6. G-code support

The dialect is controller.dialect in machine.json (section 4.2): DDCS V4.1 (the default), GRBL 1.1 or grblHAL. Section 6.1 is the syntax of each, section 6.2 the DDCS V4.1 code table, section 6.3 the GRBL 1.1 and grblHAL code table. The two families are handled in opposite ways. GRBL is precisely specified by its source, so the simulator rejects what GRBL rejects, with GRBL's error number (section 6.3; it was compared line by line with GRBL 1.1h built as grbl-sim, and the few known differences are listed there); the published page GRBL 1.1 and grblHAL in DryRun summarises it. For DDCS V4.1 the simulator follows the manual, which is short and partly contradictory; docs/research/03-ddcs-v41-dialect.md lists the sources. Where the manual is silent or contradicts itself, the simulator picks the most likely behaviour, executes it and attaches a warning ending in — behaviour on DDCS not confirmed. Unlike the simulator, the real DDCS V4.1 stops at a line it cannot parse ("parsing error prompt"); the simulator skips the line and continues so you see every problem in one pass.

6.1 Words and syntax

DDCS V4.1:

• Accepted words: G M X Y Z A I J K R F S T H D P Q L N O. Any other letter (B, C, U, V, W, E, …) is the error unsupported word "<L>". S and D are read and ignored.
• ( … ) comments, several per line. An unclosed ( or a stray ) is a parse error.
• ; comment to end of line, / block delete (the line is executed), lower-case letters, N numbers: accepted, each with one "not confirmed" warning per program (N sequence numbers: DDCS may reject them; consider disabling sequence numbers in the post).
• A line that is exactly % is ignored. After a second % line nothing is executed (warning lines after closing % ignored if anything non-empty follows).
• Decimal comma (X1,5) is the error comma is not a decimal separator.
• A line with #, [, ], = or a macro keyword (IF, GOTO, WHILE, SIN, …) is skipped with the warning macros and # variables are not simulated; line skipped.
• The same letter twice on a line (except G and M) is an error; two G codes of the same modal group on one line are an error.
• Axis, arc and radius values larger than 100 000 mm (or degrees) are the error number out of range (on GRBL and grblHAL with — GRBL accepts it and drives there; the simulator stops at 100 000 mm).
• Line endings LF, CRLF and CR are all accepted.
• At most 1 000 000 executed lines (subprogram repeats count); beyond that the run stops with execution limit reached.

GRBL 1.1 and grblHAL (research §1.3, gnea's protocol.c): everything below is documented GRBL syntax, so it gives no note unless one is named.

• Accepted words: G M F I J K L N P R S T X Y Z, and A on a machine with an A joint; grblHAL adds D H O Q. Any other letter rejects the line with error:20 (O1001, H1 and Q2 on GRBL 1.1; B, C, E, … on both).
• ( … ) and ; comments, lower case, packed words (G0X0.00Y0.00Z6.35), / (the line runs) and N numbers: accepted without a note. N above 10 000 000 is error:27. GRBL 1.1 ignores a / anywhere in the line, an unclosed ( comments out the rest of the line, a ) without ( is error:1.
• A % line gives one note per program, "%" is stripped by senders; GRBL itself rejects it (error:1), and does not end the program.
• A line starting with $ ($H, $J=…) is a system command: the warning $H: a GRBL system command, not simulated — senders run it outside the job, no motion.
• A letter without a number is error:2; a character that is not part of a word is error:1.
• GRBL 1.1 keeps at most 79 characters of a line, not counting whitespace, /, ( … ) and ; comments (gnea protocol.c, LINE_BUFFER_SIZE 80); a longer line is error:11, GRBL rejects the line: 80 characters without spaces and comments, GRBL keeps 79. grblHAL's 256-character buffer is not modelled.
• Axis words before any motion code run as G0, GRBL's power-on mode, without a warning.
• There is no program has no M30 warning: GRBL runs lines as they arrive.
• # variables, [ expressions: GRBL 1.1 has none and reads them as bad words: #1=5 is error:1, G0 X#1 and G0 X[1+2] error:2 (the message adds GRBL has no # variables or expressions); on grblHAL the line is skipped with macros and # variables are not simulated; line skipped, as on the DDCS.

