Releases: MathiasStokholm/alkymi
Releases · MathiasStokholm/alkymi
Release v0.3.1
Fixed
- Fixed readthedocs.org documentation build now requiring a .readthedocs.yaml config file
- Fixed a bug where the actual error message from an exception happening during execution through a
Lab
instance would
be omitted - Fixed an issue where exceptions in recipes could cause asyncio to print warning information about the
_schedule
coroutine never being awaited - Fixed an issue where asyncio would warn about exceptions not having been retrieved from futures when more than one
exception was raised during parallel execution. Alkymi will now just raise the first exception it encounters.
Release v0.3.0
[0.3.0] - 2024-04-23
Added
- Added a new
doc
property to recipes that provides a documentation string for the recipe. If not provided upon
creation of a recipe, alkymi will try to infer the documentation string by parsing the bound function's docstring - Added a new optional
doc
argument to thearg
recipe that provides a documentation string for the argument
Changed
stderr
is now streamed by default when using theutils.call()
utility function #42- When utilizing the
Lab
command line interface, alkymi will now omit alkymi internals from tracebacks resulting from
exceptions raised by user code (recipes) - Refactored the
Lab
command line interface to provide rich information by making use of the newdoc
property on
Recipe
andarg
instances - Changed the allowed arguments to
brew X
to only contain arguments that affect recipeX
Fixed
- Remove unnecessary decimal points in fancy progress output #42
- Fixed a bug where the terminal cursor would disappear if an exception was thrown during execution
- Fixed a bug where the terminal cursor would disappear if a
Lab
call tobrew
was terminated using ctrl-c - Fixed a bug where recipes returning
None
(no return value) would not be correctly cached on subsequent runs - Fixed CI pipeline integration with codecov.io (test coverage)
- Fixed CI pipeline execution on MacOS agents
Removed
- Removed the ability to brew multiple recipes from a
Lab
by listing them in succession
Release v0.2.1
Release v0.2.0
Added
- Added
rich
as a runtime dependency - Added fancy progress output mode that shows a summary of the execution at the bottom of the console using Rich.
- Added support for Python 3.11
Changed
- Rewrote parallel execution to only run bound functions in parallel, but use asyncio for the main execution logic
- Bumped
mypy
version to1.2.0
(only used for development) - Bumped
coverage
version to6.4.2
(only used for development)
Fixed
- Fixed a bug where a deadlock could occur when evaluating a number of
ForeachRecipes
with the same number of jobs - Fixed a bug where a certain graph configuration could prevent parallelization from happening correctly
- Fixed a regression where all cached recipe outputs would always be loaded during execution, regardless of whether they
were needed to produce the requested output.
Removed
- Dropped support for Python 3.6
Release v0.1.0
Added
- Added
networkx
as a runtime dependency - Added support for parallel execution - the
brew()
function now takes ajobs
parameter that controls
parallelization of the recipe evaluation. - Added support for parallel execution to the
Lab
CLI - use--jobs
or-j
when callingbrew
to run pipeline in
parallel - Added
twine
andwheel
todev-requirements.txt
Changed
- Rewrote the internals of alkymi to create and use a
networkx
graph (DAG) for execution. This change will enable
future changes, like running recipes in parallel. - Renamed
alkymi.py
tocore.py
- Converted
ForeachRecipe
to have the same functional interface as a regularRecipe
. This change enables the
functions incore.py
to not have special handling forForeachRecipe
, which makes the code simpler and easier to
maintain. - Coverage step
labfile.py brew coverage
will not print which lines lack test coverage - The
invoke
andinvoke_foreach
functions have been moved fromrecipe.py
andforeach_recipe.py
intocore.py
to allow more control over the execution flow.
Removed
- Removed the special
MappedInputsDirty
status, which only applied toForeachRecipe
. If aForeachRecipe
has only
completed a partial evaluation of its enumerable input, its status will now showInputsChanged
instead.
Release v0.0.7
Added
- Added a
file_checksum_method
config option that can be used to select whether to use file content hashing (default)
or file modification timestamps for calculating the checksum of an external file represented by aPath
object
Fixed
- Fixed a bug where CLI arguments containing a hyphen would not be parsed correctly when used in a
Lab
- Fixed a bug where the implementation of coverage in
labfile.py
would break the debugger for the entire script. Now
only thecoverage
steps breaks debugging (which is expected, since coverage replaces the tracing function used by the
debugger)
Changed
- Made the
call()
utility function print program outputs (stdout) to stdout by default - this can be controlled using
theecho_output_to_stream
argument.
Release v0.0.6
Added
- Added py.typed file to signal to downstream users that alkymi has type annotations
- Added optional dependency on xxhash to speed up hashing of large files (
Path
objects outside of alkymi's cache) - Added option to supply ingredients to a recipe by using argument names that match recipes (ala pytest's fixtures)
- Added support for explicit naming of recipes (if not provided, the bound function name will be used)
- Added support for providing CLI arguments to recipes through a
Lab
using the newregister_arg()
function - Added tests for the
Lab
class
Changed
- Updated Sphinx and associated packages to fix documentation build errors
- Updated development libraries (pytest, coverage, sphinx, mypy, flake8)
- Made the
call()
utility function print error stack traces to stderr by default - this can be controlled using the
echo_error_to_stream
argument.
