CHANGELOG.md

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

unreleased

0.8.1 - 2024-08-13

Added

• CRESTAdapter that supports CalcType.conformer_search.
• mkdocs documentation.

0.8.0 - 2024-07-19

Changed

• Updated to qcio 0.11.0 which removed NoResults and requires explicit passing of a .results value to ProgramOutput rather than implicitly passing None and having qcio set .results = NoResult. It also moved .files from ProgramOutput to the results objects.

0.7.5 - 2024-07-17

Changed

• Only respect collect_file for for Adapter.compute if the adapter has .uses_files=True.
• Moved the program_version = self.program_version(stdout) call from BaseAdapter.compute(...) to outside the try block so that program version is collected even if the calculation fails.
• Cleaned up xtb cached version implementation.
• Added confirm_version check to release.py.

0.7.4 - 2024-07-16

Changed

• Added threading.Lock() to capture_sys_stdout so that it is thread safe.
• Cached XTBAdapter.program_version result since calls to importlib.metadata.version(...) call os.listdir() and when doing very many xtb calls these became substantial (almost 1/2 the execution time). We have to use importlib.metadata.version rather than xtb.__version__ directly because their __version__ string is wrong.

0.7.3 - 2024-07-12

Added

• release.py script for automated releases.

Changed

• Upgraded to qcio 0.10.2 which reverts back to the default of Structure.identifiers instead of Structure.ids.
• Fix depreciated call to qcparse.parse_results for the TeraChem adapter.

0.7.2 - 2024-07-12

Added

• xtb to optional installs and error message for XtbAdapter to indicate it can be installed with pip install qcop[xtb].

• qcop.adapters.utils.set_env_variable context manager to address the fact that one must set the env var OMP_NUM_THREADS=1 before importing any of the xtb library or else performance is severely impacted (around 10x). They spawn threads within threads so if you leave the default value (which is the number of cores on your machine) you get way more threads than this! E.g., on my 16 core machine I'll get 47 threads spawned for a single calculation. If I set this to 1 I actually get the correct number of threads, which is 16. One must set this value before importing any piece of the xtb library, so if your script imports anything from xtb before calling qcop.compute("xtb", ...) wrap this import with the qcop.adapters.utils.set_env_variable wrapper like this:

  from qcop.adapters.utils import set_env_variable
  
  with set_env_variable("OMP_NUM_THREADS", "1"):
      import xtb
      import xtb.interface
      import xtb.libxtb
      from xtb.utils import Solvent
  
  ### The rest of your script

0.7.1 - 2024-07-10

Fixed

• Incorrect check in BaseAdapter.compute() using isinstance(inp_obj, FileInput). Updated to type(inp_obj) is FileInput to work with updated qcio inheritance hierarchy.

0.7.0 - 2024-07-10

Added

• Added validation check for programs that do not support file inputs if files are passed as an input.

Changed

• 🚨 Updated to qcio 0.10.1 which uses Structure in place of Molecule.
• Changed Adapter.write_files to Adapter.uses_files to more clearly define purpose of this variable.
• tmpdir context manager only creates a temporary directory if mkdir is passed as the first argument. This works in conjunction with Adapter.uses_files so that adapters that do not use file I/O (e.g., pure Python programs) do not have the additional overhead of creating and removing temporary directories for their calculations.

0.6.2 - 2024-06-13

Added

• try/except to top-level compute function that will append the .program_output to the exception if the get_adapter function raises an exception.

Changed

• Modified exception hierarchy to include fewer **kwarg arguments, use mostly args instead, and fixed issue where some exceptions still didn't have .program_output set on them.

0.6.1 - 2024-06-10

Fixed

• *args and **kwargs passing to parent classes in AdapterError and subclasses to fix celery serialization issues in BigChem.

Changed

• 🚨 Changed compute kwarg collect_wavefunction to collect_wfn to match BaseAdapter nomenclature. (Cheating in this breaking change since I don't think this feature is used by outsiders yet...)

0.6.0 - 2024-04-23

Fixed

• Fixed all mypy errors that were previously not raised because qcio did not contain a py.types file.

Changed

• Updated to latest qcio data structures eliminating SinglePointOutput, OptimizationOutput, and ProgramFailure in favor of a Generic ProgramOutput object. Backwards compatibility is maintained by qcio so that old SinglePointOutput, OptimizationOutput, and ProgramFailure objects are still accepted in end user code with FutureWarning messages.
• BaseAdapter is generic over InputType and ResultsType.
• ProgramAdapter is generic over StructuredInputType and ResultsType.
• Adapters now inherit from these Generic classes and specify compatible types explicitly.
• Updated examples and tests to reflect the new qcio data structures nomenclature, emphasizing pi for ProgramInput and po for ProgramOutput objects.

0.5.5 - 2024-04-12

Added

• AttributeError check to capture_sys_stdout so that when running inside celery and sys.stdout is LoggingProxy object, we can still capture the stdout correctly.

0.5.4 - 2024-04-12

Added

• try/finally blocks to tmpdir context managers to ensure that temporary directories are always cleaned up and the original cwd is restored, even if an exception is raised.
• Properly capture qcparse encoder/parser exceptions in TeraChemAdapter and raise qcop.AdapterInputError.

