#4846 - Allow importing Python package with C-extension e.g. numpy in the labs CLI #4868

jmarrec · 2023-04-25T20:59:54Z

Pull request overview

Fixes [Enhancement Request] Allow importing Python package with C-extension e.g. numpy #4846
Add some ctest for using numpy with the CLI. 3 flavors
- Explicitly insert to sys.path the site-packages and lib-dynload folders from the system python
- Pass --python_path twice (site-packages and lib-dynload)
- Pass --python_home
Model API Changes / Additions
Any new or modified fields have been implemented in the EnergyPlus ForwardTranslator (and ReverseTranslator as appropriate)
Model API methods are tested (in src/model/test)
EnergyPlus ForwardTranslator Tests (in src/energyplus/Test)
If a new object or method, added a test in NREL/OpenStudio-resources: Add Link
If needed, added VersionTranslation rules for the objects (src/osversion/VersionTranslator.cpp)
Verified that C# bindings built fine on Windows, partial classes used as needed, etc.
All new and existing tests passes
If methods have been deprecated, update rest of code to use the new methods

Labels:

If change to an IDD file, add the label IDDChange
If breaking existing API, add the label APIChange
If deemed ready, add label Pull Request - Ready for CI so that CI builds your PR

Review Checklist

This will not be exhaustively relevant to every PR.

Perform a Code Review on GitHub
Code Style, strip trailing whitespace, etc.
All related changes have been implemented: model changes, model tests, FT changes, FT tests, VersionTranslation, OS App
Labeling is ok
If defect, verify by running develop branch and reproducing defect, then running PR and reproducing fix
If feature, test running new feature, try creative ways to break it
CI status: all green or justified

jmarrec · 2023-04-26T01:11:59Z

CMakeLists.txt

@@ -1,5 +1,5 @@
 message(STATUS "Using CMake ${CMAKE_VERSION}")
-cmake_minimum_required(VERSION 3.19.0)
+cmake_minimum_required(VERSION 3.20.0)


using cmake_path which is new in 3.20

https://cmake.org/cmake/help/latest/command/cmake_path.html

jmarrec · 2023-04-26T01:12:40Z

python/SetupPython.cmake

@@ -7,9 +7,9 @@ set(Python_USE_STATIC_LIBS OFF)
 # find_package(Python) has the problem that on github actions in particular it'll pick up the most recent python (eg 3.9) from the tool cache
 # even if you have used the setup-python action and set it to 3.8
 if (PYTHON_VERSION)
-  find_package(Python ${PYTHON_VERSION} EXACT COMPONENTS Interpreter Development REQUIRED)
+  find_package(Python ${PYTHON_VERSION} EXACT REQUIRED COMPONENTS Interpreter Development OPTIONAL_COMPONENTS NumPy)


I'm using NumPy to test loading a native gem because the FindPython (cmake) has a NumPy component so it's easy to test whether the user actually has it installed.

jmarrec · 2023-04-26T01:13:08Z

src/cli/CMakeLists.txt

+  # ============ Native Ruby Gems / Python Modules - C extensions =============
+  if (Python_NumPy_FOUND)


Only enable the tests if numpy found

jmarrec · 2023-04-26T01:13:28Z

src/cli/CMakeLists.txt

+    if(NOT Python_ROOT_DIR)
+      # Python_STDLIB: we expect it to be
+      # Unix: ~/.pyenv/versions/3.8.12/lib/python3.8
+      # Windows C:\Python38\Lib
+      cmake_path(GET Python_STDLIB PARENT_PATH Python_ROOT_DIR)
+      if(UNIX)
+        cmake_path(GET Python_ROOT_DIR PARENT_PATH Python_ROOT_DIR)
+      endif()
+    endif()


I don't like these shenanigans too much, but not sure how to do otherwise.

jmarrec · 2023-04-26T01:14:18Z

src/cli/CMakeLists.txt

+    add_test(NAME OpenStudioCLI.Labs.execute_python_script.numpy.explicit_sys_path_insert
+      COMMAND $<TARGET_FILE:openstudio> labs execute_python_script execute_python_script_with_numpy.py ${Python_STDLIB}
+      WORKING_DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}/test/"
+    )
+
+    if(UNIX)
+      add_test(NAME OpenStudioCLI.Labs.execute_python_script.numpy.python_path
+        COMMAND $<TARGET_FILE:openstudio> labs
+                  --python_path "${Python_SITELIB}" # == "${Python_STDLIB}/site-packages"
+                  --python_path "${Python_STDLIB}/lib-dynload"
+                  execute_python_script execute_python_script_with_numpy.py
+        WORKING_DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}/test/"
+      )
+    else()
+      add_test(NAME OpenStudioCLI.Labs.execute_python_script.numpy.python_path
+        COMMAND $<TARGET_FILE:openstudio> labs
+                  --python_path "$<SHELL_PATH:${Python_SITELIB}>"
+                  execute_python_script execute_python_script_with_numpy.py
+        WORKING_DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}/test/"
+      )
+    endif()
+
+    add_test(NAME OpenStudioCLI.Labs.execute_python_script.numpy.python_home
+      COMMAND $<TARGET_FILE:openstudio> labs
+                --python_home "$<SHELL_PATH:${Python_ROOT_DIR}>"
+                execute_python_script execute_python_script_with_numpy.py
+      WORKING_DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}/test/"
+    )


Three different tests to cover the scenarios listed in OP

jmarrec · 2023-04-26T01:14:55Z

python/engine/PythonEngine.cpp

  }
 }

 PythonEngine::PythonEngine(int argc, char* argv[]) : ScriptEngine(argc, argv), program(Py_DecodeLocale(pythonProgramName, nullptr)) {
+  // TODO: modernize and use PyConfig (new in 3.8): https://docs.python.org/3/c-api/init_config.html


TODO for later

jmarrec · 2023-04-26T01:15:34Z

python/engine/PythonEngine.cpp

+  // The PYTHONPATH / PYTHONHOME should be set before initializing Python
+  // If this Py_SetPath is called before Py_Initialize, then Py_GetPath won't attempt to compute a default search path
+  // The default search path is affected by the Py_SetPythonHome
+  // * if the user passed --python_home, we use that as the Python Home, and do not use Py_SetPath. But later we add the E+ standard_lib anyways
+  //   so it takes precedence (to limit incompatibility issues...)
+  // * If the user didn't pass it, we use Py_SetPath set to the E+ standard_lib
+
+  std::vector<std::string> args(argv, std::next(argv, static_cast<std::ptrdiff_t>(argc)));
+  bool pythonHomePassed = false;
+  auto it = std::find(args.cbegin(), args.cend(), "--python_home");
+  if (it != args.cend()) {
+    openstudio::path pythonHomeDir(*std::next(it));
+    wchar_t* h = Py_DecodeLocale(pythonHomeDir.make_preferred().string().c_str(), nullptr);
+    Py_SetPythonHome(h);
+    pythonHomePassed = true;
+  } else {
+    wchar_t* a = Py_DecodeLocale(pathToPythonPackages.make_preferred().string().c_str(), nullptr);
+    Py_SetPath(a);
+  }


Deal with --python_home . Can't do it later, it has to be done before Py_Initialize... so grab it from the argc/argv...

If feels like a code smell that we are parsing argc / argv at this point instead of having everything parsed out by the CLI11 library. I think I am guilty of establishing this pattern in ScriptEngine and I think I did it because the ruby init functions (at least one of them) accept argc and argv.

ruby_sysinit does take argc, argv yes.

Plus, here this is an order thing, and specific to Python while the CLI only sees a ScriptEngineInstance, so adding an explicit PythonHome argument isn't great either.

If you see another cleaner way to do it, this can be addressed later. There are other things that probably need cleanup too (including simplifying the folder / Cmake structure probably).

Another thing: we need to process the argv data manually at the start of the CLI anyways since we do support (for legacy reasons) some usage like openstudio labs test.rb that is translated as being openstudio labs execute_ruby_script test.rb . cf https://github.com/NREL/OpenStudio/blob/2af6123b91304e12060af360d0afb24083336a4c/src/cli/main.cpp#L45C1-L89

jmarrec · 2023-04-26T01:16:00Z

python/engine/PythonEngine.cpp

+  if (pythonHomePassed) {
+    addToPythonPath(pathToPythonPackages);
+  }
+#if defined(__APPLE__) || defined(__linux___) || defined(__unix__)
+  addToPythonPath(pathToPythonPackages / "lib-dynload");
+#endif


lib-dynload is present in the E+ python_standard_lib, we should add it

jmarrec · 2023-04-26T01:16:48Z

src/cli/main.cpp

@@ -126,28 +126,33 @@ int main(int argc, char* argv[]) {
    experimentalApp
      ->add_option("-I,--include", includeDirs, "Add additional directory to add to front of Ruby $LOAD_PATH (may be used more than once)")
      ->option_text("DIR")
+      ->check(CLI::ExistingDirectory)


