dataset argument

All commands which operate on datasets have a dataset argument (-d or --dataset for the CLI) to identify a single dataset as the context of an operation. If --dataset argument is not provided, the context of an operation is command-specific. For example, clone command will consider the dataset which is being cloned to be the context. But typically, a dataset which current working directory belongs to is the context of an operation. In the latter case, if operation (e.g., get) does not find a dataset in current working directory, operation fails with an NoDatasetFound error.

Impact on relative path resolution

With one exception, the nature of a provided dataset argument does not impact the interpretation of relative paths. Relative paths are always considered to be relative to the process working directory.

The one exception to this rule is passing a Dataset object instance as dataset argument value in the Python API. In this, and only this, case, a relative path is interpreted as relative to the root of the respective dataset.

Special values

There are some pre-defined “shortcut” values for dataset arguments:


Represents to the topmost superdataset that contains the dataset the current directory is part of.


Represents the root directory of the dataset the current directory is part of.


Represents the “default” dataset located under $HOME/datalad/.

Use cases

Save modification in superdataset hierarchy

Sometimes it is convenient to work only in the context of a subdataset. Executing a datalad save <subdataset content> will record changes to the subdataset, but will leave existing superdatasets dirty, as the subdataset state change will not be saved there. Using the dataset argument it is possible to redefine the scope of the save operation. For example:

datalad save -d^ <subdataset content>

will perform the exact same save operation in the subdataset, but additionally save all subdataset state changes in all superdatasets until the root of a dataset hierarchy. Except for the specification of the dataset scope there is no need to adjust path arguments or change the working directory.