QCoDeS 0.61.0

[Jens Hedegaard Nielsen (jenshnielsen)]

QCoDeS 0.61.0 (2026-10-08)

Breaking Changes:

• Remove the deprecated TypeVars and type aliases that were retained as runtime
  compatibility imports after the conversion to PEP 695 type parameter syntax.
  Imports of typing.TypeVar and typing.ParamSpec are now banned: use PEP
  695 type parameter syntax unless a default value requires the corresponding
  typing_extensions type on Python 3.12. Remove deprecated typing compatibility exports #8543

Improved:

• The documentation on naming instrument drivers (in the Contributor guide and in the
  "Creating Instrument Drivers" example notebook) has been updated to state that instrument driver
  modules should be named using lower case snake_case, e.g. weinschel_8320.py. Vendor and
  model capitalization belongs in the class name (Weinschel8320) and not in the module name. Enable ruff lint N999 #8356

• Exceptions raised by QCoDeS drivers are now more specific.

  The ruff rule TRY002 (raise-vanilla-class) has been enabled. A number of drivers raised a
  bare Exception; these now raise a specific builtin exception instead. Invalid arguments raise
  ValueError, using the instrument in an unsupported state raises RuntimeError and selecting an
  unimplemented acquisition mode on the AlazarTech ATS raises NotImplementedError. Code that
  catches Exception is unaffected since all of these are subclasses of Exception.

  In addition, two catch-all exception handlers have been narrowed to the specific exceptions that the
  guarded code can actually raise: reading a screenshot from the Keysight Infiniium now only handles
  OSError and VisaIOError, and waiting for the Tektronix AWG5014 to become ready only handles
  VisaIOError. Both previously swallowed unrelated errors.

  Finally, the coding style section of the contributor documentation now points at ruff check and
  ruff format, configured in pyproject.toml, rather than the long removed pycodestyle and
  flake8 based workflow. Implement remaining ruff 0.16 lints #8368

• Adding metadata that SQLite cannot store in a column now fails with a message naming the tag and its type.

  A nested dict or a sequence previously surfaced as sqlite3.ProgrammingError: Error binding parameter 1: type 'list' is not supported, which said nothing about which tag was at fault. validate_dynamic_column_data
  now rejects such values with a TypeError suggesting serialization, alongside its existing checks for invalid
  tags and None. Values that NumPy registers a SQLite adapter for, such as scalars and arrays, are unaffected. Fail meaningfully on metadata a column cannot hold #8518

• Fixed NetCDF export of measurements without setpoint dependencies, including
  scalar and array-valued parameters measured with do0d or dond.
  The exported data retains its shape using parameter-specific dimensions. fix: support NetCDF export without dependencies #8570

• A second interrupt during a database write rolls the open transaction back instead of leaving the file locked.

Improved Drivers:

• The Infiniium driver now tests explicitly for the type of parent instrument when required and
  used cast where pyvisa types the
  return of read_binary_values/query_binary_values as a Sequence[float]
  regardless of the requested container. This replaces five type checker
  suppressions. More type safe Infiniium driver #8457

New:

• Added split raw data storage for the DataSet: an opt-in results
  backend selected via the dataset.raw_data_backend config option
  (set it to "sqlite_per_dataset_db") that writes raw measurement
  data into individual per-dataset SQLite files while keeping all metadata
  in the main database. This prevents the main DB file from growing
  excessively large and allows users to have a single main SQLite QCoDeS
  database on their PC as opposed to many that grow infinitely. When
  selected, no results table is created in the main database at all - it
  holds only metadata - and all public DataSet APIs continue to work
  unchanged. Backend-specific settings (such as the per-dataset file
  folder raw_data_path) live under dataset.raw_data_backend_config.