Add some CLI11 validators to ensure the various args are valid directories to begin with and catch mistakes

jmarrec · 2023-04-26T01:17:17Z

src/utilities/core/DynamicLibraryPOSIX.hpp

+  explicit DynamicLibrary(openstudio::path location) : m_location{std::move(location)} {
+    int flags = RTLD_LAZY | RTLD_LOCAL;  // NOLINT(misc-const-correctness, hicpp-signed-bitwise)
+
+    // This seems to work on Mac without RTLD_GLOBAL...
+#ifdef __linux__
+    if (m_location.filename().generic_string().find("python") != std::string::npos) {
+      // https://stackoverflow.com/questions/67891197/ctypes-cpython-39-x86-64-linux-gnu-so-undefined-symbol-pyfloat-type-in-embedd
+      flags = RTLD_LAZY | RTLD_GLOBAL;
+    }
+#endif
+    m_handle = {dlopen(m_location.c_str(), flags), m_handle_deleter};


This is the bit that bothers me a bit @kbenne

Are we sure we actually need these flags ever? I'm not sure about that. I think there is a decent chance they got added unnessarily.

Which flag(s)? RTLD_LAZY?

jmarrec · 2023-04-26T13:02:56Z

src/cli/CMakeLists.txt

+    if(NOT Python_ROOT_DIR)
+      # Python_STDLIB: we expect it to be
+      # Unix: ~/.pyenv/versions/3.8.12/lib/python3.8 or
+      #       or on CI: /usr/lib/python3.8/ ... with numpy if install via pip3 and not apt install python3-numpy in `/usr/local/lib/python3.8/dist-packages/`
+      # Windows C:\Python38\Lib
+      cmake_path(GET Python_STDLIB PARENT_PATH Python_ROOT_DIR)
+      if(UNIX)
+        cmake_path(GET Python_ROOT_DIR PARENT_PATH Python_ROOT_DIR)
+      endif()
+    endif()
+
+    if(UNIX)
+      set(PYTHON_PATH "${Python_SITELIB}" "${Python_STDLIB}/lib-dynload")
+
+      if(NOT APPLE)
+        set(EXTRA_LOCAL_DIST "/usr/local/lib/python3.8/dist-packages")
+        if (EXISTS "${EXTRA_LOCAL_DIST}")
+          list(APPEND PYTHON_PATH "${EXTRA_LOCAL_DIST}")
+        endif()
+      endif()
+    else()
+      set(PYTHON_PATH "$<SHELL_PATH:${Python_SITELIB}>")
+    endif()
+
+    message(DEBUG "PYTHON_PATH=${PYTHON_PATH}")
+
+    add_test(NAME OpenStudioCLI.Labs.execute_python_script.numpy.explicit_sys_path_insert
+      COMMAND $<TARGET_FILE:openstudio> labs execute_python_script execute_python_script_with_numpy.py ${Python_STDLIB}
+      WORKING_DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}/test/"
+    )
+
+    add_test(NAME OpenStudioCLI.Labs.execute_python_script.numpy.python_path
+      COMMAND $<TARGET_FILE:openstudio> labs
+                "$<$<BOOL:${PYTHON_PATH}>:--python_path;$<JOIN:${PYTHON_PATH},;--python_path;>>"
+                execute_python_script execute_python_script_with_numpy.py
+      COMMAND_EXPAND_LISTS
+      WORKING_DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}/test/"


CI on ubuntu uses the system python... which on debian uses a special directory for pip3 installed stuff (apt-get install python3-numpy would have worked out of the box).... so I had to do great cmake horrors with cmake generator expressions being expanded a list... sorry

* Explicitly insert to sys.path the site-packages and lib-dynload folders from the system python * Pass --python_path twice (site-packages and lib-dynload) * Pass --python_home

… avoid issues.

…er passed

…work

…python3.8/dist-packages/`

…2.04

…more reliable (Ubuntu 22.04 had this issue)

jmarrec · 2023-06-30T07:12:27Z

38eed72, which installs jinja2 numpy pandas pip if -DPYTHON_INSTALL_NATIVE_MODULES:BOOL=ON is passed can be reverted, I'm just trying it out.