Fixed
- Fixed a bug where file names were not taken into account during checksumming (only file content was taken into
account) - Fixed a bug where a recipe's cache path would be different when a script was invoked as a relative or absolute path
- Fixed a bug where item validity was not being taken into account in
ForeachRecipe
(e.g., an invalid Path such as a
deleted file would not cause reevaluation) - Fixed a bug where a
ForeachRecipe
wouldn't be completely reevaluated if its bound function changed - Fixed a bug where changing the default value of an argument to a bound function did not cause the checksum of the
bound function to change - Fixed coverage to correctly include alkymi imports when executed through
labfile.py
Removed
kwargs
/NamedArgs
built-in recipe removed (functionality can now be mimicked using the built-inArg
)- Dropped support for Python 3.5
Release v0.0.5
Added
- Added a new
zip_results()
built-in recipe generator to zip together outputs from multiple recipes - Documentation in the
docs/
directory. The documentation is built using Sphinx and hosted on
https://alkymi.readthedocs.io/en/latest/. The documentation can be built by runningpython labfile.py brew docs
. - Added checksum unit tests for the
Path
type - Added unit test for caching of recipes that do not return anything (None)
- Added
allow_pickling
configuration option to let the user turn off pickling for serialization, deserialization and
checksumming and a unit test for it - Added test coverage using
coverage
and https://codecov.io (see thecoverage
recipe inlabfile.py
)
Changed
- Regular checks for cleanliness are now run even if a custom cleanliness check passes (e.g. for the
glob_files()
built-in recipe generator). This ensures that changes to external files and others are correctly caught and handled. - Broke
Status.Dirty
into several more explicit causes of dirtiness (e.g.OutputsInvalid
) to make it more clear to
the user why a given recipe has been marked dirty - Moved functions for checking Recipe and ForeachRecipe dirtiness to alkymi.alkymi to reduce the complexity of the
classes - Moved the core logic of the
Recipe.brew()
function to alkymi.alkymi - Provide names explicitly to built-in recipe generators to avoid name clashes when a built-in recipe generator is used
multiple times in a single module - Greatly simplified the serialization and deserialization logic and got rid of generators
- Converted the built-in recipe types
Args
andNamedArgs
to be subclasses ofRecipe
to avoid the clunky.recipe
property. - Updated
README.md
to reflect the new documentation page at https://alkymi.readthedocs.io/en/latest/. - Replaced the usage of
tuple
with a newOutputs
type that keeps most of the behavior, but ensures that all return
types are valid - Made
Recipe
generic in the return type and forwarded return type information from decorators to allowbrew
to
return valid type information - Updated built-in recipes to supply return type information to their
Recipe
objects - Use the highest available protocol when pickling for serialization and checksumming
Fixed
- Converted several captured variables inside tests to globals to avoid them interfering with hashing of the bound
functions - Fixed an issue where a bound function changing between evaluations could cause the status to be reported as
"NotEvaluated" instead of "BoundFunctionChanged". Added a test step to check this. - Fixed
utils.call()
to work regardless of operating system and added a small unit test - Fixed caching of dictionaries output by ForeachRecipe
- Dictionaries are now serialized using a dict with key and value entries supporting arbitrary nesting, instead of being
pickled - Fixed a bug where non-existent
Path
objects would result in the same checksum - Fixed a bug where recipes without return values would remain
NotEvaluated
even though they had been evaluated and
cached
Release v0.0.4
Added
- Updated README.md with "Command Line Usage", "Upcoming Features" and "Known Issues" sections
- Included
labfile.py
in static type checking and linting - Added a unit test for the built-in
file()
recipe generator - Made cache directory configurable using
AlkymiConfig
and added a test for it - Added a new test to verify that partially executed ForeachRecipes can be reloaded from cache and continue from point
of interruption/failure
Removed
- "clean" and "clean-cache" commands from Lab (these will be reimplemented properly at a later point)
Changed
- Renamed alkymi's labfile from
lab.py
tolabfile.py
to fit the new "labfile" naming scheme - ForeachRecipe now supports saving/restoring partial evaluation states to the cache. This fixes the issue where a
failure in one of the foreach evaluations could cause all the work to be lost - Recipes will now be marked dirty if their bound function has changed between invocations
- ForeachRecipe now only serializes each output once
- Outputs are now loaded lazily from the cache. This means that actual deserialization is deferred until an output is
actually needed for computation. - alkymi's labfile now uses
exit()
instead of raising exceptions in order to not clutter the output of tests etc. - Adapted the way that external (outside alkymi's cache) files are checked for validity to see if they have changed
between alkymi evaluations - Updated README.md to reflect recently merged features
Fixed
- Fixed type annotations in alkymi's labfile (
labfile.py
) - Fixed reloading of recipe checksums from cache (these needed to be converted from lists to tuples due to JSON
serialization format). This would cause some recipes to be marked dirty even though they weren't. - Fixed a bug where the aggregate checksum for all mapped inputs to a ForeachRecipe weren't being saved and restored
from cache
Initial release
alkymi 0.0.3 Release Notes
Initial version of alkymi open sourced and released to PyPI!