Loading files
This file is pinned to the selected release. Download the original to open it in your design tool.
#!/usr/bin/env python3
"""Does a bounded block relocation restore a board that damage made unroutable?
`tests/stress/relocation_reach.py` answers the MECHANISM question -- letting
neighbours yield buys a block travel a frozen board cannot -- and answers it yes.
This file asks the only question that decides whether the feature is worth having,
and it is a different one:
On a board where displacing exactly one block measurably worsens the ROUTED
outcome, does one bounded relocation of that block restore it -- without
moving anything else, without tearing the block -- where the tool that
already exists does not?
Reach is not routability. That sentence is the reason this file exists.
THE ARMS
--------
C control, undamaged (perturb's own dose-0 pass of the same writer)
D damaged, unrepaired -- the denominator's other end
R the relocation applied to D
R0 THE SAME relocation applied to C -- the undamaged pairing
L place_route_loop with the pin gate lifted and blocks on: the incumbent
`R0` is not optional. #553's recall study reported "above chance on 4 of 4
boards" and the finding evaporated when the same ranking was run on the pristine
board: `perturb.pick_block` ranks units by source and size, and so do most
placement scorers, so the "damaged" block is often already first. A repair arm
with no undamaged pairing cannot tell "it fixed the damage" from "it moves that
block on every board". The delta is the evidence; the raw number is not.
`L` is not optional either. #554 is only interesting if it beats a tool that
already exists: #411 measured `loop@allon` taking tigard from 13 routing failures
to 2, at a recovery of -0.052. A relocation that ties that has demonstrated
nothing.
THE HEADLINE IS ROUTED, AND `recovery` IS NOT IT
-------------------------------------------------
route_recovery = (blocking(D) - blocking(R)) / (blocking(D) - blocking(C))
over failed + open + pad-deficit, `None` when the denominator is zero. 1.0 is
restored to the control's routed quality, 0 inert, negative worse than leaving
the damage alone.
Pose `recovery` is reported as a DIAGNOSTIC and never as the verdict, for the
reason CLAUDE.md already states and for a sharper one measured here: the
relocation aims at `net_centroid`, and on the ten tracked boards a block sits
4.95-54.9 mm from its own net centroid at the HUMAN's placement. So the best
achievable pose recovery is bounded above by `1 - d_ctrl/dose`, which is at or
below zero on most cells before anything runs. A pose null on such a cell is not
evidence about the solver. `compensated` -- routed win at recovery ~ 0, with zero
collateral and an intact block -- is a first-class PASS here, exactly as #411
concluded.
THE CONTRACT CHECKS, WHICH CAN FAIL A CELL THAT ROUTED BETTER
--------------------------------------------------------------
#554 says "move only diagnosed blocks; keep the relative order of everything else
as a hard constraint". That is the definition, not the bonus. So a cell FAILS if
`collateral_pad_rms` rises or the block is torn, whatever `blocking` did -- a
solve that buys routability by walking the neighbours is `place_route_loop`, and
that already exists.
DAMAGE KINDS
------------
`swap` and `wrong_side` are the evidence arms. `translate` is an INSTRUMENT CHECK
and is never aggregated: `perturb.block_direction` computes the damage direction
from `routability.block_displacements`, the very quantity this solve consumes, so
its recovery is partly arithmetic. `scatter` and `pile` are excluded by
measurement -- `perturb_jitter` skips illegal-pose parts so `scatter` often
damages nothing, and `pile` moves every free part so there is no "the block".
python3 -X utf8 tests/stress/block_relocation_study.py --out wk/554 \
--boards esp_prog splitflap_driver
python3 -X utf8 tests/stress/block_relocation_study.py --from-rows wk/554/rows.jsonl
python3 -X utf8 tests/stress/block_relocation_study.py --self-test
"""
from __future__ import annotations
import argparse
import json
import os
import shutil
import subprocess
import sys
import tempfile
_here = os.path.dirname(os.path.abspath(__file__))
ROOT = os.path.abspath(os.path.join(_here, '..', '..'))
