Skip to content

Fix periodic isosurface coordinates for integer supercells - #377

Merged
janosh merged 1 commit into
janosh:multi-volume-isosurfacesfrom
Stardust0831:fix/periodic-volume-display-range
Jul 12, 2026
Merged

janosh merged 1 commit into
janosh:multi-volume-isosurfacesfrom
Stardust0831:fix/periodic-volume-display-range

Conversation

@Stardust0831

@Stardust0831 Stardust0831 commented Jul 11, 2026 •

Copy link
Copy Markdown
Contributor

Thank you for the remarkably comprehensive implementation in #376. While testing the integer-supercell and cross-volume coloring paths, I noticed a coordinate-consistency edge case and prepared this small follow-up on top of the feature branch.

What changes

  • Route periodic integer supercells through the same endpoint-inclusive range extraction used by fractional display ranges.
  • Preserve the original periodic voxel spacing when finite marching cubes is used for clipping.
  • Keep finite volumes bounded instead of implicitly repeating them with the atom supercell.
  • Use N - 1 source intervals when cropping finite grids.
  • Harden finite range sanitization when requested bounds lie completely outside the volume.
  • Export the finite/periodic range-resolution utility and its configuration type through the public library API.

Why

A periodic grid with N samples spans [0, 1). Repeating it S times produces N * S samples ending at S - 1/N. Passing that array to marching cubes as finite maps it over [0, S], changing the voxel spacing from 1/N to S / (N*S - 1). Repeated surfaces can therefore drift relative to the atomic supercell, while cross-volume coloring still samples in the correct Cartesian frame.

Using the existing extract_volume_range() path for both integer and fractional ranges produces N*S + 1 endpoint-inclusive samples and keeps geometry and scalar sampling in the same frame.

@coderabbitai

coderabbitai Bot commented Jul 11, 2026 •

Copy link
Copy Markdown

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 684b943f-81ef-4c04-8359-64b04ce4e5c9

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

  • ✅ Review completed - (🔄 Check again to review again)
✨ Finishing Touches
🧪 Generate unit tests (beta)
  • Create PR with unit tests

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

Comment @coderabbitai help to get the list of available commands.

janosh added a commit that referenced this pull request Jul 12, 2026
After #377 unified integer tiling and fractional bounds through range
extraction, the range fully determines a volume's prepared grid: it
encodes tiling and halo for periodic volumes and both are irrelevant
for finite ones. Remove the now-redundant tiling_key/halo fields from
PreparedGrid and geometry_key — also stops finite-volume caches from
needlessly invalidating on supercell changes.
@janosh
janosh merged commit c602e78 into janosh:multi-volume-isosurfaces Jul 12, 2026
2 checks passed
@janosh

janosh commented Jul 12, 2026

Copy link
Copy Markdown
Owner

thanks for the sharp catch and the clean fix! verified the drift empirically (finite marching cubes on a tiled periodic 8-point grid put vertices up to 0.75 off the requested isovalue — your self-sampling test recovers it to 1e-5) and merged into the feature branch via a19c657, with a small follow-up commit dropping the tiling/halo cache-key fields your range unification made redundant. this will ship with #376

janosh added a commit that referenced this pull request Jul 13, 2026
* Add multi-volume sampling and vertex-coloring core for isosurfaces

Lay the engine groundwork for #375 (multi-volume
isosurfaces with cross-volume scalar coloring):

- sampling.ts: world-coordinate volume sampler (origin offsets,
  non-orthogonal lattices, periodic wrap, clamp/fallback out-of-bounds
  policies), strict VESTA-style grid compatibility check with mismatch
  diagnostics, and fractional display-range extraction for non-integer
  supercells that clip exactly at fractional bounds
- coloring.ts: linear-RGB colormap LUTs (d3 sRGB converted for Three.js
  vertex colors), scalar-to-vertex-color mapping with inverted-range and
  non-finite handling, auto colormap/range selection with symmetric
  ranges for signed fields
- types.ts: IsosurfaceLayer gains volume_idx/color_volume_idx/colormap/
  color_range, IsosurfaceSettings gains display_range; helpers for layer
  materialization, volume removal with index remapping, and source-file
  based import merging (append same-cell fields, replace reimports)
- file-drop.ts: process all dropped files sequentially; one corrupt file
  no longer aborts the batch

trilinear_interpolate moved from slice.ts to sampling.ts (public export
path unchanged).

* Render multi-volume isosurfaces with cross-volume vertex coloring

Rework the isosurface renderer and controls for #375:

- Isosurface.svelte renders all loaded volumes at once: each layer picks
  a geometry-source volume and optionally a different color-source
  volume sampled per vertex through a colormap (e.g. density colored by
  ESP). Geometry is cached per layer signature so colormap/color-range
  changes never rerun marching cubes; sampled scalars are cached and
  only remapped through the LUT. Supercell tiling moved into the
  renderer so colors sample the untiled full-res volume and stay
  continuous across cell boundaries and tiles. Fractional display
  ranges (VESTA-style non-integer supercells) clip surfaces exactly at
  the requested bounds, independent of the atom supercell.
- IsosurfaceControls groups surfaces under their geometry volume with
  add/remove-surface and remove-volume buttons, per-surface color-source
  select, colormap picker, min/max range inputs (empty = auto-fit to
  surface values) with inline colorbar, grid-compatibility warning, and
  per-axis fractional display-range inputs.
- Structure.svelte appends volumes from files describing the same cell
  (in-place replace on reimport of the same source file) instead of
  replacing the scene; supports multi-file drops; volumetric_data prop
  stays backward compatible (single volume or array).

