Skip to content

YACC1 — machine state as of 2026-09-20

What is physically in the machine and what is loaded into it, so the tree can be read against the hardware. Sources: hardware/FABRICATED.md (per-card confirmation), the bring-up session of 2026-09-18/19 with the bus tester, the confirmations at the machine of 2026-09-19/20 (the open questions were answered on 2026-09-20); both ATmega firmwares were settled 2026-09-21 by reading the flash out (bus tester = 2020 bus-driver, since re-flashed; sequencer = Sequencer3). only a rebuild-and-reflash from the tree can settle.

Cards on the bus

Card Revision in the machine Design Notes
Backplane V2.0 (2021-08) hardware/bus/backplane/eagle/v2.0 8 slots; bus signal naming = Bus Template V3.2
Bus jumpers none fitted hardware/bus/bus-jumper-* horizontal V3.2 and vertical V3.0 were built for an older bus arrangement and are obsolete (2026-09-20)
Sequencer logic v2.1 (2020-12) hardware/cards/sequencer-logic/eagle/v2.1 the 218 mm "V2.1l" board
Sequencer memory V2.1 (2020-12) + EEPROM adaptor hardware/cards/sequencer-memory/eagle/v2.1, accessories/eeprom-adaptor adaptor fitted in IC9, carries two 24Cxx at I2C 6 and 7 (confirmed 2026-09-20)
ALU V3.2 (2020-11) hardware/cards/alu/eagle/v3.2
Index registers 1.1 (2020-08) hardware/cards/register/eagle/v1.1 two cards: R0–R3 and R4–R7, selected by ADDR-REG-ID2..3 via each card's J3 (confirmed 2026-09-20). Only card 0 was in the bring-up build until 2026-09-22 evening: tests/assembler/romdiag read R7 as $FF, and the monitor ROM needs both (R7 = every string pointer); card 1 fitted 2026-09-22 evening: tests/assembler/romcount (R3 + TMP build) then ran overnight into 2026-09-23 on the reloaded microcode, counting on the LEDs/TIL311s without a fault (seen at the machine)
I/O V1.1 (2020-11) hardware/cards/io/eagle/v1.1 UART behind P0/P1 at 38400, switch/LED port, TIL311s
Memory v1.3 (boards ordered 2025-06-27, JLCPCB 2000765A); out of the machine since 2026-11-09 (replaced by v2.0) hardware/cards/memory/eagle/v1.3 (+ KiCad kicad/v1.3) see jumpers below. The built design is the Fusion 360 export of 2026-09-24 (proven against the order's gerbers); eagle/deprecated/v1.3-do-not-use is an earlier save without IC15 (the 74245 enable) and with the TMP registers off the board 2026-10-09: IC5 (74LS245, the data-bus buffer) replaced while chasing BASIC's RUN failures (BACKLOG.md).
Memory v2.0 v2.0 (ordered 2026-10-09) hardware/cards/memory/kicad/v2.0 2026-11-09: the memory side built and in the memory slot in place of the v1.3 card, R15 (470 Ω, the ADDR15 pull-down) fitted, JP3 on 2-3 (ROM write-protected). The same evening the CF section was back in use on the finished card: O boots Y1/OS from the CF through the card's own memory side (the last step of the review's "CF first" plan). The ROM chip (ROM 2026-10-21) moved across from