From: KHMan <kei...@gm...> - 2015-06-01 19:01:20
|
On 6/1/2015 11:25 AM, Yongwei Wu wrote: > On 1 June 2015 at 10:22, KHMan wrote: >> >> On 6/1/2015 9:03 AM, マーズ ho han keng wrote: >>> http://developers.slashdot.org/story/15/05/31/1311254/mingw-and- >>> msvcrt-conflict-causes-floating-point-value-corruption >> >> And your point is? >> >> Blog posting linked to the Slashdot item is a LOT of rubbish. >> >> Don't assume that people who sound authoritative on blogs know >> what they are doing. > > BTW, the blog reports a legitimate bug of MinGW, but the summary on > Slashdot is less clear about where the problem is. The blog report has plenty of rubbish. Having said that... (and the following is only my personal opinion and has nothing to do with any MinGW developer, so flame me and not them) I disagree on calling it a bug, that is a simplistic tag. This is an impedance matching problem, the price of using a compiler with System V ABI pedigree and trying to fit it into a Microsoft world that always like to do things differently probably for strategic and tactical reasons. It was always going to be messy. From Slashdot see [1] then see the linked posting [2], it is one of the many tweaks available, so it's up to you to pick the poison you want. So they have picked System V behaviour. Because of the many matching/fitting issues, who's to say this is right or that is wrong? Personally I'd avoid 80 bit thingies unless it's really really really really really unavoidable -- but in doing so it means the program is sensitive to tiny differences >16 digits, and so I'd better check it with 128 bit thingies too. Meh. [1] http://www.diusrex.com/2015/05/long-double-bug-in-mingw/ [2] http://lists.cs.uiuc.edu/pipermail/cfe-commits/Week-of-Mon-20130930/090167.html -- Cheers, Kein-Hong Man (esq.) Kuala Lumpur, Malaysia |