Vitest controls suite renamed to .svelte.ts for $state props and
extended to the new multi-volume UI.

* Add multi-volume demo page, paired fixtures, and e2e coverage

Complete #375 with testable demos:

- /structure/multi-volume demo page with seven scenarios: glycine
  density colored by ESP (strictly matching 50^3 cube pair), caffeine
  HOMO+LUMO rendered together, Fe charge colored by magnetization
  (2x2x2 supercell), Al(111) slab density colored by LOCPOT (3x3x1),
  hBN density colored by ELF on a non-orthogonal lattice, a VESTA-style
  fractional display range (a, b in [-0.15, 2.15]) with an independent
  integer atom supercell, and an 80x80x96 grid-pair perf stress test.
  Clicking picker files appends same-cell volumes; scenario loads are
  race-guarded and URL-addressable.
- New matching-grid fixtures: glycine density+ESP cubes (softened
  Coulomb potential of neutral partial charges), Al-slab-CHGCAR pairing
  the existing LOCPOT, hBN-ELFCAR pairing hBN-CHGCAR, and a large-grid
  LOCPOT for perf testing. Fixture generator gained per-file CLI args,
  a slab fix (zero lattice vectors no longer triple-count images), and
  shared Al slab geometry.
- Playwright suite covering every scenario plus controls interactions
  (volume grouping, color-source-only note, add-surface).

* Address review: empty-file errors, stale scenario label, dedupe, strict zip

- file-drop: report empty decompressed content as a batch failure with
  the filename instead of silently skipping the file
- multi-volume demo: leave preset-scenario mode (overlay label + URL
  param) after any user file import, not just scene replacement
- sampling: reuse exported create_cart_to_frac_matrix from math.ts
  instead of duplicating the inverse-transpose construction
- generate_examples: zip(..., strict=True) in pbc_gaussian_sum to fail
  loudly on mismatched center/weight/sigma lists (no fixture changes)

* Remove accidentally committed __pycache__ and gitignore it

* Fix scenario clicks, drop serialization, and same-cell import edge cases

Second review round on #375:

- multi-volume demo: the URL-sync effect tracked active_scenario, so
  clicking a scenario card reran the effect against the still-old URL
  and reloaded the previous scenario, discarding the click as a stale
  load. Track only the URL, fall back to the first scenario when no
  preset is loaded, and absorb the old onMount. Playwright regression
  test clicks a card and asserts the switch.
- file-drop: queue overlapping drop batches so concurrent on_drop state
  mutations can't interleave; capture DataTransfer contents
  synchronously (only readable during event dispatch) via a new
  dropped_file_url helper split out of handle_url_drop
- Structure: parse errors propagate to the batch aggregator instead of
  last-write-wins; dropping the first volumetric file onto an existing
  same-cell structure appends instead of replacing the scene; structure
  files describing the same cell keep loaded volumes and camera (mixed
  CHGCAR + POSCAR batches survive either drop order)
- merge_imported_volumes: pin implicit layers to active_volume_idx
  before index remapping instead of assuming volume 0
- Isosurface: cancel pending debounced rebuild when volumes clear;
  cache samplers per volume object
- generate_examples.py: friendly unknown-filename CLI error, rename
  shadowed single-letter var
- tests: drop-queue serialization, plural failure aggregation, merge
  active-index regression, six new controls interactions; Playwright
  suites drop flaky networkidle waits for wait_for_3d_canvas and share
  open_settings_pane via helpers.ts

* Runtime-check dropped URL payload instead of trusting FileInfo cast

Drop payloads are external input: a JSON body with a truthy non-string
url (number, object) previously flowed through dropped_file_url's type
assertion into fetch(). Return the url only when it is a non-empty
string; cover empty-string, numeric, and object urls in the
handle_url_drop suite and assert fetch is never reached.

* Fold URL-drop failures into the batch failure report

A drop carrying both a URL payload and plain files silently swallowed
the URL error when the URL fetch failed — the files were processed but
the failure never reached on_error. The URL failure now joins the
aggregated failures message alongside per-file errors, with a test
covering the mixed URL+files failure path.

* Dedupe color-volume lookups and lattice extraction, compact tests

Simplification sweep across the branch:

- Isosurface + IsosurfaceControls: one color_vol_of helper each replaces
  four hand-rolled color_volume_idx lookups (plus a template condition)
- Structure: current_lattice_matrix helper replaces the twice-repeated
  lattice extraction ternary in the import seam
- controls tests: three tests that manually mounted with $state props
  now use the mount_controls helper's returned reactive props

No behavior changes; 495 unit + 18 e2e tests pass.

