[tomneiser]

Problem

FUSE CI (test/runtests_study.jl) started failing after ProjectTorreyPines/FUSE.jl#1124 added ActorSOLPSNN.Psol with default=NaN:

Psol had different value:
act.ActorSOLPSNN.Psol ... value: NaN
act.ActorSOLPSNN.Psol ... value: NaN

diff() compares values through equals_with_missing, which fell through to a == b. Since NaN == NaN is false, two identical NaN values were reported as different and diff errored.

(FUSE side is independently fixed by ProjectTorreyPines/FUSE.jl#1138, which drops the NaN sentinel — this PR remains worthwhile as a robustness fix: parameter values can legitimately contain NaN, e.g. inside arrays.)

Fix

• equals_with_missing(a, b) = isequal(a, b) — isequal treats missing == missing and NaN == NaN as true (including inside arrays), and otherwise matches ==.
• Fix a copy-paste bug in diff(::AbstractParametersVector, ...): v2 was read from p1 instead of p2, so vector-parameter diffs silently compared p1 against itself.

Note on the src/io.jl call site

equals_with_missing is also used in par2ystr's skip_defaults sub-tree check. The scalar-parameter path below it already uses value === default, and for floats === treats NaN === NaN as true and -0.0 === 0.0 as false. The old ==-based check diverged from that on exactly those two cases (an untouched NaN-defaulted parameter was wrongly considered "not default" and written out); isequal matches === float semantics, so this change aligns the sub-tree check with the existing scalar check.

Testing

• Targeted check: NaN-defaulted Entry diffs cleanly, real differences still throw, missing/NaN-array semantics correct.
• Full Pkg.test() suite passes.

Version bump to be done after merge per repo convention.

🤖 Generated with Claude Code

[jmcclena]

Human @tomneiser, why does the test return act.ActorSOLPSNN.Psol ... value: NaN?

[tomneiser]

Thanks, before being fed any data I made the actor return missing rather than NaN here: ProjectTorreyPines/FUSE.jl#1138. The NaN was over-complicating things