fastgeojson 0.3.0

Breaking changes

• sf_geojson_str() and df_json_str() are removed. as_json() replaces both and dispatches on its input.
• as_json() takes jsonlite::toJSON()’s arguments in the same order, so positional arguments after x change meaning, and follows jsonlite’s output conventions. Two defaults differ: numbers are lossless (digits = Inf; digits = 4 gives toJSON()’s output) and sf objects become GeoJSON (sf = "geojson").
• Requires R 4.5 or later.

Fixes

• Memory safety: worker threads read no attributes and hold no pointers into objects the call created; gctorture tests cover the cases that used to fail.
• Date, POSIXt, factor, complex, raw, matrices, nested and list-column frames, integer64 and non-UTF-8 encodings now follow toJSON(); jsonlite’s own test suite runs against as_json() with no failures.

Performance

• 20–31% faster than 0.2.2 at identical output, and faster than yyjsonr single-threaded on every shape measured. Shortest-digit output uses Żmij.
• as_bytes = TRUE returns a raw vector, skipping R’s string interning; fastgeojson_threads() sets the worker count.

Build

• All Rust crates are vendored; minimum Rust 1.71; no non-API R entry points; Windows on aarch64 builds.

fastgeojson 0.2.2

• Dataframe Orientation: Added the dataframe argument to toggle between row-oriented output (default, [{...}]) and column-oriented output ({...}).
• Context-Aware NA Handling: Implemented “Smart” default logic for missing values to match standard R conventions:
  • Default: In column mode, numeric NAs are coerced to "NA" strings to maintain array type homogeneity, while other types default to null. In row mode, NA values are omitted to reduce payload size.
  • Explicit: Specifying na = "null" or na = "string" strictly enforces the requested format regardless of the data type or structure.
• Null Value Control: Added the null argument to control the serialization of NULL (empty) values in lists, supporting coercion to empty containers ("list", default) or explicit JSON null.

fastgeojson 0.2.1

• Scalar Serialization Support: Extended the as_json() interface to include the auto_unbox argument, enabling the serialization of length-one atomic vectors as native JSON scalars. This provides deterministic control over data typing, allowing specific distinction between singleton arrays and primitive scalar values in the output payload.

fastgeojson 0.2.0

• New Function: Added as_json(), a high-performance, generic serializer that handles sf objects, data frames, lists, and atomic vectors. It serves as a parallelized, drop-in replacement for jsonlite::toJSON().
• Performance Engineering:
  • Geometry Arena: Implemented a contiguous memory arena for spatial data, flattening nested R lists into a linear structure to eliminate allocation overhead during parallel processing.
  • Direct-Heap Writing: Switched to ryu::raw for floating-point formatting, writing bytes directly to the final memory buffer to bypass stack copies.
  • LUT Escaping: Implemented a static Look-Up Table (LUT) for string escaping, enabling O(1) scanning of characters.
  • Loop Batching: Optimized vector writes by batching JSON tokens (e.g., ",["), to minimize capacity checks.

fastgeojson 0.1.3

• Fixed a parallel build race during Rust compilation on some CRAN check platforms by ensuring Rust build artifact cleanup runs only after package linking completes.

fastgeojson 0.1.2

• Fixed _abort symbol warnings on macOS and Linux by implementing proper Rust build artifact cleanup.
• Achieved clean compilation (0 warnings) across Windows, macOS, and Linux.

fastgeojson 0.1.1

• Build Stability: Pinned the compilation environment to Rust 1.75 via rust-toolchain.toml. This ensures strictly reproducible builds and maximizes compatibility with deployment servers like shinyapps.io and Posit Connect.
• Documentation: Added a “Deploying to shinyapps.io” guide to the README.

fastgeojson 0.1.0

• Initial Release: Introduced fastgeojson, a high-performance JSON and GeoJSON serializer backed by Rust.
• Core Functions:
  • Added sf_geojson_str(): Converts sf objects to GeoJSON FeatureCollections.
  • Added df_json_str(): Converts data frames to JSON arrays of objects.
• Performance: Implemented multi-threaded processing using the Rust rayon crate for massive datasets.
• Integration: Output strings are assigned the json and geojson classes to enable zero-copy transfer in Shiny and Plumber applications.