Front page | perl.perl5.porters |
Postings from April 2018
On Fri, Apr 27, 2018 at 02:42:21PM -0400, James E Keenan wrote: > On 04/27/2018 09:29 AM, Dave Mitchell wrote: > > On Thu, Apr 26, 2018 at 12:12:04AM +1000, sisyphus1@optusnet.com.au wrote: > > > -----Original Message----- From: Dave Mitchell > > > Sent: Wednesday, April 25, 2018 11:58 PM > > > To: sisyphus1@optusnet.com.au > > > Cc: perl5-porters@perl.org > > > Subject: Re: 5.27.11 fails ../lib/File/Copy.t on linux (ubuntu-16.04) with > > > -Dusequadmath > > > > > > > > not ok 53 - mtime preserved by copy() while testing cross-partition > > > > > # Failed test 'mtime preserved by copy() while testing cross-partition' > > > > > # at lib/File/Copy.t line 128. > > > > > # got: '1000000000.123449998' > > > > > # expected: '1000000000.123449999' > > > > > > > > > > Configure args were '-des -Duse64bitint -Duse64bitall -Dusethreads > > > > > -Dusequadmath -Dusedevel -Uversiononly'. > > > > > > > > It seems to be an issue with Time::HiRes; that module's > > > > nanosec-supporting stat() and utimes() calls are in some fashion not > > > > preserving an exact round-trip mtime. I'm still investigating... > > > > > > If it helps, that test passes when I run that test file (ie the > > > lib/File/Copy.t that ships with 5.27.11) from perl-5.27.5 and earlier, but > > > fails when I run it from perl-5.27.6 and later. > > > > I've found it. It's a rounding error (using truncating rather than rounding) > > I've pushed a fix for smoking as smoke-me/davem/time_hires_nsec. > > > > What would be the arguments for Configure needed in that branch? The important one's -Dusequadmath. I don't think the bug is limited to qudmath builds, but the particular combination of floating-point literal stored as a quad float and then converted to long, tickled the bug, and is the value I used in hte added test. I expect specific other time values would trigger the bug on "normal" builds. -- I before E. Except when it isn't.Thread Previous