feat: Phase 5 — GitHub Actions for lint + profile re-run

.github/workflows/README.md

@@ -0,0 +1,59 @@
+# `.github/workflows/`
+
+CI for `autolens_profiling`. Two workflows, deliberately split.
+
+## `lint.yml` — PR + push-to-main gate
+
+Runs on every PR and every push to `main`. CPU-only, target wall time under 5 minutes.
+
+What it checks:
+
+| Step | Purpose |
+|------|---------|
+| `ruff check .` | Pyflakes + pycodestyle + isort + pyupgrade + flake8-bugbear (see `ruff.toml`) |
+| `ruff format --check .` | Formatting parity with sister PyAutoLabs repos (black-compatible defaults) |
+| `python scripts/build_readme.py --check` | Dashboard idempotence — the auto-generated tables in every section README must match what `build_readme.py` would generate from the current `results/` artifacts. Catches the "forgot to rerun the dashboard generator after dropping a new result" class of bug |
+| `lychee` | Markdown link-rot across every `README.md` |
+| Smoke — one script per section | Runs `likelihood/imaging/mge.py`, `simulators/imaging.py`, and `searches/nautilus/simple.py` with `AUTOLENS_PROFILING_SMOKE=1`. Every profile script reads that env var at module top and exits 0 after the import + setup section. Catches import-graph breakage (broken `sys.path` injection, missing dependency, renamed module) without running the full profile |
+
+The smoke step does **not** produce real result artifacts — every script short-circuits before the JIT compile / sampling / FITS writes. If you need full smoke output, run `profile.yml` manually instead.
+
+## `profile.yml` — manual + on-release profile re-run
+
+Triggered by:
+
+- `workflow_dispatch` (manual via the GitHub UI). Optional `sections` input lets you scope a run to one of `likelihood`, `simulators`, `searches`, or any comma-separated combination. Leave blank to run everything.
+- `release: published` — when a new GitHub release is published.
+
+What it does:
+
+1. Runs every script under `likelihood/`, `simulators/`, and `searches/nautilus/`, producing JSON+PNG artifacts under `results/`. `continue-on-error: true` per section so a single regression doesn't block the dashboard refresh for the remaining 16+ scripts; failures emit a `::warning::` annotation and the matching dashboard cell will show `ERR`.
+2. Skips `simulators/point_source.py` in the simulator loop because its default `dataset_name="simple"` overwrites the Phase 1 likelihood input JSONs (see `simulators/README.md`). Run that one manually with a non-conflicting `dataset_name` when needed.
+3. Runs `python scripts/build_readme.py` to refresh every auto-generated table from the latest artifacts.
+4. Commits the diff back to `main` as `github-actions[bot]` with `[skip ci]` in the subject (prevents the lint workflow from re-triggering on the auto-generated commit).
+
+Hardware: GitHub-hosted `ubuntu-latest` (CPU). Expect Nautilus's `n_live=200` runs to take 30–60 minutes each on CPU; total job time can approach the 4-hour `timeout-minutes` budget on a full run. Self-hosted GPU runners can be added later as a separate job that appends `*_gpu*.json` artifacts to `results/` without restructuring this workflow — the dashboard's hardware-tier column extension (top-level README "Future enhancements") is the matching reader-side change.
+
+## How to trigger `profile.yml` manually
+
+From the GitHub UI:
+
+1. Repo → **Actions** → **profile** workflow.
+2. Click **Run workflow**.
+3. (Optional) Enter a sections filter, e.g. `likelihood,simulators`, or leave blank for all.
+4. Click **Run workflow**.
+
+Or via `gh`:
+
+```bash
+gh workflow run profile --repo PyAutoLabs/autolens_profiling -F sections=likelihood
+gh workflow run profile --repo PyAutoLabs/autolens_profiling  # all sections
+```
+
+## Design decisions (captured for future maintainers)
+
+- **CPU-only runners** to start. GitHub-hosted is free and the dashboard's CPU column is the most-requested baseline. Future GPU laptop / A100 columns require either self-hosted runners or external upload of `*_<hardware>_<version>.json` files; both are additive on this workflow's shape.
+- **No matrix across Python versions / OSes**. Lens-modelling profiling is Linux-only in practice, and a single Python version (3.12) keeps the matrix lean.
+- **No coverage reporting**. This repo has no unit tests by design — it's a scripts collection, and the smoke step + dashboard idempotence check are the practical equivalents.
+- **Why `[skip ci]` rather than path filters**: the lint workflow's smoke step does run scripts, and a commit that touches only `results/` and `README.md` files could in principle re-trigger the lint workflow (which is cheap, but pointless). Subject-line `[skip ci]` is the simplest fix and is honoured by GitHub Actions natively.
+- **`continue-on-error: true`** on each profile section, rather than fail-fast: a regression in one script shouldn't block the dashboard refresh for the other 16. The `::warning::` annotation surfaces the failure in the run UI and the dashboard cell will show `ERR` until the next successful refresh.

