Closed mahrud closed 1 year ago
Does defining the BOOST_MATH_DISABLE_FLOAT128
macro help at all? We use MFPR for all the Boost special functions, so there's no reason to link against libquadmath for 128-bit floats.
It doesn't seem like it. Can you try seeing if your executables are linked with libquadmath
, and if so does passing that actually remove it?
I was able to reproduce this using brew's Boost, which is version 1.80. The Debian/Ubuntu packages are still at 1.74, which is probably why I never saw this (plus the autotools build was already adding -lquadmath
to the linker flags -- I'm not 100% sure why...)
This is https://github.com/boostorg/math/issues/733, and linking with -lquadmath
appears to be the correct solution.
Not the ideal fix, and I'm not sure if it is universal, but closing this for now.
I kept getting linking errors for a while, and I just noticed that it's also causing issues building brew bottles on ubuntu:
On my machine linking with
quadmath
resolves this, but apparently this fails on MacOS. @d-torrance any ideas what might be behind this?