…n in the test itself. Resolves rt.cpan.org #61727.
…ecretly broken it. :)
…n someone needs IPv6. Reported by Apocalypse in irc.perl.org.
…can study what the heck happens
…e into wtf
…be delayed. Resolves rt.cpan.org ticket #63242.
…_data.t can do their thing.
Roll back most of the lexical FIFO event arrays. Going through a POE::Queue for this will be a bit slower, but POE::XS::Queue::Array may be fast enough. Continue to guarantee FIFO event ordering despite the rollback to dumping them in a time-ordered queue. Detect minor time regressions, and use the last time value plus a small number when needed.
…entation, spotted by Paul G Webster
aborted, rather than ECONNABORTED everywhere else. Add ESPIPE to the list of harmless accept() errors.
based things. nmap disconnects before accept() can be called, so accept() returns ECONNABORTED. Technically applications can handle this themselves and not destroy the wheel, but the wheel maybe shouldn't report harmless errors.
events' latency to at 1sec. First reported by Getty, then Avar opened the ticket. Fixed with Hinrik's copious valuable assistance.