arrayview 0.26.3__tar.gz → 0.27.2__tar.gz

{arrayview-0.26.3 → arrayview-0.27.2}/.mex/AGENTS.md

@@ -1,7 +1,7 @@
 ---
 name: agents
 description: Lightweight pointer to ROUTER.md for task routing.
-last_updated: 2026-05-16
+last_updated: 2026-06-19
 ---
 
 # arrayview

{arrayview-0.26.3 → arrayview-0.27.2}/.mex/ROUTER.md

@@ -24,7 +24,7 @@ edges:
     condition: when starting a task — check the pattern index for a matching pattern file
   - target: ../DESIGN.md
     condition: when the task touches design philosophy — new features, UI changes, mode additions, or interaction model decisions
-last_updated: 2026-06-05
+last_updated: 2026-06-19
 ---
 
 # arrayview — Router

{arrayview-0.26.3 → arrayview-0.27.2}/.mex/SETUP.md

@@ -1,6 +1,6 @@
 # Setup — Populate This Scaffold
 
-Last reviewed: 2026-05-16
+Last reviewed: 2026-06-19
 
 This file contains the prompts to populate the scaffold. It is NOT the dev environment setup — for that, see `context/setup.md` after population.
 

{arrayview-0.26.3 → arrayview-0.27.2}/.mex/SYNC.md

@@ -1,6 +1,6 @@
 # Sync — Realign This Scaffold
 
-Last reviewed: 2026-05-16
+Last reviewed: 2026-06-19
 
 ## Recommended: Use sync.sh
 

{arrayview-0.26.3 → arrayview-0.27.2}/.mex/context/conventions.md

@@ -18,7 +18,7 @@ edges:
     condition: when applying frontend conventions (section separators, reconcilers, command registry)
   - target: context/render-pipeline.md
     condition: when applying render pipeline conventions (pipeline order, cache patterns, LUT usage)
-last_updated: 2026-05-16
+last_updated: 2026-06-19
 ---
 
 # Conventions

{arrayview-0.26.3 → arrayview-0.27.2}/.mex/context/decisions.md

@@ -14,7 +14,7 @@ edges:
     condition: when a decision relates to technology choice
   - target: context/frontend.md
     condition: when a decision relates to the frontend architecture or viewer modes
-last_updated: 2026-05-16
+last_updated: 2026-06-19
 ---
 
 # Decisions

{arrayview-0.26.3 → arrayview-0.27.2}/.mex/context/lifecycle.md

@@ -16,7 +16,7 @@ edges:
     condition: when component boundaries or display routing need broader context
   - target: context/stack.md
     condition: when VS Code, FastAPI, WebSocket, or packaging details are needed
-last_updated: 2026-06-05
+last_updated: 2026-06-19
 ---
 
 # Lifecycle

{arrayview-0.26.3 → arrayview-0.27.2}/.mex/context/project-state.md

@@ -25,8 +25,9 @@ last_updated: 2026-06-05
 - Colormap picker: `c` opens an expanded colorbar-island grid without changing the colormap; subsequent `c` taps cycle, hover/hjkl/arrows live-preview, Enter/click commits, Esc cancels, and auto-dismiss pauses while hovered
 - Cold-start loading spinner in VS Code and native shell
 - Tool menu (`/` menu) supports multi-select where allowed: spacebar toggles tools, Enter applies selection. Mutual exclusion enforced (ROI ↔ Segmentation, and overlay/vectorfield ↔ everything else). Cursor indicator shows focused tile via yellow background + left accent bar
-- Dynamic island renders sections for all active plugins simultaneously (qMRI pills + ROI shapes/stats separated by divider), replacing the old single-plugin priority chain
-- ROI mode works alongside qMRI: drawing on any pane mirrors the ROI to all panes in real-time via per-pane overlay canvases; stats fetched per parameter map and shown as sub-rows in the island
+- Dynamic island renders sections for active plugins such as qMRI and segmentation, replacing the old single-plugin priority chain
+- ROI mode is a lightweight measurement tool: `Shift+R` shows/hides session ROIs, defaults to true circles, uses slim colorbar controls for shape/stats/hide/clear, and manages stats/rename/delete/export from the ROI manager modal
+- ROI mode works alongside qMRI: drawing on any pane mirrors the ROI to all panes in real-time via per-pane overlay canvases; stats are fetched per visible parameter map and shown in the ROI manager
 - qMRI map toggle (`_islandToggleQmriMap`) fade animation now covers dimbar and array-name in addition to panes and colorbars
 - Segmentation menu shares the ROI layout (yellow accent, magnifier action icon, common `#export-overlay` modal). Pre-activation shows a pulsing "nnInteractive · connecting" row that morphs into the normal shape toolbar once `/seg/activate` resolves
 - Overlays are a plugin tile (`OV`): per-overlay row with colour swatch, editable label, eye toggle, × delete; shared opacity slider; `+ add overlay` opens a filesystem picker rooted at the launched file's directory
@@ -44,7 +45,7 @@ last_updated: 2026-06-05
 - Crossfade animation added to rotate/transpose transitions
 - `--version` flag and version string in help overlay
 - In-viewer array picker for multi-array `.mat` and `.npz` files
-- ROI stats tooltips for qMRI canvases (per-parameter-map sub-rows in the island)
+- ROI manager modal for qMRI canvases (per-parameter-map stats rows, CSV export, label-mask export)
 - Invocation lifecycle contract defined and hardened: local VS Code CLI launches use transient daemon shutdown, remote/tunnel launches remain persistent only where transport requires it, URL webview backend checks run in the extension host against `/ping`, URL panel disposal releases all sessions encoded in the viewer URL, quick viewer connect/disconnect races detected by monotonic connection counter, ambiguous multi-window tunnel routing fails closed, stale viewer SID retry state cleared on WebSocket disconnect, bundled opener extension rebuilt as v0.14.12
 - Hover info wrong values fixed
 - Native launcher startup restored

{arrayview-0.26.3 → arrayview-0.27.2}/.mex/context/render-pipeline.md

@@ -20,7 +20,7 @@ edges:
     condition: when writing new render functions or extending the pipeline
   - target: patterns/debug-render.md
     condition: when a render produces wrong output or the wrong cache is hit
-last_updated: 2026-05-16
+last_updated: 2026-06-19
 ---
 
 # Render Pipeline

{arrayview-0.26.3 → arrayview-0.27.2}/.mex/context/setup.md

@@ -13,7 +13,7 @@ edges:
     condition: when specific technology versions or library details are needed
   - target: context/architecture.md
     condition: when understanding how components connect during setup
-last_updated: 2026-05-16
+last_updated: 2026-06-19
 ---
 
 # Setup
@@ -62,7 +62,7 @@ No `.env` file is needed. All env vars are optional overrides; the server runs w
 
 **Port already in use:** The server binds to a random free port by default. If you see a port conflict, check for orphan `uvicorn` processes: `lsof -i :<port>`. Kill with `kill -9 <PID>`. The server is supposed to shut down automatically when all viewer windows close.
 
-**VS Code extension not loading:** Signal-file IPC can fail if the extension version is mismatched. Check `_VSCODE_EXT_VERSION` in `_vscode.py` matches the installed extension.
+**VS Code extension not loading:** Signal-file IPC can fail if the extension version is mismatched. Check `_VSCODE_EXT_VERSION` in `_vscode_extension.py` matches the installed extension.
 
 **Playwright tests fail on CI:** Ensure `uv run playwright install chromium` has been run. Browser tests require a display; on headless servers they need `DISPLAY=:0` or Xvfb.
 

