[nivednarayan]

Fixes #1113

While looking at this issue I found that brute_force_decomp_helper keeps track of used terms in a uint64_t mask, but it was updating it with 1 << start and 1 << k. Those shift a 32-bit int, so once the shift count reaches 31 or more the result is undefined or gets sign-extended and the mask is wrong. In practice, with decompose_errors=True, an error touching 33 or more detectors would lose its higher targets (D32 in the example from the issue).

The fix is just changing those three shifts to uint64_t{1} << .... The existing check that rejects problems with 64 or more terms is untouched, so the biggest shift is 62 and stays in range. The fix and the regression test follow what was suggested in the issue.

I added test_decomposition_preserves_large_single_qubit_error to circuit_pybind_test.py. It builds the circuit from the issue for 32, 33 and 63 measurements and checks that the decomposed 0.25 error still flips every detector. Without the fix it fails at 33 (AssertionError: 33, mask is 0xFFFFFFFF so D32 is missing), and with the fix it passes.

Tests I ran:

• python -m pytest src/stim/circuit src/stim/simulators: 222 passed
• ./build/out/stim_test --gtest_filter='ErrorAnalyzer.*': 57 passed

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[Strilanc]

nice