the v1.3 card. Bus tester, no CPU cards (tests/memory/full-run-2026-11-09-2211-v2.0.log): memory_status.py ALL OK (boot remap, the ROM identical to ROM 2026-10-21, low RAM, the block map $8000-$CFFF RAM / $D000 undecoded / $E000-$FFFF ROM); tmp_test.py ALL PASS; memory_full_test.py address lines and RAM $0000-$CFFF both patterns 0 bad (its ROM lines compared the unburned tree image ROM 2026-10-22 and differ only where 10-21 and 10-22 differ; its video lines fail with no video RAM on the bus). The R15 check: with the address bus floating (it reads $7FFF: A15 held low against the tester's 10k pull-ups), 20 -VMA pulses leave FORCE-ROM set ($0000 still reads the ROM's $F000 page) and a read of $8000 releases it; review M1 is fixed in the machine. With the CPU cards back in (2026-11-09): the banner after a reset with the 1 MHz clock running (no single-step workaround), and tests/bench/run.py 15 of 15 PASS (tests/bench/logs/bench-2026-11-09-2236.log), as with the v1.3 card. JP3: inverted bytes written to $E000, $E001, $F000 and $FFFF leave the ROM identical. 2026-11-03 before it: the card was built with the CF section only (IC30-IC34, RN9, J2) beside the v1.3 card, a CF-to-IDE adapter on J2 set to master (on slave every register reads $FF: docs/procedures/CF-CARD.md); tests/memory/cf_test.py ALL PASS, the disk image written with tools/cfwrite_bus.py, Y1/OS booted on the machine from it (2026-11-03).
Video V1.0 (Fusion, export 2026-09-18); the card to build next is kicad/v4.0 (the 6845 on VGA: designed and routed, not ordered, docs/cards/video-v4.md; the V2.0 design named here until 2026-11-18 was removed that day, never built); the V1 line's KiCad designs (kicad/v1.3 = the DOTS jumper - 6 or 8 dots a character - on a 4-layer board; kicad/v1.2 = v1.1 + the bench fixes; v1.1 from 2026-09-21) hardware/cards/video/eagle/v1.0-fusion-export-2026-09-18 installed; a picture with the CPU out on 2026-10-21 (80 x 24 on a TV-rate monitor with an 11.0592 MHz crystal, bus tester); MC6845 fitted 2026-10-21; two bent pins straightened 2026-09-18; RN2 = 10k as designed (was 1k from the 2026-09-18 tests until 2026-09-21); +5V and VCC joined by a wire (2026-09-21) — the design left the +5V rail unfed
Bus tester v1.1 (2016 board, 2020 rework) hardware/cards/bus-tester/eagle/v1.1 plugged in for bring-up sessions, FTDI 19200
Mem Switch / Mem Register re-fitted 2026-09-21 for CPU bring-up hardware/cards/mem-switch, mem-register the bring-up cards (switch ROM at $0000, 16-byte RAM at $0010); removed once the memory card worked (2026-09-20), back in 2026-09-21 with the ALU and one index card: the CPU executes the 16-byte switch-ROM program as expected on a function-generator TTL clock in the logic card's oscillator socket (pin 8 clock, pin 7 GND), with the tree's microcode in the sequencer RAM (seen at the machine 2026-09-21)

