Skip to main content

Beta Release Notes

Current version: 0.3.0

The changes on this page are relative to 0.1.0 (and earlier).

This page tracks API and naming changes since the GeoBrix project started. After the project is approved, formal release notes will take over; until then, use this as the single place to look up what changed and why.

What's new in v0.3.0​

Released 2026-05-26. Per-version highlights; full migration tables are in the per-component sections below.

  • rst_clip CRS axis-order fix (all-black clips). GDAL 3+ defaults EPSG-imported SpatialReferences to authority-compliant axis order (lat/lon for EPSG:4326), which silently swapped axes against JTS/Databricks WKT/WKB cutlines so the clip missed the raster entirely. The reprojection now clones the source/destination SpatialReferences and forces OAMS_TRADITIONAL_GIS_ORDER before the OGR transform; caller-owned SpatialReferences are not mutated.
  • EWKT / EWKB support for rst_clip. JTS.fromWKT / JTS.fromWKB auto-detect EWKT/EWKB; new JTS.toEWKT / JTS.toEWKB helpers emit SRID-preserving forms. rst_clip reprojects the cutline when its SRID differs from the raster CRS, and falls back to the raster's CRS (Mosaic-compatible) when the SRID is 0 / unresolvable.
  • rst_transform rejects invalid SRIDs. targetSrid <= 0 and unresolvable EPSG codes now surface a clear error via tile metadata error_message instead of returning a raster with an uninitialized CRS.
  • /vsimem/ path-handling hardening. rst_memsize / rst_unlink / GDAL writer in-memory byte fetch now use startsWith("/vsimem/") (not contains) and null-check GetMemFileBuffer, so datasets whose description embeds the substring (e.g. NetCDF subdataset selectors) aren't mis-routed through the in-memory branch.
  • tile.raster bytes are always self-contained (no VRT payloads). Three RasterX operations — MergeRasters (gbx_rst_merge, gbx_rst_merge_agg), MergeBands (gbx_rst_frombands), and PixelCombineRasters (gbx_rst_derivedband, gbx_rst_derivedband_agg, gbx_rst_combineavg, gbx_rst_combineavg_agg) — used to return tiles whose metadata("driver") claimed VRT even though the on-disk file was a materialized GTiff. That mis-tag propagated through RasterDriver.writeToBytes (which keys both the tempfile extension AND the -of flag in the inner gdal_translate call off metadata.driver), causing the serialized tile.raster payload to be VRT XML referencing a /vsimem/ tempfile only reachable on the producing executor. Single-node testing passed by accident; multi-executor clusters hit file not found when the VRT was opened elsewhere. Fix: GDALTranslate.executeTranslate now records the output dataset's driver in its returned metadata (not the input's), and RasterDriver.writeToBytes defensively coerces VRT to GTiff on serialization + sniffs the result to refuse shipping VRT bytes. Regression coverage in RST_NoVrtPayloadTest.
  • PixelCombineRasters pixel function now actually fires (combineavg / derivedband were silently returning one of the inputs). gbx_rst_combineavg, gbx_rst_combineavg_agg, gbx_rst_derivedband, and gbx_rst_derivedband_agg build a multi-source VRT, inject a <PixelFunctionLanguage>Python</...> band, and re-open it for gdal_translate. The previous implementation re-opened the VRT before mutating the XML file, so the in-memory Dataset handle never saw the pixel function; gdal.Translate then fell back to a default multi-source mosaic (last-source-wins per pixel). On co-extensive inputs (e.g. a monthly EO time-series), the output silently equaled one of the inputs — non-deterministic per partition in a distributed setting, producing visible tile-of-different-years patchwork on multi-executor clusters. Fix: PixelCombineRasters.combine now injects the pixel function before the VRT is re-opened, and pre-creates the per-JVM NodeFilePathUtil.rootPath staging dir itself (previously only ClipToGeom did, so combineavg would file not found if it was the first op to hit a fresh JVM). Regression coverage: RST_AggregationsTest "CombineAvg actually averages pixel values" (two constant rasters 50 + 100 → output 75).
  • Friendly error on ARRAY<tile>-function misuse. Calling gbx_rst_combineavg, gbx_rst_merge, gbx_rst_frombands, or gbx_rst_mapalgebra on a single tile column (instead of an ARRAY<tile> like collect_list(tile)) used to surface as a raw ClassCastException: StructType cannot be cast to ArrayType from inside Catalyst analysis — untraceable from a notebook. The four expressions now route through RST_ExpressionUtil.arrayOfTileRasterType, which raises a clean IllegalArgumentException naming the function, the actual type received, and (where applicable) the aggregator companion the user likely wanted, e.g. gbx_rst_combineavg expects ARRAY<tile> (e.g. collect_list(tile) or array(t1, t2, ...)), but received STRUCT<...>. To aggregate the column across rows, use gbx_rst_combineavg_agg(tile).
  • Docs: GDAL_VRT_ENABLE_PYTHON for custom GDAL code paths. Built-in combineavg / derivedband calls auto-enable VRT Python via the in-process GDALManager.withVrtPython bracket — no cluster config needed. The new RasterX § VRT Python pixel functions section documents how to enable the same evaluation in your own GDAL calls (Python gdal.SetConfigOption, cluster spark.executorEnv, or the JVM withVrtPython helper) and points to the TRUSTED_MODULES variant for less-trusted VRT sources. A cross-reference is added in Security § 6 explaining why GeoBrix ships the option NO by default.
  • gbx_rst_derivedband / gbx_rst_derivedband_agg numerical-correctness regression coverage. These functions share the PixelCombineRasters code path with combineavg, so they were silently no-opping in the same way (returning one of the inputs unchanged on co-extensive stacks). The ordering fix above repairs both call sites, but the existing tests only checked that the result wasn't null — they would have passed either way. This release adds explicit pixel-value assertions: RST_AggregationsTest covers the in-process RST_DerivedBand path with a doubling pyfunc and a 3-input numpy-mean pyfunc, and RST_AggEvalTest covers the Spark-aggregation rst_derivedband_agg path end-to-end (three constant-Byte tiles 10/20/30 with a "mean × 2" pyfunc must yield 40 across the result tile). Two previously-passing tests used def myfunc(x): return x * 2 — an invalid VRT pixel-function signature — and were updated to the canonical (in_ar, out_ar, xoff, yoff, xsize, ysize, raster_xsize, raster_ysize, buf_radius, gt, **kwargs) shape; they only "passed" before because the pyfunc never actually ran.
  • gbx_rst_combineavg / gbx_rst_combineavg_agg math corrected (NoData, valid zeros, rounding). With the pixel function now firing (previous bullet), several latent bugs in the average kernel surface and are fixed in this release. The pyfunc used to sum every source value blindly — including each band's NoData sentinel (e.g. 255 on Byte EO products) — and counted only strictly-positive cells in the divisor (np.sum(stacked > 0, axis=0)), which (a) inflated the numerator with NoData and (b) wrongly excluded valid 0 measurements from the divisor. It also used np.divide(..., casting='unsafe'), which truncates rather than rounds when casting back to an integer output dtype (Byte / UInt16), producing systematic underbias on integer EO stacks. Now the kernel reads each source band's declared NoData (via BandAccessors.getNoDataValue, baked into the pyfunc source as a literal list at VRT-write time), masks NoData cells out of both sum and divisor, includes valid 0s, uses float64 internally, and rounds-to-nearest-even before the unsafe cast when the output dtype is integer. The bogus np.clip(out_ar, stacked.min(), stacked.max(), ...) (the bounds were contaminated by NoData sentinels) is removed. When at least one input declares NoData, that value is also stamped on the output band so downstream GetNoDataValue reports all-NoData pixels. Regression coverage in RST_AggregationsTest: "excludes declared NoData from both sum and divisor", "counts valid 0 cells in the divisor", "rounds (not truncates) when casting to integer output".
  • Scalar args without f.lit(...). Python wrappers auto-wrap bool / int / float / bytes; Scala adds typed overloads. SQL was already natively-typed. String literals still wrap in f.lit(...) per pyspark's column-ref convention. Details and migration examples in Scalar values vs lit(...) wrapping.
  • Example notebooks — EO Series, xView, and enablement diagrams. New end-to-end walkthroughs under docs/examples/ covering EO time-series, xView object-detection rasters, and RasterX architecture diagrams.
  • Supply-chain hardening (lockdown). Jobs pinned to the Databricks-hardened runner group (org-level allowlist, ephemeral VMs, constrained secret access); every Maven dependency, transitive dep, plugin, and plugin dependency is PGP-verified against .maven-keys.list before any compile or test execution; pip and Maven routed through JFrog with OIDC; init script + pinned package versions vetted; new Security page in the docs.
  • Pre-built, hash-verified GDAL bundle. The GDAL native install path is now a CI-built tarball (geobrix-gdal-artifacts-v<version>-noble.tar.gz + matching .sha256 sidecar, attached to each release alongside a versioned geobrix-gdal-init.sh). Cluster start drops from ~15 minutes (legacy PPA dance per boot) to ~30–90 seconds (verify sidecar → extract → dpkg -i). Trust chain is now four layers: CI-side GPG fingerprint pin → per-file SHA256SUMS inside the tarball → outer .sha256 sidecar in the staging Volume → the Volume's write ACL. The legacy on-cluster path is preserved as scripts/geobrix-gdal-init-ppa.sh for bundle bootstrapping. Bundle is amd64 / x86_64 only (Intel or AMD CPUs); ARM-based instance types — AWS Graviton, Ampere, Apple Silicon — are not supported. See Installation and the rationale on the Security page.

Conventions:

  • baseline — Name or behavior before the change (what to search for in old code or docs).
  • Notes — Short reason (e.g. standardize across languages, underscore standardization, _geometry → _geom).

General​

BaselineCurrentNotes
Python import geobrix.*databricks.labs.gbx.*Match Scala package and published artifact; avoid namespace clashes.
Extra underscores in function names (multi-word parts spelled with _)Single underscore between prefix and compound (e.g. rst_pixelwidth, gbx_bng_cellarea)Underscore standardization: one leading prefix, then one compound word; no _ inside the operation name.
Non-Column value args required f.lit(...) / lit(...) wrapping (e.g. rst_clip(tile, geom, f.lit(True)), bng_pointascell(pt, f.lit(1)))Plain Python/Scala non-string scalars accepted directly (e.g. rst_clip(tile, geom, True), rst_transform(tile, 4326), bng_pointascell(pt, 1))Matches Mosaic/DBR built-in ergonomics for booleans/numerics. Python wrappers auto-wrap bool/int/float/bytes via f.lit; Scala adds typed overloads. Strings still follow pyspark's column-ref convention — rx.rst_width("tile") is still f.col("tile"); wrap in f.lit(...) for string literals (e.g. driver=f.lit("GTiff")).

All specific function renames from that standardization are listed in the component tables below.

RasterX​

BaselineCurrentNotes
(GDAL reader output column) pathsourceDocs/tests aligned to GDAL reader output column name.
rst_band_metadata / gbx_rst_band_metadatarst_bandmetadata / gbx_rst_bandmetadataUnderscore standardization.
rst_bounding_box / gbx_rst_bounding_boxrst_boundingbox / gbx_rst_boundingboxUnderscore standardization.
rst_pixel_width / gbx_rst_pixel_widthrst_pixelwidth / gbx_rst_pixelwidthUnderscore standardization.
rst_pixel_height / gbx_rst_pixel_heightrst_pixelheight / gbx_rst_pixelheightUnderscore standardization.
rst_num_bands / gbx_rst_num_bandsrst_numbands / gbx_rst_numbandsUnderscore standardization.
rst_pixel_count / gbx_rst_pixel_countrst_pixelcount / gbx_rst_pixelcountUnderscore standardization.
rst_scale_x / gbx_rst_scale_xrst_scalex / gbx_rst_scalexUnderscore standardization.
rst_scale_y / gbx_rst_scale_yrst_scaley / gbx_rst_scaleyUnderscore standardization.
rst_upper_left_x / gbx_rst_upper_left_xrst_upperleftx / gbx_rst_upperleftxUnderscore standardization.
rst_upper_left_y / gbx_rst_upper_left_yrst_upperlefty / gbx_rst_upperleftyUnderscore standardization.
rst_geo_reference / gbx_rst_geo_referencerst_georeference / gbx_rst_georeferenceUnderscore standardization.
rst_get_nodata / gbx_rst_get_nodatarst_getnodata / gbx_rst_getnodataUnderscore standardization.
rst_get_subdataset / gbx_rst_get_subdatasetrst_getsubdataset / gbx_rst_getsubdatasetUnderscore standardization.
rst_mem_size / gbx_rst_mem_sizerst_memsize / gbx_rst_memsizeUnderscore standardization.
rst_sub_datasets / gbx_rst_sub_datasetsrst_subdatasets / gbx_rst_subdatasetsUnderscore standardization.
rst_combine_avg_agg / gbx_rst_combine_avg_aggrst_combineavg_agg / gbx_rst_combineavg_aggUnderscore standardization.
rst_derived_band_agg / gbx_rst_derived_band_aggrst_derivedband_agg / gbx_rst_derivedband_aggUnderscore standardization.
rst_from_content / gbx_rst_from_contentrst_fromcontent / gbx_rst_fromcontentUnderscore standardization.
rst_from_file / gbx_rst_from_filerst_fromfile / gbx_rst_fromfileUnderscore standardization.
rst_from_bands / gbx_rst_from_bandsrst_frombands / gbx_rst_frombandsUnderscore standardization.
rst_make_tiles / gbx_rst_make_tilesrst_maketiles / gbx_rst_maketilesUnderscore standardization.
rst_re_tile / gbx_rst_re_tilerst_retile / gbx_rst_retileUnderscore standardization.
rst_separate_bands / gbx_rst_separate_bandsrst_separatebands / gbx_rst_separatebandsUnderscore standardization.
rst_to_overlapping_tiles / gbx_rst_to_overlapping_tilesrst_tooverlappingtiles / gbx_rst_tooverlappingtilesUnderscore standardization.
rst_init_nodata / gbx_rst_init_nodatarst_initnodata / gbx_rst_initnodataUnderscore standardization.
rst_is_empty / gbx_rst_is_emptyrst_isempty / gbx_rst_isemptyUnderscore standardization.
rst_map_algebra / gbx_rst_map_algebrarst_mapalgebra / gbx_rst_mapalgebraUnderscore standardization.
rst_raster_to_world_coord / gbx_rst_raster_to_world_coord (and X/Y variants)rst_rastertoworldcoord / gbx_rst_rastertoworldcoord (and X/Y)Underscore standardization.
rst_world_to_raster_coord / gbx_rst_world_to_raster_coord (and X/Y variants)rst_worldtorastercoord / gbx_rst_worldtorastercoord (and X/Y)Underscore standardization.
rst_as_format / gbx_rst_as_formatrst_asformat / gbx_rst_asformatUnderscore standardization.
rst_combine_avg / gbx_rst_combine_avgrst_combineavg / gbx_rst_combineavgUnderscore standardization.
rst_h3_raster_to_grid_avg (and Count/Max/Min/Median)rst_h3_rastertogridavg (and Count/Max/Min/Median)Underscore standardization.
rst_bandmetadata(tile) (single arg)rst_bandmetadata(tile, band)Required band parameter added; use e.g. rst_bandmetadata("tile", f.lit(1)).
rst_fromfile raster field was StringType (path) with metadata.size = -1rst_fromfile raster field is BinaryType (file bytes) with real metadata.sizerst_fromfile now reads the file into the tile, so tiles are self-contained and downstream ops (e.g. rst_clip) no longer produce orphan temp paths. Matches rst_fromcontent and the GDAL reader.
Default output compression was ZSTD (TIFF tag 50000)Default output compression is DEFLATE (baseline TIFF)ZSTD output was not decodable by Java ImageIO and broke the Databricks image preview after operators like rst_clip. DEFLATE is universally previewable and (with PREDICTOR=2/3) still compresses well. Override per-call via tile metadata compression key.

GridX (BNG)​

BaselineCurrentNotes
bng_eastnortasbng (Python) / gbx_bng_eastnortasbng (SQL)bng_eastnorthasbng / gbx_bng_eastnorthasbngStandardize across languages (Python had typo; Scala already eastnorth).
bng_cell_area / gbx_bng_cell_areabng_cellarea / gbx_bng_cellareaUnderscore standardization.
bng_cell_intersection / gbx_bng_cell_intersectionbng_cellintersection / gbx_bng_cellintersectionUnderscore standardization.
bng_cell_union / gbx_bng_cell_unionbng_cellunion / gbx_bng_cellunionUnderscore standardization.
bng_euclidean_distance / gbx_bng_euclidean_distancebng_euclideandistance / gbx_bng_euclideandistanceUnderscore standardization.
bng_point_as_bng / gbx_bng_point_as_bngbng_pointascell / gbx_bng_pointascellUnderscore standardization; Renamed for clarity: point → cell (not "point as BNG").
bng_cell_intersection_agg / gbx_bng_cell_intersection_aggbng_cellintersection_agg / gbx_bng_cellintersection_aggUnderscore standardization.
bng_cell_union_agg / gbx_bng_cell_union_aggbng_cellunion_agg / gbx_bng_cellunion_aggUnderscore standardization.
bng_geometry_kring / gbx_bng_geometry_kringbng_geomkring / gbx_bng_geomkring_geometry → _geom in name.
bng_geometry_kloop / gbx_bng_geometry_kloopbng_geomkloop / gbx_bng_geomkloop_geometry → _geom in name.
bng_geometry_kring_explode / gbx_bng_geometry_kring_explodebng_geomkringexplode / gbx_bng_geomkringexplode_geometry → _geom + underscore standardization.
bng_geometry_kloop_explode / gbx_bng_geometry_kloop_explodebng_geomkloopexplode / gbx_bng_geomkloopexplode_geometry → _geom + underscore standardization.
bng_k_ring / gbx_bng_k_ringbng_kring / gbx_bng_kringUnderscore standardization.
bng_k_loop / gbx_bng_k_loopbng_kloop / gbx_bng_kloopUnderscore standardization.
bng_k_ring_explode / gbx_bng_k_ring_explodebng_kringexplode / gbx_bng_kringexplodeUnderscore standardization.
bng_k_loop_explode / gbx_bng_k_loop_explodebng_kloopexplode / gbx_bng_kloopexplodeUnderscore standardization.
bng_tessellate_explode / gbx_bng_tessellate_explodebng_tessellateexplode / gbx_bng_tessellateexplodeUnderscore standardization.

VectorX​

BaselineCurrentNotes
(Schema/column) _geometry_geomStandardize geometry column suffix across readers and examples.
st_legacy_as_wkb / gbx_st_legacy_as_wkbst_legacyaswkb / gbx_st_legacyaswkbUnderscore standardization.

Readers​

BaselineCurrentNotes
shapefileshapefile_ogrReader namespace: format + engine to avoid conflicts with other Spark extensions.
geojsongeojson_ogrSame.
ogr_gpkggpkg_ogrSame; consistent format_engine order.
file_gdbfile_gdb_ogrSame.
(none)gtiff_gdalNew reader: named GDAL reader for GeoTIFF; use instead of gdal with option("driver", "GTiff").
info

Reader renames above are planned for 0.2.0. Beta (0.1.x) may still expose the baseline names in some contexts.

Scalar values vs lit(...) wrapping​

Previously, every non-Column argument had to be wrapped in f.lit(...) (Python) or lit(...) (Scala). That was a regression from Mosaic/DBR built-ins, where booleans and numerics can be passed as plain values. In 0.3.0, plain scalars are accepted across Python, Scala, and SQL bindings.

Python — wrappers accept Column or scalar (bool/int/float/bytes); non-string scalars are auto-wrapped with f.lit(...). Strings still follow pyspark's column-reference convention (bare string ≈ f.col(name)); wrap in f.lit("...") to pass a string literal.

# ✅ Before 0.3.0 — required f.lit for every value
rx.rst_clip("tile", "geom", f.lit(True))
rx.rst_transform("tile", f.lit(4326))
bx.bng_pointascell("pt", f.lit(1))
bx.bng_pointascell("pt", f.lit("1km"))

# ✅ 0.3.0 — scalars accepted directly
rx.rst_clip("tile", "geom", True)
rx.rst_transform("tile", 4326)
bx.bng_pointascell("pt", 1)
bx.bng_pointascell("pt", f.lit("1km"))   # string literal — still wrap in f.lit

Scala — typed overloads added for Boolean / Int / Double / String value parameters. Column args (e.g. geometry, tile) still take Column.

// ✅ 0.3.0 — scalar overloads resolve without lit(...)
rst_clip(col("tile"), col("geom"), cutlineAllTouched = true)
rst_transform(col("tile"), 4326)
bng_pointascell(col("pt"), 1)
bng_pointascell(col("pt"), "1km")

SQL — values are already natively accepted by Spark SQL; no change needed:

SELECT gbx_rst_clip(tile, geom, true) FROM ...;
SELECT gbx_bng_pointascell(pt, 1) FROM ...;
SELECT gbx_bng_pointascell(pt, '1km') FROM ...;

When you still need f.lit(...) in Python:

  • String literals: rx.rst_fromfile(f.lit("/path/to.tif"), f.lit("GTiff")) — a bare string is treated as a column reference.
  • Nulls / explicit typing: e.g. f.lit(None).cast("double").