6.2 Code table (DDCS V4.1)

"Executed" = simulated. "Ignored" = the code is dropped with a warning and the rest of the line runs. "Line skipped" = the whole line is not executed (warning). "Error" = the whole line is not executed (red).

Other behaviour:

• Axis words without a motion code before any G0/G1/G2/G3: moved as G0, "not confirmed".
• F (when ignoreProgramFeed is false) is modal and in the current units per minute; F ≤ 0 is an error.
• Program start: all axes at machine 0, the lowest-numbered tool in the spindle, no length offset, G54, G90, G17, G21. The controller starts wherever the machine stands, so the first move's path in the simulator is from machine zero. (A GRBL machine starts at its controller.home instead, section 4.2.)
• The listing keeps every line, including skipped ones; the toolpath simply has no motion for them.

6.3 GRBL 1.1 and grblHAL

GRBL is precisely specified, so instead of warnings the simulator rejects what GRBL rejects, with GRBL's error number: a red entry GRBL rejects G81 (error:20, unsupported command) (on grblHAL grblHAL rejects …), the line not executed. A block is checked completely before it runs, as GRBL does, so a rejected line changes no modal state. Sources: docs/research/05-grbl-dialects.md §1 and §2 (gnea/grbl 1.1h gcode.c, grblHAL core build 20261004); the published page GRBL 1.1 and grblHAL in DryRun has GRBL's full error table.

The DO NOT RUN rule. GRBL itself only discards the bad line and answers error:N, but Carbide Motion, gSender, UGS and CNCjs stop the job on that reply: on the machine the job ends part-way, with the spindle on and the tool possibly in the material. So any rejected line makes the verdict DO NOT RUN (section 2), with the count and the first one: 4 rejected lines — GRBL rejects line 2 (error:20); the sender stops the job there, and the Check tag Problems, even when nothing collides. The first rejected line of a program adds — the sender stops the job here; the simulation goes on as if you continued; the simulation does continue, so one pass shows every problem. Rejected lines are tagged Not executed and do not count in the INCOMPLETE banner's N lines were not executed (section 7.11). Whether your sender really stops is not confirmed (section 7.12, G1).

Code table. "Executed" = simulated. "No effect" = accepted and executed without motion or state the check uses. "Line skipped" = a warning, the line not executed. "Error n" = rejected with GRBL's error n (the hint, if any, is part of the message).

Other rejections, by number: error:21 two codes of one modal group (G0 G80, M8 M9, M3 M5, G4 G53, G90 G91); 24 two codes that use axis words (G0 G1 X1, G28 G1 X1, G92 G0 X1; GRBL checks this before the modal groups, and on GRBL 1.1 before it refuses a G43, G38.x or Gxx.x that follows a motion code on the line: G0 G43 Z15 H1 is 24, G43 Z15 H1 alone 20); 25 a letter twice (X1 X2); 4 a negative F, N, P, S or T; 38 T > 255 (grblHAL with a tool table: a T, H or M61 Q that is not a tool in setup.json); 27 N > 10 000 000; 36 unused value words (G0 X1 R2, G1 X1 P1, a lone P5); the motion rules below (26, 29, 30, 32–35, 37); 2 a letter without a number (G1 X); 1 a stray character (G0 X1 ,); 11 a line over 79 characters (section 6.1). grblHAL adds 19 (non-positive value) and 54 (retract position below the hole bottom) for canned-cycle words.

Known differences from GRBL 1.1h (a line-by-line comparison with GRBL 1.1h built as grbl-sim, tools/grbl-oracle/): values over 100 000 mm (GRBL accepts them and drives there; the simulator refuses the line with number out of range: X200000 — GRBL accepts it and drives there; the simulator stops at 100 000 mm); M2/M30 (GRBL goes on with lines the sender still streams; the simulation stops); % and $ lines (a note and a warning, section 6.1).

