TLS-Anvil 1.5 possible bug relating to ECC curve selection

[randombit]

In Botan we recently added support for the TLS 1.3 Brainpool point identifiers, and this seems to have broken TLS-Anvil 1.5 in some way. Backtrace

2026-06-25T04:59:42.4229568Z �[m�[1;31m04:59:42: Test failed without AssertionError [5] Combination{INCLUDE_EXTENDED_MASTER_SECRET_EXTENSION=INCLUDE_EXTENDED_MASTER_SECRET_EXTENSION = true, TCP_FRAGMENTATION=TCP_FRAGMENTATION = false, CIPHER_SUITE=CIPHER_SUITE = TLS_ECDHE_ECDSA_WITH_AES_128_CBC_SHA, INCLUDE_ENCRYPT_THEN_MAC_EXTENSION=INCLUDE_ENCRYPT_THEN_MAC_EXTENSION = true, CERTIFICATE=Public Key Type: ECDH_ECDSA,Named Curve: SECP384R1,Certificate Signature Type: SHA256_WITH_RSA_ENCRYPTION, SIG_HASH_ALGORIHTM=SIG_HASH_ALGORIHTM = ECDSA_SHA512, NAMED_GROUP=NAMED_GROUP = BRAINPOOLP512R1TLS13, RECORD_LENGTH=RECORD_LENGTH = 16384}
2026-06-25T04:59:42.4234743Z 
2026-06-25T04:59:42.4236546Z java.lang.NullPointerException: Cannot invoke "de.rub.nds.tlsscanner.serverscanner.probe.invalidcurve.point.InvalidCurvePoint.getPublicPointBaseX()" because "invalidCurvePoint" is null
2026-06-25T04:59:42.4239223Z 	at de.rub.nds.tlstest.suite.tests.client.tls12.rfc8422.TLSExtensionForECC.rejectsInvalidCurvePoints(TLSExtensionForECC.java:161)
2026-06-25T04:59:42.4241836Z 	at java.base/
...

Full log is here: https://productionresultssa1.blob.core.windows.net/actions-results/06b01599-d02f-4e8b-b2df-3dd93dc92fd2/workflow-job-run-eeaedcd9-dcd8-5cb5-9734-867135927676/logs/job/job-logs.txt

[mmaehren]

Hey,

I couldn't access the log (Public access is not permitted on this storage account.). We just released version 1.5.2 which should address the specific issue shown in the log excerpt. Did any other tests fail for brainpool curves? The missing Invalid Curve test vector should not affect any other ECC related tests.

[randombit]

I'll try 1.5.2 asap. If I had to guess the problem was that when we added the TLS 1.3 Brainpool ids, we started being willing to negotiate every curve ID TLS-Anvil thought to try, it took its internal list and subtracted the curves that were offered, the set was empty leading to the null deref (but I haven't looked at the code, this is just a guess). In the short term I resolved it by modifying the policy we use for TLS-Anvil testing to skip the 1.3 ids.

[mmaehren]

The specific issue shown above was due to how we handle Invalid Curve tests. For each curve, we list points that are known to produce small subgroups of prime order on a different curve. This list simply missed that TLS 1.3 brainpool curve definitions have been added and that the points we have for legacy brainpool groups can be re-used. So the code returned null and the test template tried to deref it. This is why I would assume that other tests should not be affected. If other tests fail for brainpool, it should be a separate problem.