RC 0.2.0

Changes based on code review feedback.

requested review from @normann

assigned to @salid

added 1 commit

• fbe086dd - Version bump to 0.2.0

Compare with previous version

Hi Daniel,

Thanks a lot for your MR. You did a very good job!

For the record, going through each point raised in the code review and in our email exchange:

• Dynare preprocessor options are now supported via the dynare_options argument in the dynare() function.

• PyIndices has been unified in a new indices.py module.

• No remaining use of Self: all class methods now use either quoted class names or omit return annotations.

• Docstrings have been added to all from_julia routines.

• PyGsSolverWs.from_julia is now a fully functional method. The former placeholder convert_jl_distribution has been removed, and a clean distribution_from_julia routine now returns proper scipy.stats objects. Much better!

• Logging is now configurable via setup_logging() and respects the DYNARE_PYTHON_LOGLEVEL environment variable (DEBUG, INFO, WARNING, ERROR, CRITICAL).

• Enum parsing has been cleaned up: fragile repr()-based parsing was replaced by cls[str(jl_enum)] in SymbolType.from_julia, EstimatedParameterType.from_julia, and PySGSolver.from_julia.

• Dynamic loading of external libraries (e.g., Tasmanian) is still a bit fragile:

  • If LIBTASMANIAN_PATH is wrong or the file is corrupted, the current implementation may crash without a clear error.
  • Required functions (like tsgConstructTasmanianSparseGrid) aren’t checked explicitly.
  • The library is loaded at module scope, which introduces side effects on import even if Tasmanian features aren’t used.
  • Although Tasmanian handles its own memory, the global ctypes.CDLL reference keeps the library loaded for the entire Python session, which could limit flexibility in modular or long-running applications.

A lazy-loading helper such as the following would improve safety and control:

_TASlib = None

def get_tasmanian_lib():
    global _TASlib
    if _TASlib is not None:
        return _TASlib

    libtas_path = os.getenv("LIBTASMANIAN_PATH")
    if not libtas_path:
        logger.warning("LIBTASMANIAN_PATH not set. Tasmanian-dependent features will be unavailable.")
        return None

    try:
        _TASlib = ctypes.CDLL(libtas_path)
        for name in ["tsgConstructTasmanianSparseGrid", ...]:
            getattr(_TASlib, name)
        return _TASlib
    except Exception as e:
        logger.error(f"Failed to load Tasmanian library: {e}")
        return None

Each method needing Tasmanian would then call:

lib = get_tasmanian_lib()
if lib is None:
    raise RuntimeError("Tasmanian is not available.")

I noticed you implemented a related pattern in PyTasmanianSG.__del__, which is great, but we may want to consolidate this logic into a central function like get_tasmanian_lib() for reuse and extensibility.

• Unused TypeVar declarations in PyGeneralizedFirstOrderAlgorithm were removed.
• The typo List(int) was corrected to List[int] in PyLUWs.
• Global initialization is now deferred via lazy_init. Initialization happens only on first use of dynare() (via ensure_initialized() and the _has_run flag). Users can also call ensure_initialized() manually to trigger setup earlier or outside of dynare().

Let me know what you think of the remaining points, especially regarding the Tasmanian loading strategy. Once that's settled, I believe the code will be ready for release. Thanks again for the quality of your work and your responsiveness throughout! It's a real pleasure working with you on this.

left review comments