Skip to main content

Known Limitations

tile57 runs the official IHO S-101 Portrayal Catalogue, and on the test charts (US4MD81M.000, US5MD1MC.000 + updates) every feature portrays with zero rule errors — depth areas and contours, soundings, coastline and land, buoys and beacons, lights (including sector legs and arcs), dangers (obstructions / wrecks / rocks), data-quality zones, restricted/anchorage areas, and text labels.

But "no rule errors" is not the same as "pixel-perfect S-52" — do not use it for navigation. This page is an honest list of what is still incomplete, taken from the engine code.

Not for navigation

See the warning on the introduction. NOAA ENC charts are U.S. public domain and not for navigation; this renderer adds its own gaps on top.

Native S-101 charts​

A native S-101 (S-100 Part 10a) chart is read directly into the S-101 portrayal model with no conversion — the format is auto-detected from the file, so png, bake, and the C API accept a native S-101 .000 transparently, and its sequential .001… update files are applied (record-level insert / delete / modify by identity). On the SHOM France test datasets and the S-164 update conformance set every feature portrays (no errors), including complex attributes such as light sectors and buoy topmarks. The cursor-pick report serves each feature's S-101 class and full attribute tree (nested complex attributes included), with text preserved as UTF-8. One area is not yet wired for native S-101:

  • Feature-to-feature associations. FASC/INAS records are parsed but the StructureEquipment relationship (e.g. a light supportedBy its beacon) is not yet resolved for the query report's related-feature grouping. It carries no portrayal weight for the test charts, so rendering is unaffected.

S-57 → S-101 conversion​

An S-57 chart's portrayal rules are S-101, so an S-57 chart passes through an S-57 → S-101 adapter (src/s101/adapter.zig) before the rules run (a native S-101 chart skips this — it is already in the S-101 model). S-57 has no perfect S-101 translation; the adapter follows the IHO S-65 conversion guidance, and the result is best effort:

  • Missing S-101 content stays missing. S-101 attributes and features with no S-57 source are never invented; rules that test them take their fallback branch.
  • Unconvertible objects fall back or drop. An S-57 object class with no S-101 equivalent portrays as the S-52 question mark (QUESMRK1) — or, where the S-65 guidance says the object is simply not carried into S-101 (e.g. an area-geometry recommended track), it is dropped.

Portrayal gaps​

  • Ignored portrayal-instruction keys. The instruction translator (src/s101/instructions.zig) lowers the drawing vocabulary the catalogue actually leans on (point/line/text instructions, colour fills, area-fill and viewing-group references, AugmentedRay / ArcByRadius / AugmentedPoint construction — that is how light-sector legs and arcs render). A few keys are skipped: AugmentedPath (its path-grouping semantics — the grouped figures stroke individually instead), AlertReference (non-visual ECDIS alert metadata), DisplayPlane, Hover, LinePlacement, TextVerticalOffset, and the rare standalone ScaleMinimum / ScaleMaximum / Dash forms.
  • Full-length sector lines are not baked. Sector legs render at their fixed S-52 display length; the mariner's full-length sector line variant (legs drawn to the light's nominal range) is not emitted, so the show_full_sector_lines setting currently has nothing to act on.
  • Sector figures can stop at a tile boundary beyond the owning chart. The compositor keeps a light's sector legs and arcs WHOLE across ownership boundaries (they are fixed-size decorations anchored at the light, exempt from face clipping). The remaining gap: a figure reaching into a tile where the owning chart holds no ground at all is absent there — the compositor never consults that chart for the tile — so a figure within roughly one tile of the chart's owned ground can cut at that tile's edge (directional ground-length legs can reach further).
  • Single-primitive rules vs. non-conformant geometry. Some S-101 rules handle only one primitive (e.g. RecommendedTrack is Curve-only); a chart that encodes the feature with another primitive (an area-encoded recommended track) errors in the rule and the feature is suppressed.
  • Low-accuracy sounding ring via spatial QUAPOS. SNDFRM04's low-accuracy ring triggers on the direct quality attributes (QUASOU/STATUS); the rule's fallback to the spatial quality-of-position (QUAPOS on the geometry) is not wired, so a sounding whose only low-accuracy signal is spatial QUAPOS misses its ring. (Approximate-position line dashing from spatial QUAPOS is applied.)
  • M_NSYS direction-of-buoyage arrow. The IALA-A/IALA-B system boundary (MARSYS51 / NAVARE51 linestyles) is drawn; the ORIENT-driven DIRBOY01/A1/B1 arrow (CentreOnArea) is not yet ported.

Display / style gaps​

  • Overscale hatch occlusion is tile-path only. The S-52 §10.1.10 overscale indication (OVERSC01, see architecture) is gated correctly everywhere, but only the generated MapLibre style sandwiches the hatch under finer at-scale fills. The PNG/PDF/ASCII surfaces draw in S-52 priority order without that sandwich, so where a finer chart overlaps a coarser one the hatch can show through fills that should occlude it.
  • SCAMIN gating snaps to integer zooms outside bucket mode. With a SCAMIN manifest, the style builds per-value layers with exact fractional native minzooms. The no-manifest zoom-filter fallback and the filter-gate mode still honour each feature's value, but MapLibre re-evaluates filters at integer zoom steps, so features appear/disappear on whole-zoom boundaries rather than at their precise 1:N display scale.

The tile57 CLI baker runs the same full S-101 portrayal as the live chart path, so a baked archive matches live generation. The native PNG/PDF renderer has its own short list of deliberate gaps — see The Rendering Engine.

ENC_ROOT loading​

Opening an ENC_ROOT (Chart.openPath / openChartsStreaming) builds a cheap spatial index (band + bbox per chart) and reads a chart's bytes only when a metadata or feature query needs them — the catalogue opens in seconds and memory stays bounded. It serves metadata and extraction only: tiles and views always come from a bake (bake each chart once, then compose on demand). Caveats:

  • Baking is the import step. The whole NOAA catalogue is a multi-minute one-time bake (tile57 bake / tile57_bake_tree); a region is far quicker, and re-runs are incremental — only charts whose source changed re-bake.
  • Index-scan cost. Opening the whole catalogue as a streaming chart pays a one-time index scan (a few seconds, parsing every chart header).
  • Low zoom is style-gated. The generated style's vector-source minzoom is the bake's tile floor (default 8), and MapLibre never requests tiles below a source's minzoom — nothing draws below it regardless of the data.