Hi Arthur, On 13/03/2019 17:43, Arthur Eubanks wrote:
Martin suggested looping through all available loopback addresses rather than hardcoding them. In that case we'll need helper test functions to return loopback addresses (and test that it actually returns all available loopback addresses).
Though I agree that in a ideal world that's what we would like to do, I am afraid that this would increase the possibility of getting random intermittent failures when running our tests on tests farms. It would certainly increase the opportunity of stumbling on some bad configuration issues. Also having several tests running concurrently when one test binds on the wildcard while another test concurrently binds on the local host address (or the loopback) is a known recipe for instability due to the underlying system port reuse strategy. In addition some of our tests (in httpclient) are already pretty greedy trying to cover the matrix of http / https / http 1.1 / http 2 multiplied by all the different test cases that the test wants to test - so running them over all possible loopbacks would probably become a maintenance hassle: We already had to split several of them (sometimes a bit artificially) because they were going over the reasonable resource quota when run on some exotic configuration (like fastdebug + Xcomp for instance). So I'm not sure how feasible such an idea would be. best regards, -- daniel