.github/workflows/lint.yml

@@ -0,0 +1,114 @@
+# lint.yml — PR / push-to-main gate.
+#
+# Cheap, fast (<5 min), CPU-only. Runs on every PR and on push to main.
+# Does NOT produce real result artifacts — the smoke step short-circuits
+# every profile script via AUTOLENS_PROFILING_SMOKE=1 (lands as part of
+# Phase 5; see scripts under likelihood/, simulators/, searches/nautilus/).
+#
+# Profile re-runs that actually produce results live in `profile.yml`.
+
+name: lint
+
+on:
+  pull_request:
+  push:
+    branches: [main]
+
+permissions:
+  contents: read
+
+jobs:
+  lint:
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    steps:
+      - name: Checkout autolens_profiling
+        uses: actions/checkout@v4
+        with:
+          path: autolens_profiling
+
+      - name: Checkout PyAutoConf
+        uses: actions/checkout@v4
+        with:
+          repository: PyAutoLabs/PyAutoConf
+          path: PyAutoConf
+      - name: Checkout PyAutoFit
+        uses: actions/checkout@v4
+        with:
+          repository: PyAutoLabs/PyAutoFit
+          path: PyAutoFit
+      - name: Checkout PyAutoArray
+        uses: actions/checkout@v4
+        with:
+          repository: PyAutoLabs/PyAutoArray
+          path: PyAutoArray
+      - name: Checkout PyAutoGalaxy
+        uses: actions/checkout@v4
+        with:
+          repository: PyAutoLabs/PyAutoGalaxy
+          path: PyAutoGalaxy
+      - name: Checkout PyAutoLens
+        uses: actions/checkout@v4
+        with:
+          repository: PyAutoLabs/PyAutoLens
+          path: PyAutoLens
+
+      - name: Set up Python 3.12
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+
+      - name: Install runtime dependencies
+        run: |
+          python -m pip install --upgrade pip
+          pip install numpy scipy matplotlib jax jaxlib
+          pip install nautilus-sampler==1.0.5
+          pip install numba
+          # autoconf / autofit / autoarray / autogalaxy / autolens are imported
+          # from the sibling checkouts via PYTHONPATH below — no install needed.
+
+      - name: Install lint tools
+        run: |
+          pip install ruff
+          # lychee is a Rust binary; fetch the prebuilt release.
+          curl -sSfL https://github.com/lycheeverse/lychee/releases/latest/download/lychee-x86_64-unknown-linux-gnu.tar.gz \
+            | tar -xz -C /usr/local/bin lychee
+
+      - name: ruff check
+        working-directory: autolens_profiling
+        run: ruff check .
+
+      - name: ruff format --check
+        working-directory: autolens_profiling
+        run: ruff format --check .
+
+      - name: build_readme.py --check (dashboard idempotence)
+        working-directory: autolens_profiling
+        run: python scripts/build_readme.py --check
+
+      - name: lychee — markdown link-rot
+        working-directory: autolens_profiling
+        run: |
+          # Conservative scope: only README.md files in this repo, exclude
+          # GitHub UI links that lychee tends to false-positive on.
+          lychee \
+            --exclude '^https://github\.com/.*/(issues|pull|commit|releases)/' \
+            --no-progress \
+            --accept '200..=299,429' \
+            --max-redirects 5 \
+            $(find . -name 'README.md' -not -path './results/*' -not -path './dataset/*')
+
+      - name: Smoke — one script per section
+        working-directory: autolens_profiling
+        env:
+          AUTOLENS_PROFILING_SMOKE: "1"
+          NUMBA_CACHE_DIR: /tmp/numba_cache
+          MPLCONFIGDIR: /tmp/matplotlib
+          PYTHONPATH: ${{ github.workspace }}/PyAutoConf:${{ github.workspace }}/PyAutoFit:${{ github.workspace }}/PyAutoArray:${{ github.workspace }}/PyAutoGalaxy:${{ github.workspace }}/PyAutoLens
+        run: |
+          # Each script reads AUTOLENS_PROFILING_SMOKE at module top and
+          # exits 0 immediately after the import + setup section. Catches
+          # import-graph breakage without running the full profile.
+          python likelihood/imaging/mge.py
+          python simulators/imaging.py
+          python searches/nautilus/simple.py

