Remove temporary BORINGSSL_YYYYMM #ifdefs. #1166
Conversation
It seems to have compilation problems with BoringSSL:
And a plain old test failing with OpenSSL? There's actually multiple issues, it seems... |
That's maybe because I merged the DSA related PR yesterday? I think this Martin On 5 May 2016 at 11:09, Pierre Phaneuf notifications@github.com wrote:
|
His branch is from right before. He's going to have to rebase it and fix the conflicts, for sure. |
Yeah, I can rebase it. The patch pretty clearly cannot affect OpenSSL though. The failure seems to be
which looks like an infrastructure flake. Unless there's a failure elsewhere I can't find. (Your build logs are rather insane long... you might want to improve the error-reporting here.) The BoringSSL failure is because you're pinned against a much older BoringSSL than the internal repository. What else uses this code with BoringSSL? If you legitimately need to support something older, we can keep the temporarily #ifdefs around for you. If not, we need to point |
The logs are rather insane long "because gclient (and OpenSSL and BoringSSL and libICU)". 😞 If you search for "exited with", you'll find the A bit above, you'll find some relevant error messages. Use the same "exited with" technique with this other one, it has a build failure somewhere a bit before: https://travis-ci.org/google/certificate-transparency/jobs/126699939 We use You shouldn't need to maintain anything in BoringSSL, because we're just "stuck" on some arbitrary older version that happened to work when we added it. 😉 |
I see. Thanks! So it's
Has that test flaked before, or is this the first time? The CL should only have an effect in Alright, well, if you all don't mind my bumping (To be honest, that arrangement hasn't worked out that well because of this kind of skew. You actually aren't on inherently that old of a BoringSSL---what Chrome is shipping---but Chrome snapshots the consumer atomically with BoringSSL. It might be worth adding a single |
I've seen tests in etcd flake because they rely on hardcoded sleeps. I The massive build log is a real pain but not sure there's an easy solution. Martin On 5 May 2016 at 14:55, David Benjamin notifications@github.com wrote:
|
The build failure of the ASan one was that flaky test, I just re-ran it and it passed. |
LGTM for the code changes, assuming they haven't broken Travis (seems like it was flakiness). |
David, mind re-basing ? If the rebased commit passes travis and nobody else has any objections, I'll merge. |
@davidben ping - please rebase... |
Sync to the BoringSSL revision used in current Chromium stable.
Hrmf. So I rebased and bumped BoringSSL up to a new enough revision, but Travis is unhappy: It looks like it's hitting the case for where your
But since you all cache the build directory, it doesn't end up doing that. That said, I'm rather puzzled because we don't have this problem in Chromium. Are you all doing something weird to |
I cleared the Travis caches and triggered a rebuild, which looks to have worked OK. |
@davidben happy for me to merge, then? |
If the tests pass, sure. Sorry I took to long to revisit this CL. |
We can assume a new enough BoringSSL now.