1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376
|
# Copyright (c) 2019-2025, Saransh Chopra, Henry Schreiner, Eduardo Rodrigues, Jonas Eschle, and Jim Pivarski.
#
# Distributed under the 3-clause BSD license, see accompanying file LICENSE
# or https://github.com/scikit-hep/vector for details.
"""
Ensures that new or modified vector.compute.* functions don't break any existing
or future backends by using unsupportable Python language features.
Compute functions are highly restricted, a least common denominator for all the
backends we *ever* want to support. The functions themselves are duck-typed:
arguments could be numbers, NumPy arrays, Awkward Arrays, and potentially
TensorFlow/Torch/JAX/etc. An ``if`` statement on individual numbers would have
to be ``np.where` or a masked assignment in NumPy, so ``if`` is not allowed.
JAX traces a function for JIT-compilation and autodifferentiation by passing a
"tracer" object through it, and that object can only follow one code path, another
reason to exclude ``if`` statements. Loops are even more problematic.
This suite of tests statically analyzes all of the compute functions by decompiling
their bytecode with uncompyle6 (on Python 3.8; will have to be modified slightly
every few years). Some compute functions are dynamically generated, so they don't
all have an AST to inspect.
The sieve has been defined narrowly: compute functions can use more functions,
binary operators, and possibly more language features than are allowed here.
Expanding this set of rules is therefore allowed and encouraged. The test failure
and requirement to expand the rules is intended to force you to think about
new features, to ask yourself if they can be supported by all current and hoped-for
backends, and whether a (formally) simpler implementation is possible.
"""
from __future__ import annotations
import collections
import contextlib
import inspect
import sys
import pytest
import vector._compute.lorentz
import vector._compute.planar
import vector._compute.spatial
uncompyle6 = pytest.importorskip("uncompyle6")
spark_parser = pytest.importorskip("spark_parser")
pytestmark = pytest.mark.dis
Context = collections.namedtuple("Context", ["name", "closure"])
functions = dict(
[
(
f"{y.__name__}({', '.join(repr(v) if isinstance(v, str) else v.__name__ for v in w)})",
z[0],
)
for x, y in inspect.getmembers(
vector._compute.planar, predicate=inspect.ismodule
)
if hasattr(y, "dispatch_map")
for w, z in y.dispatch_map.items()
]
+ [
(
f"{y.__name__}({', '.join(repr(v) if isinstance(v, str) else v.__name__ for v in w)})",
z[0],
)
for x, y in inspect.getmembers(
vector._compute.spatial, predicate=inspect.ismodule
)
if hasattr(y, "dispatch_map")
for w, z in y.dispatch_map.items()
]
+ [
(
f"{y.__name__}({', '.join(repr(v) if isinstance(v, str) else v.__name__ for v in w)})",
z[0],
)
for x, y in inspect.getmembers(
vector._compute.lorentz, predicate=inspect.ismodule
)
if hasattr(y, "dispatch_map")
for w, z in y.dispatch_map.items()
]
)
python_version = f"{sys.version_info[0]}.{sys.version_info[1]}"
is_pypy = "__pypy__" in sys.builtin_module_names
try:
uncompyle6.scanner.get_scanner(python_version, is_pypy=is_pypy)
except RuntimeError as err:
is_unsupported = True
unsupported_message = str(err)
else:
is_unsupported = False
unsupported_message = ""
@pytest.mark.skipif(is_unsupported, reason=unsupported_message)
@pytest.mark.slow
@pytest.mark.parametrize("signature", functions.keys())
def test(signature):
analyze_function(functions[signature])
def analyze_function(function):
if function not in analyze_function.done:
closure = dict(function.__globals__)
if function.__closure__ is not None:
for var, cell in zip(function.__code__.co_freevars, function.__closure__):
# the cell has not been filled yet, so ignore it
with contextlib.suppress(ValueError):
closure[var] = cell.cell_contents
analyze_code(function.__code__, Context(function.__name__, closure))
analyze_function.done.add(function)
analyze_function.done = set()
def analyze_code(code, context):
# this block is all uncompyle6
parser = uncompyle6.parser.get_python_parser(
python_version,
debug_parser=dict(spark_parser.DEFAULT_DEBUG),
compile_mode="exec",
is_pypy=is_pypy,
)
scanner = uncompyle6.scanner.get_scanner(python_version, is_pypy=is_pypy)
tokens, customize = scanner.ingest(code, code_objects={}, show_asm=False)
parsed = uncompyle6.parser.parse(parser, tokens, customize, code)
# now the disassembled bytecodes have been parsed into a tree for us to walk
analyze_body(parsed, context)
def analyze_body(node, context):
assert node.kind == "stmts"
assert len(node) >= 1
for statement in node[:-1]:
analyze_assignment(statement, context)
analyze_return(node[-1], context)
def analyze_assignment(node, context):
assert node.kind == "sstmt"
assert len(node) == 1
assert node[0].kind == "assign", (
"only assignments and a final 'return' are allowed (and not tuple-assignment)"
)
assert len(node[0]) == 2
assert node[0][1].kind == "store"
if node[0][1][0].kind == "STORE_FAST":
analyze_expression(expr(node[0][0]), context)
elif node[0][1][0].kind == "unpack":
assert len(node[0][1][0]) >= 2
assert node[0][1][0][0].kind.startswith("UNPACK_SEQUENCE")
for item in node[0][1][0][1:]:
assert item.kind == "store"
assert len(item) == 1
assert item[0].kind == "STORE_FAST"
else:
print(node[0][1][0])
raise AssertionError("what is this?")
def expr(node):
assert node.kind == "expr"
assert len(node) == 1
return node[0]
def is_pi(node):
return (
node.kind == "attribute"
and len(node) == 2
and expr(node[0]).kind == "LOAD_FAST"
and expr(node[0]).attr == "lib"
and node[1].kind == "LOAD_ATTR"
and node[1].attr == "pi"
)
def is_nan_to_num(node):
if node.kind != "call_kw36" or len(node) < 3:
return False
function = expr(node[0])
return (
function.kind == "attribute"
and expr(function[0]).attr == "lib"
and function[1].attr == "nan_to_num"
)
def analyze_return(node, context):
assert node.kind == "sstmt"
assert len(node) == 1
assert node[0].kind == "return", "compute function must end with a 'return'"
assert len(node[0]) == 2
assert node[0][0].kind in ("ret_expr", "return_expr")
assert len(node[0][0]) == 1
expr(node[0][0][0])
assert node[0][1].kind == "RETURN_VALUE"
if node[0][0][0][0].kind == "tuple":
assert len(node[0][0][0][0]) >= 2, "returning an empty tuple?"
assert node[0][0][0][0][-1].kind.startswith("BUILD_TUPLE")
for item in node[0][0][0][0][:-1]:
analyze_expression(expr(item), context)
else:
analyze_expression(node[0][0][0][0], context)
def analyze_expression(node, context):
if node.kind == "LOAD_FAST":
# Don't bother checking to see if this variable has been defined.
# Unit checks test that if the coverage is complete.
pass
elif node.kind == "LOAD_CONST":
assert isinstance(node.attr, (int, float))
elif is_pi(node):
pass
elif node.kind == "unary_op":
assert len(node) == 2
analyze_expression(expr(node[0]), context)
assert node[1].kind == "unary_operator"
assert len(node[1]) == 1
analyze_unary_operator(node[1][0], context)
elif node.kind == "bin_op":
assert len(node) == 3
analyze_expression(expr(node[0]), context)
analyze_expression(expr(node[1]), context)
assert node[2].kind == "binary_operator"
assert len(node[2]) == 1
analyze_binary_operator(node[2][0], context)
elif node.kind == "compare":
assert len(node) == 1
assert node[0].kind == "compare_single", "only do single comparisons"
assert len(node[0]) == 3
analyze_expression(expr(node[0][0]), context)
analyze_expression(expr(node[0][1]), context)
assert node[0][2].kind == "COMPARE_OP"
assert node[0][2].attr in allowed_comparisons, (
f"add {node[0][2].attr!r} to allowed_comparisons"
)
elif node.kind == "call":
assert len(node) >= 2
assert node[-1].kind.startswith("CALL_METHOD") or node[-1].kind.startswith(
"CALL_FUNCTION"
)
analyze_callable(expr(node[0]), context)
for argument in node[1:-1]:
expr_arg = argument[0] if argument.kind == "pos_arg" else argument
assert expr_arg.kind == "expr", "only positional arguments"
analyze_expression(expr(expr_arg), context)
elif is_nan_to_num(node):
analyze_expression(expr(node[1]), context)
else:
print(node)
raise AssertionError("what is this?")
def analyze_unary_operator(node, context):
assert node.kind in allowed_unary_operators, (
f"add {node.kind!r} to allowed_unary_operators"
)
def analyze_binary_operator(node, context):
assert node.kind in allowed_binary_operators, (
f"add {node.kind!r} to allowed_binary_operators"
)
def analyze_callable(node, context):
if node.kind == "attribute37":
assert len(node) == 2
module = expr(node[0])
assert module.kind in {"LOAD_FAST", "LOAD_GLOBAL"}
assert node[1].kind == "LOAD_METHOD"
if module.attr == "lib":
assert node[1].attr in allowed_lib_functions, (
f"add {node[1].attr!r} to allowed_lib_functions"
)
else:
module_name = ".".join(
context.closure.get(module.attr).__name__.split(".")[:-1]
)
assert module_name in (
"vector._compute.planar",
"vector._compute.spatial",
"vector._compute.lorentz",
)
elif node.kind in {"LOAD_GLOBAL", "LOAD_DEREF"}:
function = context.closure.get(node.attr)
assert function is not None, f"unrecognized function in scope: {node.attr!r}"
analyze_function(function)
else:
print(node)
raise AssertionError("what is this?")
allowed_unary_operators = [
"UNARY_NEGATIVE",
]
allowed_binary_operators = [
"BINARY_ADD",
"BINARY_SUBTRACT",
"BINARY_MULTIPLY",
"BINARY_TRUE_DIVIDE",
"BINARY_MODULO",
"BINARY_POWER",
"BINARY_AND",
]
allowed_comparisons = [
"==",
"!=",
"<",
">",
]
allowed_lib_functions = [
"absolute",
"sign",
"copysign",
"maximum",
"minimum",
"sqrt",
"exp",
"log",
"sin",
"cos",
"tan",
"arcsin",
"arccos",
"arctan",
"arctan2",
"sinh",
"cosh",
"tanh",
"arcsinh",
"arccosh",
"arctanh",
"isclose",
# TODO: https://github.com/scikit-hep/vector/issues/615
# remove once https://github.com/cupy/cupy/issues/9143
# is fixed.
"where",
]
|