.github/workflows/profile.yml

@@ -0,0 +1,172 @@
+# profile.yml — manual + on-release profile re-run + dashboard refresh.
+#
+# Runs the real profile scripts (NumPy + JAX paths on CPU runners),
+# regenerates the JSON+PNG artifacts under results/, runs
+# `scripts/build_readme.py` to refresh the dashboard tables in every
+# README, and commits the diff back to main as `github-actions[bot]`
+# with `[skip ci]` in the subject to avoid loops.
+#
+# Design decisions (deliberate, captured here for future maintainers):
+#
+#   - **CPU-only on GitHub-hosted ubuntu-latest**. Self-hosted GPU runners
+#     can be added later as a separate job that appends *_gpu*.json / *_a100*.json
+#     artifacts without restructuring this workflow. The dashboard's
+#     hardware-tier column extension (see top-level README "Future
+#     enhancements") is the matching reader-side change.
+#
+#   - **Manual + on release tag**, not a weekly cron. Profile scripts on
+#     CPU take O(minutes) each; running them on every PR or every commit
+#     burns CI minutes for noise. Releases are the natural cadence for
+#     cross-version comparison anyway.
+#
+#   - **`github-actions[bot]` with `[skip ci]`**: needs `contents: write`.
+#     The `[skip ci]` tag prevents the lint workflow from re-running on the
+#     auto-generated commit, which would loop.
+#
+#   - **`continue-on-error` per script**: a single regression shouldn't
+#     block the dashboard refresh for the other 16 scripts. The build_readme
+#     step still runs over whatever artifacts landed, so partial data is
+#     better than no data.
+
+name: profile
+
+on:
+  workflow_dispatch:
+    inputs:
+      sections:
+        description: "Comma-separated section list to run (likelihood,simulators,searches) — leave blank for all"
+        required: false
+        default: ""
+  release:
+    types: [published]
+
+permissions:
+  contents: write
+
+jobs:
+  profile:
+    runs-on: ubuntu-latest
+    timeout-minutes: 240  # 4 hours — generous upper bound for CPU runs
+    steps:
+      - name: Checkout autolens_profiling
+        uses: actions/checkout@v4
+        with:
+          path: autolens_profiling
+          token: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Checkout PyAutoConf
+        uses: actions/checkout@v4
+        with:
+          repository: PyAutoLabs/PyAutoConf
+          path: PyAutoConf
+      - name: Checkout PyAutoFit
+        uses: actions/checkout@v4
+        with:
+          repository: PyAutoLabs/PyAutoFit
+          path: PyAutoFit
+      - name: Checkout PyAutoArray
+        uses: actions/checkout@v4
+        with:
+          repository: PyAutoLabs/PyAutoArray
+          path: PyAutoArray
+      - name: Checkout PyAutoGalaxy
+        uses: actions/checkout@v4
+        with:
+          repository: PyAutoLabs/PyAutoGalaxy
+          path: PyAutoGalaxy
+      - name: Checkout PyAutoLens
+        uses: actions/checkout@v4
+        with:
+          repository: PyAutoLabs/PyAutoLens
+          path: PyAutoLens
+
+      - name: Set up Python 3.12
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip
+          pip install numpy scipy matplotlib jax jaxlib
+          pip install nautilus-sampler==1.0.5
+          pip install numba
+
+      - name: Run likelihood/ scripts
+        if: ${{ inputs.sections == '' || contains(inputs.sections, 'likelihood') }}
+        working-directory: autolens_profiling
+        continue-on-error: true
+        env:
+          NUMBA_CACHE_DIR: /tmp/numba_cache
+          MPLCONFIGDIR: /tmp/matplotlib
+          PYTHONPATH: ${{ github.workspace }}/PyAutoConf:${{ github.workspace }}/PyAutoFit:${{ github.workspace }}/PyAutoArray:${{ github.workspace }}/PyAutoGalaxy:${{ github.workspace }}/PyAutoLens
+        run: |