Edit: separated onto another branch

ci-commercialbuildings · 2023-06-30T07:42:19Z

CI Results for 3a3175d:

Ubuntu 20.04:
Ubuntu 22.04:
- OpenStudio-3.7.0-alpha+3a3175dbe4-Ubuntu-22.04-x86_64.deb
- OpenStudio-3.7.0-alpha+3a3175dbe4-Ubuntu-22.04-x86_64.tar.gz
Darwin:
- OpenStudio-3.7.0-alpha+3a3175dbe4-Darwin-x86_64.dmg
- OpenStudio-3.7.0-alpha+3a3175dbe4-Darwin-x86_64.tar.gz
Windows:
- OpenStudio-3.7.0-alpha+3a3175dbe4-Windows.exe
- OpenStudio-3.7.0-alpha+3a3175dbe4-Windows.tar.gz
Code Coverage:

kbenne · 2023-07-05T16:04:55Z

python/engine/PythonEngine.cpp

@@ -25,9 +25,12 @@ void addToPythonPath(const openstudio::path& includePath) {
  if (!includePath.empty()) {
    PyObject* sys = PyImport_ImportModule("sys");
    PyObject* sysPath = PyObject_GetAttrString(sys, "path");
+    Py_DECREF(sys);  // PyImport_ImportModule returns a new reference, decrement it


I wonder how many instances there are where we miss these decrements. Or we have one that should not be. I bet there are a few remaining.

kbenne · 2023-07-05T16:09:26Z

python/engine/PythonEngine.cpp

+  // The PYTHONPATH / PYTHONHOME should be set before initializing Python
+  // If this Py_SetPath is called before Py_Initialize, then Py_GetPath won't attempt to compute a default search path
+  // The default search path is affected by the Py_SetPythonHome
+  // * if the user passed --python_home, we use that as the Python Home, and do not use Py_SetPath. But later we add the E+ standard_lib anyways
+  //   so it takes precedence (to limit incompatibility issues...)
+  // * If the user didn't pass it, we use Py_SetPath set to the E+ standard_lib
+
+  std::vector<std::string> args(argv, std::next(argv, static_cast<std::ptrdiff_t>(argc)));
+  bool pythonHomePassed = false;
+  auto it = std::find(args.cbegin(), args.cend(), "--python_home");
+  if (it != args.cend()) {
+    openstudio::path pythonHomeDir(*std::next(it));
+    wchar_t* h = Py_DecodeLocale(pythonHomeDir.make_preferred().string().c_str(), nullptr);
+    Py_SetPythonHome(h);
+    pythonHomePassed = true;
+  } else {
+    wchar_t* a = Py_DecodeLocale(pathToPythonPackages.make_preferred().string().c_str(), nullptr);
+    Py_SetPath(a);
+  }


If feels like a code smell that we are parsing argc / argv at this point instead of having everything parsed out by the CLI11 library. I think I am guilty of establishing this pattern in ScriptEngine and I think I did it because the ruby init functions (at least one of them) accept argc and argv.

kbenne · 2023-07-05T16:17:26Z

src/utilities/core/DynamicLibraryPOSIX.hpp

+  explicit DynamicLibrary(openstudio::path location) : m_location{std::move(location)} {
+    int flags = RTLD_LAZY | RTLD_LOCAL;  // NOLINT(misc-const-correctness, hicpp-signed-bitwise)
+
+    // This seems to work on Mac without RTLD_GLOBAL...
+#ifdef __linux__
+    if (m_location.filename().generic_string().find("python") != std::string::npos) {
+      // https://stackoverflow.com/questions/67891197/ctypes-cpython-39-x86-64-linux-gnu-so-undefined-symbol-pyfloat-type-in-embedd
+      flags = RTLD_LAZY | RTLD_GLOBAL;
+    }
+#endif
+    m_handle = {dlopen(m_location.c_str(), flags), m_handle_deleter};


Are we sure we actually need these flags ever? I'm not sure about that. I think there is a decent chance they got added unnessarily.

kbenne · 2023-07-05T16:20:49Z

src/cli/CMakeLists.txt

+
+    message(DEBUG "PYTHON_PATH=${PYTHON_PATH}")
+
+    add_test(NAME OpenStudioCLI.Labs.execute_python_script.numpy.explicit_sys_path_insert


At first this series of tests kind of seemed off to me, because they are testing the environment of the developer/build environment, but when I think more carefully, I guess that is the point. The dev environment is a proxy for whatever environment a user may have, and since you already have the scaffolding in place to find a random numpy installation it turns out to be a convenient way to formumulate a test. Seems nice to me.

