Pre-processing stays out of scope. Work top to bottom: each step builds on the one before it, so no step rewrites code an earlier one wrote. One commit per step.
- #28: answered with 0.1.18 (tag, folded Inspect tab, pick kinds as checkboxes, face / edge selection and totals); close it once the reporter confirms.
- Extras (optional). Pin (freeze Inspect to compare two picks); back / forward through recent picks.
- Then decide on the other floating windows. Keep as centred dialogs: Units, Formula, Strength materials, Mesh quality, About, Messages. Keep floating: History, Path, Convergence, Integrals (a bottom dock could come later). Candidates for more pane tabs: Find, Contact settings, Legend settings, Title block.
- Known limits. A face's area and normal, an edge's length and the selection's
totals (volume, area, length) are taken on the undeformed mesh; the named
selections and the selection's CSV and deck lines hold elements and nodes, not
the faces and edges selected; on shells the normal is not oriented and the surface match uses
the deck's face number.
--findwith a restored.ccxviewends with nothing probed.
- All time steps to VTU for ParaView, on all cores (#43): a big feature, for later.
- Reinforcement: transverse shear (the sandwich's core) and compression steel are not designed; the facet method (Capra-Maury) of Code_Aster is not offered.
- The sandwich's lever arm is fixed at h - 2c; Annex F iterates the inner layer (about 19 % more steel in pure bending than Code_Aster SSLS134).
- Qx Qy at the mid-side nodes of quadratic shells come from two points through the thickness only.
- The ramp of a load inside a static step is not drawn (the end value is).
- LOAD CASE=2 (the imaginary part of a steady state dynamics load) is left out; a switch to show it instead of load case 1 could follow.
- Surfaces defined on shell elements use the shell's face numbers, which may not be the expanded solid's.
- The moment point (nodes:SET@x,y,z) is in model coordinates, not in the length shown.
- See-through STL layers are sorted as wholes, by their box centres: one layer inside another's box, or two interleaved, can still blend the wrong way (sorting triangles would fix it, at a cost per frame).
--field NAME:COMPwith an unknown component falls back to the default field and says so only in the message window; print it to stderr too, with the valid labels.
- Backends are in, none tested on a device: macOS (3DconnexionClient by dlopen), Linux (spacenavd's socket,
ndof.c), Windows (Raw Input); web has none (WebHID is blocked while 3DxWare runs). Buttons 1 / 2 = Fit / iso view. The camera motion (app_ndof_apply, rates scaled to the distance,ndof_on,ndof_speed) is ready and tested; Linux protocol tested against a fake daemon. Left: a hand test per platform with a real device (axis signs, roll direction, the 0.2 s release rule on Linux, whether Windows needs the same), newer devices' button-number reports on Windows, free flight (G), invert per axis / lock rotation or translation / dominant axis, button mapping options (Fit, Reset, named views), a View panel group.