* Replace manual signature tracking with derived strings, dedupe load guards

Deeper simplification pass:

- Isosurface: geometry and color signatures are now $derived strings —
  Svelte only notifies dependents when a derived value changes, so the
  rebuild/recolor effects skip unrelated updates natively, replacing
  three module-level last-signature/last-entries trackers and their
  compare-and-store choreography. Recolor-without-marching-cubes
  contract verified in the browser (colormap switch recolors the
  glycine ESP surface without a rebuild).
- multi-volume demo: run_load() centralizes the stale-load token,
  loading flag, and error handling shared by scenario loads and file
  imports (was duplicated try/catch/finally in both).
- generate_examples: hBN geometry (lattice, frac→cart, site fractions)
  shared between the CHGCAR and ELFCAR generators via hbn_geometry(),
  mirroring the Al slab helper. Generator output verified byte-identical
  via content hashes before/after.

* Fix periodic isosurface supercell coordinates

* Drop cache-key fields subsumed by the extraction range

After #377 unified integer tiling and fractional bounds through range
extraction, the range fully determines a volume's prepared grid: it
encodes tiling and halo for periodic volumes and both are irrelevant
for finite ones. Remove the now-redundant tiling_key/halo fields from
PreparedGrid and geometry_key — also stops finite-volume caches from
needlessly invalidating on supercell changes.

* fix CmdPalette with search text not navigating on option click

* Inline single-use halo helper and tiling-range construction

effective_halo had one caller left after the range unification (the
resolver already clamps and gates halo on periodicity), and the
integer-tiling DisplayRange collapses to a map over the tiling factors.

* Tighten demo scenario descriptions and make card text selectable

Scenario blurbs cut to their essentials (what renders + what it
demonstrates); user-select: text on the card buttons since browsers
suppress selection inside <button> by default.

* Clarify that cache reuse rebinds entry identity to current loop values

* Harden 3D viewer overlays and WebGL lifecycle for desktop embedders

- Overlay chrome that must stack above WebGL canvases (atom legend,
  cell-select dropdown, glass chips, carousel pager, viewer controls,
  draggable panes) now gets its own compositing layer: WKWebView (Tauri)
  reattaches canvas layers above non-composited overlays after its
  window is hidden and re-shown, ignoring z-index.
- Document the in-viewer overlay stack as --z-index-viewer-* custom
  properties in app.css and wire components to them (values unchanged).
- Pause auto-rotation while the page is hidden: desktop embedders
  disable background throttling, which left hidden windows re-rendering
  every auto-rotating canvas indefinitely.
- StructureViewport remounts its <Canvas> when the browser evicts the
  WebGL context without restoring it (bounded retries), so blank
  carousel cards self-heal.
- CellSelect opens on sustained hover (200ms) instead of instantly on
  mouseenter, so scrubbing the pointer across compact embeds no longer
  pops the menu open.

* Add arbitrary volume slices and scalable isosurface rendering

Generalize plane sampling with reusable filled and contour rendering, and move geometry extraction to an instrumented worker for responsive large-volume updates. Add benchmark and regression coverage while hardening related file-drop and UI state handling.

* Polish slice responsiveness and WebGL recovery

Render high-resolution slices without deep-proxy or contour stalls, surface exhausted WebGL recovery, and strengthen portable performance and cache regression coverage.

* Fix packaged isosurface worker resolution

Use an extensionless worker URL so source and compiled package layouts both resolve correctly.

* Stabilize isosurface E2E tests under CI load

Serialize the CPU-heavy suite while preserving four-worker parallelism for unrelated tests.

* Preserve structure camera rotation across hover

Keep the orientation gizmo mounted so hover visibility changes do not reinitialize orbit controls.

* Refine file picker chip corners

* Simplify isosurface pipeline and tests

Centralize geometry preparation and remove duplicated control, demo, and test setup while preserving performance-sensitive paths.

* Polish isosurface controls and viewer defaults

Use compact labeled color controls, simplify file picking behavior, and improve initial structure framing.

* Fix CI E2E hangs on unstable canvas screenshots

Playwright locator.screenshot waits for element stability; under CI load
continuous WebGL/layout updates never settle and burn the test timeout.
Capture canvases via clipped page screenshots instead.

* Avoid Playwright screenshots for 2D canvas E2E checks

page.screenshot still hung on the busy isosurface demo under CI load.
Read 2D canvas pixels via toDataURL; keep clipped page screenshots only
for WebGL where blit would be blank without preserveDrawingBuffer.

* Polish file-picker clicks, isosurface controls, and viewer chrome

Key pending clicks by URL so same-name files stay distinct, restore hover/focus
gizmo gating, and tighten labeled isosurface layer controls plus theme-aware UI.

* Require matching pending click for HeatmapMatrix dblclick

Align delayed click/dblclick disambiguation with FilePicker so orphaned or
cross-cell dblclicks schedule a single-click instead of firing ondblclick.

* Drop underscore prefixes on isosurface internal modules.

Prefer plain names and keep internals out of the package barrel instead of _* marking; also remove the unused bench:isosurface script.

* Fold page_visibility into scene props and cut low-value test fat.

Merge the tiny visibility module into props.svelte.ts and drop redundant FilePicker, controls, types, and perf harness coverage that mostly restated defaults.

