fix(tri27): decoder takes the T736 field table from the t27 spec (Refs gHashTag/t27#6507)

[gHashTag]

Refs gHashTag/t27#6507

Owner approval, 2026-10-06, on gHashTag/t27#6507: "add the label yourself, do the work, do not ask about labels". Label owner-approved-foreign (created here, color B60205).

What

src/tri27/emu/decoder.zig read src2 for ADD..XOR (and BUNDLE3) only, so DOT, BIND and BUNDLE2 decoded with src2 = 0. That breaks the T736 field table of TRI-27. The decoder now takes the table from the t27 spec instead of keeping a second hand-copied one:

• src/tri27/emu/tri27_encoding.zig (new) is t27c gen output of gHashTag/t27 specs/isa/ternary_encoding.t27, byte for byte. Do not hand-edit it.
  • spec sha256: 5e2c265a0972189ccf887d523abb86c13f9fa7ae6a1f35a59c20f1a1fc4796cd
  • generated sha256: da08bb8d44888dd747b11de4476d75395921f3fbc11754acb9532a218ffa3fad, which equals gen_hash_zig in t27 .trinity/seals/isa_Tri27Encoding.json.
• decode reads fields through t27a_src1, t27a_src2, t27a_imm and t27a_imm_form, and puts BUNDLE3's third register (bits 27:23) in cond.
• encode calls t27a_encode.

Tests

New decoder tests:

• DOT, BIND and BUNDLE2 keep src2.
• BUNDLE3 keeps its third register.
• The six words that gHashTag/t27 specs/isa/t27a.t27 pins decode to the same fields: ADD, LDI -5, MOV, JGT 7, BUNDLE3 and NOP.

All results are zig 0.15.2, the CI version.

Suite	Passed
decoder	24/24
tri27_encoding	14/14
test_encoder_fuzz	31/31
test_comprehensive	66/66
executor	33/33
smoke_tests	37/37
encoder_simple	35/35
emit_verilog	2/2
emit_zig	2/2

tri_exec.zig does not compile on main because of a two-argument std.debug.assert call. That is a pre-existing bug, not touched here.

gitleaks was run by hand on the commit: no leaks.

The t27 side, t27c asm / t27c disasm on the same spec, is in a separate gHashTag/t27 PR. That repository's T27A_CONSUMER_GAP marker is removed once this PR lands.

Work report

{
  "version": 1,
  "head_sha": "e40776afc9981a0af5e7a3e6598df1744bd58966",
  "summary": "The TRI-27 emulator decoder now reads its instruction field table from the t27 spec ternary_encoding.t27 instead of a hand-copied table, so DOT, BIND and BUNDLE2 keep their second source register.",
  "changes": [
    "src/tri27/emu/tri27_encoding.zig is new: the byte-for-byte t27c gen output of gHashTag/t27 specs/isa/ternary_encoding.t27, recorded with its spec and generated sha256.",
    "decode in src/tri27/emu/decoder.zig reads src1, src2, the immediate and the immediate form through t27a_src1, t27a_src2, t27a_imm and t27a_imm_form.",
    "encode in src/tri27/emu/decoder.zig is t27a_encode, so encode and decode share one field table.",
    "Three new decoder tests: DOT, BIND and BUNDLE2 keep src2, BUNDLE3 keeps its third register, and six words pinned by t27 specs/isa/t27a.t27 decode to the same fields."
  ],
  "tests": [
    {
      "command": "zig test src/tri27/emu/decoder.zig",
      "result": "All 24 tests passed under zig 0.15.2",
      "status": "passed",
      "evidence": "local run on the PR head, zig 0.15.2 as in CI"
    },
    {
      "command": "zig test src/tri27/emu/tri27_encoding.zig",
      "result": "All 14 tests of the generated table passed",
      "status": "passed",
      "evidence": "local run on the PR head, zig 0.15.2 as in CI"
    },
    {
      "command": "zig test src/tri27/emu/test_encoder_fuzz.zig",
      "result": "All 31 encoder fuzz tests passed",
      "status": "passed",
      "evidence": "local run on the PR head, zig 0.15.2 as in CI"
    },
    {
      "command": "zig test src/tri27/emu/test_comprehensive.zig",
      "result": "All 66 comprehensive tests passed",
      "status": "passed",
      "evidence": "local run on the PR head, zig 0.15.2 as in CI"
    },
    {
      "command": "zig test src/tri27/emu/executor.zig",
      "result": "All 33 executor tests passed",
      "status": "passed",
      "evidence": "local run on the PR head, zig 0.15.2 as in CI"
    },
    {
      "command": "zig test smoke_tests, encoder_simple, emit_verilog, emit_zig",
      "result": "37, 35, 2 and 2 tests passed respectively",
      "status": "passed",
      "evidence": "local run on the PR head, zig 0.15.2 as in CI"
    },
    {
      "command": "gitleaks detect --log-opts=-1",
      "result": "one commit scanned, no leaks found",
      "status": "passed",
      "evidence": "run by hand on the PR head commit"
    }
  ],
  "limitations": [
    "tri_exec.zig does not compile on main because of a two-argument std.debug.assert call; that pre-existing bug is not touched here.",
    "The generated table is vendored as a file: when the t27 spec changes, tri27_encoding.zig must be regenerated from t27 and replaced, never hand-edited."
  ],
  "tags": [
    "tri27",
    "t27",
    "isa",
    "emulator"
  ],
  "blog": {
    "title": "The TRI-27 decoder now reads its field table from the spec",
    "summary": "The emulator decoder used to keep its own copy of the TRI-27 field table, and that copy dropped the second register of DOT, BIND and BUNDLE2. It now imports the code t27c generates from the ternary encoding spec.",
    "outline": [
      "The TRI-27 instruction word has one field table, written down in the t27 spec ternary_encoding.t27, but the Zig emulator decoder carried a second hand-written copy of it.",
      "The copy disagreed with the spec: it read the second source register only for ADD through XOR, so DOT, BIND and BUNDLE2 decoded with that register set to zero.",
      "Instead of fixing the copy, the decoder now imports tri27_encoding.zig, which is exactly what t27c gen writes from the spec, with both hashes recorded next to the import.",
      "New tests decode the six words the t27 assembler spec pins and check DOT, BIND, BUNDLE2 and BUNDLE3 keep every register, alongside the existing encoder and executor suites."
    ]
  }
}

🤖 Generated with Claude Code

[github-actions]

X Brain Health Check

Score: 100.0/100
Status: 🟢 HEALTHY
Threshold: 80/100

X Brain is above merge threshold

[github-actions]

🧪 Brain Stress Test Results

Tests: 261/261 passed, 0 leaked
Status: 🟢 PASS
Gate: every test in zig build test-brain-stress passes and none leaks