Skip to content

Tracking: Theme A file/query parity (before recipes B and dynamic C) #2

Open

Description

@jimbrig

Why this exists

This fork exists so tiling can stay out of R memory (Rust DuckDB + GeoParquet always on, including Windows). Recipes, overview GeoParquet, and cartographic tuning are later. The encoder already tiles Vec<LayerData> with per-layer zoom bands; the gaps are how we feed that kernel and how much RAM it holds.

Do first

Later (not this epic)

Recipes / overviews / serving upgrades should be separate issues so this list stays encoder-shaped.

Notes

Kernel: engine.rs generate_tiles, encode_tile_multilayer. File ingest: file_input.rs. Point-only spooling: streaming.rs. Default freestile_file(engine = "geoparquet") is the API lie.

Context: dev/docs/ideas.md (root-level workflow gaps).

Activity

  1. jimbrig commented on Nov 13, 2026

    @jimbrig
    OwnerAuthor

    Product themes (post-establishment)

    Fork work beyond repo hygiene / upstream sync collapses to three themes.

    A — API parity (in-memory sf vs file/query Rust path)

    Do first (this epic):

    Technical foundations (still relevant, not a separate theme):

    B — Recipes / cartographic knobs / source-to-map gap

    C — Dynamic MVT/MLT from DuckDB/GeoParquet

    Sequencing: A (+ tech) then B then C.

  2. changed the title [-]Tracking: file-to-file tiling before recipes and overviews[/-] [+]Tracking: Theme A file/query parity (before recipes B and dynamic C)[/+] on Nov 13, 2026
  3. jimbrig commented on Nov 13, 2026

    @jimbrig
    OwnerAuthor

    Product themes (post-establishment)

    Fork work beyond repo hygiene / upstream sync collapses to three themes.

    A — API parity (in-memory sf vs file/query Rust path)

    Do first (this epic): #3, #7, #4.

    Technical foundations still relevant: #5 (memory/spool; upstream streaming merged), #6 (GeoParquet 1.1/2.0 + CRS errors).

    B — Recipes / cartographic knobs / source-to-map gap

    #8 (blocked on A). Docs/examples/vignettes that prove maps. #9 is static large-file HTTP for B, not dynamic tiles.

    C — Dynamic MVT/MLT from DuckDB/GeoParquet

    Distinct from #9. New tracking issue when A is far enough (single-tile encode + bbox SQL + thin httpuv XYZ).

    Sequencing: A (+ tech) then B then C.

    Also watching upstream PR stack walkerke#31/walkerke#32/walkerke#33 (LayeredTiler) for Theme A memory parity.

  4. jimbrig commented on Nov 14, 2026

    @jimbrig
    OwnerAuthor

    Status update (2026-11-14)

    Working workbook: dev/docs/issues-review.md on branch wip/layered-tiler-absorb.

    Architecture plan: Themes A → B → C, Rust-first. LayeredTiler absorb (upstream PRs walkerke#31–walkerke#33) is on that branch — enables #3 memory story (peak ≈ largest layer). Not merged to main until CI/R rebuild prove it.

    Checklist refresh

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions