NGen
Loading...
Searching...
No Matches
CHANGELOG

All notable changes to this project will be documented in this file. We follow the Semantic Versioning 2.0.0 format.

x.y.z - YYYY-MM-DD

Added

  • An output configuration block for realization configs: root, per-domain (catchment/nexus) enable/format/grouping/rank_subdir, and a global value precision. See Realization Configuration.
  • Catchment output routed through a pluggable output manager (CSV backend), with an optional grouping: per_formulation that aggregates a formulation's catchments into one file with a leading catchment_id column (columns uniform per file by construction).
  • Per-domain rank_subdir option placing each MPI rank's CSV output under a rank_<N>/ subdirectory (applied automatically for per_formulation under MPI).
  • Initial hydrofabric v4.0 support, alongside the existing v2.2 and v4.0beta1 schemas.
  • A HydrofabricReader abstraction covering both supported hydrofabric formats, with GeoJSON and GeoPackage implementations selected by make_hydrofabric_reader(). Both drivers now obtain their catchment and nexus features through it instead of dispatching on the file extension themselves.

Changed

  • A GeoPackage with no nexus layer is no longer read as a hydrofabric. It was previously treated as v2.2 and read anyway; it is now reported as not a hydrofabric. Reading such a file as a plain GeoPackage of arbitrary layers is still supported, through ngen::geopackage::GeoPackageReader, which carries no hydrofabric knowledge at all.
  • Catchment and nexus data paths must name the same format. Passing a GeoPackage for one and a GeoJSON file for the other is now an error rather than being honored per path.
  • Hydrofabric data files are recognized by their contents rather than their file names. A GeoPackage is identified by the SQLite header it begins with, so it is read as one whatever it is called, and anything else is read as GeoJSON. The previous test required a name ending in a lowercase gpkg, which missed .GPKG, matched unrelated names ending in those letters, and — because loading a subdivided hydrofabric under MPI appends a rank suffix to both data paths — missed every GeoPackage in that mode.
  • The INFO: hydrofabric detected: ... line now reports for either format and fires exactly once per run. GeoJSON hydrofabrics previously reported nothing, and a single-file GeoPackage hydrofabric reported twice, once per file opened.

Changed

  • Models referenced as git submodules under extern/ are now built in the CMake binary tree under extern-libs/, not in the source tree as cmake_build/ directories under their respective source roots

Deprecated

  • Top-level realization config keys output_root, disable_catchment_output, and per_formulation_nexus_files, superseded by the output block (still honored when no output block is present).

Removed

  • The output_precision formulation parameter in realization configs. It has had no effect since catchment output moved to the output manager; text precision is the output block's precision value.

Fixed

  • Hydrofabric schema detection now reads each layer's columns from the file that holds it. A v4.0 hydrofabric split across a catchment file and a nexus file was previously detected as v4.0beta1, because divides.flowpath_toid – the column that separates the two – is absent from the nexus file; its divides were then joined through a flowpaths table that need not be there, leaving them with no downstream reference.