* Keep Structure gizmo mounted across hover cycles

Gate visibility with the gizmo-visible CSS class only. Tying the gizmo prop
to viewer_active remounted Threlte Gizmo on every leave and broke the
hover-chrome E2E regression test.

---------

Co-authored-by: Stardust0831 <169599847+Stardust0831@users.noreply.github.com>
janosh added a commit that referenced this pull request Jul 14, 2026
* Add multi-volume sampling and vertex-coloring core for isosurfaces

Lay the engine groundwork for #375 (multi-volume
isosurfaces with cross-volume scalar coloring):

- sampling.ts: world-coordinate volume sampler (origin offsets,
  non-orthogonal lattices, periodic wrap, clamp/fallback out-of-bounds
  policies), strict VESTA-style grid compatibility check with mismatch
  diagnostics, and fractional display-range extraction for non-integer
  supercells that clip exactly at fractional bounds
- coloring.ts: linear-RGB colormap LUTs (d3 sRGB converted for Three.js
  vertex colors), scalar-to-vertex-color mapping with inverted-range and
  non-finite handling, auto colormap/range selection with symmetric
  ranges for signed fields
- types.ts: IsosurfaceLayer gains volume_idx/color_volume_idx/colormap/
  color_range, IsosurfaceSettings gains display_range; helpers for layer
  materialization, volume removal with index remapping, and source-file
  based import merging (append same-cell fields, replace reimports)
- file-drop.ts: process all dropped files sequentially; one corrupt file
  no longer aborts the batch

trilinear_interpolate moved from slice.ts to sampling.ts (public export
path unchanged).

* Render multi-volume isosurfaces with cross-volume vertex coloring

Rework the isosurface renderer and controls for #375:

- Isosurface.svelte renders all loaded volumes at once: each layer picks
  a geometry-source volume and optionally a different color-source
  volume sampled per vertex through a colormap (e.g. density colored by
  ESP). Geometry is cached per layer signature so colormap/color-range
  changes never rerun marching cubes; sampled scalars are cached and
  only remapped through the LUT. Supercell tiling moved into the
  renderer so colors sample the untiled full-res volume and stay
  continuous across cell boundaries and tiles. Fractional display
  ranges (VESTA-style non-integer supercells) clip surfaces exactly at
  the requested bounds, independent of the atom supercell.
- IsosurfaceControls groups surfaces under their geometry volume with
  add/remove-surface and remove-volume buttons, per-surface color-source
  select, colormap picker, min/max range inputs (empty = auto-fit to
  surface values) with inline colorbar, grid-compatibility warning, and
  per-axis fractional display-range inputs.
- Structure.svelte appends volumes from files describing the same cell
  (in-place replace on reimport of the same source file) instead of
  replacing the scene; supports multi-file drops; volumetric_data prop
  stays backward compatible (single volume or array).

Vitest controls suite renamed to .svelte.ts for $state props and
extended to the new multi-volume UI.

* Add multi-volume demo page, paired fixtures, and e2e coverage

Complete #375 with testable demos:

- /structure/multi-volume demo page with seven scenarios: glycine
  density colored by ESP (strictly matching 50^3 cube pair), caffeine
  HOMO+LUMO rendered together, Fe charge colored by magnetization
  (2x2x2 supercell), Al(111) slab density colored by LOCPOT (3x3x1),
  hBN density colored by ELF on a non-orthogonal lattice, a VESTA-style
  fractional display range (a, b in [-0.15, 2.15]) with an independent
  integer atom supercell, and an 80x80x96 grid-pair perf stress test.
  Clicking picker files appends same-cell volumes; scenario loads are
  race-guarded and URL-addressable.
- New matching-grid fixtures: glycine density+ESP cubes (softened
  Coulomb potential of neutral partial charges), Al-slab-CHGCAR pairing
  the existing LOCPOT, hBN-ELFCAR pairing hBN-CHGCAR, and a large-grid
  LOCPOT for perf testing. Fixture generator gained per-file CLI args,
  a slab fix (zero lattice vectors no longer triple-count images), and
  shared Al slab geometry.
- Playwright suite covering every scenario plus controls interactions
  (volume grouping, color-source-only note, add-surface).

* Address review: empty-file errors, stale scenario label, dedupe, strict zip

- file-drop: report empty decompressed content as a batch failure with
  the filename instead of silently skipping the file
- multi-volume demo: leave preset-scenario mode (overlay label + URL
  param) after any user file import, not just scene replacement
- sampling: reuse exported create_cart_to_frac_matrix from math.ts
  instead of duplicating the inverse-transpose construction
- generate_examples: zip(..., strict=True) in pbc_gaussian_sum to fail
  loudly on mismatched center/weight/sigma lists (no fixture changes)

* Remove accidentally committed __pycache__ and gitignore it

* Fix scenario clicks, drop serialization, and same-cell import edge cases

Second review round on #375:

