amos 2f32624d3b Batch 18 (20261026-C-hwe-codegen-18): add 100 calibrated HWE code-gen tasks; pause production for rollout
100 tasks from the HWE arm-linux code-gen pool (outside the lk/man/pth
range given to batch 22), each with at least one independent
contract-edge defect and plain public docs. Claude reviewed 100/100
USABLE against current digests; no ID or source overlap with main or
batch 22. Difficulty is unmeasured.

progress.md records the wrap-up: all Codex sessions and Qwen schedulers
stopped for the user's rollout, Qwen results so far (57 HARD of 373),
and what is left on each batch branch.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 17:52:13 +08:00

287 lines
14 KiB
Python

"""Private independent integer model and raw-observer comparison."""
from pathlib import Path
import json
import random
import subprocess
import sys
import tempfile
LOW = -(1 << 63)
HIGH = (1 << 63) - 1
OK, RANGE, EXISTS, DEAD, STALE, TOKEN, ENGAGED, WAIT, OWNER, QUEUED, TURN, NOT_OWNER, NOT_WAITING, BUSY, OVERFLOW = range(15)
class Model:
def __init__(self):
self.slots = [dict(active=0, epoch=0, token=0, owner=-1, value=0, queue=[]) for _ in range(8)]
self.actors = [[0, -1, 0] for _ in range(16)]
def bind(self, p, kind, slot):
self.actors[p] = [kind, slot, self.slots[slot]['epoch']]
def valid(self, args, participants=()):
s, e, t = args[:3]
if not 0 <= s < 8 or any(not 0 <= p < 16 for p in participants):
return RANGE
node = self.slots[s]
if not node['active']:
return DEAD
if node['epoch'] != e:
return STALE
if node['token'] != t:
return TOKEN
return OK
def step(self, command):
op, *v = command
reply = op in ('I', 'R', 'Z', 'Q')
def result(code, value=0):
return [op, code, value] if reply else [op, code]
if op == 'I':
s, t = v
if not 0 <= s < 8:
return result(RANGE)
node = self.slots[s]
if node['active']:
return result(EXISTS)
node.update(active=1, epoch=node['epoch'] + 1, token=t, owner=-1, value=0, queue=[])
return result(OK, node['epoch'])
if op == 'V':
s = v[0]
if not 0 <= s < 8:
return [op, RANGE, 0, 0, 0, -1, 0, 0]
n = self.slots[s]
return [op, OK, n['active'], n['epoch'], n['token'], n['owner'], n['value'], len(n['queue']), *n['queue']]
if op == 'P':
p = v[0]
return [op, OK, *self.actors[p]] if 0 <= p < 16 else [op, RANGE, 0, -1, 0]
if op == 'S':
return [op, sum(n['active'] for n in self.slots), sum(n['owner'] != -1 for n in self.slots), sum(len(n['queue']) for n in self.slots)]
participants = v[3:5] if op == 'H' else v[3:4] if op in ('L', 'N', 'U', 'C', 'X', 'A', 'Q') else ()
code = self.valid(v, participants)
if code != OK:
return result(code)
s, e, t = v[:3]
n = self.slots[s]
p = v[3] if len(v) > 3 else None
if op == 'D':
if n['owner'] != -1 or n['queue']:
return result(BUSY)
n.update(active=0, token=0, owner=-1, value=0, queue=[])
elif op == 'T':
if n['owner'] != -1 or n['queue']:
return result(BUSY)
n['token'] = p
elif op == 'R':
return result(OK, n['value'])
elif op in ('L', 'N'):
if n['owner'] == p:
return result(OWNER)
if p in n['queue']:
return result(QUEUED)
if self.actors[p][0]:
return result(ENGAGED)
if n['owner'] == -1 and not n['queue']:
n['owner'] = p
self.bind(p, 1, s)
elif op == 'N':
return result(BUSY)
else:
n['queue'].append(p)