Not in the machine: the two bring-up cards (Mem Switch, Mem Register), the two bus jumper boards (obsolete), Address+TMP (retired 2021), all deprecated/ revisions, Bus Tester V3.1/V3.11 (never built), Blank V3.2 (design only).

Memory card v1.3 settings (verified with the bus tester 2026-09-18)

  • Low RAM $0000–$7FFF; high half decoded per 4K block by IC7 into the 3x8 jumper header: jumper up = high RAM, down = ROM, none = undecoded.
  • Fitted: $8000–$CFFF → RAM (5 jumpers up), $D000–$DFFF → no jumper (undecoded, reserved for the video card), $E000–$FFFF → ROM (2 jumpers down).
  • FORCE-ROM boot remap: after -RESET the ROM appears at every address until the first access with ADDR15 high.
  • The 470 Ω pull-down on ADDR15 (the FORCE-ROM race fix, review M1; docs/cards/memory.md 4.1) was never fitted on this card: the v1.3 card left the machine on 2026-11-09, and the v2.0 card that replaced it has it as R15.
  • Reading an undecoded block returns the last value left on the bus (looks like RAM that echoes the last write).
  • A jumper wire from IC7 pin 4 is present but unconnected; purpose not remembered (2026-09-20) — see BACKLOG.

What is loaded

Store Contents Proof
Memory card 28C64 EEPROM (an Atmel AT28C64B, from its marking 2026-10-07) 2026-10-22: the tree's rom.bin is ROM 2026-10-22 (the video geometry read from the card's latch at $D801: 80 x 24 on the card in the machine, 80 or 64 columns on a V1.3; MD5 ffee916318157f829a75cba09e6b2139) — NOT BURNED; before burning it, fit a jumper on the video card's JP1 pins 3-4 (latch bit 2 held high = the V1.0-V1.2 card). 2026-10-21: the tree's rom.bin is ROM 2026-10-21 (the video unit's CRTC table for the 11.0592 MHz crystal, 80 x 24; MD5 68b584fc04885ac6abbfc26a96548696) — burned 2026-10-21 on the T48 (minipro -p AT28C64B -u -P -w, verify OK, read back and cmp identical); the chip in the machine. 2026-10-09: ROM 2026-09-23 on both AT28C64B chips (one fitted) with the 2026-09-23 microcode: the pair that worked on 2026-09-23 - BASIC hangs on its first line, so the machine has a hardware fault (BACKLOG.md). Before: a second AT28C64B with ROM 2026-10-07 (MD5 122f1a93...); the first chip was powered reversed 2026-10-07. Before: ROM 2026-10-07, burned 2026-10-07, re-burned 2026-10-08 with the 2026-10-08 layout (MD5 122f1a937f23db61c2b53893613375e6, the first build 3a4ff07923bff1cf752a5f67c9df3526; burned 2026-10-07 with the TL866II+ (minipro -p AT28C64B -u -P -w: erase, write, verify OK, software data protection on) and read back identical; the chip was once fitted reversed and powered that day - afterwards its pin test passed, a second burn verified and the read-back was identical, so it is in service; on the machine: the monitor's help with V, BASIC takes lower case, backspace and junk (firmware/basic/README.md), tests/bench hello/brur/isa/arith PASS, bench-2026-10-07-2115.log). Before it: the 2026-09-23 build, burned 2026-09-23 (Visual Minipro, firmware/rom/shipped/rom.bin, MD5 d2d7b027e7c6951d7dd93412a8fd9cd8): BASIC unchanged; the monitor with G = JSRUR R7, the CompactFlash driver and O, the : Intel-hex loader, vectors CFINIT/CFREAD/CFWRITE/CONST/UARTINNE at $FFEC..$FFFC; banner YACC 2020: HELLO WORLD ROM 2026-09-23 first boot on the machine 2026-09-23: banner, register dump and the full help menu (with O and :) over the UART at 38400 (screen /dev/cu.usbserial-AB0MVHSQ 38400); then I used the monitor commands and ROM BASIC interactively, all working (2026-09-23). The ROM tool chain (assembler -> image -> Visual Minipro -> 28C64 -> boot) is proven. tests/bench on the machine, 2026-09-23 evening: 14 of 14 PASS (tests/bench/logs/bench-2026-09-23-1855.log): the : loader, BRUR, the full ISA sweep and eleven compiled C programs, byte-identical to the microcode emulator, after the SHIFT-OUT carry fix (docs/cards/alu.md 3.3). The first run (bench-2026-09-23-1809.log) found that fault: hello, brur and isa passed, arith hung. The 2021 chip image is kept as firmware/rom/eprom-captured-2026-09-18.*. The tree's shipped/rom.bin was this image until 2026-09-25 (a ROM 2026-09-23B rebuild with the CF on P4/P5 was withdrawn unburned on 2026-09-24). 2026-09-25: the tree's image is ROM 2026-09-25 (the video unit: $D000 probe, screen driver, V command, mirroring off by default; MD5 3ebc67898f70d319853fe42abdfd2cb9) — NOT BURNED; 2026-10-07: ROM 2026-10-07, the same monitor + BASIC's line input fixed (backspace, lower case, SYNTAX ERROR; MD5 122f1a937f23db61c2b53893613375e6 since the 2026-10-08 layout) — the chip in the machine; the chip keeps working with everything in the tree meanwhile
Sequencer microcode (RAM/EEPROM on the sequencer-memory card) 2026-11-06: the tree's firmware/microcode/ucode-generator2/test.hex (the hold-step pass, docs/system/MICROCODE.md 5.8) is NOT loaded; the card holds the image before it (make keephold, e48d869). Below, the history of what was loaded: the 2026-09-22 image (BRUR at $AD, the H-2 fix in BRZ/BRNZ/BR16Z/BR16NZ, the H-1 fix in PUSHR: six records differed from the 2026-09-21 image; all 256 sent with tools/ucode_send.py --all on 2026-09-22, the card verified RAM == EEPROM and its five boot dumps equal test.hex, tests/sequencer/boot-run-mode-2026-09-22-reload.log) the loader's cache = what was last sent. 2026-10-10 evening: POPR ($08) fixed and reloaded (it read [SP] while the sequencer drove no ADDR-REG-ID; now through TMP1; --boot-check 0 mismatches, tests/sequencer/boot-run-mode-2026-10-10-popr.log). Before: 2026-10-10: the tree's test.hex (stage 2) loaded again (--all through the sequencer's own FTDI AL00FSLF; --boot-check: RAM == EEPROM, 320 dump lines, 0 mismatches). 2026-10-09: the 2026-09-23 image loaded back (the cache at 9000a22; boot check 0 mismatches) for the baseline test (BACKLOG.md). 2026-10-08 evening: stage 1 for a test. Before: 2026-10-04: the tree's test.hex is in the EEPROM (loaded in two stages, item 7 below): LDZ/STZ/ADDIW/SHL16, the HALT records for the undefined opcodes, the three-step fetch prologue, M-1 and the idle steps removed; --boot-check: RAM == EEPROM, 320 dump lines, 0 mismatches; tests/bench 15/15 (tests/bench/logs/bench-2026-10-04-1719.log)
Sequencer-memory ATmega328P embedded/sequencer-card/sequencer4 flashed 2026-09-21 (400 kHz I2C, block EEPROM reads, page writes, the copy read back before READY) run-mode boot: copy 16 s, verify 29 s, READY at 54 s, RAM == EEPROM for all 256 instructions (tests/sequencer/boot-run-mode-2026-09-21-sequencer4.log). Sequencer3 (what it ran before, 156 s unverified) is in deprecated/sequencer3; the pre-2026 flash is in readback/
Bus tester ATmega328P 2026-11-03: embedded/bus-tester/bus-monitor (listen only, every pin an input, 19200) - the card stays in with the CPU running; flash bus-driver (blocks-2, the I/O block commands) back for the test scripts. Before: embedded/bus-tester/bus-driver blocks-1 (2026-09-21), flashed by arduino-cli as an Uno at 115200 through the FTDI Before the flash the chip held the 2020 bus-driver (same strings and signal table, older toolchain build; read out to embedded/bus-tester/readback/), so the MCP23X17 port was never loaded. Banner bus-driver blocks-1 2026-09-21
Video card 2026-11-09: back in the machine with the memory card v2.0; working. IC28 was found missing (no picture, "no signal") and refitted with a 74x163 (a 160/161 divides by 5: the wrong line rate); from the CPU vram 0 errors with the 6845 stopped and running, crtcreg all OK, the test card on the screen (BACKLOG.md). MC6845 fitted 2026-10-21 (RS on A1, LPSTB to GND; register interface passes through the bus tester); Q2 = 11.0592 MHz, fitted 2026-10-21, SV4 on 1-2 (a 1.8432 MHz character clock: 15.75 kHz / 60.1 Hz, 80 x 24; a 9 MHz crystal before it the same day fitted only 64 columns); IC1 pin 2 (N$4, the one-shot) bent out of the socket and wired to IC1 pin 14 (VCC), 2026-10-21: E = the strobe XOR (IC26/C), high for the whole of every memory read and write (finding 6.1); IC2 is a 74LS373 since 2026-10-21 (a 74LS374 was fitted); 4.7k pull-ups fitted on the bench 2026-10-21: IC2 pin 1 to pin 20 (N$16, the latch's -OC) and IC27 pin 8 to pin 14 (N$5, /A0); IC28 is a 74163 since 2026-10-18 (was the drawn 74HC160): a synchronous clear, so the character clock is the dot clock ÷ 6 with a clean one-dot CHARCLOCK (finding 6.2 gone) and a blank 6th dot between characters - the part the ELF's 6845 card was built with. IC25 holds the remapped character EPROM (burned 2026-10-20; the original was the ELF's gr2 font, letters beyond this card's reach: firmware/video/README.md) the whole 2K of dual-port RAM, crtc_regs.py and crtc_picture.py pass through the bus tester (2026-10-21); Phase B with the CPU passed 2026-10-21: vram, crtcreg, testcard and vstress at 1 MHz, the ROM's VI/VW/VM1 (the console mirrored on screen). ROM support written 2026-09-25 (in ROM 2026-09-25, not burned): probe at reset, the monitor's V command for bring-up by hand (docs/cards/video.md section 8), mirroring off until the card is debugged