  Management helpers included:

  • update_raw_data_paths(): update stored paths after raw data files
    have been moved to a new folder.
  • purge_orphaned_datasets(): find and remove dataset records whose
    raw data files no longer exist on disk (dry_run=True by default).
  • cleanup_datasets(): remove datasets (DB records and raw data files)
    by age, sample name, or file size (dry_run=True by default). feat: split raw data storage into per-dataset SQLite files #8219

• The method for setting a parameter's cache from a raw value as reported by the
  instrument is now public as Parameter.cache.set_from_raw_value. It is the
  raw-side counterpart of Parameter.cache.set and applies the same
  conversions as a get (get_parser, scale, offset and
  val_mapping). This is useful for drivers that populate the caches of many
  parameters from a single bulk instrument reply without issuing a get per
  parameter. The previous private _set_from_raw_value is kept as a
  backwards-compatible alias. Make Parameter cache.set_from_raw_value public #8484

• Instruments can now opt in to a logger of their own, rather than sharing a
  single logger with every other instrument in the process. Setting the class
  attribute default_logger_scope to "instrument" makes Instrument.log
  (and VisaInstrument.visa_log) use a logger named after the driver class and
  the instrument's name_parts, so that levels, handlers and propagate can
  be configured per instrument::

  class MyDriver(VisaInstrument):
      default_logger_scope = "instrument"
  

  The scoped loggers mirror the instrument hierarchy below the module-qualified
  driver class. For an instrument myinst of MyDriver, defined in the
  module vendor.driver, the loggers are named::

  vendor.driver.MyDriver                    the driver class
  vendor.driver.MyDriver.myinst             Instrument.log
  vendor.driver.MyDriver.myinst.my_module   log of a submodule
  vendor.driver.MyDriver.myinst.com.visa    VisaInstrument.visa_log
  

  A submodule is named after the key it is added with in add_submodule, for
  example my_module for self.add_submodule("my_module", MyModule(self)),
  and not after its own name or class. A submodule that is not added as an
  individual submodule, such as a channel that only exists in a channel list, is
  named after its own name.

  Each logger is a child of the previous one, so a level or handler can be set
  for a whole driver class, for everything that belongs to one instrument
  (its own messages, its submodules and its VISA traffic), or for a single
  submodule or the VISA traffic of one instrument. Setting the level for a whole
  driver also works before any of its instruments exist, so it can be configured
  via logger_levels in qcodesrc.json.

  Because module-qualified scoped loggers are no longer below
  qcodes.instrument.instrument_base, opted-in drivers do not inherit levels
  configured on that shared logger or on qcodes.instrument.instrument_base.com.visa.
  Configure the module-qualified driver class logger instead.

  The default is unchanged: without opting in, all instruments keep sharing one
  logger exactly as before. Add opt-in per-instrument logger scope (default_logger_scope) #8523

Under the hood:

• Enable the ruff rule N999 (invalid-module-name). Instrument driver modules that are part of the
  public qcodes namespace are exempted via lint.pep8-naming.extend-ignore-names in pyproject.toml
  so that no public module names change. Test modules with invalid names have been renamed to lower case. Enable ruff lint N999 #8356

• CombinedParameter.parameter is now a small dataclass holding the name,
  full_name, label and unit of the combined parameter, replacing a
  lambda that had those attributes attached to it. It remains callable, and
  calling it returns None as before. Bump github/codeql-action/upload-sarif from 4.37.3 to 4.37.4 #8364

• Two additional ruff lint rules have been enabled.

  TRY004 (type-check-without-type-error) ensures that new code raises TypeError when
  rejecting a value because of its type. Existing type checks that raise ValueError or
  RuntimeError are unchanged and carry an explicit noqa comment, since TypeError is not a
  subclass of either and changing them would break user code that catches the current exception.

  S110 (try-except-pass) flags exception handlers that silently pass. No behaviour changed as a
  result: the only remaining silent handlers are in teardown paths that must never raise or log, and
  they are now explicitly marked as intentional. Implement remaining ruff 0.16 lints #8368

• Enable the ruff rule BLE001 (blind-except). Places where QCoDeS intentionally catches a
  broad exception are now explicitly marked, and several catch-all handlers now log the full traceback
  using Logger.exception rather than only the exception message. The private module level loggers
  named _LOG have been renamed to _LOGGER. style: Enable ruff BLE001 (blind-except) #8498

• The test suite no longer depends on deepdiff. The tests that used it now compare the data directly
  using InterDependencies equality and numpy.testing.assert_array_equal. Update agent instructions to prefer prek and uv run #8552

---

This discussion was created from the release QCoDeS 0.61.0.