Evaluation time is exponential w.r.t. the depth of the dependency graph

[BurNiinTRee]

In particular the size of the nickel-res.json derivation. This gist contains code for the following dependency graph with variable depth, both created with native nix and organist. Evaluation of former seems to have the expected linear complexity, while for the one created with organist, it has exponential complexity.

This seems inevitable when we're passing the dependency graph from nickel to nix by serializing it to a tree with nickel export.

[thufschmitt]

That's indeed a limiting factor for scalability. There's some work on the Nickel side to add some experimental support for directly calling Nix from Nickel (though it's not good-enough for our use-case yet), that plus RFC 92 could hopefully sidestep this problem altogether eventually