{arrayview-0.26.3 → arrayview-0.27.2}/.mex/context/stack.md

@@ -14,7 +14,7 @@ edges:
     condition: when understanding how to use a technology in this codebase
   - target: context/architecture.md
     condition: when understanding how a library fits into the overall system
-last_updated: 2026-04-29
+last_updated: 2026-06-19
 ---
 
 # Stack

{arrayview-0.26.3 → arrayview-0.27.2}/.mex/patterns/INDEX.md

@@ -1,5 +1,7 @@
 # Pattern Index
 
+Last reviewed: 2026-06-19
+
 Lookup table for all pattern files in this directory. Check here before starting any task — if a pattern exists, follow it.
 
 <!-- This file is populated during setup (Pass 2) and updated whenever patterns are added.

{arrayview-0.26.3 → arrayview-0.27.2}/.mex/patterns/README.md

@@ -1,6 +1,6 @@
 # Patterns
 
-Last reviewed: 2026-05-16
+Last reviewed: 2026-06-19
 
 This folder contains task-specific guidance — the things you would tell your agent if you were sitting next to it. Not generic instructions. Project-specific accumulated wisdom.
 

{arrayview-0.26.3 → arrayview-0.27.2}/.mex/patterns/add-server-endpoint.md

@@ -16,7 +16,7 @@ edges:
     condition: when the endpoint triggers a render (slice, mosaic, projection)
   - target: context/conventions.md
     condition: for lazy import rules and Verify Checklist
-last_updated: 2026-05-16
+last_updated: 2026-06-19
 ---
 
 # Add Server Endpoint

{arrayview-0.26.3 → arrayview-0.27.2}/.mex/patterns/animation-verify.md

@@ -16,7 +16,7 @@ triggers:
 edges:
   - target: patterns/frontend-change.md
     condition: always — animation changes are always frontend changes
-last_updated: 2026-05-16
+last_updated: 2026-06-19
 ---
 
 # Animation Verify

{arrayview-0.26.3 → arrayview-0.27.2}/.mex/patterns/debug-render.md

@@ -22,7 +22,7 @@ edges:
     condition: only after Step 1 points to `_viewer.html` rather than the Python pipeline
   - target: context/conventions.md
     condition: for the Verify Checklist after fixing
-last_updated: 2026-05-16
+last_updated: 2026-06-19
 ---
 
 # Debug Render

{arrayview-0.26.3 → arrayview-0.27.2}/.mex/patterns/debug-vscode-extension-python.md

@@ -12,7 +12,7 @@ triggers:
 edges:
   - target: context/conventions.md
     condition: when making code changes after confirming the failure mode
-last_updated: 2026-05-16
+last_updated: 2026-06-19
 ---
 
 # Debug VS Code Extension Python

{arrayview-0.26.3 → arrayview-0.27.2}/.mex/patterns/extend-compare-mode.md

@@ -13,7 +13,7 @@ edges:
     condition: when the work is primarily a local `_viewer.html` edit
   - target: context/render-pipeline.md
     condition: when compare changes require backend render or diff parameter changes
-last_updated: 2026-05-13
+last_updated: 2026-06-19
 ---
 
 # Extend Compare Mode

{arrayview-0.26.3 → arrayview-0.27.2}/.mex/patterns/extract-server-route-module.md

@@ -11,7 +11,7 @@ edges:
     condition: when deciding whether `_server.py` should keep or delegate a route cluster
   - target: context/project-state.md
     condition: when the extraction changes the active architecture workstream
-last_updated: 2026-05-16
+last_updated: 2026-06-19
 ---
 
 # Extract Server Route Module

{arrayview-0.26.3 → arrayview-0.27.2}/.mex/patterns/frontend-change.md

@@ -16,7 +16,7 @@ edges:
     condition: for shared conventions and the Verify Checklist
   - target: patterns/debug-render.md
     condition: when the change produces wrong visual output
-last_updated: 2026-05-16
+last_updated: 2026-06-19
 ---
 
 # Frontend Change

{arrayview-0.26.3 → arrayview-0.27.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: arrayview
-Version: 0.26.3
+Version: 0.27.2
 Summary: Fast multi-dimensional array viewer
 Project-URL: Home, https://github.com/oscarvanderheide/arrayview
 Project-URL: Source, https://github.com/oscarvanderheide/arrayview
@@ -38,20 +38,20 @@ Description-Content-Type: text/markdown
 
 _This is my array viewer. There are many like it, but this one is mine._ 
 
-And I think you might like it too. Fast, responsive scrolling
-through slices of multi-dimensional arrays. Feature rich with a modern, minimal UI that only shows what needs to be shown.
+Arrayview lets you scroll through multi-dimensional arrays.
 