Changed

• Moved self.validate_input(inp_obj), self.program_version(stdout), and self.collect_wfn(...) inside of try/exceptblock inBaseAdapter.compute(). This is more correct by catching all possible qcop exceptions and returning a ProgramFailure object on the raised exception. This helps eliminate 500 errors in ChemCloud when a user passes an invalid input object to a program adapter or requests a wavefunction from a program that doesn't support it.
• Redefined signature to collect_wfn to collect_wfn() -> Dict[str: Union[str, bytes]] so that it can run before constructing the ProgramFailure object in BaseAdapter.compute().

0.5.3 - 2024-04-11

Added

• capture_sys_stdout context manager to capture stdout from non-python libraries (e.g., C/C++). This is necessary for processes that do not write to Python's sys.stdout object, such as xtb.
• xtb to developer dependencies and a xtb tests to the test suite.

Changed

• xtb does not close file handles correctly so we'd get OSError: [Errno 24] Too many open files after running a few hundred jobs. XTBAdapter now uses the new capture_sys_stdout context manager to capture stdout.
• xtb.program_version now returns importlib.metadata('xtb') because xtb incorrectly hardcodes the version in the __init__.py file. Issue here.

0.5.2 - 2024-04-10

Added

• Added xtb adapter. qcengine.get_program adds almost >1 second of overhead to each call, which makes xtb unusable via qcengine since a single xtb execution make take only ~0.003 seconds.

0.5.1 - 2024-03-29

Changed

• Upgraded to qcio>=0.8.1 to fix QCElemental behavior that auto-rotates Structures without user consent.
• Modified all example scripts to be standalone without referencing an external .xyz file and to use try/except statements now that raise_exc=True is the default.

Added

• py.typed file to enable type checking in other projects that use qcop.

0.5.0 - 2024-03-16

Changed

• Changed default compute behavior to raise_exc=True instead of raise_exc=False.

Removed

• Non operational placeholder psi4 integration test.

0.4.8 - 2024-01-12

Added

• GeometricError to qcop.exceptions module. Without this when geometric raised an exception it wouldn't be caught by the except QCOPBaseError block in adapter.compute() and the ProgramFailure object wouldn't be returned to the user. This fixes a 500 error in ChemCloud Server when geometric fails to converge and raises an exception (a geometric exception) that isn't found by the server.

Changed

• QCEngineError now a subclass of ExternalProgramError instead of QCOPBaseError.
• ExternalProgramExecutionError renamed to ExternalSubprocessError to better reflect its purpose.

0.4.7 - 2023-09-27

Changed

• Removed TeraChem input file encoding as this was refactored to the qcparse library.
• Updated qcparse dependency from >=0.5.1 to >=0.5.2 to support the new encoding API.
• Updated qcio dependency to reflect the same minimum version as qcparse.

0.4.6 - 2023-09-22

Changed

• adapter.collect_wfn() now raises AdapterInputError rather than NotImplementedError if propagate_wfn=True is passed. This change allows these errors to be captured by the except QCOPBaseError as e: block in adapter.compute() so that the ProgramFailure object can be returned to the user. This fixes a 500 error in ChemCloud Server when a user passes propagate_wfn=True to a program that doesn't support it.

0.4.5 - 2023-09-19

Added

• README.md documentation.

Changed

• Dropped Python version from ^3.8.1 to 3.8.
• Updated qcparse dependency from >=0.5.0 to >=0.5.1 to support Python ^3.8 change.

0.4.4 - 2023-09-08

Added

• Added ProgramFailure object to exception.args in BaseAdapter.compute() method so that Celery will correctly serialize the object and pass the exception.program_failure object from workers to clients.

Changed

• Updated qcio from >=0.5.0 to >=0.6.0.
• Dropped multiple inheritance on ExternalProgramExecutionError by removing CalledProcessError. Retained the original attributes.

0.4.3 - 2023-09-03

Fixed

• Extras specification in pyproject.toml so that qcengine, qcelemental, and geometric don't get installed by default but may be optionally installed if compatibility is desired.

0.4.2 - 2023-09-02

• TeraChem adapter was writing CalcType.energy instead of energy for runtype in tc.in file.

0.4.1 - 2023-09-02

Added

• Added a ProgramFailure object to all program related exceptions raised by adapter.compute(). Now every exception returned from .compute() will have the full traceback context and provenance data for error handling. Very helpful so that BigChem can operate on a raise_exc=True paradigm to take advantage of python standard failure/retry logic based on exceptions but still return the entire ProgramFailure object for returning to end users via ChemCloud.

Fixed

• stdout now gets correctly added to ProgramFailure object.

0.4.0 - 2023-08-31

Changed

• Updated pydantic from v1 -> v2.

0.3.2 - 2023-08-18

Added

• GeometricEngine now appends all computed trajectory results to the raised exception if subprogram crashes so a user can inspect the failed results.

0.3.1 - 2023-07-19

Fixed

• Tiny bug in ensure_geometric test.

0.3.0 - 2023-07-19

Added

• GeometricAdapter
• Added .collect_wfn and .propagate_wfn methods to adapters to facilitate wavefunction passing for sequential program adapters--like geomeTRIC.

Changed

• Refactored top level main.compute() function to mostly live in BaseAdapter.compute(). This makes consumption of adapters by dual program adapters, like GeometricAdapter, much cleaner.

0.2.0 - 2023-07-17

Changed

• Updated to flattened qcio v0.3.0 data models.
• Changed order of args to compute() to allow for the same pattern to compute_args.

0.1.0 - 2023-06-29

Added

• Core compute capabilities exposed in top-level compute() function, including QCEngine fallback for programs without a qcop adapter.
• Created adapter for TeraChem.
• Exposed general purpose computing for any subprocess with FileAdaptor