code wiki / _hdl_build / nx_blake2b_extvec_gate.nx
`nx_blake2b_extvec_gate.nx` source

↩ module page · 249 lines · 12158 B
1// nx_blake2b_extvec_gate.nx -- validated against RFC7693-AppendixA, read from the pinned+corroborated RFC7693 document.
2// ⚠HEADER CORRECTED 2026-08-01: this file was CLONED from nx_hmac_extvec_gate.nx and inherited its
3// header verbatim, so it claimed to be that gate validating HMAC-SHA-256 against RFC 4231. Nine of
4// twenty-three extvec gates carried the same wrong self-description. ★★★★★A CLONED FILE INHERITS ITS
5// PARENT'S CLAIMS, AND ON AN EVIDENCE ARTIFACT THE HEADER IS A PROVENANCE CLAIM, NOT A COMMENT --
6// an auditor reading headers would have concluded RFC 4231 validated all nine subjects.
7//
8// SUPERSEDES the single-case version. Going from 1 vector to 7 matters because the cases are deliberately
9// chosen by the authority to hit DIFFERENT code paths, and the ones I was NOT running are the interesting
10// ones: case 3 uses a 50-byte repeated data block, cases 6 and 7 use a 131-BYTE KEY (longer than the 64-byte
11// SHA-256 block, so the key must be HASHED first), and case 5 publishes a TRUNCATED 128-bit MAC.
12// ★A gate that ran only case 1 would never touch the key-longer-than-block branch -- the single most
13// commonly botched part of HMAC. Running one vector from a seven-vector suite is not "validated against
14// RFC 4231"; it is validated against one line of it.
15//
16// ⚠VARIABLE-LENGTH FIELDS, HANDLED BY TERMINATOR NOT BY LENGTH. Key/Data lengths differ per case and their
17// annotations are inconsistent -- "(20 bytes)" for keys but ("Hi There") for data -- so a length cannot be
18// read uniformly. Instead the hex run is read until the first `(`, which terminates both forms.
19// ⚠THE MAC HAS NO `(` TERMINATOR and case 5's is TRUNCATED to 16 bytes, so it is read as hex pairs until a
20// pair is not both-hex, capped at 32. That correctly stops at the section heading that follows -- note
21// "4.3." begins with '4', a HEX DIGIT, and is only rejected because '.' is not: the PAIR rule saves this,
22// a single-nibble rule would have swallowed it.
23//
24// Construction unchanged: no expected value in this source, document pinned to a socket-time digest, every
25// key/data/MAC read from that pinned document, and a completeness check that refuses GREEN below 7.
26// license_tier: ORIGINAL  expect_exit: 0
27import "nx_syscalls.nx"
28import "nx_sha256_wasm.nx"
29import "nx_blake2b.nx"
30import "nx_gate_verdict.nx"
31
32func w(s: *u8) -> i64 { var n: i64 = 0; while s[n] != (0 as u8) { n = n + 1 } sys_write(1, s, n); return 0 }
33func wb(b: *u8, n: i64) -> i64 { sys_write(1, b, n); return 0 }
34
35func nn(v: i64) -> i64 {
  var m: i64 = v
  if m < 0 { w("-" as *u8); m = 0 - m }
  let t: *u8 = sys_mmap(32)
  var k: i64 = 0
  if m == 0 { t[0] = 48 as u8; k = 1 }
  while m > 0 { t[k] = (48 + (m % 10)) as u8; m = m / 10; k = k + 1 }
  let b: *u8 = sys_mmap(32)
  var j: i64 = 0
  while j < k { b[j] = t[k - 1 - j]; j = j + 1 }
  sys_write(1, b, k)
  return 0
47}
48
49func hexnib(v: i64) -> i64 { if v < 10 { return 48 + v } return 87 + v }
50
51func hexval(c: i64) -> i64 {
  if c >= 48 { if c <= 57 { return c - 48 } }
  if c >= 97 { if c <= 102 { return c - 87 } }
  if c >= 65 { if c <= 70 { return c - 55 } }
  return 0 - 1
56}
57
58func isws(c: i64) -> i64 {
  if c == 32 { return 1 }
  if c == 10 { return 1 }
  if c == 13 { return 1 }
  if c == 9 { return 1 }
  return 0
64}
65
66func starts(b: *u8, n: i64, at: i64, s: *u8) -> i64 {
  var i: i64 = 0
  while s[i] != (0 as u8) {
      if at + i >= n { return 0 }
      if b[at + i] != s[i] { return 0 }
      i = i + 1
  }
  return 1
74}
75
76func findfrom(b: *u8, n: i64, s: *u8, from: i64) -> i64 {
  var p: i64 = from
  while p < n {
      if starts(b, n, p, s) == 1 { return p }
      p = p + 1
  }
  return 0 - 1
83}
84
85// Read hex PAIRS (whitespace between pairs is skipped) until a pair is not both-hex, or `(` is reached,
86// or cap is hit. Returns the byte count.
