Autoconf-style tests would be nice but I think it'll be painful to write tests for obscure asm issues (like, do binutils support SSE4?). Or compliler releases that die in an ICE after compiling pari for a while. Maybe we should have a combination of both, first autoconf tests and then supplement them with a list of specific (distribution, gcc version) pairs that don't work.
On Monday, April 30, 2012 5:51:57 PM UTC-4, Dr. David Kirkby wrote: > > On 30 April 2012 16:23, Volker Braun <vbraun.n...@gmail.com> wrote: > > Essentially by maintaining a list of gcc versions / architectures that > work > > well enough with reduced optimizations, and that are hopelessly broken. > This > > can just be some shell script that shitlists specific compilers... > > A problem with that approach is that a compiler can be broken on one > platform, or verions of linux, but ok on another. A better approach is > probaby the one taken by autoconf - test the compiler. If you know a > compiler computes 1+1=3, then write a test for that. Autoconf is good > for this. > > Dave > -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org