Plan for the disk operating system, CF card and the port/memory maps: OS-PLAN.md (2026-09-22).

Software that talks to it (Mac)

  • tools/busdrv.py — drives the bus tester (CMD:OPERAND#, >> prompt); tools/alias_min.py reproduces the (now resolved) video-card fault; tests/video/video_ram_test.py is the video RAM test; tests/memory/memory_status.py the memory-card status check (boot remap, ROM image, RAM, block map).
  • embedded/command-sender/command_sender_8 (Processing) — sends tests/bus-tester-scripts/.
  • embedded/sequencer-card/microcode-loader/simple_microcode_sender_64 (Processing, 115200) — loads test.hexz.
  • software/assembler (asm file -d=yacc1), software/emulator, software/ubasic-c, software/disassembler/disasm2 (needs its include restored).
  • os/ (2026-09-22) — Y1/OS v0: make -C os run boots it on the microcode emulator from os/disk.img (P8XFS v2, tools/p8xfs.py); both emulators take -c disk.img for the CF interface on ports P8/P9. The CF hardware is the memory card v2.0's CF section (the CF card v1.0 circuit, moved onto the memory card 2026-09-24, docs/cards/cf.md); Y1/OS boots on the machine from it since 2026-11-03. Y1/OS ends lines with a bare LF: tools/y1term.py shows it as CR LF (BACKLOG.md).
  • software/ucemu/y1ucemu (2026-09-22) — the microcode-level emulator (steps test.hex through a model of the cards); the monitor boots on it from reset and runs compiled programs via G3000.
  • software/compiler/y1cc.py (2026-09-22) — the C cross-compiler: prog.c → YACC1 assembly → .img; images load at $3000 (BASIC's token buffer is $1000-$1FFF); with the rebuilt monitor G3000 calls main and RET comes back to the prompt (--vector for the 2021 chip). Not yet run on the machine: loading RAM needs the E-command loader (BACKLOG).

Known faults / open on the hardware

  1. Video card: 6845 data register unreachable as drawn (-CS includes /A0 while RS = A0); bench fix = RS to A1, see hardware/cards/video/docs/fix-6845-register-select.md.
  2. Video card: 7416 open-collector outputs with no pull-ups - RESOLVED on the bench 2026-10-21 (4.7k on N$16 and N$5, 10k on N$3): without the N$16 pull-up the JP1 latch drove the data bus on every memory read in the machine.
  3. Video card: the block-0/9 write-through fault of 2026-09-18 is RESOLVED (2026-09-21): its cause was the unpowered +5V rail (IC1, IC2, RN2 and all decoupling on a net with no source); joined to VCC by a wire, 1K RAM test passes 8/8.
  4. Blank V3.1 template (and the cards drawn on it) label C3–C6 with the pre-V3.2 names; harmless, documented.
  5. RESOLVED 2026-09-23: the monitor's G command (BRVR R7 on the 2021 chip, so G AAAA never ran the code). The 2026 ROM with JSRUR R7 + BR cmdloop is burned and boots; y1cc.py --vector is only needed for a 2021 chip.
  6. RESOLVED 2026-09-22/23: the sequencer EEPROM holds the tree's microcode (BRUR $AD, the H-1/H-2 bus-fight fixes, and the ALU SHIFT-OUT carry fix), loaded and verified with tools/ucode_send.py --all; tests/bench passes 14/14 on the machine (tests/bench/logs/bench-2026-09-23-1855.log).
  7. RESOLVED 2026-10-04 (loaded in two stages; stage 1 = the old prologue + the new instructions: 14/15, xisa lost to M-1 at $3Fxx, tests/bench/diag/zero_readback.py; stage 2 = the tree's image: no $00 refused in $3E00-$3FFF, bench 15/15, bench-2026-10-04-1719.log). Was: 2026-09-24: the tree's microcode gained LDZ Rn,d / STZ Rn,d ($80-$8F, the word at R6.hi:d), ADDIW Rn,#w ($C0-$C7) and SHL16 Rn ($C8-$CF) (docs/programming/ISA-REFERENCE.md); until the EEPROM is reloaded those 32 records are the old all-zero words on the machine (review H-4), and so are $A5, $AE, $F8-$FA (given the HALT record in the tree 2026-09-29, docs/system/MICROCODE.md 5.4). The tree's image also has the three-step fetch prologue (2026-09-29, section 5.6 there; the machine still runs the six-step one) the M-1 fix (no count step reads memory) and 152 idle steps fewer; BACKLOG.md "Sequencer microcode: five changes" loads it in two stages. Reload with tools/ucode_send.py --all (at the machine), then run tests/bench/run.py --port ...: isa now checks the four instructions.

Last updated 18 November 2026