IO Support Matrix

Intent

Define which formats are GPU-native targets, which are acceptable hybrid paths, and which remain explicit fallback adapters.

Request Signals

  • io support matrix

  • format classification

  • gpu native format

  • hybrid path

  • fallback format

Open First

  • docs/architecture/io-support-matrix.md

  • docs/architecture/io-arrow.md

  • docs/architecture/io-files.md

Verify

  • uv run python scripts/check_docs.py --check

Risks

  • Compatibility formats becoming the design center instead of GeoArrow and GeoParquet.

  • Missing explicit fallback events on legacy GDAL formats hides work that bypasses the GPU stack.

Decision

  • GeoArrow is the canonical GPU-native interchange format.

  • GeoParquet is the canonical GPU-native persisted format.

  • WKB is a hybrid compatibility bridge.

  • GeoJSON and Shapefile are explicit hybrid pipelines.

  • Untargeted GDAL formats remain fallback-only until justified.

Matrix

Format

Default Path

Read

Write

GPU Canonical

GeoArrow

GPU-native

GPU-native

GPU-native

yes

GeoParquet

GPU-native

GPU-native

hybrid

yes

WKB

hybrid

hybrid

hybrid

no

GeoJSON

hybrid

GPU geometry + CPU properties

hybrid

no (geometry GPU-accelerated)

Shapefile

hybrid

hybrid

hybrid

no

GDAL legacy

fallback

fallback

fallback

no

Performance Notes

  • Most serious IO throughput work should center on GeoArrow and GeoParquet.

  • Compatibility formats must not become the design center.

  • Hybrid paths are acceptable when transport or container parsing is host-heavy but geometry decode, filtering, or encode can still move toward GPU stages.