canary 2026.06.08 release notes #

SYNOPSIS #

This release covers the set of changes from April 2026 (sha 98de7fb3d99f66647c90a0e96fb37f1f6be59b15) to June 8, 2026 (sha 80f99b2b3c61622d856d2be0f45f7fe4ae3065f8).
Work in this range focuses on the TestCase → Job migration, introduction of a new JobSpec IR + dependency resolver pipeline, a major status model refactor, improved workspace/view semantics, HPC batching enhancements, and substantial documentation and CI updates.

Migrated core execution terminology and types from TestCase to Job across the codebase.
Introduced :py:`_canary.job.BaseJob` and standardized queue/executor interactions around a job-phase lifecycle.
Backward compatibility is preserved: :py:`canary.TestCase`` remains available as an alias of :py:`canary.Job`, so most existing client code does not need to change.

Added :py:`_canary.ir.JobSpecIR` for generator outputs that contain unresolved dependency selectors.
Added :py:`_canary.jobspec.JobSpec` as the resolved, runnable spec format.
Added :py:`_canary.resolve_dependency` to resolve dependency patterns/selectors into concrete spec dependencies.

Replaced legacy status tuple concepts with: * :py:`_canary.status.Category` (PASS/FAIL/SKIP/CANCEL/NONE) * :py:`_canary.status.Outcome` (SUCCESS/FAILED/TIMEOUT/…) * Updated :py:`_canary.status.Status` API (category/outcome/reason/code).
Updated return-code aggregation and reporting logic to use the new model.

Workspace DB now stores specs as JSON (instead of pickled blobs).
Results schema updated to store job_state and new status fields; includes a migration helper for older DBs.

Added canary exec subcommand to execute a single job/spec directly.
Added view configuration controls: * New view settings (mode/when/only/name). * canary run --view key=value,... for per-run view behavior. * canary view refresh enhanced (mode/only/name).
Improved parser behavior: * Nested subparser lookup via Parser.get_subparser("a::b::c"). * New hooks: canary_cmdline_parse and canary_cmdline_modifyargs.

Extracted former builtin PYT support into a new plugin package: ``canary_pyt``.
Refactored canary_vvtest to a similar “model/loader/adapter/emitter” architecture.
Added builtin archive plugin: --archive NAME to tar/tgz selected artifacts post-run.
GPU selection improvements: * --gpu-backend default changed to none. * Added GPU selection config section and stricter backend selection logic.

Batch specification defaults updated (notably default batching node constraints).
Added explicit batch dependency construction from job dependency graphs.
Batch objects now behave as jobs (implementing BaseJob interface).
Added canary hpc log helper to page batch logs.

README updated to describe Canary as a general-purpose job/workflow execution framework.
Added a large Sphinx reveal.js presentation under docs/presentations/HL.
Expanded tutorials: * Added batch spec documentation and extending-canary tutorial series. * Improved language around workspace/view behavior and job terminology.
Added/updated API docs for new modules (job, jobspec, ir, resolve_dependency, util modules, etc.).

Updated pytest paths and CI execution order; added src/canary_pyt/tests.
Enabled Ruff security selection (S324) in both GitHub Actions and GitLab CI.
GitHub coverage steps currently commented out in workflow.
Added bin/collect-release-data.sh helper script to generate deterministic release inputs from local git history.

Added :py:`_canary.job.Job`, :py:`_canary.job.BaseJob`, :py:`_canary.job.JobPhase`, :py:`_canary.job.JobState`.
Updated conductor/executor/queue to operate on job phases and readiness rather than legacy testcase status states.
Updated job execution path: * robust setup/run/finish error handling; * explicit job event recording (submitted/started/finished) for queue executor.

Added :py:`_canary.ir.JobSpecIR` and :py:`_canary.resolve_dependency.resolve`.
Updated generator system: * generators now emit JobSpecIR/JobSpec rather than legacy UnresolvedSpec/ResolvedSpec. * dependency resolution is handled in a dedicated module (with parallel resolution support).
Workspace DB changes: * specs stored as JSON; * dependency reconstruction uses stored ordering to preserve edge positions; * migration routine for legacy results.status_state → results.job_state.

Implemented enum-based status model in :py:`_canary.status`.
Updated reporters (JUnit/Markdown/HTML/email/GitLab) to group and display based on new status fields.
Updated returncode computation to use status category/outcome and job completion.

New command: canary exec to execute a single job from a spec ID.
View improvements: * new ViewSettings and ResultsView abstractions; * configurable view creation (mode/when/only) and rebuild behavior.
Parser improvements: * nested subparser addressing via :: notation; * new hooks for command line parse/modify steps.

Added canary_pyt plugin: * loader parses directives from .pyt files without executing tests; * adapter applies directives into a model; * lock emitter generates JobSpecIR entries (including analyze/composite base job).
Refactored canary_vvtest to match the same architecture.
Added builtin archive plugin: * archives job artifacts into tar/tgz; de-duplicates payload.
GPU detection improvements: * updated selection logic; added config schema; clarified backend conflict behavior.
YAML example generator updated: * uses schema validation; * emits command directly; * template substitution via :py:`string.Template`.

Batch spec defaults updated and documented (including layout/duration/count/nodes constraints).
Introduced batch dependency construction and optional backend dependency usage.
Batch objects implement BaseJob interface and emit job events consistent with queue executor.
Added canary hpc log to display batch logs.

README refreshed with updated framing and improved developer instructions.
Added HL presentation in Sphinx reveal.js format.
Expanded tutorial and user docs: * clarified job terminology and updated examples; * added batch spec doc page; * added extending-canary tutorial series (job generator concepts, YAML generator worked example); * reorganized usage docs into “Integrations” sections (CMake/HPC/VVTest/GitLab).

GitHub workflow: * expanded pytest suites (canary_gitlab/canary_pyt/canary_vvtest ordering). * ruff: added --extend-select S324. * coverage steps commented out.
GitLab CI: * ruff-check updated to include --select S324.
Added bin/collect-release-data.sh utility for deterministic release data extraction.
Updated internal check command to run pytest suites in parallel and extend ruff checks across more directories.