diff options
Diffstat (limited to 'research/experiments/044-specs/json_tree.py')
| -rw-r--r-- | research/experiments/044-specs/json_tree.py | 164 |
1 files changed, 164 insertions, 0 deletions
diff --git a/research/experiments/044-specs/json_tree.py b/research/experiments/044-specs/json_tree.py new file mode 100644 index 0000000..8807cdc --- /dev/null +++ b/research/experiments/044-specs/json_tree.py @@ -0,0 +1,164 @@ +# A JSON text read into a tree and walked breadth first for a digest: the design of aup's examples/json_tree.aup in +# plain Python (the reference 140 counts its lines against; the tree is Python's own values, a dict kept as pairs). +def space(b): + return b in (0x20, 0x09, 0x0a, 0x0d) + + +def digit(b): + return 0x30 <= b <= 0x39 + + +def hexdigit(b): + return digit(b) or 0x61 <= b <= 0x66 or 0x41 <= b <= 0x46 + + +class Pairs(list): + pass + + +def read(d): + try: + d.decode('utf-8') + except UnicodeDecodeError: + return None + n, i, ok, mode, root, open_, v = len(d), 0, True, 0, [], [], [] + while ok: + while i < n and space(d[i]): + i += 1 + if i == n: + ok = mode == 1 and not open_ + break + b = d[i] + if mode == 1: + if not open_: + ok = False + elif b == 0x2c: + mode = 2 if open_[-1]['object'] else 0 + elif b == 0x5d and not open_[-1]['object']: + v.append(open_[-1]['items']) + elif b == 0x7d and open_[-1]['object']: + v.append(Pairs(open_[-1]['members'])) + else: + ok = False + if v: + open_.pop() + i += 1 + elif mode == 3 and b == 0x7d or mode == 4 and b == 0x5d: + v.append(Pairs(open_[-1]['members']) if b == 0x7d else open_[-1]['items']) + open_.pop() + i += 1 + elif mode in (2, 3): + ok = b == 0x22 + mode = 5 + elif b in (0x5b, 0x7b): + open_.append({'items': [], 'members': [], 'name': '', 'object': b == 0x7b}) + mode = 4 if b == 0x5b else 3 + i += 1 + elif b == 0x22 or mode == 5: + i += 1 + s = bytearray() + while ok and i < n and d[i] != 0x22: + start = i + while i < n and d[i] != 0x22 and d[i] != 0x5c and d[i] >= 0x20: + i += 1 + s += d[start:i] + if i < n and d[i] < 0x20: + ok = False + elif i + 1 < n and d[i] == 0x5c: + e = d[i + 1] + if e == 0x75: + ok = i + 5 < n and all(hexdigit(x) for x in d[i + 2:i + 6]) + u = int(d[i + 2:i + 6], 16) if ok else 0 + i += 6 + if 0xd800 <= u < 0xdc00 and i + 5 < n and d[i] == 0x5c and d[i + 1] == 0x75 and all(hexdigit(x) for x in d[i + 2:i + 6]): + low = int(d[i + 2:i + 6], 16) + if 0xdc00 <= low < 0xe000: + u = 0x10000 + ((u - 0xd800) << 10 | low - 0xdc00) + i += 6 + s += chr(u).encode('utf-8', 'surrogatepass') + else: + ok = e in b'"\\/bfnrt' + s += {0x62: b'\b', 0x66: b'\f', 0x6e: b'\n', 0x72: b'\r', 0x74: b'\t'}.get(e, bytes([e])) + i += 2 + elif i < n and d[i] == 0x5c: + ok = False + ok = ok and i < n + i += 1 + try: + t = s.decode('utf-8') + except UnicodeDecodeError: + return None + if mode == 5: + open_[-1]['name'] = t + while i < n and space(d[i]): + i += 1 + ok = ok and i < n and d[i] == 0x3a + i += 1 + mode = 0 + else: + v.append(t) + elif b == 0x2d or digit(b): + start = i + if b == 0x2d: + i += 1 + while i < n and (digit(d[i]) or d[i] in b'.eE+-'): + i += 1 + text = d[start:i].decode() + import re + if not re.fullmatch(r'-?(0|[1-9]\d*)(\.\d+)?([eE][+-]?\d+)?', text) or float(text) in (float('inf'), -float('inf')): + return None + v.append(float(text)) + else: + word = b'true' if b == 0x74 else b'false' if b == 0x66 else b'null' + ok = d[i:i + len(word)] == word + i += len(word) + if ok: + v.append({b'true': True, b'false': False, b'null': None}[word]) + if ok and v: + if not open_: + ok = not root + root.append(v[-1]) + elif open_[-1]['object']: + open_[-1]['members'].append((open_[-1]['name'], v[-1])) + else: + open_[-1]['items'].append(v[-1]) + v.pop() + mode = 1 + return root if ok else None + + +def digest(tree): + d = dict(nulls=0, trues=0, falses=0, numbers=0, strings=0, arrays=0, objects=0, sum=0.0, text_bytes=0, key_bytes=0) + q, at = [tree], 0 + while at < len(q): + x = q[at] + if x is None: d['nulls'] += 1 + elif x is True: d['trues'] += 1 + elif x is False: d['falses'] += 1 + elif isinstance(x, float): d['numbers'] += 1; d['sum'] += x + elif isinstance(x, str): d['strings'] += 1; d['text_bytes'] += len(x.encode('utf-8')) + elif isinstance(x, Pairs): + d['objects'] += 1 + for k, y in x: d['key_bytes'] += len(k.encode('utf-8')); q.append(y) + else: d['arrays'] += 1; q.extend(x) + at += 1 + return d + + +if __name__ == '__main__': + import json, os, sys, time + suite = sys.argv[1] + for name in sorted(os.listdir(suite)): + tree = read(open(os.path.join(suite, name), 'rb').read()) + assert not name.startswith('y_') or tree is not None, name + assert not name.startswith('n_') or tree is None, name + big = b'[' + b','.join(b'{"key%d": [1.5e3, true, null, "text \\u00e9 %d"]}' % (k, k) for k in range(2000)) + b']' + runs, loads = [], [] + for _ in range(7): + start = time.perf_counter() + digest(read(big)[0]) + runs.append((time.perf_counter() - start) * 1e9 / len(big)) + start = time.perf_counter() + digest(json.loads(big, object_pairs_hook=Pairs)) + loads.append((time.perf_counter() - start) * 1e9 / len(big)) + print('{"size":%d,"python_ns_per_byte":%.1f,"python_json_module_ns_per_byte":%.1f}' % (len(big), sorted(runs)[3], sorted(loads)[3])) |