Spin the stock.Read the signal.

A controller name alone does not establish compatibility with your machine. Confirm firmware, enabled axes, units, and feed mode on your setup. The posts in the middle column exist, and they do not yet have a machine-cut qualification published here. Prove them out before you trust them with a real job.

Stock GRBL covers flat work and engraving. Do not use the grbl post for rotary motion. A custom post drops in at ~/.rotatocam/posts/. Full table, evidence, and setup notes.

CE is not a crippled trial. It stays useful on its own: a suitable rotary part, from model to G-code, with the four core strategies and the full safety and setup workflow. The advanced strategies are where most of the late-night work went, so they live in Pro. If RotatoCAM saves you time, Pro is how you back it. If it does not, the free edition is yours to keep.

A rough-then-finish queue could be refused outright with a message about not being able to establish clearance for a rapid near the stock, even though each operation generated fine on its own. Nothing was going to hit anything. The pass that lowers a non-turning rapid to the local material radius plus your clearance had already checked the worst radius across the whole move, and the safety check that runs afterwards was judging that same move against the worst radius anywhere on the blank, which on a square billet is the swept half diagonal. The two could never agree, so the better the rougher cleared, the more of the finisher got stopped. A reported 6 mm flat plus 4 mm ball queue produced 473 such moves. It generates now, and it keeps the shorter retracts: on that job 1,692 of 1,897 finisher rapids stay down, about 15 m less vertical rapid travel, roughly two minutes at 7,500 mm/min.

When clearance cannot be established, the message carries the operation number, the tool, the operation name, the rapid's from and to in X Y Z A, and by how much it misses the stock surface. Three crash logs for the same refusal previously carried no coordinates at all.

The window reads Safe Z as a clearance above the stock surface. The command-line runner read the same saved number as an absolute height, so one job sheet meant two different retracts depending on how you ran it, and because the window saves the number you typed, its own sheets were the ones that disagreed. On a 32 by 32 billet, 8 meant 30.627 in the window and 8 from the command line, which is 14 mm inside the blank. One conversion now, used by both.

It was taking the model's largest radius, which on a square billet sits well inside the corners the stock actually sweeps, so a job run outside the window was checked against a datum up to 9.6 mm lower than the same job inside it. Round stock, bar diameter, a billet's swept half diagonal and the wrapped-relief touch-off are all honoured now, as they already were in the window.

It used to refuse with one tool per program. Posting the operations as separate jobs instead is the one thing you cannot recover from, because the finisher would then plan against a fresh blank rather than against what the rougher actually left. The queue now builds together and splits on the way out, one single-tool file per tool in cutting order, the same as the window has always done.

It is the sideways spacing of finishing passes, so it can never be zero, and in inch mode its default reads as 0.00079: a number you cannot clear, and with Stepover % doing the job nothing told you it was idle. It greys out now.

Every setting a strategy uses is on its page all the time. Spring passes stays conditional, because it is a shared row that only some strategies read and showing it elsewhere would be a control that changes nothing.

The display unit no longer resets between launches. Restoring your saved theme was writing the module default over your saved choice, because the theme restore saves your preferences and ran before the unit was put back.

A rough pass winds the rotary forward as it cuts, and the operation after it used to start by turning all of that back. On a reported two-operation job that was one rapid of 419 turns, minutes of the chuck spinning with Z parked at the retract and nothing else moving, which at the machine looks exactly like a dead Z axis. Whole turns now come out of a rapid before the program is written, in the same direction and leaving the part in the same place. Cutting moves are untouched. On that job, total rotary travel went from 839 turns to 287 and the longest single rotation from 419 turns to 356 degrees, with every cut identical.

Ramp, arc from side and helix did nothing at all on Indexed Surfacing and 3+1 Adaptive: inside a face the rotary is parked and Z is a real height, but the engine was treating it as a bar radius. They work now. Generation also reports how often a program drops straight down into an enclosed pocket, how deep, and how many of those drops go past the cutter's flute length.

Fixed a spindle and holder collision check that could exhaust system memory during toolpath generation, especially on indexed 3+1 jobs. The check keeps its existing sample spacing and collision margin while processing the work in smaller batches. Generation can also be cancelled between batches.

