Roadmap

This roadmap lists remaining work. For implemented behavior, see the ecosystem status and language reference.

Completed foundations

The current monorepo already provides:

• indentation-sensitive Arx parsing with typed functions and variables
• structured control flow, imports, classes, templates, assertions, and tests
• ASTx as the shared node model
• IRx semantic analysis, structured diagnostics, LLVM lowering, and linking
• on-demand native runtime features
• Apache Arrow C++ arrays, tensors, tables, Series, and RecordBatch IPC
• static-schema Arx DataFrames and fixed-shape numeric tensors
• ArxPy and ArxJIT package foundations

Native Arrow and collections

• [ ] Add dynamic indexing for runtime-shaped Tensor parameters.
• [ ] Define ownership and return semantics for runtime-shaped Tensor values.
• [ ] Add runtime validation when an unknown shape or row count flows into a statically sized target.
• [ ] Support partial Tensor constraints such as tensor[f64, 2, ...].
• [ ] Add symbolic shape parameters for generic algorithms.
• [ ] Expand the Arx DataFrame surface to UTF-8, nullable, and temporal columns already represented by lower-level Arrow facilities.
• [ ] Define an Arx-facing RecordBatch and streaming surface.
• [ ] Add selected Arrow Compute operations without turning IRx into a general query engine.
• [ ] Complete dynamic-list storage reclamation and collection ownership.

Arx language

• [ ] Stabilize language semantics and publish a versioned language specification beyond the lexical manifest.
• [ ] Decide the final const declaration behavior; it is currently reserved lexically but not a general parsed declaration form.
• [ ] Add break and continue to the Arx surface after aligning parser, ASTx, IRx, and documentation behavior.
• [ ] Expand string, temporal, and collection operations.
• [ ] Define constructor arguments and richer class lifecycle behavior.
• [x] Remove the unimplemented arx --shell option from the stable CLI. An interactive shell can return later as a separately designed feature.
• [ ] Grow the bundled pure-Arx standard library.

IRx and ASTx

• [ ] Continue separating stable public APIs from internal helper nodes.
• [ ] Expand lowering coverage without implying that every ASTx node must have an LLVM implementation.
• [ ] Stabilize native runtime ownership and ABI compatibility guarantees.
• [ ] Improve API documentation generated from Douki Python docstrings.
• [ ] Add explicit compatibility documentation for Python, llvmlite, LLVM, and native C++ toolchains.

Python entry points

ArxPy

• [ ] Add parse-from-string and parse-from-file APIs.
• [ ] Add compile, artifact, and execution result objects.
• [ ] Translate all Arx and IRx failures into the public ArxPy hierarchy.

ArxJIT

• [ ] Connect source extraction and validation to @jit.
• [x] Add fail-closed ASTx lowering for the scalar function shell and straight-line literal, parameter, arithmetic, unary, and comparison expressions.
• [ ] Extend ASTx lowering to assignments and validated control flow.
• [ ] Compile and call native functions through IRx.
• [ ] Implement signature inference, runtime marshalling, and caching.
• [ ] Add Arrow-backed array and tensor signatures after scalar compilation is stable.

Release readiness

• [ ] Define stability levels and compatibility policies for each package.
• [ ] Publish complete migration notes for breaking language or API changes.
• [ ] Expand cross-platform native-runtime coverage.
• [ ] Provide end-user binary/toolchain installation guidance for supported platforms.