+          set +e  # don't abort the whole step on one script's failure
+          for script in likelihood/imaging/*.py likelihood/interferometer/*.py \
+                        likelihood/point_source/*.py likelihood/datacube/delaunay.py; do
+            echo "::group::$script"
+            python "$script" || echo "::warning::$script failed — see log; dashboard cell will show ERR"
+            echo "::endgroup::"
+          done
+
+      - name: Run simulators/ scripts
+        if: ${{ inputs.sections == '' || contains(inputs.sections, 'simulators') }}
+        working-directory: autolens_profiling
+        continue-on-error: true
+        env:
+          NUMBA_CACHE_DIR: /tmp/numba_cache
+          MPLCONFIGDIR: /tmp/matplotlib
+          PYTHONPATH: ${{ github.workspace }}/PyAutoConf:${{ github.workspace }}/PyAutoFit:${{ github.workspace }}/PyAutoArray:${{ github.workspace }}/PyAutoGalaxy:${{ github.workspace }}/PyAutoLens
+        run: |
+          set +e
+          for script in simulators/imaging.py simulators/interferometer.py \
+                        simulators/cluster.py simulators/group.py simulators/multi.py; do
+            # point_source.py default dataset_name="simple" overwrites tracked
+            # likelihood inputs; skip in CI (see simulators/README.md). Manual
+            # runs with a non-default dataset_name remain supported.
+            echo "::group::$script"
+            python "$script" || echo "::warning::$script failed — see log; dashboard cell will show ERR"
+            echo "::endgroup::"
+          done
+
+      - name: Run searches/nautilus/ scripts
+        if: ${{ inputs.sections == '' || contains(inputs.sections, 'searches') }}
+        working-directory: autolens_profiling
+        continue-on-error: true
+        env:
+          NUMBA_CACHE_DIR: /tmp/numba_cache
+          MPLCONFIGDIR: /tmp/matplotlib
+          PYTHONPATH: ${{ github.workspace }}/PyAutoConf:${{ github.workspace }}/PyAutoFit:${{ github.workspace }}/PyAutoArray:${{ github.workspace }}/PyAutoGalaxy:${{ github.workspace }}/PyAutoLens
+        run: |
+          set +e
+          # Nautilus at n_live=200 on CPU takes O(minutes) to O(hours) per
+          # script. The full pair runs sequentially; expect 30-60 min total
+          # on the GitHub-hosted runner.
+          python searches/nautilus/simple.py || echo "::warning::nautilus/simple failed"
+          python searches/nautilus/jax.py    || echo "::warning::nautilus/jax failed"
+
+      - name: Refresh dashboard tables
+        working-directory: autolens_profiling
+        run: python scripts/build_readme.py
+
+      - name: Commit and push refreshed artifacts + READMEs
+        working-directory: autolens_profiling
+        run: |
+          git config user.name  "github-actions[bot]"
+          git config user.email "41898282+github-actions[bot]@users.noreply.github.com"
+          if [ -z "$(git status --porcelain)" ]; then
+            echo "No changes — exiting clean."
+            exit 0
+          fi
+          # Force-add tracked-but-gitignored result artifacts. Phase 4
+          # currently keeps results/{likelihood,simulators,searches}/
+          # gitignored; future hardware-tier work may flip a subset to
+          # tracked, in which case this `add -f` becomes redundant but
+          # harmless.
+          git add -f results/ likelihood/README.md likelihood/*/README.md \
+                     simulators/README.md searches/README.md \
+                     searches/nautilus/README.md README.md
+          git commit -m "chore: refresh profile artifacts + dashboard tables [skip ci]
+
+          Triggered by ${{ github.event_name }} (${{ github.event.release.tag_name || github.run_id }})."
+          git push origin HEAD:main