-It works wherever you work. The shell, scripts (Python/Julia/Matlab), or Jupyter notebooks. Local or remote. Easy to install.
+Open it from the shell, from Python, Julia, or Matlab, or inside a Jupyter notebook. Use it locally or over SSH. 
 
-VS Code users can even open the viewer just by clicking an array in the explorer tab. 
+If you work in VS Code, you can open arrays directly from the explorer; with Remote SSH, it works the same way.
 
-And if you're using the VS Code remote extension to connect to a remote, it just works—no additional setup required.
+It is meant to feel simple but there's more to it than meets the eye. 
 
 Curious? Give it a try with
 ```bash
 uvx arrayview your_array.npy
 ```
-Explore, there's more to it than meets the eye. 
+
 
 Check the [docs](https://oscarvanderheide.github.io/arrayview/) to learn more. 
 
+**Warning**: Arrayview is still under active development. Things may break or change without warning.

arrayview-0.27.2/README.md

@@ -0,0 +1,21 @@
+# <img src="docs/logo.png" height="36"> arrayview
+
+_This is my array viewer. There are many like it, but this one is mine._ 
+
+Arrayview lets you scroll through multi-dimensional arrays.
+
+Open it from the shell, from Python, Julia, or Matlab, or inside a Jupyter notebook. Use it locally or over SSH. 
+
+If you work in VS Code, you can open arrays directly from the explorer; with Remote SSH, it works the same way.
+
+It is meant to feel simple but there's more to it than meets the eye. 
+
+Curious? Give it a try with
+```bash
+uvx arrayview your_array.npy
+```
+
+
+Check the [docs](https://oscarvanderheide.github.io/arrayview/) to learn more. 
+
+**Warning**: Arrayview is still under active development. Things may break or change without warning.

{arrayview-0.26.3 → arrayview-0.27.2}/docs/measurement.md

@@ -2,15 +2,19 @@
 
 ## ROI
 
-`Shift+R` opens ROI options. Draw on the canvas to measure a region — live stats (mean, std, min, max) appear in the floating panel.
+`Shift+R` shows or hides ROI mode. Hiding ROI mode keeps the session's ROIs.
 
-Shapes: rectangle, ellipse, freehand, flood fill. Switch via the shape icons in the panel.
+Draw on the canvas to measure a region. The default shape is a circle.
+
+Shapes: circle, rectangle, freehand, flood fill. Switch via the colorbar controls.
 
 For flood fill, `[` / `]` adjusts tolerance.
 
 Click an existing ROI to select it. `Delete` / `Backspace` removes it.
 
-`N` exports the ROI mask as `.npy`.
+`Stats` opens the ROI manager: rename/delete ROIs, adjust extent, export CSV, or export a label mask.
+
+`N` exports the active ROI or segmentation mask as `.npy`.
 
 ## Ruler
 

{arrayview-0.26.3 → arrayview-0.27.2}/docs/viewing.md

@@ -38,6 +38,10 @@ Rendered as direct color via `--rgb` on the CLI or `rgb=True` in Python. The fir
 
 `m` — cycles through magnitude, phase, real, and imaginary parts for complex-valued arrays.
 
