mpir build failures with some Haswell CPUs

[embray]

See wbhart/mpir#252

Upstream: Fixed upstream, but not in a stable release.

Component: packages: standard

Author: Erik Bray

Branch/Commit: 50c1e0a

Reviewer: Travis Scrimshaw

Issue created by migration from https://trac.sagemath.org/ticket/25858

[embray]

Commit: 50c1e0a

[embray]

New commits:

50c1e0a

add a hand-customized version of the patch from the upstream PR which only patches the configure script and the relevant .c sources

[embray]

Author: Erik Bray

[embray]

Branch: u/embray/ticket-25858

[tscrim]

Reviewer: Travis Scrimshaw

[tscrim]

comment:2

You can set this to a positive review if upstream accepts the PR.

[embray]

comment:3

I believe this issue can reasonably be addressed for Sage 8.4.

[embray]

comment:4

Got confirmation from Odile that this patch works.

If possible I'd like to get this in 8.3 actually, but admittedly adding a patch to MPIR is a big rebuild, and it also affects only a few rare machines and has an easy workaround e.g. building MPIR with --build=skylake or some other earlier architecture.

[tscrim]

comment:5

I am much more split on this ticket (than the one that only affects Cygwin) as it could have more unintended side-effects. I am slightly on the side of waiting until 8.4 and having this go through beta cycles. If we get posts on sage-devel, you have at least seen the error and know the workaround. Plus this will only affect people building it from source on these rare machines, so the chances of someone being afflicted are marginal.

[embray]

Changed upstream from Reported upstream. Developers acknowledge bug. to Fixed upstream, but not in a stable release.

[embray]

comment:6

I agree, it's pretty marginal.

[vbraun]

Changed branch from u/embray/ticket-25858 to 50c1e0a