- multi-volume demo: the URL-sync effect tracked active_scenario, so
  clicking a scenario card reran the effect against the still-old URL
  and reloaded the previous scenario, discarding the click as a stale
  load. Track only the URL, fall back to the first scenario when no
  preset is loaded, and absorb the old onMount. Playwright regression
  test clicks a card and asserts the switch.
- file-drop: queue overlapping drop batches so concurrent on_drop state
  mutations can't interleave; capture DataTransfer contents
  synchronously (only readable during event dispatch) via a new
  dropped_file_url helper split out of handle_url_drop
- Structure: parse errors propagate to the batch aggregator instead of
  last-write-wins; dropping the first volumetric file onto an existing
  same-cell structure appends instead of replacing the scene; structure
  files describing the same cell keep loaded volumes and camera (mixed
  CHGCAR + POSCAR batches survive either drop order)
- merge_imported_volumes: pin implicit layers to active_volume_idx
  before index remapping instead of assuming volume 0
- Isosurface: cancel pending debounced rebuild when volumes clear;
  cache samplers per volume object
- generate_examples.py: friendly unknown-filename CLI error, rename
  shadowed single-letter var
- tests: drop-queue serialization, plural failure aggregation, merge
  active-index regression, six new controls interactions; Playwright
  suites drop flaky networkidle waits for wait_for_3d_canvas and share
  open_settings_pane via helpers.ts

* Runtime-check dropped URL payload instead of trusting FileInfo cast

Drop payloads are external input: a JSON body with a truthy non-string
url (number, object) previously flowed through dropped_file_url's type
assertion into fetch(). Return the url only when it is a non-empty
string; cover empty-string, numeric, and object urls in the
handle_url_drop suite and assert fetch is never reached.

* Fold URL-drop failures into the batch failure report

A drop carrying both a URL payload and plain files silently swallowed
the URL error when the URL fetch failed — the files were processed but
the failure never reached on_error. The URL failure now joins the
aggregated failures message alongside per-file errors, with a test
covering the mixed URL+files failure path.

* Dedupe color-volume lookups and lattice extraction, compact tests

Simplification sweep across the branch:

- Isosurface + IsosurfaceControls: one color_vol_of helper each replaces
  four hand-rolled color_volume_idx lookups (plus a template condition)
- Structure: current_lattice_matrix helper replaces the twice-repeated
  lattice extraction ternary in the import seam
- controls tests: three tests that manually mounted with $state props
  now use the mount_controls helper's returned reactive props

No behavior changes; 495 unit + 18 e2e tests pass.

* Replace manual signature tracking with derived strings, dedupe load guards

Deeper simplification pass:

- Isosurface: geometry and color signatures are now $derived strings —
  Svelte only notifies dependents when a derived value changes, so the
  rebuild/recolor effects skip unrelated updates natively, replacing
  three module-level last-signature/last-entries trackers and their
  compare-and-store choreography. Recolor-without-marching-cubes
  contract verified in the browser (colormap switch recolors the
  glycine ESP surface without a rebuild).
- multi-volume demo: run_load() centralizes the stale-load token,
  loading flag, and error handling shared by scenario loads and file
  imports (was duplicated try/catch/finally in both).
- generate_examples: hBN geometry (lattice, frac→cart, site fractions)
  shared between the CHGCAR and ELFCAR generators via hbn_geometry(),
  mirroring the Al slab helper. Generator output verified byte-identical
  via content hashes before/after.

* Fix periodic isosurface supercell coordinates

* Drop cache-key fields subsumed by the extraction range

After #377 unified integer tiling and fractional bounds through range
extraction, the range fully determines a volume's prepared grid: it
encodes tiling and halo for periodic volumes and both are irrelevant
for finite ones. Remove the now-redundant tiling_key/halo fields from
PreparedGrid and geometry_key — also stops finite-volume caches from
needlessly invalidating on supercell changes.

* fix CmdPalette with search text not navigating on option click

* Inline single-use halo helper and tiling-range construction

effective_halo had one caller left after the range unification (the
resolver already clamps and gates halo on periodicity), and the
integer-tiling DisplayRange collapses to a map over the tiling factors.

* Tighten demo scenario descriptions and make card text selectable

Scenario blurbs cut to their essentials (what renders + what it
demonstrates); user-select: text on the card buttons since browsers
suppress selection inside <button> by default.

* Clarify that cache reuse rebinds entry identity to current loop values

* Harden 3D viewer overlays and WebGL lifecycle for desktop embedders

- Overlay chrome that must stack above WebGL canvases (atom legend,
  cell-select dropdown, glass chips, carousel pager, viewer controls,
  draggable panes) now gets its own compositing layer: WKWebView (Tauri)
  reattaches canvas layers above non-composited overlays after its
  window is hidden and re-shown, ignoring z-index.
- Document the in-viewer overlay stack as --z-index-viewer-* custom
  properties in app.css and wire components to them (values unchanged).
- Pause auto-rotation while the page is hidden: desktop embedders
  disable background throttling, which left hidden windows re-rendering
  every auto-rotating canvas indefinitely.
- StructureViewport remounts its <Canvas> when the browser evicts the
  WebGL context without restoring it (bounded retries), so blank
  carousel cards self-heal.
- CellSelect opens on sustained hover (200ms) instead of instantly on
  mouseenter, so scrubbing the pointer across compact embeds no longer
  pops the menu open.

