Brian Groenke

> The expression you timed does not touch the file system. Right, that makes sense, but surely it has to query the file system at least to build the `FileTree`...

Ok, I see. What about the `values` associated with each node? Would they also be transferred for every `Thunk`? Edit: > When loading there is ofc not alot of data...

Any updates on this?

Is it type unstable in all cases or just in specific cases?

I am a bit confused as to the relationship between this PR and `getall` to what is implemented in #23. Could someone clarify this?

But you are able to make it type stable for up to 4 levels of nesting currently? Could this be extended further? There probably is a reasonable limit (maybe

You may want to look at how this problem was solved in Flatten.jl `reconstruct`. I used generated functions to recursively construct a double-linked tree data structure that allowed `reconstruct` to...

I also got this (resolved by specifying `FullSpecialize`) and the same error from #370.

> In general one has to condition on the existing values which is not always as trivial as for a multivariate normal distribution with iid components. Sure, but this is...

`CVODE_BDF` has `jac_upper` and `jac_lower` fields which can be used to take advantage of the tridiagonal structure. I have not tried `FBDF` on this MWE, but on my actual nonlinear...