+## qMRI
+
+`q` — opens qMRI for a parameter dimension. Hold Alt/Option while qMRI is active to choose quantitative parameters and synthesised contrasts. Synthetic contrasts show TE/TR/TI sliders below each pane when available.
+
 ## Navigation
 
 Scroll through slices. `/` opens the tool launcher for ortho, compare, qMRI, ROI, segmentation, FFT, projections, and overlays. The direct keys still work. `0` resets the zoom. `?` shows the full key guide.

arrayview-0.27.2/plans/roi-analysis-requirements.md

@@ -0,0 +1,126 @@
+# ROI Analysis Requirements
+
+## Purpose
+
+ROI analysis should be a lightweight MRI/qMRI measurement tool. It should make
+quick measurements easy, support repeated phantom-well measurements, and export
+the resulting statistics and label masks.
+
+ROI analysis is not a replacement for segmentation. Full white matter / gray
+matter segmentation, voxel-label editing, automatic well detection, and mask
+roundtripping belong outside ROI v1.
+
+The current uncommitted ROI upgrade work is superseded by this document. Useful
+backend or mask-export ideas may be reused, but the UI and interaction model
+should restart from these requirements.
+
+## Primary Use Cases
+
+- Draw a quick circular or freehand ROI in in-vivo MRI data to inspect mean,
+  standard deviation, minimum, and maximum values.
+- Draw multiple true circular ROIs inside cylindrical gel phantoms in transverse
+  slices and compare their measurements.
+- In qMRI mode, draw one ROI and measure the same region across visible
+  parameter maps.
+- Export ROI statistics as CSV.
+- Export a single integer `.npy` label mask matching the source array shape.
+
+## Interaction Model
+
+- `Shift+R` toggles ROI mode visibility and interaction.
+- When ROI mode turns off, ROI shapes, buttons, and hover affordances disappear.
+- Turning ROI mode back on restores the session's existing ROI shapes.
+- ROI geometry persists only for the current viewer session.
+- The default ROI shape is a true circle.
+- Additional ROI shapes are rectangle/square, freehand, and flood fill.
+- Click an existing ROI to select it.
+- `Delete` or `Backspace` deletes the selected ROI.
+- No automatic or suggest-only circle detection in ROI v1.
+
+## UI Model
+
+- Everyday ROI controls live on a colorbar flip surface, not in the right tool
+  drawer.
+- The colorbar flip should expose only the common drawing controls: shape
+  picker, stats popup button, and minimal clear/hide actions if needed.
+- Statistics open in a popup/modal.
+- The stats popup is the ROI manager for v1:
+  - one row per ROI by default
+  - qMRI rows expand or reveal per-map measurements
+  - rename ROI
+  - delete ROI
+  - export CSV
+  - export label mask `.npy`
+  - advanced dimensional extent controls
+- Do not put rename/delete controls directly on the canvas in v1.
+- Keep the canvas visually quiet; ROI outlines and labels are the only persistent
+  ROI marks while ROI mode is active.
+
+## Statistics and Dimensional Scope
+
+- Normal mode default: a new ROI applies only to the current displayed slice.
+- qMRI default: a new ROI applies to the visible qMRI parameter maps.
+- Advanced dimensional extent controls live in the stats popup per ROI.
+- Advanced controls must use plain-language labels, not debug-like labels such
+  as `d2:135` or `d3:all`.
+- Stats should report at least count, mean, standard deviation, minimum, and
+  maximum.
+- CSV export should include ROI name, shape, dimensional extent, and qMRI map
+  labels where applicable.
+
+## Masks and Segmentation Boundary
+
+- ROI mask export writes one `.npy` integer label mask matching the source array
+  shape.
+- Label `0` is background.
+- ROI labels are `1..N` in list order.
+- ROI mask loading and roundtrip editing are out of scope for ROI v1.
+- Segmentation remains the tool for voxel-label editing, external segmentation
+  workflows, brush/lasso editing, and future mask roundtripping.
+
+## Visibility and Color
+
+- ROI outlines must remain visible across colormaps, themes, and local image
+  brightness.
+- Keep per-ROI identity colors, but do not rely on color alone for visibility.
+- Use an adaptive high-contrast halo or dual-outline rendering for ROI borders.
+- ROI labels should use contrast-aware text/halo rendering.
+
+## Implementation Direction
+
+- Rebuild ROI around a small session-local model:
+  `id`, `name`, `shape`, `geometry`, `scope`, `visible`, `selected`.
+- Use true shape masks for stats and export. Do not fall back to bounding boxes
+  for circle, freehand, or flood fill.
+- Route ROI visibility through existing UI reconcilers and colorbar flip
+  patterns.
+- Update the command registry and `GUIDE_TABS` when finalizing shortcuts and
+  help text.
+- Update `docs/measurement.md` only after implementation matches this spec.
+
+## Test Plan
+
+- Spec review first: confirm this document captures the intended ROI direction
+  before rebuilding the implementation.
+- Backend tests:
+  - true circle mask stats
+  - freehand and floodfill mask stats
+  - qMRI per-map stats
+  - `.npy` label mask shape and label values
+- Browser tests:
+  - `Shift+R` hide/show preserves session ROIs
+  - default drawing shape is circle
+  - stats popup supports rename, delete, CSV export, and mask export
+  - qMRI mode shows per-map stats
+  - ROI outlines remain readable across themes/colormaps
+- Regression tests:
+  - simple ROI drawing does not require the right drawer
+  - no circle suggestion UI or API
+  - no cryptic dimensional labels
+
+## Assumptions
+
+- ROI and segmentation remain separate tools.
+- ROI v1 prioritizes fast measurement and export over full mask management.
+- Session-only ROI geometry is acceptable.
+- Exported CSV and `.npy` masks are the persistence boundary for ROI v1.

