Notes on Py 3.13, update pyproject

.github/workflows/test.yml

@@ -70,7 +70,7 @@ jobs:
   test:
 
     runs-on: ${{ matrix.os }}
-    continue-on-error: ${{ matrix.python-version == '3.13' }}
+    # continue-on-error: ${{ matrix.python-version == '3.13' }}
 
     strategy:
       matrix: # https://github.com/actions/runner-images
@@ -81,6 +81,8 @@ jobs:
             python-version: '3.8'
           - os: macos-latest # aarch64
             python-version: '3.9'
+          - os: windows-latest # SCS missing windows wheels for 3.13, https://github.com/bodono/scs-python/pull/117
+            python-version: '3.13'
 
     steps:
 

README.rst

@@ -35,6 +35,13 @@ The documentation of the library is at
 
 *News:*
 
+   Python 3.13 is partially supported. `One dependency is missing compiled wheels
+   <https://github.com/bodono/scs-python/pull/117>`_: Pip installation via source distribution
+   (local compilation) should work on Mac out of the box and on Linux if you
+   have BLAS/LAPACK development libraries installed. Conda installation may work
+   on Windows too. Also, Numpy 2 is currently causing some issues with CVXPY on
+   Python 3.13, so we're temporarily restricting Numpy to < 2 on Python >= 3.13.
+
    Since end of 2023 we're running daily `example strategies
    <https://github.com/cvxgrp/cvxportfolio/tree/master/examples/strategies>`_
    using the `development (master) branch

pyproject.toml

@@ -24,9 +24,9 @@ maintainers = [{name = "Enzo Busseti", email = "[email protected]"}]
 # from some combination of dependency versions that break Cvxportfolio code.
 
 dependencies = ["pandas",
-    # July 2024, Numpy 2 causes issues with CVXPY's conic interface on Py 3.12
-    "numpy; python_version<'3.12'",
-    "numpy<2.0.0; python_version>='3.12'",
+    # October 2024, Numpy 2 causes issues with CVXPY on Python 3.13
+    "numpy; python_version<'3.13'",
+    "numpy<2.0.0; python_version>='3.13'",
     "matplotlib", "requests", "cvxpy",
     # "multiprocess", # robustifies usage w/ 3rd party modules, used if installed
     "scs" # it's hardcoded as fallback solver if numerical errors