Make package installation automatic and deprecate manual installation

README.md

@@ -31,14 +31,6 @@ To install diffeqpy, use pip:
 pip install diffeqpy
 ```
 
-To install Julia packages required (and Julia if needed) for diffeqpy, open up Python
-interpreter then run:
-
-```pycon
->>> import diffeqpy
->>> diffeqpy.install()
-```
-
 and you're good!
 
 ## Collab Notebook Examples
@@ -171,7 +163,7 @@ sol = de.solve(prob)
 #### Limitations
 
 `de.jit`, uses ModelingToolkit.jl's `modelingtoolkitize` internally and some
-restrictions apply. Not all models can be jitted. See the 
+restrictions apply. Not all models can be jitted. See the
 [`modelingtoolkitize` documentation](https://docs.sciml.ai/ModelingToolkit/stable/tutorials/modelingtoolkitize/#What-is-modelingtoolkitize?)
 for more info.
 
@@ -552,29 +544,19 @@ sol = de.solve(ensembleprob,de.Tsit5(),de.EnsembleSerial(),trajectories=10000,sa
 ```
 
 To add GPUs to the mix, we need to bring in [DiffEqGPU](https://github.com/SciML/DiffEqGPU.jl).
-The `diffeqpy.install_cuda()` will install CUDA for you and bring all of the bindings into the returned object:
+The command `from diffeqpy import cuda` will install CUDA for you and bring all of the bindings into the returned object:
 
-```py
-diffeqpy.install_cuda()
-```
-
-then run the cuda import:
-
-```py
-from diffeqpy import cuda
-```
-
-#### Note: `diffeqpy.install_cuda()` and `from diffeqpy import cuda` can take awhile to run the first time as it installs the drivers!
+#### Note: `from diffeqpy import cuda` can take awhile to run the first time as it installs the drivers!
 
 Now we simply use `EnsembleGPUKernel(cuda.CUDABackend())` with a
-GPU-specialized ODE solver `cuda.GPUTsit5()` to solve 10,000 ODEs on the GPU in 
+GPU-specialized ODE solver `cuda.GPUTsit5()` to solve 10,000 ODEs on the GPU in
 parallel:
 
 ```py
 sol = de.solve(ensembleprob,cuda.GPUTsit5(),cuda.EnsembleGPUKernel(cuda.CUDABackend()),trajectories=10000,saveat=0.01)
 ```
 
-For the full list of choices for specialized GPU solvers, see 
+For the full list of choices for specialized GPU solvers, see
 [the DiffEqGPU.jl documentation](https://docs.sciml.ai/DiffEqGPU/stable/manual/ensemblegpukernel/).
 