{arrayview-0.26.3 → arrayview-0.27.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "arrayview"
-version = "0.26.3"
+version = "0.27.2"
 description = "Fast multi-dimensional array viewer"
 readme = { file = "README.md", content-type = "text/markdown" }
 requires-python = ">=3.12"

{arrayview-0.26.3 → arrayview-0.27.2}/src/arrayview/_analysis.py

@@ -67,7 +67,7 @@ def _safe_float(v) -> float | None:
     return f if np.isfinite(f) else None
 
 
-def _histogram_payload(finite: np.ndarray, bins: int) -> dict:
+def _histogram_payload(finite: np.ndarray, bins: int, log_bins: bool = False) -> dict:
     """Build a histogram response from finite values."""
     if finite.size == 0:
         return {"counts": [], "edges": [], "vmin": 0.0, "vmax": 1.0}
@@ -81,7 +81,13 @@ def _histogram_payload(finite: np.ndarray, bins: int) -> dict:
             "vmax": vmax,
         }
     bins = max(8, min(int(bins), 512))
-    counts, edges = np.histogram(finite, bins=bins)
+    if log_bins and vmin > 0:
+        log_data = np.log10(finite)
+        _, log_edges = np.histogram(log_data, bins=bins)
+        edges = np.power(10.0, log_edges)
+        counts, _ = np.histogram(finite, bins=edges)
+    else:
+        counts, edges = np.histogram(finite, bins=bins)
     return {
         "counts": counts.tolist(),
         "edges": [float(e) for e in edges],
@@ -113,7 +119,8 @@ def _slice_histogram(
     data = apply_complex_mode(raw, complex_mode)
     finite = data.ravel()
     finite = finite[np.isfinite(finite)]
-    return _histogram_payload(finite, bins)
+    log_bins = qmri_role in ("t1", "t2")
+    return _histogram_payload(finite, bins, log_bins=log_bins)
 
 
 def _parse_fixed_indices(fixed_indices: str) -> dict[int, int]:
@@ -172,6 +179,7 @@ def _volume_histogram(
         session, dim_x, dim_y, scroll_dim, scroll_dims
     )
 