* Add arbitrary volume slices and scalable isosurface rendering

Generalize plane sampling with reusable filled and contour rendering, and move geometry extraction to an instrumented worker for responsive large-volume updates. Add benchmark and regression coverage while hardening related file-drop and UI state handling.

* Polish slice responsiveness and WebGL recovery

Render high-resolution slices without deep-proxy or contour stalls, surface exhausted WebGL recovery, and strengthen portable performance and cache regression coverage.

* Fix packaged isosurface worker resolution

Use an extensionless worker URL so source and compiled package layouts both resolve correctly.

* Stabilize isosurface E2E tests under CI load

Serialize the CPU-heavy suite while preserving four-worker parallelism for unrelated tests.

* Preserve structure camera rotation across hover

Keep the orientation gizmo mounted so hover visibility changes do not reinitialize orbit controls.

* Refine file picker chip corners

* Simplify isosurface pipeline and tests

Centralize geometry preparation and remove duplicated control, demo, and test setup while preserving performance-sensitive paths.

* Polish isosurface controls and viewer defaults

Use compact labeled color controls, simplify file picking behavior, and improve initial structure framing.

* Fix CI E2E hangs on unstable canvas screenshots

Playwright locator.screenshot waits for element stability; under CI load
continuous WebGL/layout updates never settle and burn the test timeout.
Capture canvases via clipped page screenshots instead.

* Avoid Playwright screenshots for 2D canvas E2E checks

page.screenshot still hung on the busy isosurface demo under CI load.
Read 2D canvas pixels via toDataURL; keep clipped page screenshots only
for WebGL where blit would be blank without preserveDrawingBuffer.

* Polish file-picker clicks, isosurface controls, and viewer chrome

Key pending clicks by URL so same-name files stay distinct, restore hover/focus
gizmo gating, and tighten labeled isosurface layer controls plus theme-aware UI.

* Require matching pending click for HeatmapMatrix dblclick

Align delayed click/dblclick disambiguation with FilePicker so orphaned or
cross-cell dblclicks schedule a single-click instead of firing ondblclick.

* Drop underscore prefixes on isosurface internal modules.

Prefer plain names and keep internals out of the package barrel instead of _* marking; also remove the unused bench:isosurface script.

* Fold page_visibility into scene props and cut low-value test fat.

Merge the tiny visibility module into props.svelte.ts and drop redundant FilePicker, controls, types, and perf harness coverage that mostly restated defaults.

* Keep Structure gizmo mounted across hover cycles

Gate visibility with the gizmo-visible CSS class only. Tying the gizmo prop
to viewer_active remounted Threlte Gizmo on every leave and broke the
hover-chrome E2E regression test.

---------

Co-authored-by: Stardust0831 <169599847+Stardust0831@users.noreply.github.com>
janosh added a commit that referenced this pull request Jul 31, 2026
* Add multi-volume sampling and vertex-coloring core for isosurfaces

Lay the engine groundwork for #375 (multi-volume
isosurfaces with cross-volume scalar coloring):

- sampling.ts: world-coordinate volume sampler (origin offsets,
  non-orthogonal lattices, periodic wrap, clamp/fallback out-of-bounds
  policies), strict VESTA-style grid compatibility check with mismatch
  diagnostics, and fractional display-range extraction for non-integer
  supercells that clip exactly at fractional bounds
- coloring.ts: linear-RGB colormap LUTs (d3 sRGB converted for Three.js
  vertex colors), scalar-to-vertex-color mapping with inverted-range and
  non-finite handling, auto colormap/range selection with symmetric
  ranges for signed fields
- types.ts: IsosurfaceLayer gains volume_idx/color_volume_idx/colormap/
  color_range, IsosurfaceSettings gains display_range; helpers for layer
  materialization, volume removal with index remapping, and source-file
  based import merging (append same-cell fields, replace reimports)
- file-drop.ts: process all dropped files sequentially; one corrupt file
  no longer aborts the batch

trilinear_interpolate moved from slice.ts to sampling.ts (public export
path unchanged).

* Render multi-volume isosurfaces with cross-volume vertex coloring

Rework the isosurface renderer and controls for #375:

- Isosurface.svelte renders all loaded volumes at once: each layer picks
  a geometry-source volume and optionally a different color-source
  volume sampled per vertex through a colormap (e.g. density colored by
  ESP). Geometry is cached per layer signature so colormap/color-range
  changes never rerun marching cubes; sampled scalars are cached and
  only remapped through the LUT. Supercell tiling moved into the
  renderer so colors sample the untiled full-res volume and stay
  continuous across cell boundaries and tiles. Fractional display
  ranges (VESTA-style non-integer supercells) clip surfaces exactly at
  the requested bounds, independent of the atom supercell.
- IsosurfaceControls groups surfaces under their geometry volume with
  add/remove-surface and remove-volume buttons, per-surface color-source
  select, colormap picker, min/max range inputs (empty = auto-fit to
  surface values) with inline colorbar, grid-compatibility warning, and
  per-axis fractional display-range inputs.
- Structure.svelte appends volumes from files describing the same cell
  (in-place replace on reimport of the same source file) instead of
  replacing the scene; supports multi-file drops; volumetric_data prop
  stays backward compatible (single volume or array).

