it); CMake 3.28 configure unaffected.
The lint deps step's system-wide pip install fails on the current ubuntu-24.04 runner image (exit 1 in ~15s, both repos, while the identical commands succeed on a stock Ubuntu 24.04 image and the test jobs' smaller pip installs succeed on the runner). Install the lint toolchain (pre-commit, black, flake8, pylint, test deps) into an isolated venv instead of fighting the runner image's system-python state, and put it on PATH for the later steps via GITHUB_PATH. Adds python3-venv to the apt line so the venv can be created. Verified in a stock ubuntu:24.04 container: apt, venv creation, pip install, and pre-commit all succeed.
…kout failure
- pom.xml: define a default <revision> property (0.9.4-SNAPSHOT). Maven >= 3.9.10 (the ubuntu-24.04 runner's 3.9.11) refuses to read the POM at all when ${revision} is undefined; -Drevision override still works. Exact CI error reproduced locally pre-fix with Maven 3.9.11; POM parses post-fix.
- bindings/rust/cargo + test/cargo_smoke_test/cargo (identical wrappers): git describe --tags dies under set -e in tag-less checkouts (actions/checkout fetches no tags), failing cargo_test in 0.00s. Fall back to the latest release version with the short SHA as prerelease suffix (valid semver) when no tag is found. Verified in a tag-less shallow clone: full cargo test passes; tagged-checkout behavior unchanged.
Mike's formatter ruling (2026-10-17): one-time reformat with current tools instead of pinning era formatter versions. Tool versions (the same unpinned installs the CI lint venv makes): clang-format 18.1.3, black 26.10.1, flake8 7.4.1, pylint 4.1.3, pre-commit 4.7.0. - clang-format: src/genomicsqlite.cc, loaders/vcf_into_sqlite.cc - black: bindings/python/genomicsqlite/__init__.py, bindings/python/version.py, test/genomicsqlite_smoke_test.py, test/test_bam.py, test/test_ext.py, test/test_gri.py, test/test_vcf_lines.py - flake8 E226 (missing whitespace around arithmetic operator): test/test_txdb.py, test/test_gri.py - pylint R0917 too-many-positional-arguments (new in pylint 3.3) flags three established public API functions in bindings/python/genomicsqlite/__init__.py; added to the existing -d list in .pre-commit-config.yaml next to the already-disabled sibling check too-many-arguments. No signature changes (positional callers must keep working). - pylint E0401 in bindings/python/setup.py: line-targeted import-error disable; setuptools is present by construction when setup.py runs, but is not installed in the lint venv. Verified locally in the CI lint-venv shape: pre-commit run --all-files green (all hooks pass); Release build + smoke/capi/pytest ctest entries 100% passed.
…-format 1.9 cannot run on JDK >= 16); GJF reformat of GenomicSQLite.java + JdbcDriver.java (Mike ruling 2026-10-17)
…e-only attempt via create_or_update_file)
…e-only attempt via push_files)
Connector content-writes reset file modes to 100644 (verified empirically: both create_or_update_file and push_files leave the mode at 100644, and every file the connector has rewritten on this branch lost its 100755 bit). The cargo wrapper therefore cannot be executed by path (EACCES/exit 126). - CMakeLists.txt: invoke the wrapper through bash in the cargo_test ctest registration (content of the wrapper unchanged). - build.yml STAGED at repo root (rename into .github/workflows/ pending): all six by-path wrapper invocations (rust lint step, linux-x86-64-so and macOS-arm64-dylib 'test rust crate' steps) now go through bash.
…oot build.yml) can be renamed into place
The genomicsqlite CLI wrapper passed its startup commands (.load, .open, tuning PRAGMAs) to the sqlite3 shell as -cmd arguments, including SQL -cmds, under -bail. Since SQLite 3.52 the shell exits immediately -- successfully, without reading any input -- whenever -bail is combined with an SQL -cmd, so on current shells (e.g. Homebrew's) the wrapper silently did nothing. Assemble the startup sequence (.load/.open/.once/tuning/.prompt, plus the interactive version banner and .headers) into a temporary -init file instead; no SQL remains in -cmd (there are no -cmd args at all). The tuning SQL gains a terminating semicolon, which -init file context requires. os.execvp replaces the process, so the temp file is left for the OS reaper. Verified: positional/stdin/interactive shapes behave identically under sqlite 3.53.4 and 3.45.1 shells.
bundle_libgenomicsqlite() was defined only for linux+x86_64 and macos+x86_64, but open() calls it unconditionally under default features, so the crate never compiled with default features on Apple Silicon (E0425) or linux/aarch64. Drop the target_arch gates so bundling is keyed by OS; the bundled bytes are the per-platform lib placed beside the crate.
… names) for rename to .github/workflows/build.yml
…entory, fallback to library search, legible load errors)
…stage (libgenomicsqlite-linux-x86_64.so)
… historical C++ ABI; Alma stock GCC defaults to ABI=1, breaking ABI=0 loader/test executables)
…lite3 shell first on PATH) for rename to .github/workflows/build.yml
macOS CI is ARM-only (R1), so the supported bundled set is three targets: linux x86_64, linux aarch64, macOS aarch64. macos/x86_64 now takes the catch-all fallback (system library search) like any other unsupported target, with no build.rs warning.
Per the project's SQLite version policy: the bundled/portable build's pin tracks the newest upstream release at each maintenance touch (3.54.1, released 2026-10-16). Runtime floor (3.31.0) and version gates unchanged. Rehearsed locally in AlmaLinux 8 (udocker): builder builds clean, sqlite_version()=3.54.1; libsqlite3.so.0 GLIBC ceiling stays exactly 2.28; step-7-shaped ctest 5/5 with the paired portable shell.
… .github/workflows/
Mike approved adopting ruff's stricter isort grouping on 2026-10-18 as a low-risk cleanup, superseding the tool swap's initial deliberate non-adoption. ruff.toml now selects I001, and the tree is import-sorted with ruff 0.17.1 --fix: pure import regrouping/sorting in 11 Python files, no semantic changes.
…est Java floor 1.8
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Modernize the
buildworkflow so CI can run again: jobs targeted retired ubuntu-20.04 and macOS-11 images, actions/checkout@v2, and v2 artifact actions (shut off).-march=ivybridgedropped; brew paths made prefix-relative--break-system-packages(PEP 668)Local validation (2026-10-17): C++ extension builds clean (Debug+Release); smoke + C-API tests pass; Python suite 29/29. Known-red expectations for the
testjob, both pre-existing and tracked as follow-ons: cargo_test fails on Rust dependency drift (unpinned rusqlite now resolves to a version whose load_extension signature changed), and JdbcTests is expected to fail because the JDBC pom targets Java 1.7, which modern JDKs reject.Prepared by @Wid-L-Hacker (Conrad) under @mlin's supervision.