• Motion (as gnea's gcode.c executes it):
  • Arcs. IJK is always incremental from the start. GRBL's radius check: the arc is refused (error:33, GRBL rejects the arc: radius differs by 0.006 mm at the end) when the end radius differs from the start radius by more than 0.005 mm and by more than 0.5 mm or 0.1 % of the radius. So r 5 accepts 0.004 and refuses 0.006; r 10 accepts 0.009; r 1000 accepts 0.4 and refuses 0.6. The check runs in single-precision floats, as on GRBL: G2 X100.05 I50 is 0.0500031 off and refused. A zero-radius arc (G2 X0 Y0 I0 J0) passes and is a straight feed to its end, as GRBL's mc_arc does. DDCS keeps its 0.002 mm. An R arc to its start point is error:33 (full circles need IJK, with an axis word: G2 X0 I5), and R below half the chord is error:34 with no tolerance. G2 I5 (no axis word) is error:26, no axis word in the plane (G17 G2 Z5 I1) error:32, no offset in the plane (G17 G2 X1 Y1 K1) error:35, IJK next to R error:36. Helical arcs and R < 0 are documented by GRBL: no note.
  • G28 / G30 go to the stored g28 / g30 (machine coordinates, section 4.2). Without axis words every axis rapids straight there. With axis words GRBL first rapids to that point (work offset, G92, tool length and G90/G91 applied), then only the named axes rapid to the stored position: G91 G28 Z0 lifts Z to the stored Z, G28 G91 X0 Y0 moves X and Y. G28.1 / G30.1 store where the machine stands for the rest of the run, with the note G28.1 changes the stored position on the controller (it persists).
  • G10 L2 Pn / G10 L20 Pn (P0 = the active system, P1–P6 = G54–G59) change the run's copy of that offset for the named axes: L2 to the given machine coordinates, L20 so the current position reads the given values. One warning per program: G10 changes the stored work offset on the controller (it persists after the job); the simulator uses it for this run only; an L20 also warns G10 L20 sets the offset from where the machine stands; the simulator starts at controller.home (X -5.000 Y -5.000 Z -5.000) (the machine's home, section 4.2). Refused: no axis words (26), neither P nor L (28; a missing P alone is P0), P > 6 (29), an L other than 2 or 20, or L2 with R (20).
  • G92 as on the DDCS (needs axis words, else error:26); G92.1 clears it. M2/M30 do not clear it.
  • G43.1 Z sets a dynamic tool length offset (added to the touch-off length, section 4.2); G49 cancels it. G43.1 with any axis word other than exactly Z, or none, is error:37.
  • G53 runs with G0 or G1 (with G1 at the feed rate, no note); under G2/G3/G80 it is error:30.
  • M2 / M30 end the run, M2 without a note. GRBL resets the motion mode (G1), plane, G90, G94 and G54 there; the simulation stops at that line, so this does not show.
• Tools. GRBL only stores T. With toolChange: "touch-off" (the GRBL default) every tool is touched off on the work zero, so the Project panel says touched off where DDCS says G43 H (sections 4.5.3, 4.5.6). A T line that names a tool other than the one in the spindle, with no M6 on the line, gives once per tool T2 without M6: GRBL and the senders do not stop here — the machine keeps cutting with T1 (the Fusion "Grbl" post writes no M6 unless "Use M6" is on), and the tool stays. A program without any M6 gets no tool change in program — GRBL ignores T; T1 from the setup is in the spindle at its first move, tagged Note (task055: one tool per file is the normal GRBL case; on the DDCS the same situation stays the warning no tool change in program — T1 from setup.json assumed). With toolChange: "touch-off" the first M6 gives the note M6: simulated as a manual change with Z touched off again, as Carbide Motion, CNCjs, UGS and grblHAL do; gSender's default (Tool Changing: Ignore) skips it and keeps the old tool; GRBL itself rejects M6 (error:20).
• grblHAL (dialect: "grblhal") runs everything above, with grblHAL's error texts (grblHAL rejects …), and adds (research §2, grblHAL core gcode.c):
