You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Current goals for establishing a new OpenBT Python package release for inclusion in BAND framework v1.0
Determine the new version number to assign to the C++ CLTs and the Python package
Note that the contents of this release will be essentially no different from the current v1.1.3 OpenBTMixing release.
If this release makes it into the OpenBT PyPI project page, then the version must be larger than the latest version that was published by that project (v0.0.5).
The safest is to just seed both the C++ CLTs and the Python package with the same starting version of v1.2.0
No known need to update minimum Meson version. Both meson.build and pyproject.toml using the same version constraints
Latest eigen version in Meson's wrapdb is 5.0.1-1, which is what we presently use
Reviewed metadata in setup.py
Updated documentation link to point to RTD docs
Updated author/maintainer list
Generalized package description thinking toward eventual inclusion of standard regression methods
Modernized external dependence version constraints
setuptools/setuptools-scm set according to current recommendations for custom configurations
The package doesn't specify minimum limits for packages such as numpy and scipy. We also don't have a means to confirm that minimum versions are actually work. So, we continue to leave it open ended for this release.
Actions are testing for Python 3.10 up to 3.14. Keep Python 3.10, which will reach end of life in October, as oldest version supported.
setup.py and pyproject.toml external dependencies are disjoint and independent. Therefore, they are consistent.
tox.ini doesn't have any hardcoded version identifiers still.
Reviewed actions completely and carefully since this is effectively the first release with this set of actions
Dependabot has been keeping action versions up to date
Cleaned up typos.toml to account for changes to repo contents made during merge
macOS actions issue warning about untrusted homebrew taps. Ignoring for now.
Anaconda-based actions have informative warning about trying to install packages using "conda-pypi beta". Ignoring.
No need to update action runners yet (though Ubuntu 26 is coming online and macOS 14 is being deprecated)
Excluding Ubuntu 24.04/MPICH tests of Python source distribution. Previous debugging did not yield success, although I note that the C++ CLT test runs this setup but without installing libmpich-dev. Added TODO to action regarding this potential avenue of exploration.
Root README looks good.
All links functioning as expected
Consistent with Python package README
Consistent with Sphinx docs landing page. Will ask Matt if he wants license, support, or citation content added to landing page.
Reviewed all sphinx docs
OpenBT SDK will be finalized when BAND is added to the bandframework repo for the BAND v1.0 release
Reviewed actions and action artifacts at commit 7048a73e
Tagged 7048a73e as v1.2.0, officially published a release associated with that version tag, and confirmed that the release candidate action ran through successfully.
Inspected manually built source distributions to confirm correct contents and correct version in filename
Confirmed that LICENSE file is included in the source distribution
Confirmed correct metadata in source distribution
Reviewed Ubuntu 24/Open MPI/Python 3.12 log carefully. The Meson build logs, wheel build logs, and installation information all looked as expected
Removed old venvs (including meson installs), old CLT installations, and homebrew installs (including pkgconfig). Cleaned out openbt_pypkg in local clone.
Confirm that C++ build script is building and installing CLTs correctly
macOS
clang++ v16.0.0
Open-MPI v5.0.9 installed by homebrew.
meson v1.11.2 installed from homebrew
ninja v1.13.2 installed from homebrew
brew cleanup --prune=all -s
Removed pre-existing builddir, subpackage downloads, & subpackage cache to ensure that we pull down eigen
Confirmed that eigen v5.0.1 is pulled down and installed automatically when eigen not in system
Confirmed that both eigen v5.0.1 and v3.4.1 installed via homebrew were found and used bymeson so long as pkgconf was also installed by homebrew to provide pkg-config. Curiously the package is referred to by eigen3 in pkg-config for both versions.
For all, confirmed correct OpenBT version printed by Meson
For all, reviewed compiler flags to ensure use of intended installations
Compiler flags reasonable and different for both release and debug builds
-DNDEBUG, -DSILENT, and -D_USEMPI appear when appropriate only
Confirm that developers can use meson to build the CLTs directly as debug builds with warnings treated as errors.