87func parserun(b: *u8, n: i64, from: i64, out: *u8, cap: i64) -> i64 {
  var p: i64 = from
  var got: i64 = 0
  var done: i64 = 0
  while done == 0 {
      if got >= cap { done = 1 }
      else {
          // skip whitespace
          var d1: i64 = 0
          while d1 == 0 {
              if p >= n { d1 = 1 }
              else { if isws(b[p] as i64) == 1 { p = p + 1 } else { d1 = 1 } }
          }
          if p + 1 >= n { done = 1 }
          else {
              // `(` ends a LINE SEGMENT, not the value: RFC 4231 annotates EVERY wrapped line, e.g.
              //   Data = 7768...6e7420  ("what do ya want ")
              //          666f...693f    ("for nothing?")
              // Treating `(` as the value terminator stopped case 2 at 16 of its 28 bytes. On `(`, skip
              // to the next line and continue ONLY if it resumes with a hex pair; otherwise stop.
              if b[p] == (40 as u8) {
                  var dl: i64 = 0
                  while dl == 0 {
                      if p >= n { dl = 1 }
                      else { if b[p] == (10 as u8) { p = p + 1; dl = 1 } else { p = p + 1 } }
                  }
                  var dw: i64 = 0
                  while dw == 0 {
                      if p >= n { dw = 1 }
                      else { if isws(b[p] as i64) == 1 { p = p + 1 } else { dw = 1 } }
                  }
                  // ⚠"Da" IS VALID HEX. The next line may be `   Data =  7768...`, and testing only
                  // "do the first two chars parse as hex" accepted D,a and read the LABEL as byte 0xDA --
                  // adding one phantom byte to every key (case 1 read 21 of 20, case 2 read 5 of 4).
                  // Third form of the same hazard today: an ASCII gutter, then English prose, now a
                  // FIELD LABEL that happens to spell hex.
                  // STRUCTURAL DISCRIMINATOR: a continuation line has NO '='; every label line has one.
                  var eqfound: i64 = 0
                  var sc: i64 = p
                  var de: i64 = 0
                  while de == 0 {
                      if sc >= n { de = 1 }
                      else { if b[sc] == (10 as u8) { de = 1 }
                      else { if b[sc] == (61 as u8) { eqfound = 1; de = 1 } else { sc = sc + 1 } } }
                  }
                  if eqfound == 1 { done = 1 }
                  else {
                      if p + 1 >= n { done = 1 }
                      else {
                          if hexval(b[p] as i64) < 0 { done = 1 }
                          else { if hexval(b[p + 1] as i64) < 0 { done = 1 } }
                      }
                  }
              }
              else {
                  let h1: i64 = hexval(b[p] as i64)
                  let h2: i64 = hexval(b[p + 1] as i64)
                  if h1 < 0 { done = 1 }
                  else { if h2 < 0 { done = 1 }
                  else {
                      out[got] = ((h1 * 16) + h2) as u8
                      got = got + 1
                      p = p + 2
                  } }
              }
          }
      }
  }
  return got
156}
157
158// Find the next "Key" FIELD LABEL at line start. Verified against the raw bytes of RFC 4231:
159//   line 192 "   Key =          0b0b..."   <- cases 1,2,4,5,6,7
160//   line 251 "   Key            aaaa..."   <- CASE 3: no '=' at all
161//   line 177 "   Keys, data, and digests..." <- PROSE. "Key" is a PREFIX of "Keys".
162// Anchoring on "\n   Key" alone matched that prose line and drove the parse to ZERO cases. So the label is
163// accepted only when the character AFTER "Key" is a SPACE or '=' -- which "Keys" fails on 's'.
164// ★A prefix match is not a token match. Every anchor in this file is now checked against the byte AFTER it.