+    log_bins = qmri_role in ("t1", "t2")
     cache_key = (
         dim_x,
         dim_y,
@@ -179,6 +187,8 @@ def _volume_histogram(
         tuple(sorted(fixed.items())),
         complex_mode,
         qmri_role,
+        bins,
+        log_bins,
     )
     if not hasattr(session, "_volume_hist_cache"):
         session._volume_hist_cache = {}
@@ -224,7 +234,7 @@ def _volume_histogram(
             pixels.append(finite)
 
     if pixels:
-        result = _histogram_payload(np.concatenate(pixels), bins)
+        result = _histogram_payload(np.concatenate(pixels), bins, log_bins=log_bins)
     else:
         result = {"counts": [], "edges": [], "vmin": 0.0, "vmax": 1.0}
     session._volume_hist_cache[cache_key] = {
@@ -241,14 +251,19 @@ def _lebesgue_slice(
     indices,
     complex_mode: int = 0,
     log_scale: bool = False,
+    qmri_role: str = "",
 ) -> np.ndarray:
     """Return float32 slice data for Lebesgue mode."""
+    from arrayview._synthetic_mri import normalize_relaxation_ms
+
     idx_tuple = (
         tuple(int(v) for v in indices.split(","))
         if isinstance(indices, str)
         else tuple(indices)
     )
     raw = extract_slice(session, dim_x, dim_y, list(idx_tuple))
+    if qmri_role:
+        raw = normalize_relaxation_ms(raw, qmri_role)
     data = apply_complex_mode(raw, complex_mode).astype(np.float32)
     if log_scale:
         data = np.log10(np.abs(data) + 1).astype(np.float32)

{arrayview-0.26.3 → arrayview-0.27.2}/src/arrayview/_launcher.py

@@ -1148,6 +1148,7 @@ def _handle_cli_spawned_daemon(
         stdout=subprocess.DEVNULL,
         stderr=subprocess.DEVNULL,
         close_fds=True,
+        start_new_session=(sys.platform != "win32"),
     )
 
     early_webview_opened = False
@@ -1245,7 +1246,7 @@ def _open_cli_spawned_view(
             _print_viewer_location(url)
             _open_browser(
                 url,
-                blocking=False,
+                blocking=(window_mode == "vscode"),
                 force_vscode=(window_mode == "vscode"),
                 title=f"ArrayView: {name}",
                 floating=floating,

{arrayview-0.26.3 → arrayview-0.27.2}/src/arrayview/_platform.py

@@ -419,19 +419,27 @@ def _can_native_window() -> bool:
 
 
 def _native_window_gui() -> str | None:
-    """Return the pywebview GUI backend name to use, or None if unavailable."""
+    """Return the pywebview GUI backend name to use, or None if unavailable.
+
+    On macOS/Windows and Linux we return ``""`` so pywebview uses its own
+    backend auto-detection, which is more robust than forcing a specific
+    backend: it probes each installed binding and falls back across them.
+    Forcing ``"qt"`` because ``find_spec("qtpy")`` succeeds is a bad probe —
+    qtpy being importable does not mean QtWebEngine can actually initialise
+    (missing libnss/libxkbcommon/QtWebEngineProcess is common on Linux), and
+    a failed forced backend hangs for ~10s before the watchdog falls back to a
+    browser. Letting pywebview pick avoids that. We still require a display
+    server on Linux and the webview package itself everywhere.
+    """
     if importlib.util.find_spec("webview") is None:
         return None
     if sys.platform in ("darwin", "win32"):
         return ""
-    # Linux/BSD: need a display server AND pywebview's GUI bindings.
+    # Linux/BSD: need a display server. pywebview probes Qt/GTK/Tk backends
+    # itself, so we don't pin one — we only confirm a display is reachable.
     if not (os.environ.get("DISPLAY") or os.environ.get("WAYLAND_DISPLAY")):
         return None
-    if importlib.util.find_spec("qtpy") is not None:
-        return "qt"
-    if importlib.util.find_spec("gi") is not None:
-        return "gtk"
-    return None
+    return ""
 
 
 # ---------------------------------------------------------------------------