Vitest controls suite renamed to .svelte.ts for $state props and
extended to the new multi-volume UI.

* Add multi-volume demo page, paired fixtures, and e2e coverage

Complete #375 with testable demos:

- /structure/multi-volume demo page with seven scenarios: glycine
  density colored by ESP (strictly matching 50^3 cube pair), caffeine
  HOMO+LUMO rendered together, Fe charge colored by magnetization
  (2x2x2 supercell), Al(111) slab density colored by LOCPOT (3x3x1),
  hBN density colored by ELF on a non-orthogonal lattice, a VESTA-style
  fractional display range (a, b in [-0.15, 2.15]) with an independent
  integer atom supercell, and an 80x80x96 grid-pair perf stress test.
  Clicking picker files appends same-cell volumes; scenario loads are
  race-guarded and URL-addressable.
- New matching-grid fixtures: glycine density+ESP cubes (softened
  Coulomb potential of neutral partial charges), Al-slab-CHGCAR pairing
  the existing LOCPOT, hBN-ELFCAR pairing hBN-CHGCAR, and a large-grid
  LOCPOT for perf testing. Fixture generator gained per-file CLI args,
  a slab fix (zero lattice vectors no longer triple-count images), and
  shared Al slab geometry.
- Playwright suite covering every scenario plus controls interactions
  (volume grouping, color-source-only note, add-surface).

* Address review: empty-file errors, stale scenario label, dedupe, strict zip

- file-drop: report empty decompressed content as a batch failure with
  the filename instead of silently skipping the file
- multi-volume demo: leave preset-scenario mode (overlay label + URL
  param) after any user file import, not just scene replacement
- sampling: reuse exported create_cart_to_frac_matrix from math.ts
  instead of duplicating the inverse-transpose construction
- generate_examples: zip(..., strict=True) in pbc_gaussian_sum to fail
  loudly on mismatched center/weight/sigma lists (no fixture changes)

* Remove accidentally committed __pycache__ and gitignore it

* Fix scenario clicks, drop serialization, and same-cell import edge cases

Second review round on #375:

- multi-volume demo: the URL-sync effect tracked active_scenario, so
  clicking a scenario card reran the effect against the still-old URL
  and reloaded the previous scenario, discarding the click as a stale
  load. Track only the URL, fall back to the first scenario when no
  preset is loaded, and absorb the old onMount. Playwright regression
  test clicks a card and asserts the switch.
- file-drop: queue overlapping drop batches so concurrent on_drop state
  mutations can't interleave; capture DataTransfer contents
  synchronously (only readable during event dispatch) via a new
  dropped_file_url helper split out of handle_url_drop
- Structure: parse errors propagate to the batch aggregator instead of
  last-write-wins; dropping the first volumetric file onto an existing
  same-cell structure appends instead of replacing the scene; structure
  files describing the same cell keep loaded volumes and camera (mixed
  CHGCAR + POSCAR batches survive either drop order)
- merge_imported_volumes: pin implicit layers to active_volume_idx
  before index remapping instead of assuming volume 0
- Isosurface: cancel pending debounced rebuild when volumes clear;
  cache samplers per volume object
- generate_examples.py: friendly unknown-filename CLI error, rename
  shadowed single-letter var
- tests: drop-queue serialization, plural failure aggregation, merge
  active-index regression, six new controls interactions; Playwright
  suites drop flaky networkidle waits for wait_for_3d_canvas and share
  open_settings_pane via helpers.ts

* Runtime-check dropped URL payload instead of trusting FileInfo cast

Drop payloads are external input: a JSON body with a truthy non-string
url (number, object) previously flowed through dropped_file_url's type
assertion into fetch(). Return the url only when it is a non-empty
string; cover empty-string, numeric, and object urls in the
handle_url_drop suite and assert fetch is never reached.

* Fold URL-drop failures into the batch failure report

A drop carrying both a URL payload and plain files silently swallowed
the URL error when the URL fetch failed — the files were processed but
the failure never reached on_error. The URL failure now joins the
aggregated failures message alongside per-file errors, with a test
covering the mixed URL+files failure path.

* Dedupe color-volume lookups and lattice extraction, compact tests

Simplification sweep across the branch:

- Isosurface + IsosurfaceControls: one color_vol_of helper each replaces
  four hand-rolled color_volume_idx lookups (plus a template condition)
- Structure: current_lattice_matrix helper replaces the twice-repeated
  lattice extraction ternary in the import seam
- controls tests: three tests that manually mounted with $state props
  now use the mount_controls helper's returned reactive props

No behavior changes; 495 unit + 18 e2e tests pass.

* Replace manual signature tracking with derived strings, dedupe load guards

Deeper simplification pass:

- Isosurface: geometry and color signatures are now $derived strings —
  Svelte only notifies dependents when a derived value changes, so the
  rebuild/recolor effects skip unrelated updates natively, replacing
  three module-level last-signature/last-entries trackers and their
  compare-and-store choreography. Recolor-without-marching-cubes
  contract verified in the browser (colormap switch recolors the
  glycine ESP surface without a rebuild).
- multi-volume demo: run_load() centralizes the stale-load token,
  loading flag, and error handling shared by scenario loads and file
  imports (was duplicated try/catch/finally in both).
- generate_examples: hBN geometry (lattice, frac→cart, site fractions)
  shared between the CHGCAR and ELFCAR generators via hbn_geometry(),
  mirroring the Al slab helper. Generator output verified byte-identical
  via content hashes before/after.