 Note that `EnsembleGPUArray` can be used as well, like:
@@ -598,20 +580,20 @@ ensemble generation:
 ```py
 import numpy as np
 from scipy.integrate import odeint
- 
+
 def lorenz(state, t, sigma, beta, rho):
     x, y, z = state
-     
+
     dx = sigma * (y - x)
     dy = x * (rho - z) - y
     dz = x * y - beta * z
-     
+
     return [dx, dy, dz]
 
 sigma = 10.0
 beta = 8.0 / 3.0
-rho = 28.0 
-p = (sigma, beta, rho)	
+rho = 28.0
+p = (sigma, beta, rho)
 y0 = [1.0, 1.0, 1.0]
 
 t = np.arange(0.0, 100.0, 0.01)

diffeqpy/__init__.py

@@ -1,17 +1,58 @@
-import os
 import shutil
-import subprocess
-import sys
-
 from jill.install import install_julia
 
-script_dir = os.path.dirname(os.path.realpath(__file__))
-
+# juliacall must be loaded after `_ensure_julia_installed()` is run,
+# so this import is in `load_julia_packages()`
+# from juliacall import Main
 
 def _find_julia():
     # TODO: this should probably fallback to query jill
     return shutil.which("julia")
 
+def _ensure_julia_installed():
+    if not _find_julia():
+        print("No Julia version found. Installing Julia.")
+        install_julia()
+        if not _find_julia():
+            raise RuntimeError(
+                "Julia installed with jill but `julia` binary cannot be found in the path"
+            )
+
+# TODO: upstream this function or an alternative into juliacall
+def load_julia_packages(*names):
+    """
+    Load Julia packages and return references to them, automatically installing julia and
+    the packages as necessary.
+    """
+    # This is terrifying to many people. However, it seems SciML takes pragmatic approach.
+    _ensure_julia_installed()
+
+    script = """import Pkg
+    Pkg.activate(\"diffeqpy\", shared=true)
+    try
+        import {0}
+    catch e
+        e isa ArgumentError || throw(e)
+        Pkg.add([{1}])
+        import {0}
+    end
+    {0}""".format(", ".join(names), ", ".join(f'"{name}"' for name in names))
+
+    # Unfortunately, `seval` doesn't support multi-line strings
+    # https://github.com/JuliaPy/PythonCall.jl/issues/433
+    script = script.replace("\n", ";")
+
+    # Must be loaded after `_ensure_julia_installed()`
+    from juliacall import Main
+    return Main.seval(script)
+
+
+
+# Deprecated (julia and packages now auto-install)
+import os
+import subprocess
+import sys
+script_dir = os.path.dirname(os.path.realpath(__file__))
 
 def install(*, confirm=False):
     """
@@ -28,7 +69,7 @@ def install(*, confirm=False):
             )
     env = os.environ.copy()
     env["PYTHON"] = sys.executable
-    subprocess.check_call([julia, os.path.join(script_dir, "install.jl")], env=env)
+    subprocess.check_call([julia, os.path.join(script_dir, "deprecated/install.jl")], env=env)
 
 def install_cuda():
     julia = _find_julia()
@@ -38,7 +79,7 @@ def install_cuda():
         )
     env = os.environ.copy()
     env["PYTHON"] = sys.executable
-    subprocess.check_call([julia, os.path.join(script_dir, "install_cuda.jl")], env=env)
+    subprocess.check_call([julia, os.path.join(script_dir, "deprecated/install_cuda.jl")], env=env)
 
 def install_amdgpu():
     julia = _find_julia()
@@ -48,17 +89,17 @@ def install_amdgpu():
         )
     env = os.environ.copy()
     env["PYTHON"] = sys.executable
-    subprocess.check_call([julia, os.path.join(script_dir, "install_amdgpu.jl")], env=env)
+    subprocess.check_call([julia, os.path.join(script_dir, "deprecated/install_amdgpu.jl")], env=env)
 
 def install_metal():
     julia = _find_julia()
     if not julia:
         raise RuntimeError(
-            "Julia must be installed before adding Metal. Please run `diffeqpy.install()` first"
+            "Julia must be installed before adding Metal. Please run `deprecated/diffeqpy.install()` first"
         )
     env = os.environ.copy()
     env["PYTHON"] = sys.executable
-    subprocess.check_call([julia, os.path.join(script_dir, "install_metal.jl")], env=env)
+    subprocess.check_call([julia, os.path.join(script_dir, "deprecated/install_metal.jl")], env=env)
 
 def install_oneapi():
     julia = _find_julia()
@@ -68,18 +109,4 @@ def install_oneapi():
         )
     env = os.environ.copy()
     env["PYTHON"] = sys.executable
-    subprocess.check_call([julia, os.path.join(script_dir, "install_oneapi.jl")], env=env)
-
-def _ensure_installed(*kwargs):
-    if not _find_julia():
-        # TODO: this should probably ensure that packages are installed too
-        install(*kwargs)
-
-# TODO: upstream this function or an alternative into juliacall
-def load_julia_packages(names):
-    # This is terrifying to many people. However, it seems SciML takes pragmatic approach.
-    _ensure_installed()
-
-    # Must be loaded after `_ensure_installed()`
-    from juliacall import Main
-    return Main.seval(f"import Pkg; Pkg.activate(\"diffeqpy\", shared=true); import {names}; {names}")
+    subprocess.check_call([julia, os.path.join(script_dir, "deprecated/install_oneapi.jl")], env=env)

diffeqpy/amdgpu.py

@@ -1,6 +1,6 @@
 import sys
 from . import load_julia_packages
-amdgpu, _ = load_julia_packages("DiffEqGPU, AMDGPU")
+amdgpu, _ = load_julia_packages("DiffEqGPU", "AMDGPU")
 from juliacall import Main
 amdgpu.AMDGPUBackend = Main.seval("AMDGPU.AMDGPUBackend") # kinda hacky
 sys.modules[__name__] = amdgpu # mutate myself

diffeqpy/cuda.py

@@ -1,6 +1,6 @@
 import sys
 from . import load_julia_packages
-cuda, _ = load_julia_packages("DiffEqGPU, CUDA")
+cuda, _ = load_julia_packages("DiffEqGPU", "CUDA")
 from juliacall import Main
 cuda.CUDABackend = Main.seval("CUDA.CUDABackend") # kinda hacky
 sys.modules[__name__] = cuda # mutate myself

diffeqpy/de.py

@@ -1,6 +1,6 @@
 import sys
 from . import load_julia_packages
-de, _ = load_julia_packages("DifferentialEquations, ModelingToolkit")
+de, _, _ = load_julia_packages("DifferentialEquations", "ModelingToolkit", "PythonCall")
 from juliacall import Main
 de.jit = Main.seval("jit(x) = typeof(x).name.wrapper(ModelingToolkit.modelingtoolkitize(x), x.u0, x.tspan, x.p)") # kinda hackey
 de.jit32 = Main.seval("jit(x) = typeof(x).name.wrapper(ModelingToolkit.modelingtoolkitize(x), Float32.(x.u0), Float32.(x.tspan), Float32.(x.p))") # kinda hackey

diffeqpy/metal.py

@@ -1,6 +1,6 @@
 import sys
 from . import load_julia_packages
-metal, _ = load_julia_packages("DiffEqGPU, Metal")
+metal, _ = load_julia_packages("DiffEqGPU", "Metal")
 from juliacall import Main
 metal.MetalBackend = Main.seval("Metal.MetalBackend") # kinda hacky
 sys.modules[__name__] = metal # mutate myself

diffeqpy/ode.py

@@ -1,3 +1,4 @@
 import sys
 from . import load_julia_packages
-sys.modules[__name__] = load_julia_packages("OrdinaryDiffEq") # mutate myself
+ode, _  = load_julia_packages("OrdinaryDiffEq", "PythonCall")
+sys.modules[__name__] = ode # mutate myself

diffeqpy/oneapi.py

@@ -1,6 +1,6 @@
 import sys
 from . import load_julia_packages
-oneapi, _ = load_julia_packages("DiffEqGPU, oneAPI")
+oneapi, _ = load_julia_packages("DiffEqGPU", "oneAPI")
 from juliacall import Main
 oneapi.oneAPIBackend = Main.seval("oneAPI.oneAPIBackend") # kinda hacky
 sys.modules[__name__] = oneapi # mutate myself