Bill, did you see the note from Giovanni? :" reverting r1823205 makes the regression test work on 3.4 and trunk for me. tested on CentOS6 (perl 5.10.1), can anybody confirm ?"
-- Kevin A. McGrail VP Fundraising, Apache Software Foundation Chair Emeritus Apache SpamAssassin Project https://www.linkedin.com/in/kmcgrail - 703.798.0171 On Mon, Jun 25, 2018 at 5:42 PM, Bill Cole < [email protected]> wrote: > On 25 Jun 2018, at 9:54, Kevin A. McGrail wrote: > > Just to be clear, others concur that with 3 uris or less, it works, 4 or >> more it fails. It's inconsistent and exists in trunk as well. >> > > Yes. Also, it does not matter whether the URIs are pure ASCII or not, > which implies that finding this in the idn_dots.t test was somewhat > accidental, as was the fact that "use utf8" masked it. I have reverted my > commit to the 3.4 branch with that change, since it was not solving the > root bug. > > Annoyingly, running the test script or the spamassassin script with the > Perl debugger also masked the error. > > It's inconsistent depending on the platform as well. >> > > I can only replicate it on CentOS 7 with the (Red Hat) stock build of Perl > 5.16.3. I'd guess that RHEL7 behaves identically but do not have a RHEL7 > system to test. I have tested on other platforms (see below.) > > >> I am not sure if it is a Perl bug or an SA bug or something we are doing >> wrong but it is a blocker. >> > > It's a Red Hat bug IMHO. > > I have been unable to reproduce the problem on any of the following > OS/Perl combinations: > > Ubuntu Trusty, Perl 5.18.2-2ubuntu1.6 > Ubuntu Xenial, Perl 5.22.1-9ubuntu0.5 > MacOS X 10.6.8, Perl 5.26.2 (MacPorts build) > MacOS X 10.6.8, Perl v5.10.0 (Apple build) > MacOS X 10.11.6, Perl 5.26.2 (MacPorts build) > CentOS 7.5.1804, Perl 5.18.4 (Perl.org source, default build config) > CentOS 7.5.1804, Perl 5.28.0 (Perl.org source, default build config) > > I had a *TRANSIENT* problem on Ubuntu Trusty at first but it was so weird > (inconsistent between tests!) that I wiped the working directory and > rebuilt from a fresh checkout, after which I could not get a failure. > >
