test_socket control-message truncation tests fail on OpenBSD and DragonFly

[serhiy-storchaka]

Bug report

Several recvmsg() control-message truncation tests in test_socket fail on OpenBSD 7 and DragonFly 6:

testCmsgTrunc1, testCmsgTrunc2Int, testCmsgTruncLen0, testCmsgTruncLen0Minus1,
testCmsgTruncLen0Plus1, testCmsgTruncLen1, testCmsgTruncLen2Minus1,
testSingleCmsgTrunc1, testSingleCmsgTrunc2Int, testSingleCmsgTruncInData,
testSingleCmsgTruncLen0Minus1, testSecondCmsgTrunc1, testSecondCmsgTrunc2Int,
testSecondCmsgTruncInData, testSecondCmsgTruncLen0Minus1

The tests pass an ancillary-data buffer deliberately too small and expect recvmsg() to return the data with MSG_CTRUNC set. On OpenBSD and DragonFly the call instead fails with OSError: [Errno 40] Message too long (EMSGSIZE):

  File ".../Lib/test/test_socket.py", line 3427, in doRecvmsg
    result = sock.recvmsg_into([buf], *args)
OSError: [Errno 40] Message too long

This is a kernel behavioral difference: these BSDs reject an undersized control buffer rather than truncating. NetBSD is unaffected (the truncation tests pass there, see gh-125860). FreeBSD and Linux are unaffected.

The tests should tolerate EMSGSIZE (or be skipped) on OpenBSD and DragonFly.

Linked PRs

• gh-154240: Skip cmsg truncation socket tests on OpenBSD and DragonFly #154244
• [3.13] gh-154240: Skip cmsg truncation socket tests on OpenBSD and DragonFly (GH-154244) #154270
• [3.14] gh-154240: Skip cmsg truncation socket tests on OpenBSD and DragonFly (GH-154244) #154285
• [3.15] gh-154240: Skip cmsg truncation socket tests on OpenBSD and DragonFly (GH-154244) #154286

[serhiy-storchaka]

This revives gh-64868 (bpo-20669), which reported the same OpenBSD recvmsg() EMSGSIZE failure in 2014 and was closed in 2017 only because there was no OpenBSD buildbot or reviewer, not because it was fixed. The behavior is now verified on OpenBSD 7 and DragonFly 6 (NetBSD is unaffected — its truncation tests pass).