likelihood/datacube/delaunay.py

@@ -106,6 +106,16 @@
 # Instrument configuration
 # ---------------------------------------------------------------------------
 
+
+# AUTOLENS_PROFILING_SMOKE=1 short-circuit (Phase 5 / CI lint smoke).
+# Verifies the import graph + module-level setup succeeded without running
+# the full profiling pipeline. Skipped entirely when the env var is unset.
+import os as _smoke_os
+import sys as _smoke_sys
+if _smoke_os.environ.get("AUTOLENS_PROFILING_SMOKE") == "1":
+    print(f"[smoke] {__file__}: imports + module setup OK; exiting.")
+    _smoke_sys.exit(0)
+
 INSTRUMENTS = {
     "sma": {"pixel_scale": 0.1, "real_space_shape": (256, 256), "mask_radius": 3.0},
     "alma": {"pixel_scale": 0.05, "real_space_shape": (256, 256), "mask_radius": 3.0},

likelihood/imaging/delaunay.py

@@ -58,6 +58,16 @@
 # Instrument configuration
 # ---------------------------------------------------------------------------
 
+
+# AUTOLENS_PROFILING_SMOKE=1 short-circuit (Phase 5 / CI lint smoke).
+# Verifies the import graph + module-level setup succeeded without running
+# the full profiling pipeline. Skipped entirely when the env var is unset.
+import os as _smoke_os
+import sys as _smoke_sys
+if _smoke_os.environ.get("AUTOLENS_PROFILING_SMOKE") == "1":
+    print(f"[smoke] {__file__}: imports + module setup OK; exiting.")
+    _smoke_sys.exit(0)
+
 INSTRUMENTS = {
     "euclid": {"pixel_scale": 0.1},
     "hst": {"pixel_scale": 0.05},

likelihood/imaging/mge.py

@@ -67,6 +67,16 @@
 # Instrument configuration
 # ---------------------------------------------------------------------------
 
+
+# AUTOLENS_PROFILING_SMOKE=1 short-circuit (Phase 5 / CI lint smoke).
+# Verifies the import graph + module-level setup succeeded without running
+# the full profiling pipeline. Skipped entirely when the env var is unset.
+import os as _smoke_os
+import sys as _smoke_sys
+if _smoke_os.environ.get("AUTOLENS_PROFILING_SMOKE") == "1":
+    print(f"[smoke] {__file__}: imports + module setup OK; exiting.")
+    _smoke_sys.exit(0)
+
 INSTRUMENTS = {
     "euclid": {"pixel_scale": 0.1},
     "hst": {"pixel_scale": 0.05},