self' is not per system? about flake-parts

[roberth]

self' goes through Nix's flake wiring whereas you could use config, which contains essentially the same information, unless you're doing something to the outputs outside of mkFlake (which you're not).

Maybe the module system induces a dependency that we don't expect. --show-trace may reveal that.

[srid]

@roberth I just got infinite recursion encountered when using a flake-parts module. It happens when evaluating self.callCabal2nix (note: this is a different 'self') from inside haskell overrides: srid/emanote#290 (comment)

(Full trace is in that link)

I'm trying to debug this. Let me know if you have any hints!

Formal bug: srid/haskell-flake#4

[roberth]

@srid Another cause could be the use of attrsOf. lazyAttrsOf is a safer default in flake modules, as they tend to be a bit more cyclical than your average NixOS configuration setting. My bad because I mentioned attrsOf in a review.

[srid]

Aha, laziness! lazyAttrsOf did the trick, thanks!

(It is as if type = is hacky way to declare types in Nix, and these types do change the behaviour of the expressions.).

[roberth]

That's how it seems, but types are also functions (from value to value) and monoid instances. Funky stuff ensues.

[roberth]

The original issue seems to have been resolved by removing uses of filterAttrs and attrsOf (as opposed to lazyAttrsOf) and/or defining more options.
If some infinite recursion still occurs, I'm going to need an example to debug it. That'd best be done in a new issue.