165func find_key_label(b: *u8, n: i64, from: i64) -> i64 {
  var p: i64 = from
  var done: i64 = 0
  while done == 0 {
      let h: i64 = findfrom(b, n, "\n   Key" as *u8, p)
      if h < 0 { return 0 - 1 }
      let c: i64 = b[h + 7] as i64
      if c == 32 { return h }
      if c == 61 { return h }
      p = h + 7
  }
  return 0 - 1
177}
178func main() -> i64 {
  w("nx_blake2b_extvec_gate -- BLAKE2b-512 vs RFC 7693 Appendix A, READ FROM THE DOCUMENT\n" as *u8)
  let lp: *i64 = sys_mmap(16) as *i64
  lp[0] = 0
  let b: *u8 = sys_read_file("knowledge/extvec/rfc7693.txt\x00" as *u8, lp)
  if lp[0] <= 0 { w("RED: fetched vector file absent -- run nx_vecfetch.\n" as *u8); return 1 }
  let ctx0: *u8 = sys_mmap(1024)
  let dg: *u8 = sys_mmap(64)
  nx_sha256_one_shot(b, lp[0], ctx0, dg)
  let hx: *u8 = sys_mmap(80)
  var i: i64 = 0
  while i < 32 { hx[i*2] = hexnib(((dg[i] as i64)/16)&15) as u8; hx[i*2+1] = hexnib((dg[i] as i64)&15) as u8; i = i + 1 }
  let wnt: *u8 = "c943754888364fe29bbd0bb3c71c6658ef392371497eaa4a9d9dde015b0721e5\x00" as *u8
  var pin: i64 = 1
  i = 0
  while i < 64 { if hx[i] != wnt[i] { pin = 0 } i = i + 1 }
  w("  acquisition digest: " as *u8); wb(hx, 64); w("\n" as *u8)
  if pin == 0 { w("RED: PIN FAILED.\n" as *u8); return 1 }
  w("  PIN OK -- bytes match the digest computed in-process at the socket\n" as *u8)
197
  // ANCHOR ON THE LABELLED ANSWER, NOT THE TRACE. Appendix A prints the internal state `h[8] = 0D4D...`
  // immediately ABOVE the answer; anchoring on the last hex block in the appendix would have grabbed the
  // internal state. Same intermediate-state trap as RFC 8439 2.3.2 -- the discriminator is the LABEL.
  let la: i64 = findfrom(b, lp[0], "BLAKE2b-512(\"abc\") =" as *u8, 0)
  if la < 0 { w("RED: no 'BLAKE2b-512(\"abc\") =' line\n" as *u8); return 1 }
  let exp: *u8 = sys_mmap(128)
  let en: i64 = parserun(b, lp[0], la + 20, exp, 64)
  if en != 64 { w("RED: expected 64 bytes, parsed " as *u8); nn(en); w(" -- the READER.\n" as *u8); return 1 }
  w("  parsed the published 64-byte digest\n" as *u8)
207
  // The message is the ASCII "abc" the document names in the label itself. Supplying it is sound because
  // the ANSWER comes from the document: a wrong message yields a different digest and the gate goes RED.
  let msg: *u8 = "abc\x00" as *u8
  let bctx: *u8 = sys_mmap(4096)
  let buf128: *u8 = sys_mmap(256)
  let sv: *i64 = sys_mmap(512) as *i64
  let sm: *i64 = sys_mmap(512) as *i64
  let got: *u8 = sys_mmap(128)
  let nokey: *u8 = sys_mmap(64)
217
  var pass: i64 = 0
  var fail: i64 = 0
  nx_blake2b_hash(msg, 3, nokey, 0, got, 64, bctx as *NxBlake2b, buf128, sv, sm)
  var same: i64 = 1
  i = 0
  while i < 64 { if got[i] != exp[i] { same = 0 } i = i + 1 }
  if same == 1 { pass = pass + 1; w("  PASS T1: BLAKE2b-512(\"abc\") == the document's published digest\n" as *u8) }
  else { fail = fail + 1; w("  FAIL T1: digest mismatch\n" as *u8) }
226
  // NEGATIVE CONTROL: a one-character message change must alter the digest.
  let msg2: *u8 = "abd\x00" as *u8
  nx_blake2b_hash(msg2, 3, nokey, 0, got, 64, bctx as *NxBlake2b, buf128, sv, sm)
  var same2: i64 = 1
  i = 0
  while i < 64 { if got[i] != exp[i] { same2 = 0 } i = i + 1 }
  if same2 == 0 { pass = pass + 1; w("  PASS NEG: \"abd\" yields a different digest (the check can fail)\n" as *u8) }
  else { fail = fail + 1; w("  FAIL NEG: message change did not alter the digest\n" as *u8) }
235
  w("\n  refsrc=https://www.rfc-editor.org/rfc/rfc7693.txt\n" as *u8)
  w("  refsrcdig=" as *u8); wb(hx, 64); w("\n" as *u8)
  w("  ref=RFC7693-AppendixA   gate=nx_blake2b_extvec_gate\n" as *u8)
  w("nx_blake2b_extvec_gate: pass=" as *u8); nn(pass); w(" fail=" as *u8); nn(fail)
  // MIGRATED onto nx_gate_verdict by nx_gate_dry_apply (D001, minimal form): every check
  // row above is untouched, so the PASS/FAIL vector cannot change; only the hand-rolled
  // verdict emission is replaced by the ONE shared base class. Proven by nx_gate_migrate verify.
  let ctr__dry: *i64 = gv_ctr()
  ctr__dry[0] = pass
  ctr__dry[1] = pass + fail
  let rc__dry: i64 = gv_verdict("BLAKE2B-EXTVEC-GATE" as *u8, ctr__dry, "teeth unchanged; verdict emission migrated onto the shared base class" as *u8)
  sys_exit(rc__dry)
  return rc__dry
249}
nx_blake2b_extvec_gate.nx source

`nx_blake2b_extvec_gate.nx` source
