You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
calculate_results takes x_airf_dist and z_airf_dist and reads neither #394
Since #393 forms alpha with inflow_angle(v, y_airf, z_airf), nothing in calculate_results reads its x_airf_dist and z_airf_dist positional arguments (src/body_aerodynamics.jl, the signature at function calculate_results(). The only in-repo callers are solve_base! in src/solver.jl and test/bench.jl, and both pass buffers that exist only for these two slots.
calculate_results is exported, so dropping the two arguments is a breaking change and was left out of #393. What I would do: remove both from the signature and the two callers, BREAKING: in the changelog, in whichever release next takes a breaking change. #94 (removing the results dict) rewrites the same function and is the natural place to fold it in.
No change needed here: #392 (open, mergeable, CI green) already does this. It removes calculate_results outright, with its export, docs entry and changelog line (BREAKING: solve and calculate_results are removed). It also drops both callers you named, the call in solve_base! and the calculate_results benchmark in test/bench.jl, and the _x_airf_dist buffer that only those two slots used. _z_airf_dist stays, and rightly so: on that branch gamma_loop! still reads it to form inflow_angle(v_rel, y_airf, z_airf). So #394 closes when #392 merges. Adding Closes #394 to the #392 description would link the two. I'm not opening a PR of my own: dropping the two arguments first would just make #392 rebase over a signature it deletes anyway.
Activity
1-Bort-1 commentedon Sep 30, 2026
Queued as
VortexStepMethod.jl-394.1-Bort-1 commentedon Sep 30, 2026
No change needed here: #392 (open, mergeable, CI green) already does this. It removes
calculate_resultsoutright, with its export, docs entry and changelog line (BREAKING: solve and calculate_results are removed). It also drops both callers you named, the call insolve_base!and thecalculate_resultsbenchmark intest/bench.jl, and the_x_airf_distbuffer that only those two slots used._z_airf_diststays, and rightly so: on that branchgamma_loop!still reads it to forminflow_angle(v_rel, y_airf, z_airf). So #394 closes when #392 merges. AddingCloses #394to the #392 description would link the two. I'm not opening a PR of my own: dropping the two arguments first would just make #392 rebase over a signature it deletes anyway.