Le dimanche 10 janvier 2016 à 09:56 -0600, Dirk Eddelbuettel a écrit : > On 8 January 2016 at 09:52, Dirk Eddelbuettel wrote: > > > > On 8 January 2016 at 16:44, Sébastien Villemot wrote: > > > Le vendredi 08 janvier 2016 à 09:35 -0600, Dirk Eddelbuettel a > > > écrit : > > > > > > > We can probably square this up. It smells like something due > > > > to gcc-5 as the > > > > same code worked for years. Every now and then we do have to > > > > tell the > > > > compiler to chill ... > > > > > > That indeed looks like a compiler problem. I am going to > > > recompile R > > > without optimization on the mips porterbox, to see if that helps. > > > > *Really* appreciate that. Am at work now and cannot dive into that. > > Did you get a chance to poke? Did you get anywhere, or did you get > stuck?
Yes I had the time to look at it. It's indeed a compiler issue. The problem goes away if one compiles with GCC 4.9 (I could not give a try to GCC 6, which is not yet compiled in experimental for mips). Optimization is already turned off for mips (and mipsel), and turning it on again does not solve the issue. So the workaround that I suggest is to use GCC 4.9 on mips. It clearly fixes the issue. That version of GCC should still remain for some time in the archive, though I guess it will be removed at some point, maybe even before the stretch release. Hopefully the bug in GCC 5 will be fixed by then. By the way, we should enable back optimization on mips (and mipsel). There seem to be no point of turning it off (I verified that R compiles with GCC 5 + optimization on mipsel). I will send you a patch probably by tomorrow. Then we should report the issue in GCC 5. I have a rather minimal testcase. Cheers, -- .''`. Sébastien Villemot : :' : Debian Developer `. `' http://sebastien.villemot.name `- GPG Key: 4096R/381A7594
signature.asc
Description: This is a digitally signed message part