How to use this page​

  • Migrating code: Search for the baseline name in your code or config; replace with Current and apply any behavior notes.
  • Docs or tests: After a change, add one row here so future readers know what changed and why.
  • After approval: Move content into formal release notes (e.g. per-version sections) and keep this page for historical beta-only changes, or retire it.

Notable improvements and fixes​

  • Python package rename: Imports changed from geobrix.* to databricks.labs.gbx.* to align with Scala and the published artifact; update all import statements and environment references.
  • Init script / NumPy: Init script updated to install NumPy 2.x so GDAL Python array operations execute correctly; fixes runtime failures in gbx_rst_mapalgebra and gbx_rst_ndvi when used with array-based paths.
  • Error handling: Functions that previously threw exceptions during execution now surface errors more clearly (e.g. return null or a controlled default with error messages captured) instead of failing with opaque stack traces.
  • RasterX rst_bandmetadata: A required band argument was added; call as rst_bandmetadata(tile, band) (e.g. rst_bandmetadata("tile", f.lit(1))) in Python/SQL/Scala.
  • GDAL reader column: Raster DataFrames from the GDAL reader use the column name source (not path) for the file path; update any code or docs that assumed path.
  • BNG aggregators (bng_cellunion_agg, bng_cellintersection_agg): Fixed a bug where aggregation buffers were shared across partitions (and across tests in the same JVM), causing incorrect core flags when running full test suites or with multiple partitions. Each partition now gets a fresh buffer. Chip fields are resolved by type/name in the union aggregator for robustness to struct field order. Test expectation corrected for “all core chips” intersection: result is now correctly documented as core=true (whole cell).
  • rst_clip axis-order fix for EPSG-imported CRS (fixes all-black clips): When the clip geometry's CRS was set via an EPSG code (plain rst_transform-style input, EWKT SRID=4326;..., or EWKB with SRID), GDAL 3+ defaults that SpatialReference to authority-compliant axis order — for EPSG:4326 that means (latitude, longitude). JTS / Databricks / most GIS tooling emit WKT/WKB coordinates in traditional (x, y) = (lon, lat) order, so the reprojection inside rst_clip was silently swapping the axes (e.g. -80 14 interpreted as lat=-80, near the south pole) and the cutline missed the raster entirely, producing all-black output. OSRTransformGeometry.transform now clones both source and destination SpatialReferences and forces OAMS_TRADITIONAL_GIS_ORDER on the clones before running the OGR transform, so JTS-origin WKB is interpreted correctly. Caller-owned SpatialReferences are not mutated.
  • EWKT / EWKB support for raster clip (CRS mismatch handling): rst_clip now accepts EWKT (SRID=<epsg>;<WKT>) and EWKB (PostGIS extended WKB) in addition to plain WKT/WKB. Semantics:
    • Plain WKT / WKB (no SRID): the geometry is assumed to already be in the raster's CRS; no reprojection is performed.
    • EWKT / EWKB (SRID set and resolvable via EPSG): the geometry's CRS is used and, if it differs from the raster's CRS, the cutline is reprojected before clipping.
    • If the SRID is 0 or not a valid EPSG code, the code falls back to the raster's CRS (same as the plain case) — this restores Mosaic-compatible behavior but no longer silently produces an empty/black clip when a caller forgets to set the SRID. JTS.fromWKT / JTS.fromWKB now auto-detect EWKT/EWKB; new JTS.toEWKT / JTS.toEWKB helpers emit SRID-preserving forms. Plain toWKT / toWKB output is unchanged (OGC, no SRID).
  • rst_transform invalid SRID: rst_transform(tile, targetSrid) now rejects targetSrid <= 0 and EPSG codes that GDAL cannot resolve with a clear error (surfaced in tile metadata error_message) instead of returning a raster with an uninitialized CRS.
  • /vsimem/ path handling hardening: rst_memsize / rst_unlink and the GDAL writer's in-memory byte fetch now use startsWith("/vsimem/") (not contains) and null-check GetMemFileBuffer, so datasets whose description happens to embed the substring (e.g. NetCDF subdataset selectors) are no longer mis-routed through the in-memory branch.