For all, otool -L on installed CLTs looks good and on intermediate library as well
Can run tests from builddir with meson test and meson test <test name> (e.g., with crn and tree)
Follow R installation guide & example using one of the C++ installations to confirm that R works
R v4.6.1 installed by homebrew
Confirmed correct version recorded in installation and LICENSE file installed as well
Confirmed correct git commit SHA as well
Installed in /opt/homebrew/lib/R/4.6/site-library
Saw four MPI processes each running at about 100% when openbt command was issued; 2 processes @ ~100%, when calculating Sobol' indices
test.obt îs 3.4M
Removed R installation and cleaned up again.
Uninstalled meson and ninja
Installed meson v1.11.2 into dedicated venv following C++ installation guides. Built C++ tools again using that meson.
Studied and tested as for other CLT installations. All good.
Installation notes don't include information about installing ninja, but the Meson error message is clear that ninja is missing. I don't want to include in this procedure a specific mechanism for installing ninja since some might prefer to install from a package manager; others, from PyPI. The need for manual installation of ninja is made clear in those same docs before getting to this point. I Don't think that we need to add more.
Deleted meson venv and uninstalled ninja. Did a full cleanup (including old meson symlink).
Follow non-PyPI Python package installation procedures using local clone
Cleaned ~/local and local clone before each test
Setup new, clean venv for each test
Normal pip install without eigen installed
Editable pip install -v -e . with (v3.4.1) and without eigen installed by homebrew
python -m pip show openbt and python -m pydoc openbt
Confirmed that LICENSE file is included in the installation.
Confirmed that installation had minimal, correct content, which includes the CLTs but not the library or headers.
Ran otool -L on installed CLTs and all looks good.
Confirmed that developer-mode installation was in expected location and had minimal, correct content (more than actual installation).
Follow PyPI Python package installation procedure but using official release candidate source distribution
Without eigen installed & tested as above
Cleaned ~/local and local clone beforehand
Setup new, clean venv for test
Install an older version of Meson and confirm that a useful message is reported to user that indicates the minimum version that is required.
Installed v1.5.2 in dedicated venv as described in README
Meson prints useful error message when using build script
Non-zero exit code returned
Confirm that pip install does not print meson setup, compile, or install output by default, but that these can be viewed by passing -v to pip install
Break setup.py in multiple ways and confirm that pip install fails with a useful message.
Typo in meson command triggered failure with both stdout/stderr printed with pip install in non-verbose mode.
Inability to find MPI installation triggered failure with both stdout/stderr printed with pip install in non-verbose mode.
Setup several different (compiler suite, MPI implementation) setups on a cluster using LMod, install Python package, and run tests.
On LCRC improv cluster (Linux) with Python v3.11.6. No ninja or meson installed. No eigen installed.
For each test setup, built clean venv and unloaded Python module. Installed openbt from release source distribution in verbose mode. Checked logs. Reviewed version and installation as above.
For each test setup, confirmed two MPI processes running during Python package test execution using top
Tested GCC v14.2.0 with Open MPI v5.0.7 (built with GCC v14.2.0) loaded by modules. Tests passed. However, Meson found the MPI installation where expected for v5.0.7, but called in 5.0.8. Compiler flags related to GCC 14.2.0 and OpenMPI 5.0.7.
Tested Intel OneAPI v2023.2.1 with Intel MPI v2021.10.0 (built with OneAPI v2023.2.1) loaded by modules. Note that Meson finds mpicxx, which is installed by Intel but curiously wraps g++. The correct wrapper is mpiicpc. As in all other cases, however, the wrappers are not called. Instead the correct Intel compiler that it detected is called with compatible flags. As in previous testing, however, the flags it used were associated with the MPI implementation used in the previous test. Reran from a clean login successfully. I have seen some issues with the OpenMPI modules still having an effect on the env vars after unloading and despite loading a new MPI installation. I wonder if what I saw previously was the effect of a bad environment.
twine check <tar ball> passed
Uploaded to TestPyPI so that I could see how the README rendering looked. All good. Confirmed that all links functioning as expected. Text reads well. Not too much, not too little.
Uploaded to PyPI. Contents look slightly different from TestPyPI rendering, but all is good.
Links function as expected.
Release history looks good with a giant jump in version number.
Only see source distribution for this release.
Followed Python installation notes for pip installing from PyPI
Clean venv
Meson complained clearly since I didn't have an MPI installation
Installed MPICH v5.0.1 with homebrew
See good information in Meson logs and compile/link commands
See v1.2.0 everywhere as expected
python -m pip show openbt and python -m pydoc openbt look good
Reviewed install carefully and ran otool -L on all installed CLTs.
