v3 L8: after a tick, the chosen card's number is applied to the gates, not card_bits(card) -- every gate one bit off #196

New Issue

2026-11-13T22:32:16Z

rajames commented

2026-11-13 22:32:16 +00:00

Found 2026-11-13 while rewriting L8 for v4 (#189 Phase 7, Task 3). Introduced by #188 (c76a742f), my own work.

The defect

Since #188, L8's table cards are numbered 0..63, and card_bits() (v3/src/ssm_jacquard.c:235) turns a card number into its seven gate bits: ((card & 0x3F) << 1) | SSM_CFG_L7.

ssm_l8_tick_score_and_select() (v3/src/ssm_jacquard.c:292) takes the card cd_config_table_tick() chose. That function returns entries[best].config_bits, which is the card number. It then tests the SSM_CFG_L1..SSM_CFG_L7 bits against that number, not against card_bits(number). So after every tick each gate is one bit off:

L1 (0x40) is never on, because a card number is below 64;
L2 gets the card's L1 bit, L3 the card's L2 bit, and so on down;
L7 gets the card's L6 bit, though #188 made L7 always on.

ssm_apply_mode_from_table() (:447) uses card_bits() correctly. #188's tests (kernel/tests/test_ssm_jacquard.c) never check the gates after a tick.

Ruled (Captain Bob, 2026-11-13)

"v4 correct, v3 as is": v4's L8 applies the card through card_bits(), and v3 is left alone, since it is being retired (#156). v4's L8 therefore departs from v3's here on purpose. This issue records why.

If v3 is ever fixed

Fix the one site: card_bits(bits) in place of bits. Add a test that fails first: after a tick, the gates equal card_bits(table->current_config).

Found 2026-11-13 while rewriting L8 for v4 (#189 Phase 7, Task 3). Introduced by #188 (c76a742f), my own work. ## The defect Since #188, L8's table cards are numbered 0..63, and `card_bits()` (`v3/src/ssm_jacquard.c:235`) turns a card number into its seven gate bits: `((card & 0x3F) << 1) | SSM_CFG_L7`. `ssm_l8_tick_score_and_select()` (`v3/src/ssm_jacquard.c:292`) takes the card `cd_config_table_tick()` chose. That function returns `entries[best].config_bits`, which is the card number. It then tests the `SSM_CFG_L1`..`SSM_CFG_L7` bits against **that number**, not against `card_bits(number)`. So after every tick each gate is one bit off: - L1 (`0x40`) is never on, because a card number is below 64; - L2 gets the card's L1 bit, L3 the card's L2 bit, and so on down; - L7 gets the card's L6 bit, though #188 made L7 always on. `ssm_apply_mode_from_table()` (`:447`) uses `card_bits()` correctly. #188's tests (`kernel/tests/test_ssm_jacquard.c`) never check the gates after a tick. ## Ruled (Captain Bob, 2026-11-13) "v4 correct, v3 as is": v4's L8 applies the card through `card_bits()`, and v3 is left alone, since it is being retired (#156). v4's L8 therefore departs from v3's here on purpose. This issue records why. ## If v3 is ever fixed Fix the one site: `card_bits(bits)` in place of `bits`. Add a test that fails first: after a tick, the gates equal `card_bits(table->current_config)`.

rajames added this to the Retire v3 milestone 2026-11-13 22:32:16 +00:00

rajames added the

labels 2026-11-13 22:32:16 +00:00

Sign in to join this conversation.