|
From: Bill H. <goo...@go...> - 2008-12-25 17:44:07
|
I've fixed the bug (the same bug caused both problems) and will issue FLINT 1.0.21 in a couple of minutes. It was as I thought, using 32 bit primes instead of 31 bits. This bug does not exist in FLINT 1.1 or 1.2. Bill. 2008/12/25 Michael Abshoff <mab...@go...>: > On Thu, Dec 25, 2008 at 8:36 AM, Bill Hart <goo...@go...> wrote: > > Hi Bill, > >> I'll reply to both posts at once. > > Ok. > >> The test failures will be cases hit because of different random seeds >> on those 32 machines. They should be trivial to fix. There is an >> outside chance I can do it tonight, but if I have access to the >> machines involved, this will be easier. > > The linux box is cicero on SkyNet, but I am currently valgrind there, > so I am keeping that box busy. The other one is varro also on SkyNet, > but I don't think your account works there, so you should ping Mariah. > >> Upgrading to 1.1 should not be painful, unless you want to incorporate >> lots of the new functionality (which I hope you will :-)). In that >> case there's the wrapper to write. But otherwise things should be >> quite stable. I've done very long runs with most of the new functions, >> and no problems. I don't think SAGE uses any of the functionality >> where the interface has changed. > > Ok. > >> I've only just noticed that these failures are being reported for >> FLINT 1.0.20. That's odd, because only bug fixes have gone into that >> branch for some time now, so presumably these bugs have been there for >> a while, and have been triggered due to different random seeds after >> rebuilding. > > Yes, I just turned tests on since I figured it cannot hurt and I was > just as surprised that on both 32 bit test boxen things didn't work > too well. > >> If the bugs are in 1.0.20 they are also likely in 1.1, just not >> triggered because of different random seeds. > > Good to know. Once I am done with 1.0.20 I will test 1.1 on the same box. > >> Bill. > > Cheers, > > Michael > > ------------------------------------------------------------------------------ > _______________________________________________ > flint-devel mailing list > fli...@li... > https://lists.sourceforge.net/lists/listinfo/flint-devel > |