Richard Jelinek <r...@petamem.com> writes: > Hi, > > we're having issues with the CPAN Testers report for > Lingua::RUS::Number (see below). Test output seems to indicate there > are encoding errors (double/tripple encoding)
I'm the submitter of the reports. I don't see any indication of encoding problems when I run the tests on the console, I see beautiful cyrillic letters. I blame either my incompetence or the cpantesters toolchain that the double encoding happens, but it is not the cause of the problem. The cause is that some of my smokers are compiled with -Duselongdouble. There is a 100% correlation between having compiled with -Duselongdouble and a failing test. One of the outputs I see is not ok 2 - 8.00 RUR # Failed test '8.00 RUR' # at t/Number.t line 70. # got: 'семь рублей ноль копеек' # expected: 'восемь рублей ноль копеек' main::(t/Number.t:71): $tests++; seven vs eight? This is a truncation problem somewhere and that the number 8.00 gets an internal representation as 7.9999999 something. Try rounding to an integer when you want an integer! > but we're unable to > reproduce this on any OS, any Perl version available here. > > Does anyone have an idea what could cause these fails on these > particular systems? (In other tests @ linux this module passes also @ > CPAN Testers) Thank you for asking. <shamelessplug> http://analysis.cpantesters.org/solved?distv=Lingua-RUS-Number-0.0708 already had the answer when I looked: Top Regressions: 1.000 conf:nvsize 1.000 conf:nvtype 1.000 conf:uselongdouble 1.000 qr:(Failed test\s+\S+.*) 1.000 conf:alignbytes 1.000 conf:archname 1.000 conf:archname+osvers If analysis has a result like it has here, the diagnose is usually correct </shamelessplug> Thanks again && Regards, -- andreas
