compute_cell_bitmask_v3 now reads the 8-bit neighbour bitmask from
the render-cell's own 4 corner paint-tiles (cardinal bit set iff
at least one of the 2 paint-tiles on that edge is painted; diagonal
bit set iff the corner paint-tile is painted), not from the
any-corner material status of the 8 neighbour render-cells.
Pre-0.5.6 used cell-neighbour-material, which let two cells share
connectivity across an empty paint-tile gap whenever any unrelated
corner of either was painted — producing connected blob shapes
where two visually separated 2x2 islands were expected.
Atlas, paint storage, override sublayer, public API and the slot
lookup table all unchanged. Doc comment and README painting-model
section rewritten to describe the dual-grid offset explicitly.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Exposes what the v3 vertex/material render path computes internally:
per (layer, x, y), returns {slot, rot, flip} for material cells,
nil for non-material. Override entry wins (normalised via existing
normalize_override_entry); otherwise resolves through
compute_cell_bitmask_v3 + SLOT_LOOKUP.
Primary consumer: debug-overlay tooling (map-editor 0.2.0c.3) that
labels each rendered cell with its blob-14 slot identity for
authoring verification.
Public accessors for palette-like consumers (map-editor 0.2.0c.2)
that need to render real tile thumbnails. All three return nil when
the atlas-idx is out of range, the slot has no tile in the atlas,
or the texture failed to load (headless test env).
atlas_tile_uv resolves the slot via the same `slot_NN_` regex used
internally by the v3 vertex/material render path, so callers see
the same slot -> tile_id mapping the renderer uses.
build_map_v3 caches an alias->atlas dict (atlas_by_alias) at map-load
time, pointing at the resolved stubs from load_tilemap. load_textures
later REPLACES m.atlases[i] in-place with the populated record
(diffuse_texture_handle + per-tile UVs from the lib-asset's
tiles.atlas.json) but never updated atlas_by_alias.
Effect: the v3 vertex/material render path resolves
m.atlas_by_alias[layer.material] to the PRE-load stub, finds no
diffuse handle, and short-circuits to the MISSING_ASSET_COLOR
fallback — every material-bound cell renders solid yellow regardless
of the bitmask/slot computation.
Fix: after each m.atlases[atlas_idx] = ... write in load_textures,
also overwrite m.atlas_by_alias[atlas_alias] with the new record.
Guarded on `if m.atlas_by_alias` so v2 maps (no autotile path) are
unaffected.
Found by the editor's freshly-bound work map after 0.2.0b.1 — vagrant
was unaffected by the bug because its hand-authored map was loaded
fine on the very first frame, masking the issue.
Vertex-grid write/read for the autotile painting path. The map-editor
0.2.0a Auto-Tile mode is the primary consumer; painting a single
vertex causes up to 4 surrounding cells to flip to material via the
any-corner rule. Coordinate-checks reject OOB writes; grid is lazily
allocated on first paint.
Three small follow-ups bundled into a single version bump:
1. Real tile.opaque consumption (cell_is_opaque_on_layer)
v0.5.0e shipped a slot-13 heuristic for vertex-painted layers
because the atlas-baker wasn't computing real opacity. Atlas-baker
v0.2.0 now sets tile.opaque via alpha-analysis (all-alpha-255
detection). Both vertex-painted and tiles[] paths now resolve the
actual slot and read its tile.opaque from atlas metadata. Pre-
v0.2.0 atlases without the flag report not-opaque (safe-
conservative: extra draws, no missed cells).
2. Manifest-schema patch (reserved fields, accept-but-ignore)
Type-checked acceptance of forthcoming manifest fields the
validator will need before consumers can ship them in real
manifests without breaking changes:
map-level: z_level (int), z_below (string)
per-layer: base_color (string #RRGGBB), tint_override (table),
collision_policy (table), traversal_modes (table),
foundation_mode (array)
Future slices (lib-core.maps v0.6 multi-z-level, lib-core.actor
movement-modes, etc.) wire the fields into actual behaviour.
3. Override format extension {slot, rot, flip} (design paper §15.1)
Override entries now accept EITHER a bare integer slot_id (compact
canonical-orientation form, backwards-compat with 0.5.0d) OR an
object `{slot, rot?, flip?}` for explicit-transform placement
(unblocks map-editor v0.2's Direct-mode Transform controls).
New normalize_override_entry helper converts disk-form to runtime
{slot, rot, flip}. Validator (validate_map_table_v3) type-checks
per-entry. set_override accepts either form and auto-stores as
compact (bare-int) when rot+flip both 0, object form otherwise —
minimises disk diff for the common canonical case. get_override
returns normalized form regardless of how the entry is stored.
draw_layer + cell_is_opaque_on_layer updated to read the
normalized form.
Closure-gate: milestone-check.sh ctest + test-all-modules GREEN.
SPOREL_CI=1 vagrant-skeleton rc=0 with 60 render_frame_ok; no
magenta-placeholders, no render-hook errors. Existing override-cells
(vagrant's 3 wall pillars) still render correctly through the new
normalize path; opaque-ceiling cache now reads real atlas data.
Consumer dep-bumps for v0.5.1 land in their own commits per repo:
sporel-lib-core.render, sporel-module-vagrant-skeleton, sporel-
module-map-editor[-test].
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three new public functions (set_cell_gid, set_roof, save_to_disk)
documented in the README alongside the existing read API. Purely
additive bump — existing consumers continue to compile and run
unchanged.
draw_layer now samples tile sub-rectangles from the atlas diffuse
texture via the ADR-0044 source-rect render primitive (13-arg
draw_sprite_transform). load_textures gains a walkable field in
tile-records (from atlas JSON) so gameplay queries work after the
tiles-dir is removed. load_tilemap falls back to an atlas-bootstrap
stub when the legacy tilemap JSON is absent, allowing maps.load to
succeed before load_textures is called. Texture loading uses pcall
to degrade gracefully in headless environments. Bumps lib version to
0.3.0 to flag the breaking change in tile-record shape (id is now
integer, name is the stable string identifier).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Sprite-mode in draw_map:
- tilemap-tiles with texture atlas-id render via
engine.render.draw_sprite_transform with rotation pivot at tile center
- tiles without texture continue to render via draw_rect with color
(Phase 1 mode preserved)
Tile rotation:
- maps support an optional tile_rotations parallel array in the map
JSON with per-cell 90-degree rotation values (0/90/180/270)
- absent or nil entries default to 0 (no rotation)
- enables 9-segment tile reuse via rotation rather than per-orientation
sprites
New API:
- maps.load_textures(asset_aliases) resolves the current tilemap's
tile atlas-ids to texture-handles via the asset-lib indirection
(asset_aliases[tilemap.asset_pack] -> lib-id -> atlas.json lookup
per tile)
Schema additions:
- tilemap.asset_pack (alias-key, optional)
- tile.texture (atlas-id, optional)
- map.tile_rotations (parallel array, optional)
All existing Phase 1 color-only tilemaps render unchanged.
Pre-existing bug: split_namespaced_id used first-dot-split. For modules
with dotted IDs (e.g. lib-core.maps-test), this misclassified
'<dotted-module-id>.<tilemap-name>' as cross-lib reference.
Replaced with split_local_tilemap() that matches by exact current-
module-id prefix. Module-id can now have arbitrary dots.
Surfaced by P.3.1 reference test-module 'lib-core.maps-test' which
has a dotted module-id.
