|
From: | Ivan Vučica |
Subject: | Re: Failing tests on MSVC in libs-base |
Date: | Wed, 3 Apr 2024 09:21:56 +0100 |
I saw this: this was recently addressed by replacing probe of example.org/silly-url-test because example.org started returning 503 or similar for that URL instead of 404. Incoming PR addressed this on every other platform by replacing example.org with another piece of infrastructure we don’t fully control, gnustep.github.io, so I merged it. (As I mentioned on that PR, worse solution would be if it depended on our infrastructure: DNS, Nginx, … on gnustep.org. Then any flake on home.gnustep.org or with Gandi for DNS would also start people’s tests failing.) Tests should not be trying to access network as part of CI. The right fix would be to spin up a temporary server on localhost etc. That itself requires carefully doing process management on all supported platforms. Or turn it from an integration test into a unit test (verify that sending and reading back HTTP over a Unix socket / windows pipe within the same process works, or similar without using OS primitives at all). Why does it fail for windows+clang? No clue. Maybe CI for windows actually blocks outgoing requests completely? This would be the cause only if we have more tests dependent on the network and they’re also failing. Maybe the CORRECT fix is not to do the test at all, at least in Ci? Sent from my iPhone On 3 Apr 2024, at 05:17, Gregory Casamento <greg.casamento@gmail.com> wrote:
|
[Prev in Thread] | Current Thread | [Next in Thread] |