stitch2d testing branch: hosted CI run (draft, do not merge)

.github/workflows/nightly.yml

@@ -8,6 +8,10 @@ on:
         description: "Teardown-crash diagnosis: also run tests/xdart WITHOUT the hard-exit wrapper, under gdb, and upload the native backtrace"
         type: boolean
         default: false
+      qt_keep_deletes:
+        description: "QT-TEST-LIFECYCLE opt-out: keep each test's deleteLater() objects instead of delivering them at its boundary (XDART_QT_DELIVER_DELETES=0)"
+        type: boolean
+        default: false
 
 jobs:
   core:
@@ -24,9 +28,12 @@ jobs:
     name: xdart-offscreen (${{ matrix.group }})
     strategy:
       fail-fast: false
-      # Same five bounded groups as pr.yml: the harness retains Qt objects to
-      # avoid unsafe DeferredDelete teardown, and one process over the whole
-      # tests/xdart tree grows that graph until theme repolishes time out.
+      # Same six bounded groups as pr.yml: windows that tests close without
+      # deleting stay alive, and one process over the whole tests/xdart tree
+      # grows that graph until theme repolishes time out. The files that
+      # change the application's theme or font restyle every live widget, so
+      # they run in their own process (inside the application group they took
+      # 9,572 s of its 10,360 s locally; alone, about 4 s).
       matrix:
         include:
           - group: execution-context
@@ -38,9 +45,21 @@ jobs:
           - group: scattering-other
             tests: tests/xdart/scattering --ignore-glob=tests/xdart/scattering/test_e[12346]*.py --ignore-glob=tests/xdart/scattering/test_p*.py
           - group: application
-            tests: tests/xdart --ignore=tests/xdart/scattering
+            tests: >-
+              tests/xdart --ignore=tests/xdart/scattering
+              --ignore=tests/xdart/test_app_appearance_options.py
+              --ignore=tests/xdart/test_app_font_scaling.py
+              --ignore=tests/xdart/test_text_selection_contrast.py
+              --ignore=tests/xdart/pages/test_synthetic_page_host.py
+          - group: appearance
+            tests: >-
+              tests/xdart/test_app_appearance_options.py
+              tests/xdart/test_app_font_scaling.py
+              tests/xdart/test_text_selection_contrast.py
+              tests/xdart/pages/test_synthetic_page_host.py
     runs-on: ubuntu-latest
-    timeout-minutes: 60
+    # Match PR headroom for the longest GUI group, including environment setup.
+    timeout-minutes: ${{ matrix.group == 'scattering-other' && 75 || 60 }}
     steps:
       - uses: actions/checkout@v6
       - uses: actions/setup-python@v6
@@ -51,6 +70,8 @@ jobs:
           QT_QPA_PLATFORM: offscreen
           XDART_SESSION_FILE: ${{ runner.temp }}/xdart-${{ github.run_id }}-${{ github.run_attempt }}-${{ matrix.group }}-session.json
           XDART_SETTINGS_FILE: ${{ runner.temp }}/xdart-${{ github.run_id }}-${{ github.run_attempt }}-${{ matrix.group }}-settings.ini
+          # Empty (the harness default: deliver) for scheduled runs; "0" only when dispatched with the opt-out.
+          XDART_QT_DELIVER_DELETES: ${{ inputs.qt_keep_deletes && '0' || '' }}
         # scripts/ci_pytest.py hard-exits with pytest's own verdict, skipping
         # the interpreter teardown where the deterministic offscreen segfault
         # (exit 139, post-summary) lives — see pr.yml for the full rationale.

.github/workflows/pr.yml

@@ -93,6 +93,13 @@ jobs:
         env:
           XDART_SESSION_FILE: ${{ runner.temp }}/xdart-pixi-browse-session.json
           XDART_SETTINGS_FILE: ${{ runner.temp }}/xdart-pixi-browse-settings.ini
+      # Average, Stitch and RSM publication on the runner's local NTFS; SMB needs an actual share.
+      - name: Publish and reopen finite results without directory descriptors
+        run: >-
+          pixi run --locked python -m pytest tests/core/test_windows_finite_publication.py -q -ra --timeout=300
+        env:
+          XDART_SESSION_FILE: ${{ runner.temp }}/xdart-pixi-finite-session.json
+          XDART_SETTINGS_FILE: ${{ runner.temp }}/xdart-pixi-finite-settings.json
 
