Tom Close

Lazy fields that point to a specific Job

1

Lazy fields that point to a specific Job, i.e. instead of a Node, to allow execution of a node with partially resolved state arrays when the upstream hasn't completed yet,...

Replace propagate_rerun with being able to pass list of node names to rerun

1

This will allow much finer grain control over which nodes of a workflow need to be run than `propagate_rerun`, which is kind of useless as if you set it to...

Better hashing of generic objects

2

Currently the hashing of generic objects is a bit brittle, for example torch.Tensor objects don't have an empty `__dict__` so they all get hashed to the same value. An option...

Hashing of functions and methods is currently done on their AST as a first pass, and falling back to the byte-code if the source isn't available. However, this doesn't pick...

There is no mention of the auditing and messaging in the docs currently (only a stub section). Would be good to have (especially as I don't know how it works!)

A detailed how to on how to use the nipype2pydra tool to convert existing nipype interfaces to pydra. This could live on separate nipype2pydra docs but might make more sense...

Replace container_ndim with split_dims arg

1

### What would you like changed/added and why? The `container_ndim` (formerly `cont_dim`) be replaced by `split_dims`. Where as container_ndim only allows you to specify the number of outer dimensions to...

In addition to the `_error.pklz` that is created when a task/workflow errors, it would improve Pydra's user friendliness if there a Jupyter notebook was generated (along with bash/batch launcher that...

Need to update the nipype2pydra package to produce the new syntax

Not only will this help when debugging crash dumps by providing a way to link the workflow directory with the node caches, we can also use this as a fallback...