* Fix periodic isosurface supercell coordinates

* Drop cache-key fields subsumed by the extraction range

After #377 unified integer tiling and fractional bounds through range
extraction, the range fully determines a volume's prepared grid: it
encodes tiling and halo for periodic volumes and both are irrelevant
for finite ones. Remove the now-redundant tiling_key/halo fields from
PreparedGrid and geometry_key — also stops finite-volume caches from
needlessly invalidating on supercell changes.

* fix CmdPalette with search text not navigating on option click

* Inline single-use halo helper and tiling-range construction

effective_halo had one caller left after the range unification (the
resolver already clamps and gates halo on periodicity), and the
integer-tiling DisplayRange collapses to a map over the tiling factors.

* Tighten demo scenario descriptions and make card text selectable

Scenario blurbs cut to their essentials (what renders + what it
demonstrates); user-select: text on the card buttons since browsers
suppress selection inside <button> by default.

* Clarify that cache reuse rebinds entry identity to current loop values

* Harden 3D viewer overlays and WebGL lifecycle for desktop embedders

- Overlay chrome that must stack above WebGL canvases (atom legend,
  cell-select dropdown, glass chips, carousel pager, viewer controls,
  draggable panes) now gets its own compositing layer: WKWebView (Tauri)
  reattaches canvas layers above non-composited overlays after its
  window is hidden and re-shown, ignoring z-index.
- Document the in-viewer overlay stack as --z-index-viewer-* custom
  properties in app.css and wire components to them (values unchanged).
- Pause auto-rotation while the page is hidden: desktop embedders
  disable background throttling, which left hidden windows re-rendering
  every auto-rotating canvas indefinitely.
- StructureViewport remounts its <Canvas> when the browser evicts the
  WebGL context without restoring it (bounded retries), so blank
  carousel cards self-heal.
- CellSelect opens on sustained hover (200ms) instead of instantly on
  mouseenter, so scrubbing the pointer across compact embeds no longer
  pops the menu open.

* Add arbitrary volume slices and scalable isosurface rendering

Generalize plane sampling with reusable filled and contour rendering, and move geometry extraction to an instrumented worker for responsive large-volume updates. Add benchmark and regression coverage while hardening related file-drop and UI state handling.

* Polish slice responsiveness and WebGL recovery

Render high-resolution slices without deep-proxy or contour stalls, surface exhausted WebGL recovery, and strengthen portable performance and cache regression coverage.

* Fix packaged isosurface worker resolution

Use an extensionless worker URL so source and compiled package layouts both resolve correctly.

* Stabilize isosurface E2E tests under CI load

Serialize the CPU-heavy suite while preserving four-worker parallelism for unrelated tests.

* Preserve structure camera rotation across hover

Keep the orientation gizmo mounted so hover visibility changes do not reinitialize orbit controls.

* Refine file picker chip corners

* Simplify isosurface pipeline and tests

Centralize geometry preparation and remove duplicated control, demo, and test setup while preserving performance-sensitive paths.

* Polish isosurface controls and viewer defaults

Use compact labeled color controls, simplify file picking behavior, and improve initial structure framing.

* Fix CI E2E hangs on unstable canvas screenshots

Playwright locator.screenshot waits for element stability; under CI load
continuous WebGL/layout updates never settle and burn the test timeout.
Capture canvases via clipped page screenshots instead.

* Avoid Playwright screenshots for 2D canvas E2E checks

page.screenshot still hung on the busy isosurface demo under CI load.
Read 2D canvas pixels via toDataURL; keep clipped page screenshots only
for WebGL where blit would be blank without preserveDrawingBuffer.

* Polish file-picker clicks, isosurface controls, and viewer chrome

Key pending clicks by URL so same-name files stay distinct, restore hover/focus
gizmo gating, and tighten labeled isosurface layer controls plus theme-aware UI.

* Require matching pending click for HeatmapMatrix dblclick

Align delayed click/dblclick disambiguation with FilePicker so orphaned or
cross-cell dblclicks schedule a single-click instead of firing ondblclick.

* Drop underscore prefixes on isosurface internal modules.

Prefer plain names and keep internals out of the package barrel instead of _* marking; also remove the unused bench:isosurface script.

* Fold page_visibility into scene props and cut low-value test fat.

Merge the tiny visibility module into props.svelte.ts and drop redundant FilePicker, controls, types, and perf harness coverage that mostly restated defaults.

* Keep Structure gizmo mounted across hover cycles

Gate visibility with the gizmo-visible CSS class only. Tying the gizmo prop
to viewer_active remounted Threlte Gizmo on every leave and broke the
hover-chrome E2E regression test.

---------

Co-authored-by: Stardust0831 <169599847+Stardust0831@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants