mirror of
https://github.com/github/codeql.git
synced 2026-07-29 14:56:51 +02:00
9b4de2bfb6
- Both the old and new CFG model this incorrectly right now. - Added ConsecutivePredecessorTimestamps.ql for the old CFG for symmetry.
38 lines
1.3 KiB
Python
38 lines
1.3 KiB
Python
"""Comparisons and evaluation order.
|
|
|
|
Comparison operands are evaluated left-to-right, followed by the comparison
|
|
node itself. Annotations record the *run-time* evaluation order, so this file
|
|
self-validates under CPython (``python3 test_comparison.py``).
|
|
|
|
Python short-circuits *chained* comparisons (in ``a < b < c`` the operand ``c``
|
|
is skipped once ``a < b`` is false), whereas the control-flow graph
|
|
conservatively evaluates every operand. That divergence cannot be expressed in
|
|
a single annotation, so the affected CFG queries report it (captured in the
|
|
ConsecutiveTimestamps and NewCfgConsecutivePredecessorTimestamps .expected
|
|
files).
|
|
"""
|
|
|
|
from timer import test, dead
|
|
|
|
|
|
@test
|
|
def test_two(t):
|
|
# Both operands, then the comparison.
|
|
(1 @ t[0] < 0 @ t[1]) @ t[2]
|
|
|
|
|
|
@test
|
|
def test_three(t):
|
|
# Chained comparison, all comparisons hold: operands left-to-right, then
|
|
# the comparison. Run time and CFG agree.
|
|
(1 @ t[0] < 2 @ t[1] < 3 @ t[2]) @ t[3]
|
|
|
|
|
|
@test
|
|
def test_three_short_circuit(t):
|
|
# ``1 > 2`` is false, so at run time Python short-circuits and never
|
|
# evaluates ``3``; the comparison is reached at timestamp 2. The CFG still
|
|
# evaluates ``3`` (dead at run time), which is why the CFG queries report a
|
|
# gap around this comparison.
|
|
(1 @ t[0] > 2 @ t[1] < 3 @ t[dead(2)]) @ t[2]
|