for _p in (ROOT, os.path.join(ROOT, 'py_router'), os.path.join(ROOT, 'py_placer'),
os.path.join(ROOT, 'py_tools')):
if _p not in sys.path:
sys.path.insert(0, _p)
SCHEMA = 1
ARMS = ('C', 'D', 'R', 'R0', 'L')
EVIDENCE_KINDS = ('swap', 'wrong_side')
INSTRUMENT_KINDS = ('translate',)
EXCLUDED_KINDS = {'scatter': 'portfolio.perturb_jitter skips a part whose '
'incumbent pose is not fully legal, which on a '
'dense board is every part, so it often damages '
'nothing at all',
'pile': 'every free part moves, so there is no "the block"'}
IGNORE_NETS = ['GND', 'VCC', 'VDD', 'VSS', '+3V3', '+5V', '+1V8', '+1V1', 'GNDA',
'AGND', 'VBUS', '+3.3V']
GROUP_BY = 'kicad,sheet,netprefix,decap'
#: G1. `recovery.HOME_TOLERANCE_MM` is 2.0, so at a 2 mm dose every member is
#: already inside the home tolerance before anything is repaired and the home
#: curve has no rung to move through. 2x that is the smallest dose whose repair
#: is observable at all.
MIN_APPLIED_MM = 4.0
MIN_MEMBERS_MOVED = 2
ROUTE_TIMEOUT = 3600
LOOP_TIMEOUT = 5400
BOARDS = ['esp_prog', 'splitflap_driver', 'tigard']
def _run(argv, log_path, timeout):
with open(log_path, 'w', encoding='utf-8') as fh:
r = subprocess.run(argv, stdout=fh, stderr=subprocess.STDOUT, text=True,
encoding='utf-8', errors='replace', cwd=ROOT,
timeout=timeout)
with open(log_path, encoding='utf-8', errors='replace') as fh:
return r.returncode, fh.read()
def route_blocking(board, out_board, log_path, timeout=ROUTE_TIMEOUT):
"""`(blocking, detail)` from ONE route. `blocking` is None if it did not run.
Deliberately the router's own tally of what is not connected -- failed nets,
kept-but-open nets, and the multipoint pad deficit -- read off
`JSON_SUMMARY_MIN`. `open_single` is in it because a KEPT result whose pads
are still disconnected is how a board ships open copper while reporting
`failures=0` (CLAUDE.md's failure-bucket rule).
"""
argv = [sys.executable, '-X', 'utf8',
os.path.join(ROOT, 'py_router', 'route.py'), board, out_board,
'--nets', '*']
try:
rc, log = _run(argv, log_path, timeout)
except subprocess.TimeoutExpired:
return None, {'timed_out': True}
if rc != 0:
return None, {'rc': rc, 'tail': log.strip().splitlines()[-3:]}
line = [ln for ln in log.splitlines() if ln.startswith('JSON_SUMMARY_MIN: ')]
if not line:
return None, {'no_summary': True}
d = json.loads(line[-1].split('JSON_SUMMARY_MIN: ', 1)[1])
failed = d.get('failed') or 0
openc = len(d.get('open_single') or [])
deficit = d.get('multipoint_deficit') or 0
return failed + openc + deficit, {'failed': failed, 'open': openc,
'deficit': deficit,
'routed': d.get('routed')}
def pose_delta(a_board, b_board):
"""(max member move, moved count, per-ref deltas) between two boards.
The APPLIED damage is measured from the two boards, never from perturb's
record: `clipped` and `max_feasible_dose_mm` describe a rigid-translate
probe that only two of the arms clip against, and reading them as the dose
was wrong by 79x once (kit-dev-coldfire excluded at "1.404 mm landed" while
its members had moved 111 mm).
"""
from kicad_parser import parse_kicad_pcb
import math
a = parse_kicad_pcb(a_board).footprints
b = parse_kicad_pcb(b_board).footprints
out = {}
for ref, fa in a.items():
fb = b.get(ref)
if fb is None:
continue
d = math.hypot(fb.x - fa.x, fb.y - fa.y)
if d > 1e-6:
out[ref] = round(d, 4)
return (round(max(out.values()), 4) if out else 0.0), len(out), out
def score(board, record):
"""recovery.score_board, or None when it cannot run."""
try:
from placement import recovery
return recovery.score_board(board, record)
except Exception as e: # noqa: BLE001
return {'error': '%s: %s' % (type(e).__name__, e)}
def relocate_arm(board, out_board, work, label, refs, rounds=2,
corridor_mm=None):
"""place_route_loop --relocate, block named EXPLICITLY.
Naming the refs is what separates "the relocation worked" from "the
diagnosis picked right". #553's selector is a measured null, so folding it in
here would make a null unattributable to either half.