   architecture-guards:
     runs-on: ubuntu-latest
@@ -108,9 +115,11 @@ jobs:
     name: xdart-offscreen (${{ matrix.group }})
     strategy:
       fail-fast: false
-      # The harness retains Qt objects to avoid unsafe DeferredDelete teardown.
-      # Bound that graph per process: application-wide theme changes otherwise
-      # repolish thousands of widgets left by unrelated tests.
+      # Windows that tests close without deleting stay alive. Bound that graph
+      # per process: application-wide theme changes otherwise repolish
+      # thousands of widgets left by unrelated tests. The files that change
+      # the application's theme or font run in their own group (nightly.yml
+      # has the measurement).
       matrix:
         include:
           - group: execution-context
@@ -122,9 +131,21 @@ jobs:
           - group: scattering-other
             tests: tests/xdart/scattering --ignore-glob=tests/xdart/scattering/test_e[12346]*.py --ignore-glob=tests/xdart/scattering/test_p*.py
           - group: application
-            tests: tests/xdart --ignore=tests/xdart/scattering
+            tests: >-
+              tests/xdart --ignore=tests/xdart/scattering
+              --ignore=tests/xdart/test_app_appearance_options.py
+              --ignore=tests/xdart/test_app_font_scaling.py
+              --ignore=tests/xdart/test_text_selection_contrast.py
+              --ignore=tests/xdart/pages/test_synthetic_page_host.py
+          - group: appearance
+            tests: >-
+              tests/xdart/test_app_appearance_options.py
+              tests/xdart/test_app_font_scaling.py
+              tests/xdart/test_text_selection_contrast.py
+              tests/xdart/pages/test_synthetic_page_host.py
     runs-on: ubuntu-latest
-    timeout-minutes: 45
+    # scattering-other took 49 minutes on hosted CI; leave room for setup/preflight.
+    timeout-minutes: ${{ matrix.group == 'scattering-other' && 75 || 45 }}
     steps:
       - uses: actions/checkout@v6
       - uses: actions/setup-python@v6

MIGRATION.md

@@ -4,8 +4,56 @@ The current distribution is **xdart**, with `xdart` and `xrd_tools` as its
 Python import packages. The legacy `ssrl_xrd_tools` import alias has been
 removed; update those imports to `xrd_tools`. See the
 [README installation guide](README.md#install) for current Pixi and pip setup.
-The notes below describe the earlier repository merger, including the
-historical `xrd-tools` distribution name and installation commands.
+The current upgrade checklist comes first; historical merger notes and older
+release records follow, including the former `xrd-tools` distribution name.
+
+## Upgrading from 1.1.5 to 1.2.0
+
+**1.2.0 is being prepared and is not yet published.** See the
+[release notes](RELEASE_NOTES.md) for the current feature summary and
+[limitations](docs/LIMITATIONS.md) for scientific and installation boundaries.
+The older release sections below describe their behavior at the time; the
+current README and installation guide are the starting points for a new install.
+
+1. **Update the environment that actually launches xdart.** Use
+   `pixi global update xdart` for a Pixi global installation, or the
+   [private-workspace update command](docs/INSTALLATION.md#updating-an-installer-workspace)
+   for the one-line installer. Keep the GUI's compiled dependencies on the
+   supported Pixi/conda-forge route; pip-only GUI installs are best-effort.
+   Restart xdart after updating.
+2. **Regenerate affected old seam-window outputs with Replace/Overwrite.**
+   This applies to Standard radial q/2θ data processed before the seam-window
+   correction. Append skips existing frames and can mix old truncated data with
+   corrected new data under the same provenance. There is no automatic data
+   migration. Retain a copy of the old output and reprocess the entire result
+   from raw inputs. See the [seam-q limitation](docs/LIMITATIONS.md#reprocess-outputs-made-before-the-seam-q-correction).
+3. **Use azimuth/χ range Auto for full-turn I(χ) and cakes.** A typed full turn
+   with a nonzero χ offset still has the documented W1 limitation. Exact 360°
+   Standard radial q/2θ windows now use Auto admission; this does not extend to
+   GI or Stitch. Extra seam-window integrators can also increase memory use.
+4. **Choose the new workspace modes explicitly.** **Stitch 1D**, **Stitch 2D**
+   and **RSM** are implemented GUI pages. Follow the README's
+   [Stitch](README.md#stitch-1d-and-stitch-2d) and
+   [RSM](README.md#reciprocal-space-mapping-rsm) instructions for raw inputs and
+   the appropriate calibration. Reopening a saved result does not reconstruct
+   the processing inputs or rerun it.
+5. **Expect stable result-folder name order.** Stitch/RSM rows are naturally
+   sorted from the start. The most recently produced result stays current but
+   may appear above older rows. No-selection fit batches follow the visible
+   order, which can change result-order trends and export order relative to a
+   previously captured arrival order. Select the intended results explicitly
+   when preparing comparisons or a fit batch.
+6. **Clear the Data Browser's filter to find other results.** Each page lists its
+   own mode's results by default; older or hand-named results without an
+   operation suffix appear once the field is cleared. A new Average writes
+   `<family>_int1d_average.nexus` or `<family>_int2d_average.nexus` beside any
+   1.1.5 `<family>_average.nexus`, which it no longer replaces; update scripts
+   that look for `_average.nexus`.
+
+There is no bulk conversion of processed files. Keep the original raw sources
+with a moved project when raw-image inspection or replay is needed. The
+historical API/format changes below remain relevant when upgrading from older
+versions, including the `ssrl_xrd_tools` → `xrd_tools` import rename.
 
 ## Repository merger (June 2026)