Closed twata1 closed 3 years ago
Thanks for the input. I'm not doubting that the proposed change fixes the problem in your specific environment. But I'm not comfortable with adding some sleep somewhere in the code with no clear explanation on why it fixes the problem. Since I cannot explain why this should fix the problem I am not confident that it actually fixes the problem for real - it might work for you but not for other cases, on other machines, ...
Given that t/sni_verify.t should basically the same as t/sni.t only with client certificates I've adapted in 3036e0bd6f85d0dafc02d8c815a1041c182b9792 the slightly more robust and verbose code from sni.t. Maybe this makes sni_verify.t more stable.
Thanks for the response. For your information, for that pull request, in fact, the test passes even if I change the sleep statement to a warn statement. I don't even know why.
Thank you,
For your information, for that pull request, in fact, the test passes even if I change the sleep statement to a warn statement. I don't even know why.
I would imagine that the output, sleep ... causes somewhere a context switch and this "stabilizes" the test. Such output should be in the changed test too, so maybe it helps.
I have confirmed that sni_verify.t passes with the Strawberry Perl versions:
Strawberry Perl 64 bit v5.30.2 v5.28.2 v5.26.1 v5.24.4 v5.22.3 v5.20.3 v5.18.4 v5.16.3
Strawberry Perl 32 bit v5.26.1 v5.24.0 v5.22.2 v5.20.3 v5.18.4 v5.16.3 v5.14.4