I tried to make it so that Metaflow-R would serialise everything that gets stored to

self

, rather than letting objects be converted to Python analogues and corrupted in the process. It all works fine until...

foreach

. Metaflow-Python needs to be able to see the object it's splitting on, and all Python sees is the serialised bytes. I can think of some hacky solutions but I don't like them. It'd be the same issue with Metaflow-Julia (which doesn't exist yet but I hope it will one day!) https://github.com/Netflix/metaflow/pull/619

@calm-motorcycle-38345 Quick question re:

1. if an argument is given to "foreach", save it to a non-serialized dummy variable and use that instead, or
2. if an argument is given to "foreach", save it to the same variable but don't serialise it.

For both

1.

and

2.

how would you determine when to apply the proposed behaviour to the argument? I can define a variable in the start step and use it as an argument for

foreach

in a subsequent step - which may complicate the implementation. How about storing all the variables as singletons or lists of raw bytes?

Oops, you’re right. Scratch those. We might be able to serialise in such a way that Python sees one object for each R object. Let me think on this.