Bump cvxpy floor to 1.9.0 and add moreau extra

CLAUDE.md

@@ -96,6 +96,6 @@ Key test patterns:
 ## Dependencies
 
 - Python >= 3.11
-- CVXPY >= 1.7.4 (for native DGP→DCP reduction)
+- CVXPY >= 1.9.0 (provides `quad_form_dpp_scope` for parametric quad_form, native DGP→DCP reduction)
 - diffcp >= 1.1.0 (default solver backend)
 - Framework: PyTorch >= 2.0, JAX >= 0.4.0, or MLX

README.md

@@ -45,7 +45,7 @@ separately, which can be done by following the instructions on their websites.
 CVXPYlayers has the following dependencies:
 * Python >= 3.11
 * [NumPy](https://pypi.org/project/numpy/) >= 1.22.4
-* [CVXPY](https://github.com/cvxgrp/cvxpy) >= 1.7.4
+* [CVXPY](https://github.com/cvxgrp/cvxpy) >= 1.9.0
 * [diffcp](https://github.com/cvxgrp/diffcp) >= 1.1.0
 
 Additionally, install one of the following frameworks:
@@ -55,8 +55,21 @@ Additionally, install one of the following frameworks:
 
 ### GPU-accelerated pathway
 
-For the best performance on CPU and GPU, install [Moreau](https://docs.moreau.so/).
-Moreau is available by request — see the [installation guide](https://docs.moreau.so/installation.html) for access and setup.
+For the best performance on CPU and GPU, install [Moreau](https://docs.moreau.so/)
+via the `moreau` extra. Before installing, review the
+[Moreau installation guide](https://docs.moreau.so/installation.html) for license
+terms and access requirements.
+
+```bash
+pip install cvxpylayers[moreau]
+```
+
+For GPU support, install the matching Moreau CUDA wheel (see the
+[Moreau installation guide](https://docs.moreau.so/installation.html) for details):
+
+```bash
+pip install "moreau[cuda12]"   # or moreau[cuda13]
+```
 
 As an open-source alternative, you can use [CuClarabel](https://github.com/oxfordcontrol/Clarabel.jl/tree/CuClarabel/) for GPU acceleration. This requires installing Julia and several additional packages:
 

docs/installation.md

@@ -167,10 +167,22 @@ print("Installation successful!")
 :::{admonition} Moreau — Best Performance on CPU & GPU
 :class: tip
 
-[Moreau](https://docs.moreau.so/) is the recommended solver for best performance. Available by request through a private package index.
+[Moreau](https://docs.moreau.so/) is the recommended solver for best performance.
 :::
 
-See the [Moreau installation guide](https://docs.moreau.so/installation.html) for access and setup instructions.
+Before installing, review the [Moreau installation guide](https://docs.moreau.so/installation.html)
+for license terms and access requirements.
+
+```bash
+pip install cvxpylayers[moreau]
+```
+
+For GPU acceleration, install the matching Moreau CUDA wheel
+(see the [Moreau installation guide](https://docs.moreau.so/installation.html) for details):
+
+```bash
+pip install "moreau[cuda12]"   # or moreau[cuda13]
+```
 
 ### CuClarabel (Open-Source Alternative)
 
@@ -234,7 +246,7 @@ b_gpu = b_t.cuda()
 |---------|---------|---------|
 | Python | >= 3.11 | Runtime |
 | NumPy | >= 1.22.4 | Array operations |
-| CVXPY | >= 1.7.4 | Problem specification |
+| CVXPY | >= 1.9.0 | Problem specification |
 | diffcp | >= 1.1.0 | Differentiable cone programming |
 :::
 
@@ -244,6 +256,7 @@ b_gpu = b_t.cuda()
 | PyTorch | >= 2.0 |
 | JAX | >= 0.4.0 |
 | MLX | >= 0.27.1 |
+| Moreau | >= 0.3.0 |
 :::
 
 ---

pyproject.toml

@@ -17,7 +17,7 @@ authors = [
 maintainers = [{ name = "P. Nobel", email = "parthnobel@berkeley.edu" }]
 requires-python = ">=3.11"
 dependencies = [
-    "cvxpy>=1.9",
+    "cvxpy>=1.9.0",
     "diffcp>=1.1.0",
     "numpy>=1.22.4",
     "scipy>=1.13.0",
@@ -49,13 +49,15 @@ torch = ["torch>=2.0.0", "diffcp>=1.1.0"]
 mlx = ["mlx>=0.27.1"]
 metal = ["mlx-metal>=0.27.1"]
 jax = ["jax>=0.4.0", "jaxlib>=0.4.0"]
+moreau = ["moreau>=0.3.0"]
 all = [
     "torch>=2.0.0",
     "diffqcp>=0.4.4",
     "jax>=0.4.0",
     "jaxlib>=0.4.0",
     "mlx>=0.27.1",
     "mlx-metal>=0.27.1",
+    "moreau>=0.3.0",
 ]
 
 [tool.hatch.metadata]

src/cvxpylayers/__init__.py

@@ -1,3 +1 @@
 __version__ = "1.1.0"
-
-import cvxpylayers._quad_form_dpp  # noqa: F401  # apply monkey-patches at import time

src/cvxpylayers/utils/parse_args.py

@@ -10,9 +10,14 @@
 from cvxpy.utilities import scopes
 
 import cvxpylayers.interfaces
-from cvxpylayers._quad_form_dpp import SUPPORTS_QUAD_OBJ
 from cvxpylayers.interfaces.base import SolverInterface
 
+# Solvers whose cvxpylayers interface supports quadratic objectives directly.
+# For these solvers, quad_form_dpp_scope is entered during canonicalization
+# so that quad_form(x, P) with parametric P is accepted as DPP.
+# DIFFCP decomposes quad_form to SOC and cannot handle parametric P.
+SUPPORTS_QUAD_OBJ = frozenset({"MOREAU", "CUCLARABEL", "MPAX"})
+
 
 
 if TYPE_CHECKING:
@@ -189,7 +194,7 @@ def _build_primal_recovery(
 ) -> VariableRecovery:
     """Build recovery info for a primal variable."""
     start = param_prob.var_id_to_col[var_id_map.get(var.id, var.id)]
-    is_sym = hasattr(var, "is_symmetric") and var.is_symmetric() and len(var.shape) >= 2
+    is_sym = var.is_symmetric() and len(var.shape) >= 2
 
     if is_sym:
         n = var.shape[0]  # type: ignore[index]
@@ -522,7 +527,7 @@ def parse_args(
         parameters, param_prob, gp_param_to_log_param, param_id_map
     )
 
-    q = getattr(param_prob, "q", getattr(param_prob, "c", None))
+    q = param_prob.q
 
     # Build variable recovery info for each requested variable
     constr_id_to_slice = _build_constr_id_to_slice(param_prob)