  • Canned cycles G73, G81, G82, G83, G85, G86, G89 with G98/G99 and G80, as grblHAL's mc_canned_drill drills: below R the tool first rises to R, then rapids over the hole; G81 feeds to Z; G82 dwells P seconds there; G83 pecks Q and goes back to R between pecks, G73 pecks with a 1 mm chip-break retract (one note per program: G73 retract read as 1 mm; grblHAL uses $28); G85 feeds in and out; G86 feeds in, dwells P, rapids out (the spindle stops on the way up); G89 feeds in, dwells P, feeds out. The way out goes to R under G99, under G98 to the Z the hole started from; G80 puts the retract mode back to G98. L repeats a hole; R, Z, P, Q stay for the following holes. No DDCS-style notes. Refused: a cycle start without R or Z, a G82/G86/G89 start without P, a G73/G83 start without Q (error:28); Q ≤ 0 or a negative P (4); L ≤ 0 (19, non-positive value); L not a whole number (23); an A word in a cycle (24); R below Z (54, retract position below the hole bottom); K (36). Cycles in G91 or outside G17 are skipped with a warning, as on the DDCS.
  • Tools: M6 runs (with toolChange as above, the note naming grblHAL's tool change $341); M61 Q sets the tool in the spindle without a change (Q missing: error:28); G43 H with toolTable: true (section 4.2), and G43.2 H, which adds tool H's length to the active offset (H missing: 28; H0 or not a tool: 38). T has no 255 limit. G49 is not an axis command (only G43.1 is): G49 X5 cancels and moves to X5, G0 G49 X1 runs, G49 G43.1 Z1 is error:21. G43.1 takes any axis word: Z sets the length, X/Y give length offset on X/Y not simulated; with no axis word error:37.
  • Arcs take P, the number of turns: G2 X10 I5 P2 adds one full circle; P not a whole number is error:23, P0 error:39 (value out of range).
  • Accepted: M7, M62–M68 (no effect); D, H, O, Q as words. M70–M73 give M70: modal state save/restore is not simulated. G10 L2/L20 take P up to 9; P7–P9 (G59.1–G59.3) skip the line with a warning. Axis words of G10/G28/G30/G92 under G80 are allowed (GRBL 1.1: error:31).
  • Line skipped (warning, not simulated; line skipped, positions after this line may differ from the controller): G5, G5.1 (splines), G50/G51 (scaling), G59.1–G59.3. # parameter lines and O-word flow control: macros and # variables are not simulated; line skipped, as on the DDCS.
  • Refused (error:20): G84 (G84 needs a spindle that reports "at speed"), G87, G88, G64, G61.1, G90.1, G41/G42, lathe codes (G7, G8, G33, G76, G95–G97), G59.x other than G59.1–G59.3.
  • Not simulated: grblHAL's 0.25 s dwell at each G73/G83 peck bottom, and its G83 way back in (grblHAL retracts to R, or under G98 to the start Z, and feeds down from there; the simulator retracts to R and rapids to 1 mm above the last depth). Hole positions and depths are the same; peck-drilling times are short.
• Guessing GRBL (quick view and a new project, task055). src/gcode/sniff.ts reads a program's header and says "GRBL" when, in this order: a comment of the first 40 lines starts with Design File: or stockMin: (the header is Carbide Create's); a comment of the first 40 lines names grbl, Shapeoko, X-Carve, LongMill, gSender, OpenBuilds, Easel or Carbide 3D / Carbide Create / Carbide Motion, in any case (the header names "<word>"; plain "carbide", a tool material, does not count); a line of the first 40 has ; set units to MM (required) (the header is Kiri:Moto's GRBL device); or the first 40 lines have G28 G91 Z0 and the whole program has no G43, no O and no N word (the program looks like Fusion's GRBL post (G28 G91 Z0 and no G43)). Anything else is not guessed, including Easel's and VCarve's comment-less headers, the older Carbide Create header and Sienci's grblHAL post: they run on the DDCS demo's setup until you choose a machine. Only quick view (section 2) and a new project's Machine item (section 4.5.8) use the guess, and both say it is one; a project's machine.json always decides.