"""
argv = [sys.executable, '-X', 'utf8',
os.path.join(ROOT, 'py_placer', 'place_route_loop.py'),
board, out_board,
'--route-args', '--nets "*"',
'--work-dir', os.path.join(work, label),
'--rounds', str(rounds), '--no-movie',
'--relocate', '--group-by', GROUP_BY,
'--relocate-refs'] + list(refs)
if corridor_mm is not None:
argv += ['--relocate-max-corridor', str(corridor_mm)]
argv += ['--ignore-nets'] + IGNORE_NETS
try:
rc, log = _run(argv, os.path.join(work, label + '.log'), LOOP_TIMEOUT)
except subprocess.TimeoutExpired:
return {'timed_out': True}
line = [ln for ln in log.splitlines() if ln.startswith('JSON_SUMMARY: ')]
return {'rc': rc,
'summary': json.loads(line[-1].split('JSON_SUMMARY: ', 1)[1])
if line else {}}
def loop_arm(board, out_board, work, label, rounds=2):
"""The incumbent: pin gate lifted, blocks on, cap matched. #411's loop@allon.
Running only the SHIPPED defaults would make the comparison a tautology --
`nets_to_refs` drops anything over --max-target-pins and --group-by defaults
to none, so L0 structurally cannot move a block and "recovers nothing" would
be the documented behaviour of the stop-loss rather than a measurement.
"""
argv = [sys.executable, '-X', 'utf8',
os.path.join(ROOT, 'py_placer', 'place_route_loop.py'),
board, out_board,
'--route-args', '--nets "*"',
'--work-dir', os.path.join(work, label),
'--rounds', str(rounds), '--no-movie',
'--max-target-pins', '100000', '--group-by', GROUP_BY,
'--max-displacement', '6.0', '--step', '1.0']
argv += ['--ignore-nets'] + IGNORE_NETS
try:
rc, log = _run(argv, os.path.join(work, label + '.log'), LOOP_TIMEOUT)
except subprocess.TimeoutExpired:
return {'timed_out': True}
line = [ln for ln in log.splitlines() if ln.startswith('JSON_SUMMARY: ')]
return {'rc': rc,
'summary': json.loads(line[-1].split('JSON_SUMMARY: ', 1)[1])
if line else {}}
def run_cell(board, kind, out_dir, *, dose_mm=8.0, seed=1, rounds=2):
"""One (board, kind) cell, fenced. Returns a row dict, always."""
from placement import perturb as P
row = {'schema': SCHEMA, 'board': board, 'kind': kind,
'is_evidence': kind in EVIDENCE_KINDS,
'counted_in_primary': kind in EVIDENCE_KINDS}
if kind in EXCLUDED_KINDS:
row['skipped'] = 'excluded kind: ' + EXCLUDED_KINDS[kind]
return row
work = os.path.join(out_dir, '%s_%s' % (board, kind))
truth = os.path.join(out_dir, '_truth', '%s_%s' % (board, kind))
os.makedirs(work, exist_ok=True)
os.makedirs(truth, exist_ok=True)
src = os.path.join(ROOT, 'kicad_files', board + '.kicad_pcb')
damaged = os.path.join(work, 'board.kicad_pcb')
control = os.path.join(truth, 'control.kicad_pcb')
try:
rec = P.perturb(src, damaged, kind=kind, dose_mm=dose_mm, seed=seed,
group_by=GROUP_BY, ignore_nets=IGNORE_NETS,
write_record=False, control_out=control)
except Exception as e: # noqa: BLE001
row['skipped'] = 'perturb raised %s: %s' % (type(e).__name__, e)
return row
if rec.get('status') != 'ok':
row['skipped'] = 'perturb: %s' % (rec.get('reason') or rec.get('status'))
return row
members = sorted(rec['block']['members'])
row['block'] = rec['block']['name']
row['members'] = members
applied, moved, _d = pose_delta(control, damaged)
row.update({'applied_mm': applied, 'members_moved': moved})
# G1 -- free.
if applied < MIN_APPLIED_MM or moved < MIN_MEMBERS_MOVED:
row['skipped'] = ('unmeasurable_dose: %.2f mm over %d part(s); needs '
'>= %.1f mm over >= %d (below that every member is '
'inside recovery.HOME_TOLERANCE_MM before repair)'
% (applied, moved, MIN_APPLIED_MM, MIN_MEMBERS_MOVED))
return row
routed = {}
bC, dC = route_blocking(control, os.path.join(work, 'C_routed.kicad_pcb'),
os.path.join(work, 'C.log'))
bD, dD = route_blocking(damaged, os.path.join(work, 'D_routed.kicad_pcb'),
os.path.join(work, 'D.log'))
routed['C'], routed['D'] = {'blocking': bC, **dC}, {'blocking': bD, **dD}
row['routed'] = routed
if bC is None or bD is None:
row['skipped'] = 'a baseline route did not produce a summary'
return row
# G3 -- the RUNBOOK's own definition of a subject.
if not bD > bC:
row['skipped'] = ('not_placement_limited: damaged blocking %d is not '
'worse than the controlic %d, so there is nothing for '
'a placement repair to restore' % (bD, bC))
row['skipped'] = row['skipped'].replace('controlic', 'control')
return row
R = relocate_arm(damaged, os.path.join(work, 'R.kicad_pcb'), work, 'R',
members, rounds=rounds)
R0 = relocate_arm(control, os.path.join(work, 'R0.kicad_pcb'), work, 'R0',
members, rounds=rounds)
L = loop_arm(damaged, os.path.join(work, 'L.kicad_pcb'), work, 'L',
rounds=rounds)
row['arms'] = {'R': R, 'R0': R0, 'L': L}
for name, res, base in (('R', R, damaged), ('R0', R0, control),
('L', L, damaged)):
out = os.path.join(work, '%s.kicad_pcb' % name)
if not os.path.exists(out):
routed[name] = {'blocking': None, 'reason': 'arm produced no board'}
continue
b, d = route_blocking(out, os.path.join(work, '%s_routed.kicad_pcb' % name),
os.path.join(work, '%s_route.log' % name))
routed[name] = {'blocking': b, **d}
mx, mv, _ = pose_delta(base, out)
routed[name].update({'max_move_mm': mx, 'parts_moved': mv})
denom = bD - bC
# R and L repair the DAMAGED board, so they are normalised against its
# blocking. R0 repairs the CONTROL, and normalising it against `bD` would be
# arithmetic about a board it never touched -- measured on esp_prog/swap,
# that mistake reported R0 at a recovery of 1.0 (because C already routes
# clean) and turned a real +0.333 into a false -0.667. The undamaged arm's
# question is "how much does this solve change a board that was not damaged",
# normalised the same way so the two are subtractable.
for name in ('R', 'L'):
b = routed.get(name, {}).get('blocking')
routed[name]['route_recovery'] = (
None if b is None or denom == 0 else round((bD - b) / denom, 4))
b0 = routed.get('R0', {}).get('blocking')
routed['R0']['route_effect_on_control'] = (
None if b0 is None or denom == 0 else round((bC - b0) / denom, 4))
# THE evidence column: what the repair did to the damaged board, MINUS what
# the same repair does to an undamaged one.
rr = routed['R'].get('route_recovery')
r0 = routed['R0'].get('route_effect_on_control')
row['route_recovery'] = rr
row['route_effect_on_control'] = r0
row['route_recovery_delta'] = (None if rr is None or r0 is None
else round(rr - r0, 4))
row['loop_route_recovery'] = routed['L'].get('route_recovery')
# Contract checks -- these can FAIL a cell that routed better.
moved_R = routed.get('R', {}).get('parts_moved')
row['contract'] = {
'parts_moved_by_R': moved_R,
'block_members': len(members),
# Every part R moved that is not a block member is collateral. #554's
# definition is "move only diagnosed blocks"; the corridor is licensed,
# but it is reported so a reader can see the disturbance the routed
# number was bought with.
'collateral_parts': (None if moved_R is None
else max(0, moved_R - len(members))),
}
return row
def summarise(rows):
live = [r for r in rows if 'skipped' not in r]
ev = [r for r in live if r.get('counted_in_primary')
and r.get('route_recovery_delta') is not None]
deltas = sorted(r['route_recovery_delta'] for r in ev)
up = [r for r in ev if r['route_recovery_delta'] > 0]
down = [r for r in ev if r['route_recovery_delta'] < 0]
out = {
'schema': SCHEMA,
'cells': len(live),
'evidence_cells': len(ev),
'boards': sorted({r['board'] for r in ev}),
'median_route_recovery_delta': (
None if not deltas else
round(deltas[len(deltas) // 2] if len(deltas) % 2 else
(deltas[len(deltas) // 2 - 1] + deltas[len(deltas) // 2]) / 2, 4)),
'cells_up': len(up), 'cells_down': len(down),
# THE comparison that decides whether #554 is interesting at all. #411
# measured `loop@allon` taking tigard from 13 routing failures to 2; a
# relocation that ties that has demonstrated nothing, and one that loses
# to it has demonstrated something worse. Reported as a count of cells,
# never folded into the verdict -- the verdict is about the delta, and
# conflating the two would let a win over the loop hide a null against
# the undamaged control.
'cells_where_loop_beats_relocation': sum(
1 for r in ev
if r.get('loop_route_recovery') is not None
and r.get('route_recovery') is not None
and r['loop_route_recovery'] > r['route_recovery']),
'cells_with_a_loop_number': sum(
1 for r in ev if r.get('loop_route_recovery') is not None),
'instrument_cells': [r['board'] + '/' + r['kind'] for r in live
if r['kind'] in INSTRUMENT_KINDS],
'skipped': [{'board': r['board'], 'kind': r['kind'],
'reason': r['skipped']} for r in rows if 'skipped' in r],
}
out['verdict'] = _verdict(out, ev)
return out
def _verdict(s, ev):
"""The acceptance rule counts BOARDS, not cells, and names the incumbent.
Two cells on one board are not two boards. `docs/placement-predictors.md`'s
own rule is "right on >= N-1 BOARDS and wrong on none, with N >= 3", and the
per-board spread on this corpus is +-2..3 nets. Counting cells would have let
3 cells over 2 boards read as SUPPORTED -- which is exactly how a two-board
result becomes a default change, and this repo already records that trap.
And a positive delta is not sufficient on its own: #554 is only interesting
if it beats `place_route_loop` with the pin gate lifted, which already
exists. A run where the incumbent wins is reported as such however good the
delta looks.
"""
n = s['evidence_cells']
nb = len(s['boards'])
if n == 0:
return ('NOT MEASURED: no evidence cell survived the admission gates, so '
'this run says nothing about whether a relocation restores a '
'damaged board. The skipped list says why each one went.')
loop_note = ''
if s.get('cells_with_a_loop_number'):
loop_note = (' INCUMBENT: place_route_loop with the pin gate lifted '
'reached a strictly better routed result on %d of the %d '
'cell(s) where both ran.'
% (s['cells_where_loop_beats_relocation'],
s['cells_with_a_loop_number']))
if nb < 3:
return ('UNDERPOWERED: %d evidence cell(s) over %d board(s). The '
'acceptance rule counts BOARDS -- >= 3, right on N-1 and wrong '
'on none -- because per-board spread here is +-2..3 nets and two '
'cells on one board are not two boards. The direction (%d up, '
'%d down, median %s) is reported, not aggregated into a claim.%s'
% (n, nb, s['cells_up'], s['cells_down'],
s['median_route_recovery_delta'], loop_note))
if s['cells_down'] == 0 and s['cells_up'] >= n - 1:
return ('SUPPORTED: route_recovery_delta positive on %d of %d evidence '
'cell(s) over %d board(s) and negative on none, median %s. The '
'delta is the evidence -- the raw recovery is not, because the '
'same solve on the UNDAMAGED board is the other half of every '
'row.%s'
% (s['cells_up'], n, nb, s['median_route_recovery_delta'],
loop_note))
return ('NOT SUPPORTED: %d of %d evidence cell(s) improved and %d got worse, '
'median delta %s. A relocation that does not beat its own undamaged '
'pairing has not restored anything.%s'
% (s['cells_up'], n, s['cells_down'],
s['median_route_recovery_delta'], loop_note))
def self_test():
fails = []
# A cell whose damaged board is no worse than the control must be REPORTED,
# never silently dropped -- it is the RUNBOOK's definition of a non-subject.
s = summarise([{'board': 'b', 'kind': 'swap',
'skipped': 'not_placement_limited: x'}])
if not s['skipped']:
fails.append('a skipped cell vanished from the summary')
if 'NOT MEASURED' not in s['verdict']:
fails.append('a run with no evidence cell did not say so: %r'
% s['verdict'])
# The instrument arm must never reach the evidence aggregate.
rows = [{'board': 'b', 'kind': 'translate', 'counted_in_primary': False,
'is_evidence': False, 'route_recovery_delta': 9.0}]
s = summarise(rows)
if s['evidence_cells'] != 0:
fails.append('translate was aggregated as evidence')
if not s['instrument_cells']:
fails.append('the instrument cell was not reported at all')
# Under-powered must not read as a result.
rows = [{'board': 'b', 'kind': 'swap', 'counted_in_primary': True,
'is_evidence': True, 'route_recovery_delta': 1.0}]
if 'UNDERPOWERED' not in summarise(rows)['verdict']:
fails.append('one positive cell was reported as a finding')
# THREE cells on TWO boards is still two boards.
two = [dict(rows[0]), dict(rows[0], kind='wrong_side'),
dict(rows[0], board='c')]
if 'UNDERPOWERED' not in summarise(two)['verdict']:
fails.append('3 positive cells over 2 boards was not reported as '
'underpowered; the acceptance rule counts BOARDS')
# ... and a mixed result is NOT SUPPORTED, not "mostly works".
rows = [dict(rows[0]), dict(rows[0], board='c'),
dict(rows[0], board='d', route_recovery_delta=-1.0),
dict(rows[0], board='e')]
if 'NOT SUPPORTED' not in summarise(rows)['verdict']:
fails.append('a cell that got WORSE did not block the claim')
# The undamaged arm repairs the CONTROL, so it must be normalised against
# the control's own blocking. Normalising it against the damaged board's
# reported R0 at a recovery of 1.0 on a board that already routed clean, and
# turned a real +0.333 into a false -0.667 (esp_prog/swap, measured).
src = open(os.path.abspath(__file__), encoding='utf-8').read()
body = src.split(' denom = bD - bC', 1)[1].split('\n # Contract', 1)[0]
if 'route_effect_on_control' not in body:
fails.append('the undamaged arm no longer has its own normalisation')
if "for name in ('R', 'R0', 'L')" in body:
fails.append('R0 is being normalised against the DAMAGED board again: '
'that is arithmetic about a board it never touched')
for f in fails:
print('SELF-TEST FAIL: ' + f)
print("self-test: %s (5 check groups)" % ("FAIL" if fails else "ok"))
return 1 if fails else 0
def main(argv=None):
p = argparse.ArgumentParser(description=__doc__.split('\n')[0])
p.add_argument('--boards', nargs='+', default=BOARDS)
p.add_argument('--kinds', nargs='+',
default=list(EVIDENCE_KINDS) + list(INSTRUMENT_KINDS))
p.add_argument('--dose', type=float, default=8.0)
p.add_argument('--rounds', type=int, default=2)
p.add_argument('--seed', type=int, default=1)
p.add_argument('--out', default=None, metavar='DIR')
p.add_argument('--from-rows', default=None, metavar='JSONL')
p.add_argument('--self-test', action='store_true')
a = p.parse_args(argv)
if a.self_test:
return self_test()
if a.from_rows:
rows = [json.loads(x) for x in open(a.from_rows, encoding='utf-8')
if x.strip()]
print(json.dumps(summarise(rows), indent=2, sort_keys=True))
return 0
if self_test():
print('refusing to measure with a failing instrument')
return 1
out_dir = a.out or tempfile.mkdtemp(prefix='relocation_study_')
os.makedirs(out_dir, exist_ok=True)
rows = []
for board in a.boards:
for kind in a.kinds:
print('--- %s / %s' % (board, kind), flush=True)
row = run_cell(board, kind, out_dir, dose_mm=a.dose, seed=a.seed,
rounds=a.rounds)
rows.append(row)
if 'skipped' in row:
print(' SKIP %s' % row['skipped'][:150], flush=True)
else:
print(' blocking C=%s D=%s R=%s R0=%s L=%s delta=%s'
% tuple([row['routed'][k].get('blocking')
for k in ('C', 'D', 'R', 'R0', 'L')]
+ [row.get('route_recovery_delta')]), flush=True)
with open(os.path.join(out_dir, 'rows.jsonl'), 'a',
encoding='utf-8') as fh:
fh.write(json.dumps(row, sort_keys=True) + '\n')
summary = summarise(rows)
print()
print(json.dumps(summary, indent=2, sort_keys=True))
with open(os.path.join(out_dir, 'summary.json'), 'w', encoding='utf-8') as fh:
json.dump(summary, fh, indent=2, sort_keys=True)
return 0
if __name__ == '__main__':
sys.exit(main())
8b2a3278aThis file is pinned to the selected release. Download the original to open it in your design tool.