self.bind(p, 2, s)
return result(WAIT)
elif op == 'U':
if n['owner'] != p:
return result(NOT_OWNER)
n['owner'] = -1
self.actors[p] = [0, -1, 0]
elif op == 'C':
if n['owner'] == p:
return result(OWNER)
if p not in n['queue']:
return result(NOT_WAITING)
if n['owner'] != -1:
return result(BUSY)
if n['queue'][0] != p:
return result(TURN)
n['queue'].pop(0)
n['owner'] = p
self.bind(p, 1, s)
elif op == 'X':
if n['owner'] == p:
return result(OWNER)
if p not in n['queue']:
return result(NOT_WAITING)
n['queue'].remove(p)
self.actors[p] = [0, -1, 0]
elif op == 'H':
q = v[4]
if n['owner'] != p:
return result(NOT_OWNER)
if q not in n['queue']:
return result(NOT_WAITING)
if n['queue'][0] != q:
return result(TURN)
self.actors[p] = [0, -1, 0]
n['queue'].pop(0)
n['owner'] = q
self.bind(q, 1, s)
elif op == 'Z':
if n['owner'] != -1:
return result(BUSY)
count = len(n['queue'])
for q in n['queue']:
self.actors[q] = [0, -1, 0]
n['queue'].clear()
return result(OK, count)
elif op == 'A':
if n['owner'] != p:
return result(NOT_OWNER)
total = n['value'] + v[4]
if not LOW <= total <= HIGH:
return result(OVERFLOW)
n['value'] = total
elif op == 'Q':
if n['owner'] == p:
return result(OK, -1)
if p in n['queue']:
return result(OK, n['queue'].index(p))
return result(NOT_WAITING)
else:
raise RuntimeError(op)
return result(OK)
def encode(rows):
return b''.join((' '.join(map(str, row)) + '\n').encode('ascii') for row in rows)
def scenes():
all_scenes = []
def add_scene(name, rows):
model = Model()
expected = encode([model.step(row) for row in rows])
all_scenes.append((name, rows, encode(rows), expected))
add_scene('empty', [])
add_scene('smoke', [('I',2,17),('L',2,1,17,4),('A',2,1,17,4,9),('U',2,1,17,4),('D',2,1,17),('I',2,17),('R',2,1,17),('V',2)])
# Repeated identity cycles, independent slots, full-width tokens and stale handles.
rows = [('S',)] + [('V',s) for s in range(8)] + [('P',p) for p in range(16)]
for s in range(8):
for epoch in range(1,5):
token = [LOW, HIGH, 0, -1][epoch-1]
rows += [('I',s,token),('I',s,token),('V',s),('R',s,epoch,token),('R',s,epoch-1,token),('L',s,epoch,token,15),('P',15),('A',s,epoch,token,15,s+epoch),('U',s,epoch,token,15),('T',s,epoch,token,token),('T',s,epoch,token,7),('R',s,epoch,token),('R',s,epoch,7),('D',s,epoch,7),('D',s,epoch,7),('R',s,epoch,7),('V',s),('P',15),('S',)]
add_scene('identity_cycles',rows)
rows = [('I',0,LOW),('I',7,HIGH),('L',0,1,LOW,0),('A',0,1,LOW,0,37),('L',7,1,HIGH,15),('L',7,1,HIGH,0),('N',7,1,HIGH,0),('L',0,1,LOW,0),('N',0,1,LOW,0),('Q',0,1,LOW,0),('L',0,1,LOW,1),('L',0,1,LOW,2),('L',0,1,LOW,3),('L',0,1,LOW,2),('N',0,1,LOW,2),('C',0,1,LOW,1),('C',0,1,LOW,0),('X',0,1,LOW,0),('Z',0,1,LOW),('D',0,1,LOW),('T',0,1,LOW,0),('H',0,1,LOW,2,1),('H',0,1,LOW,0,9),('H',0,1,LOW,0,2),('H',0,1,LOW,0,1),('R',0,1,LOW),('P',0),('P',1),('Q',0,1,LOW,2),('X',0,1,LOW,2),('V',0),('L',0,1,LOW,4),('Q',0,1,LOW,4),('Q',0,1,LOW,3),('Q',0,1,LOW,9),('U',0,1,LOW,1),('V',0),('D',0,1,LOW),('V',0),('P',3),('P',4),('S',),('N',0,1,LOW,0),('T',0,1,LOW,4),('C',0,1,LOW,4),('X',0,1,LOW,3),('V',0),('C',0,1,LOW,4),('R',0,1,LOW),('U',0,1,LOW,4),('L',0,1,LOW,3),('L',0,1,LOW,4),('U',0,1,LOW,3),('Z',0,1,LOW),('Z',0,1,LOW),('P',4),('D',0,1,LOW),('V',0),('S',)]
# Full queue length and both free/owned states; no automatic handoff or bypass.
rows += [('U',7,1,HIGH,15),('D',7,1,HIGH),('I',7,0),('N',7,2,0,0)]
rows += [('L',7,2,0,p) for p in range(1,16)]
rows += [('Q',7,2,0,p) for p in range(16)] + [('V',7),('S',),('U',7,2,0,0),('D',7,2,0),('Z',7,2,0)]
rows += [('P',p) for p in range(16)] + [('D',7,2,0),('S',)]
rows += [('I',0,99),('L',0,2,99,0),('L',0,2,99,1),('U',0,2,99,0),('L',0,2,99,2),('V',0),('Q',0,2,99,2),('C',0,2,99,2),('C',0,2,99,1),('R',0,2,99),('U',0,2,99,1),('Z',0,2,99),('D',0,2,99)]
add_scene('queue_lifecycle',rows)
# Every supported Handle API observes each validation branch directly.
rows=[]
handle_ops=['D','R','Z','T','L','N','U','C','X','Q','A','H']
def operation(op,h,p=0,q=1):
extra=[]
if op in ['T','L','N','U','C','X','Q']:extra=[p]
if op in ['A','H']:extra=[p,q]
return (op,*h,*extra)
invalid=[LOW,-1,8,HIGH]
for x in invalid:
rows += [('I',x,LOW),('V',x)]
rows += [operation(op,(x,LOW,HIGH),LOW,HIGH) for op in handle_ops]
for p in [LOW,-1,16,HIGH]:rows += [('P',p)]
for op in handle_ops:rows += [operation(op,(0,LOW,HIGH))]
rows += [('I',0,LOW)]
for epoch in [LOW,-1,0,2,HIGH]:
rows += [operation(op,(0,epoch,HIGH)) for op in handle_ops]
for token in [HIGH,-1,0,1]:
rows += [operation(op,(0,1,token)) for op in handle_ops]
for op in ['L','N','U','C','X','Q','A','H']:
for p in [LOW,-1,16,HIGH]:
rows += [operation(op,(0,LOW,HIGH),p,HIGH),operation(op,(7,0,0),p,HIGH),operation(op,(0,1,LOW),p,HIGH)]
for q in [LOW,-1,16,HIGH]:rows += [('H',0,1,LOW,0,q)]
rows += [('V',0),('P',0),('S',),('N',0,1,LOW,0),('A',0,1,LOW,1,LOW),('A',0,1,LOW,0,0),('A',0,1,LOW,0,HIGH),('A',0,1,LOW,0,1),('R',0,1,LOW),('A',0,1,LOW,0,LOW),('R',0,1,LOW),('A',0,1,LOW,0,-HIGH),('A',0,1,LOW,0,-1),('R',0,1,LOW),('A',0,1,LOW,0,HIGH),('A',0,1,LOW,0,1),('R',0,1,LOW),('U',0,1,LOW,0),('A',0,1,LOW,0,0),('T',0,1,LOW,HIGH),('V',0),('D',0,1,HIGH),('V',0)]
add_scene('domains_and_errors',rows)
rows=[('I',0,17),('I',1,-17),('L',0,1,17,0),('A',0,1,17,0,37),('L',0,1,17,1),('L',0,1,17,2),('L',1,1,-17,15),('L',1,1,-17,14)]
probes=[('V',slot) for slot in range(8)]+[('P',p) for p in range(16)]+[('S',)]
rows+=probes
failures=[('I',0,99)]
for op in handle_ops:
for h in [(-1,1,17),(7,1,17),(0,0,17),(0,1,18)]:
failures.append(operation(op,h,4,5))
for op in ['L','N','U','C','X','Q','A','H']:
failures.append(operation(op,(0,1,17),-1,16))
failures += [('D',0,1,17),('T',0,1,17,0),('L',0,1,17,0),('L',0,1,17,1),('L',0,1,17,15),('N',0,1,17,0),('N',0,1,17,1),('N',0,1,17,15),('N',0,1,17,4),('U',0,1,17,1),('C',0,1,17,0),('C',0,1,17,4),('C',0,1,17,1),('X',0,1,17,0),('X',0,1,17,4),('H',0,1,17,1,2),('H',0,1,17,0,4),('H',0,1,17,0,2),('H',0,1,17,0,0),('Z',0,1,17),('A',0,1,17,1,0),('Q',0,1,17,4)]
for row in failures:
rows.append(row);rows+=probes
rows += [('U',0,1,17,0)] + probes
for row in [('D',0,1,17),('T',0,1,17,0),('N',0,1,17,4),('C',0,1,17,2)]:
rows.append(row);rows+=probes
add_scene('failure_atomicity',rows)
rng=random.Random(16020261027)
rows=[]
model=Model()
for step in range(2500):
s=rng.randrange(8); n=model.slots[s]; p=rng.randrange(16)
op=rng.choice(['I','D','L','L','N','U','C','X','H','Z','A','R','T','Q','V','P','S'])
if op=='I': row=(op,s,rng.choice([LOW,HIGH,-9,0,17]))
elif op=='V':row=(op,s)
elif op=='P':row=(op,p)
elif op=='S':row=(op,)
else:
h=(s,n['epoch'],n['token'])
if rng.randrange(8)==0:h=(s,rng.choice([LOW,HIGH,0,n['epoch']+1]),n['token'])
row=operation(op,h,p,rng.choice([LOW,HIGH,0,-1,1,2,15]))
model.step(row);rows.append(row)
if step%25==0:rows.extend([('V',s),('P',p),('S',)])
rows += [('S',)] * (4096-len(rows))
add_scene('seeded_transitions',rows)
return all_scenes
def main():
candidate=Path(sys.argv[1])
names=['ledger.hpp','storage.hpp','identities.cpp','transitions.cpp','queries.cpp']
# Finish every trusted filesystem operation before the first compiler subprocess.
sources={name:(candidate/name).read_bytes() for name in names}
observer=(Path(__file__).resolve().parent/'observer.txt').read_bytes()
cases=scenes()
scratch=Path(tempfile.mkdtemp(prefix='instance-',dir=Path.cwd()))
for name,data in sources.items():
path=scratch/name;path.write_bytes(data)
path=scratch/'observer.cpp';path.write_bytes(observer)
argv=['g++','-std=c++17','-O2','-Wall','-Wextra','observer.cpp','identities.cpp','transitions.cpp','queries.cpp','-o','observe']
record={'build':None,'scenes':[],'first_reject':None}
try:
build=subprocess.run(argv,cwd=str(scratch),capture_output=True,timeout=35)
except subprocess.TimeoutExpired:
record['build']={'exit':'timeout'};record['first_reject']='build-timeout'
print(json.dumps(record,separators=(',',':')),file=sys.stderr);print('FAIL build timeout');return 1
record['build']={'exit':build.returncode,'out':len(build.stdout),'err':len(build.stderr)}
if build.returncode:
record['first_reject']='build';record['build']['stderr']=build.stderr[-700:].decode('utf8','replace')
print(json.dumps(record,separators=(',',':')),file=sys.stderr);print('FAIL build');return 1
for name,rows,data,expected in cases:
try:
run=subprocess.run([str(scratch/'observe')],input=data,cwd=str(scratch),capture_output=True,timeout=8)
item={'name':name,'exit':run.returncode,'out':len(run.stdout),'err':len(run.stderr),'reject':None}
if run.returncode:item['reject']='exit'
elif run.stderr:item['reject']='stderr'
elif run.stdout!=expected:item['reject']='stdout-bytes'
if item['reject'] and record['first_reject'] is None:
lines=run.stdout.split(b'\n');want=expected.split(b'\n')
index=next((i for i in range(min(len(lines),len(want))) if lines[i]!=want[i]),min(len(lines),len(want)))
record['first_reject']={'scene':name,'kind':item['reject'],'line':index+1,'command':list(rows[index]) if index<len(rows) else None,'want':want[index].decode('ascii')[:140] if index<len(want) else None,'got':lines[index].decode('ascii','replace')[:140] if index<len(lines) else None}
except subprocess.TimeoutExpired:
item={'name':name,'exit':'timeout','out':None,'err':None,'reject':'timeout'}
if record['first_reject'] is None:record['first_reject']={'scene':name,'kind':'timeout'}
record['scenes'].append(item)
print(json.dumps(record,separators=(',',':')),file=sys.stderr)
if record['first_reject'] is not None:
print('FAIL '+json.dumps(record['first_reject'],separators=(',',':')));return 1
print('PASS');return 0
if __name__=='__main__':
raise SystemExit(main())