You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
test_ssl fails with forced FIPS mode -- ssl.SSLError: [SSL] internal error (_ssl.c:1176) #161932
It seems the failing tests and the feature did not exist before that commit, so this is not a regression per se.
I am out of my expertise here, but this might be a problem in OpenSSL (it picks X25519 and that fails, while it could pick something else as default (I guess, not really sure)).
ECH hasn't been included in FIPS mode so far (in OpenSSL). There is support for p256 etc in ECH but not sure that just picking that would fix the issue (I've no clue how FIPS mode is handled in python builds, sorry).
OSSL_HPKE_get_grease_value on FIPS picks P-384 with NULL but X25519 with OSSL_HPKE_SUITE_DEFAULT.
Picking something FIPS-compatible should fix the issue, but I have no idea if picking something other than X25519 is even a reasonable choice.
As far as Python is concerned, we could skip the tests when FIPS is detected. Ideally with a link to some issue or similar, that tracks ECH in FIPS mode in OpenSSL.
OSSL_HPKE_get_grease_value on FIPS picks P-384 with NULL but X25519 with OSSL_HPKE_SUITE_DEFAULT.
Picking something FIPS-compatible should fix the issue, but I have no idea if picking something other than X25519 is even a reasonable choice.
Yep, you can safely choose p256 or p384 rather than x25519, esp for GREASEing.
As far as Python is concerned, we could skip the tests when FIPS is detected. Ideally with a link to some issue or similar, that tracks ECH in FIPS mode in OpenSSL.
I don't think there is an issue;-) I'll check and make one if not.
I created #161939 to skip the two test_ssl tests failing in FIPS mode, with a link to the OpenSSL issue openssl/openssl#33724 that @sftcd just created (thanks!).
Activity
hroncok commentedon Nov 13, 2026
46edd6f is the bisected commit
hroncok commentedon Nov 13, 2026
It seems the failing tests and the feature did not exist before that commit, so this is not a regression per se.
I am out of my expertise here, but this might be a problem in OpenSSL (it picks X25519 and that fails, while it could pick something else as default (I guess, not really sure)).
sftcd commentedon Nov 13, 2026
ECH hasn't been included in FIPS mode so far (in OpenSSL). There is support for p256 etc in ECH but not sure that just picking that would fix the issue (I've no clue how FIPS mode is handled in python builds, sorry).
hroncok commentedon Nov 13, 2026
OSSL_HPKE_get_grease_value on FIPS picks P-384 with NULL but X25519 with OSSL_HPKE_SUITE_DEFAULT.
Picking something FIPS-compatible should fix the issue, but I have no idea if picking something other than X25519 is even a reasonable choice.
As far as Python is concerned, we could skip the tests when FIPS is detected. Ideally with a link to some issue or similar, that tracks ECH in FIPS mode in OpenSSL.
sftcd commentedon Nov 13, 2026
Yep, you can safely choose p256 or p384 rather than x25519, esp for GREASEing.
I don't think there is an issue;-) I'll check and make one if not.
hroncok commentedon Nov 13, 2026
Thanks
vstinner commentedon Nov 13, 2026
I created #161939 to skip the two test_ssl tests failing in FIPS mode, with a link to the OpenSSL issue openssl/openssl#33724 that @sftcd just created (thanks!).
gh-161932: Skip test_ssl ECH test with FIPS enabled (#161939)