Settings, Spindle nose and collision checks now has a separate Check spindle collisions during generation switch. It defaults on. Turning it off keeps your measured dimensions saved and reports that spindle and holder clearance was not checked during generation.

Every radial and helical roughing level used to ride the finished surface across each column a shallower pass had already brought to size: touching, removing nothing, at feed, for the full length of the pass, on every deeper level. Those spans are skipped now. The cutter retracts, rapids across, and drops back in where there is material again. Measured against the material-removal sim on a stepped shaft, air fell from 47% of cutting distance to about 1%, and the program went from 5,764 cutting moves to 3,308. Same part and stock allowance. These are simulated path-distance and move-count results for that example, not a measured machine cycle-time reduction. Results depend on the job.

New in Stock setup for rotary jobs: Chuck guard. Give it a length and a core diameter, and cutting inside that zone at the chuck end is clamped to the core radius before the backplot, the sim, and the G-code ever see it. That reserves a holding core in the programmed geometry. Its strength and grip still depend on your material, tooling, and workholding. The sim shows the nub, the warning tells you how many moves it held back, and you part it off after the job.

Most rotaries are bolted down along X. Turn yours a quarter turn on the table and the Rotary runs along setting rotates the finished program to match: the part's length comes out on Y, its cross-section on X, and Z and the rotary word stay where they were. The setting applies a rotation rather than an axis swap. Confirm direction, zero, clearance, and the final NC on your setup before cutting.

Two named controllers join the list, both driven by customers running them. fluidnc: grbl-flavoured output with G93 rotary feeds, whatever rotary letter your config names, and Z-first rapids built in. RosettaCNC: Fanuc-flavoured RS-274 with G93 rotary feeds, and the post pins RTCP off (G49) because RotatoCAM output is already fully resolved machine motion. Check the controller software, feed mode, offsets, and clearance on your setup. Setup notes.

A post for the HC-204A handheld control, whose rotary runs along Y. Written from the controller manual rather than inferred, which matters here because its accepted command set is unusually small. It does do simultaneous four-axis: the manual has a rotary speed parameter that applies "in 4-axis linkage". Setup notes.

Its rotary is treated as a real fourth axis in plain degrees, not a substituted Y and not a circumference conversion. Absolute angles past a full turn are kept so a continuous spiral keeps counting up instead of resetting every revolution. Built from a live-machine handoff by a C4 Pro owner who read his own controller.

Pick taper, give it an included angle and a tip diameter (or a corner radius for a tapered ball nose), and a fine-tipped tapered bit plans as the shape it is. Scallop height is a geometric planning target for the selected cutter model. Inspect the simulated deviation. The actual finish also depends on tooling, material, and the machine.

Queue rough then finish and the second operation begins from what the first one really took off, so the finish cuts the leftover instead of skimming the whole part again. Use one tool per program and review every operation transition. Stock-based retract planning still needs a clearance check against your workholding and machine. Older custom posts may restrict operation chaining.

Settings, Language translates the safety gate and the getting-started guide, the two places where a misunderstanding is expensive, and Windows' own file dialogs follow along. Menus and parameter names stay in English on purpose.

Drop a post file into ~/.rotatocam/posts/ to add your own controller. No rebuild. See the controller table for operation scope and known limits. Genmitsu simultaneous rotary, AtomStack C4 Pro, HC-204A, Makera Carvera, MASSO, and RosettaCNC still want a prove-out on the exact machine.

The F word changes every line. Posts using G93 inverse-time express a target block duration: 1/F minutes. That lets the planner coordinate rotary and linear motion toward a surface-feed target. Some controllers want a different feed convention. Axis limits can lengthen the actual move.

Degrees are not millimetres. Rotary timing under G94 depends on the controller. Reusing a G93 inverse-time F value as a G94 feed can produce a very different duration. Match the post to the controller you actually have.

A720 is not a typo. A continuous spiral may count through A450, A720, and A1080. The controller and the rotary have to accept those turns. Resetting the angle on your own can make the axis reverse. Confirm wrap behavior and travel limits first.

One radius per ray. A wrapped pass stores a single radius for every ray out from the rotary axis, so it can only describe shapes that are star-shaped about that axis. RotatoCAM measures the model and tells you the percentage before you generate. A smaller tool will not recover a shape the wrap cannot describe.