Support local paths for InputDataset.source
#30
Conversation
| if _get_source_type(source) == "path": | ||
| self.exists_locally = True | ||
| self.local_path = source |
There was a problem hiding this comment.
Again I wonder if this could all be abstracted behind a class. e.g.
self.source = DataSource(source)then later use self.source.exists_locally and self.source.path.
There was a problem hiding this comment.
Yes, I began implementing this for this PR but it got hairier than expected. I think it requires a lot of design considerations and should be returned to once the core feature set is complete. AdditionalCode and BaseModel also have a source_repo attribute (that I think is about to be renamed source to accommodate local files) that could be replaced with an instance of whatever this class will be. I think this could be part of the tidy-up that brings in pathlib. I'll raise an issue for now.
TomNicholas
left a comment
TomNicholas
left a comment
There was a problem hiding this comment.
Nice. I have one main comment about the idea of abstracting out the "source" of the data. It's related to #9 and drivendataorg/cloudpathlib#455.
| If InputDataset.source is... | ||
| - ...a local path: create a symbolic link to the file in `local_dir/input_datasets`. | ||
| - ...a URL: fetch the file to `local_dir/input_datasets` using Pooch | ||
| (updating the `local_path` attribute of the calling InputDataset) |
There was a problem hiding this comment.
Yeah this is another argument for abstracting out the concept of a "source" into a standalone class / type.
This is a similar idea to using cloudpathlib, but maybe we need to attach more info, in which case we would need our own Source class with a .path attribute.
| to_fetch.fetch(os.path.basename(self.source), downloader=downloader) | ||
| self.exists_locally = True | ||
| self.local_path = tgt_dir + "/" + os.path.basename(self.source) | ||
| # If the file is somewhere else on the system, make a symbolic link where we want it |
There was a problem hiding this comment.
This could all go behind a Source interface.
- add _get_source_type method in utils (returns 'url' or 'path') - add call to _get_source_type in InputDataset.__init__() which updates local_path and exists_locally if 'path' - add logic to InputDataset.get() which creates a local symbolic link to the dataset path if it is elsewhere on the system - update InputDataset.check_exists_locally() to reflect above changes
… yaml created with Case.persist()
- Modifies the blueprint created by the first case to use local paths to input datasets where available - Creates and runs a second case `roms_marbl_local_case` from this blueprint
dafyddstephenson
force-pushed
the
local_files
branch
from
18fd968 to
b032dcc
Compare
2 participants
closes #4 .
Summary of changes:
Core:
_get_source_typemethod inutils(returns 'url' or 'path')_get_source_typeinInputDataset.__init__()and setlocal_pathandexists_locallyattributes ifsourceattribute is a local pathInputDataset.get()which creates a local symbolic link to the dataset path if it is elsewhere on the systemInputDataset.check_exists_locally()to reflect above changesCI:
tests/test_roms_marbl_example.py. The first section creates and runs the caseroms_marbl_remote_casewhere all input datasets are URLs, creating a blueprinttest_blueprint.yamlas before. The second section first moves all the fetched datasets to an unrelated directory and then modifies thetest_blueprint.yamlfile to replace URLs with the path to this dir, before creating and running another caseroms_marbl_local_casefrom this modified yaml file.Other/bugfixes:
Case.persist()andCase.from_blueprint()when modifying the test routine above:valid_end_datewas overwritingvalid_start_dateentry in blueprint outputstart_dateandend_dateassociated with input datasets were not being written to blueprint outputCase.check_is_setup()was not correctly checking for the local presence of additional codetreeon system before calling it (closes bash tree: command not found #18)Case.end_dateif they want quick results (closes simplest example should run really fast #21)ci/environment.ymlto includenetCDF4andxarray(closes Missing dependencies for analysis #23) and removencoandncview(issue not yet raised but was mentioned in Tom's notes)