LoginSign Up
← Back to all sparks
R

rpymat

ANALYTICS
Velocity0.0

Easy to Configure an Isolated 'Python' Environment

github.com/dipterix/rpymat ↗

After three dormant years, rpymat returned to fix the OpenMP crash that breaks R and conda together

rpythoncondareticulateopenmpinteroperability
◆Current state
rpymat manages an isolated conda-based Python environment for R packages, providing a reproducible bridge without touching the user's system Python. It released steadily through 2022-2023 and then went quiet for nearly three years. 0.1.9 in May 2026 is the first release since, and it addresses a specific and long-standing failure mode.
◆Where it's heading
The work is about surviving the seams between two runtimes. 0.1.9 addresses OpenMP double-initialization — the error users hit when R's OpenMP and conda's disagree — by setting KMP_DUPLICATE_LIB_OK as a compromise, and adds fix_omp_conflict() to symlink over conda's built-in version as the recommended real fix. The caveat is stated plainly: users must re-run it whenever R is updated, and the ABI versions must match. Earlier releases followed the same pattern, with 0.1.2 fixing segfaults from incompatible BLAS between numpy and R.
◆Prediction
The recurring theme across releases is native library conflicts between the R and conda stacks, so further releases are likely to keep patching that surface as Python versions move. The three-year gap makes cadence unpredictable.

◆Recent moves

  1. 2mo ago

    OpenMP conflict workaround and fix_omp_conflict() helper

    The first release in nearly three years, aimed at the OpenMP double-initialization error that surfaces when R's and conda's OpenMP runtimes collide. Sets KMP_DUPLICATE_LIB_OK as an immediate compromise and adds fix_omp_conflict() to symlink over conda's built-in version as the recommended fix, with the caveat that it must be re-run after every R update. Also refreshes the Python version registry for MATLAB and forces pip into the conda environment.

    View source ↗
  2. 3y ago

    Fix installation regression from 0.1.5

    Repairs an installation problem introduced in 0.1.5 and adds small utilities for basic Python types and global variables in run_pyscript.

    View source ↗
  3. 3y ago

    File choosers, reticulate conversion ports and reusable conda detection

    A cumulative entry covering 0.1.3 through 0.1.5. Adds cross-platform file selection dialogs, ports reticulate's R/Python object conversion, and lets R_RPYMAT_CONDA_EXE and R_RPYMAT_CONDA_PREFIX point at an already-installed conda — the change that lets conda-forge distribute the package without a second conda install. Configuration caching speeds up ensure_rpymat().

    View source ↗
  4. 4y ago

    Windows support and BLAS segfault fix

    Adds Windows support and Jupyter server management, and fixes segfaults caused by incompatible BLAS between numpy and R. The same class of native-library conflict that 0.1.9 revisits for OpenMP four years later.

    View source ↗