The only thing this leaves on the table are custom extensions that we choose to ship by default, but I believe that will be addressed in a different issue.

jmarrec added component - CLI Pull Request - Ready for CI This pull request if finalized and is ready for continuous integration verification prior to merge. labels Apr 25, 2023

jmarrec self-assigned this Apr 25, 2023

jmarrec commented Apr 26, 2023

View reviewed changes

jmarrec marked this pull request as ready for review April 26, 2023 01:23

jmarrec requested a review from kbenne April 26, 2023 01:23

jmarrec commented Apr 26, 2023

View reviewed changes

jmarrec added this to the OpenStudio SDK 3.7.0 milestone May 31, 2023

jmarrec added 11 commits June 30, 2023 09:10

#4846 - Add some ctest for using numpy with the CLI. 3 flavors

ba18689

* Explicitly insert to sys.path the site-packages and lib-dynload folders from the system python * Pass --python_path twice (site-packages and lib-dynload) * Pass --python_home

CLI: validate that the --xxx_home --xxx_path are valid DIRECTORIES to…

e5b4f90

… avoid issues.

Handle python layout being different on windows and forward/back slashes

60a7294

Correctly pass --python_home

db941e8

Misc for PythonEngine: Py_DECREF a few things, add includeDirs in ord…

454d337

…er passed

Use RTLD_GLOBAL | RTLD_NOW on Ubuntu for python, so that native exts …

e581821

…work

RTLD_LAZY | RTLD_GLOBAL seems to do it on ubuntu

c00a57b

Should add lib-dynload from E+ python_standard_lib too on unix

1d1e7c3

Deal with Ubuntu CIs which have the numpy package in `/usr/local/lib/…

7fd28c8

…python3.8/dist-packages/`

woohoou /usr/lib/python3.8/site-packages' doesn't exist on ubuntu 2…

11397c6

…2.04

Catch the case when the Python_ROOT_DIR was cached but doesn't exist …

3a3175d

…more reliable (Ubuntu 22.04 had this issue)

jmarrec force-pushed the 4846_labs_python_native branch from e131f0f to 38eed72 Compare June 30, 2023 07:12

jmarrec force-pushed the 4846_labs_python_native branch from 38eed72 to 3a3175d Compare June 30, 2023 07:14

jmarrec mentioned this pull request Jul 4, 2023

Support Testing of Python Measures with standard OpenStudio install #4906

Closed

kbenne approved these changes Jul 5, 2023

View reviewed changes

jmarrec merged commit 82304dc into develop Jul 6, 2023

jmarrec deleted the 4846_labs_python_native branch July 6, 2023 07:23

jmarrec mentioned this pull request Jul 17, 2023

Python plugin jinja template #4858

Merged

19 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

#4846 - Allow importing Python package with C-extension e.g. numpy in the labs CLI #4868

#4846 - Allow importing Python package with C-extension e.g. numpy in the labs CLI #4868

jmarrec commented Apr 25, 2023

jmarrec Apr 26, 2023

jmarrec Apr 26, 2023

jmarrec Apr 26, 2023

jmarrec Apr 26, 2023

jmarrec Apr 26, 2023

jmarrec Apr 26, 2023

jmarrec Apr 26, 2023

kbenne Jul 5, 2023

jmarrec Jul 6, 2023

jmarrec Apr 26, 2023

jmarrec Apr 26, 2023

jmarrec Apr 26, 2023

kbenne Jul 5, 2023

jmarrec Jul 6, 2023

jmarrec Apr 26, 2023

jmarrec commented Jun 30, 2023 •

edited

Loading

ci-commercialbuildings commented Jun 30, 2023 •

edited

Loading

kbenne Jul 5, 2023

kbenne Jul 5, 2023

kbenne Jul 5, 2023

kbenne Jul 5, 2023

		# ============ Native Ruby Gems / Python Modules - C extensions =============
		if (Python_NumPy_FOUND)


		message(DEBUG "PYTHON_PATH=${PYTHON_PATH}")

		add_test(NAME OpenStudioCLI.Labs.execute_python_script.numpy.explicit_sys_path_insert

#4846 - Allow importing Python package with C-extension e.g. numpy in the labs CLI #4868

#4846 - Allow importing Python package with C-extension e.g. numpy in the labs CLI #4868

Conversation

jmarrec commented Apr 25, 2023

Pull request overview

Review Checklist

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jmarrec commented Jun 30, 2023 • edited Loading

ci-commercialbuildings commented Jun 30, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jmarrec commented Jun 30, 2023 •

edited

Loading

ci-commercialbuildings commented Jun 30, 2023 •

edited

Loading