feat(exe): add support for @venv and relative paths in JULIA_PYTHONCALL_EXE

docs/src/faq.md

@@ -1,6 +1,6 @@
 # FAQ & Troubleshooting
 
-## Can I use PythonCall and PyCall together?
+## [Can I use PythonCall and PyCall together?](@id faq-pycall)
 
 Yes, you can use both PyCall and PythonCall in the same Julia session. This is platform-dependent:
 - On most systems the Python interpreter used by PythonCall and PyCall must be the same (see below).

docs/src/pythoncall.md

@@ -240,17 +240,21 @@ to your current Julia project containing Python and any required Python packages
 ENV["JULIA_CONDAPKG_BACKEND"] = "Null"
 ENV["JULIA_PYTHONCALL_EXE"] = "/path/to/python"  # optional
 ENV["JULIA_PYTHONCALL_EXE"] = "@PyCall"  # optional
+ENV["JULIA_PYTHONCALL_EXE"] = "@venv" # optional
 ```
 
 By setting the CondaPkg backend to Null, it will never install any Conda packages. In this
 case, PythonCall will use whichever Python is currently installed and in your `PATH`. You
 must have already installed any Python packages that you need.
 
 If `python` is not in your `PATH`, you will also need to set `JULIA_PYTHONCALL_EXE` to its
-path.
+path. Relative paths are resolved relative to the current active project.
 
 If you also use PyCall, you can set `JULIA_PYTHONCALL_EXE=@PyCall` to use the same Python
-interpreter.
+interpreter. [See here](@ref faq-pycall).
+
+If you have a Python virtual environment at `.venv` in your current active project, you
+can set `JULIA_PYTHONCALL_EXE=@venv` to use it.
 
 #### If you already have a Conda environment
 

docs/src/releasenotes.md

@@ -5,6 +5,8 @@
 * Added `pyconvert` rules for NumpyDates types.
 * Added `PyArray` support for NumPy arrays of `datetime64` and `timedelta64`.
 * Added `juliacall.ArrayValue` support for Julia arrays of `InlineDateTime64` and `InlineTimeDelta64`.
+* If `JULIA_PYTHONCALL_EXE` is a relative path, it is now considered relative to the active project.
+* Added option `JULIA_PYTHONCALL_EXE=@venv` to use a Python virtual environment relative to the active project.
 * Bug fixes.
 * Internal: switch from Requires.jl to package extensions.
 

src/C/context.jl

@@ -75,14 +75,38 @@ function init_context()
             exe_path = PyCall.python::String
             CTX.lib_path = PyCall.libpython::String
             CTX.which = :PyCall
+        elseif exe_path == "@venv"
+            # load from a .venv in the active project
+            exe_path = abspath(dirname(Base.active_project()), ".venv")
+            if Sys.iswindows()
+                exe_path = abspath(exe_path, "Scripts", "python.exe")::String
+            else
+                exe_path = abspath(exe_path, "bin", "python")::String
+            end
         elseif startswith(exe_path, "@")
             error("invalid JULIA_PYTHONCALL_EXE=$exe_path")
         else
             # Otherwise we use the Python specified
             CTX.which = :unknown
+            if isabspath(exe_path)
+                # nothing to do
+            elseif '/' in exe_path || '\\' in exe_path
+                # it's a relative path, interpret it as relative to the current project
+                exe_path = abspath(dirname(Base.active_project()), exe_path)::String
+            else
+                # it's a command, find it in the PATH
+                given_exe_path = exe_path
+                exe_path = Sys.which(exe_path)
+                exe_path === nothing &&
+                    error("Python executable $(repr(given_exe_path)) not found.")
+                exe_path::String
+            end
         end
 
         # Ensure Python is runnable
+        if !ispath(exe_path)
+            error("Python executable $(repr(exe_path)) does not exist.")
+        end
         try
             run(pipeline(`$exe_path --version`, stdout = devnull, stderr = devnull))
         catch