Trigger full set of Taweret GH actions, confirm that they are installing this latest version, and confirm that all tests are passing.
Reviewed trees notebooks in Taweret Jupyter book
Confirm that stable version of Guides was built by RTD.
Confirm that users see the stable version by default, but can choose to see the latest as well.
Confirm that users can also download the PDF version and that this has the correct version identifier.
Activity
(Issue #27) We will publish v1.2.0 on PyPI.
jared321 commentedon Jul 28, 2026
Pre-release review (all tasks finished)
meson.buildandpyproject.tomlusing the same version constraintseigenversion in Meson's wrapdb is 5.0.1-1, which is what we presently usesetup.pysetuptools/setuptools-scmset according to current recommendations for custom configurationsnumpyandscipy. We also don't have a means to confirm that minimum versions are actually work. So, we continue to leave it open ended for this release.setup.pyandpyproject.tomlexternal dependencies are disjoint and independent. Therefore, they are consistent.tox.inidoesn't have any hardcoded version identifiers still.typos.tomlto account for changes to repo contents made during mergelibmpich-dev. Added TODO to action regarding this potential avenue of exploration.bandframeworkrepo for the BANDv1.0releasev1.2.0Ropenbtversion already set tov1.2.0(Issue #27) Clean Python package metadata.
(Issue #27) Updated external dependence version identifiers.
(Issue #27) Udpate typos config after finalizing repo structure.
(Issue #27) General cleaning as part of review.
(Issue #27) Cleaning as part of reviewing actions for release.
(Issue #27) Cleaning as part of prerelease review.
(Issue #27) Let typos check R wrapper's manual pages
jared321 commentedon Jul 28, 2026
Release Notes
Since this is the first release in this new and significantly expanded repo, the level of release testing is high.
7048a73e7048a73easv1.2.0, officially published a release associated with that version tag, and confirmed that the release candidate action ran through successfully.openbt_pypkgin local clone.mesonv1.11.2 installed from homebrewninjav1.13.2 installed from homebrewbrew cleanup --prune=all -seigeneigenv5.0.1 is pulled down and installed automatically wheneigennot in systemeigenv5.0.1 and v3.4.1 installed via homebrew were found and used bymesonso long aspkgconfwas also installed by homebrew to providepkg-config. Curiously the package is referred to byeigen3inpkg-configfor both versions.-DNDEBUG,-DSILENT, and-D_USEMPIappear when appropriate onlyotool -Lon installed CLTs looks good and on intermediate library as wellbuilddirwithmeson testandmeson test <test name>(e.g., withcrnandtree)/opt/homebrew/lib/R/4.6/site-libraryopenbtcommand was issued; 2 processes @ ~100%, when calculating Sobol' indicestest.obtîs 3.4Mninja, but the Meson error message is clear thatninjais missing. I don't want to include in this procedure a specific mechanism for installingninjasince some might prefer to install from a package manager; others, from PyPI. The need for manual installation ofninjais made clear in those same docs before getting to this point. I Don't think that we need to add more.pip installwithout eigen installedpip install -v -e .with (v3.4.1) and without eigen installed by homebrewpython -m pip show openbtandpython -m pydoc openbtotool -Lon installed CLTs and all looks good.-vto pip installsetup.pyin multiple ways and confirm thatpip installfails with a useful message.topmpicxx, which is installed by Intel but curiously wraps g++. The correct wrapper ismpiicpc. As in all other cases, however, the wrappers are not called. Instead the correct Intel compiler that it detected is called with compatible flags. As in previous testing, however, the flags it used were associated with the MPI implementation used in the previous test. Reran from a clean login successfully. I have seen some issues with the OpenMPI modules still having an effect on the env vars after unloading and despite loading a new MPI installation. I wonder if what I saw previously was the effect of a bad environment.twine check <tar ball>passedv1.2.0everywhere as expectedpython -m pip show openbtandpython -m pydoc openbtlook goodotool -Lon all installed CLTs.(Issue #27) Propagate changes from last commit to related files.
jared321 commentedon Aug 7, 2026
Integration into BAND
bandframeworkrepo