[jbgi]

#29 was merged following a regression I noticed, but I now think it was due to me wrongly the api, hence the revert.

Indeed the idiom of returning another input as the outputs
nixpkgs = with-inputs.from.flake inputs.home-manager { } (i: i.nixpkgs) is much better written:
inherit (with-inputs.from.flake inputs.home-manager {}) nixpkgs;

So reverting #29 in order to keep the same override logic inside and outside the "outputs" function, which is much cleaner imo.

Added doc to document this.

[sini]

LGTM. Using the same override logic inside and outside the outputs function is cleaner, and the inherit (… from.flake inputs.home-manager {}) nixpkgs idiom covers the #26 case. topLevelInputs and nonEmptyInputs were indeed unused. just ci passes (51/51), and so does this branch merged with #32 and #33 (52/52).

Nits:

• Typo inline.
• Worth a test for the documented inherit idiom, so the #26 case stays covered now that #29's reproducer is gone.
• The removed "Flake backed by non-flake pins" section was inaccurate, but the mechanism it was describing still exists: indirect, built from builtins.functionArgs flake.outputs. Maybe give it a one-line mention so it isn't undocumented.

[jbgi]

• Typo inline.

fixed.

• Worth a test for the documented inherit idiom, so the fix: restore npins/niv pin metadata on flake inputs, and expose npins' revision as rev #26 case stays covered now that fix: let outputs keep their outPath in the value with-inputs returns #29's reproducer is gone.

I'll add the test in #33.

• The removed "Flake backed by non-flake pins" section was inaccurate, but the mechanism it was describing still exists: indirect, built from builtins.functionArgs flake.outputs. Maybe give it a one-line mention so it isn't undocumented.

the AI got confused here, the obsolete removed section had nothing to do with builtins.functionArgs flake.outputs...