MAGEMin v1.6.9 #10530
MAGEMin v1.6.9 #10530
Conversation
NicolasRiel
commented
Feb 14, 2025
NicolasRiel
commented
- Changed bound values for compositional variables from 1e-10 to 1e-6, which largely improves accuracy and performances (15% faster)
- Improved calculation results for extended metapelite database
- Increased maximum size of possible considered phases (max value was hit in some cases)
- Fixed NLopt version to 2.8.0 as more recent version leads to large variation in calculation results between platforms
M/MAGEMin/build_tarballs.jl
Outdated
| @@ -85,7 +85,7 @@ products = [ | |||
|
|
|||
| # Dependencies that must be installed before this package can be built | |||
| dependencies = [ | |||
| Dependency(PackageSpec(name="NLopt_jll", uuid="079eb43e-fd8e-5478-9966-2cf3e3edb778")) | |||
| Dependency(PackageSpec(name="NLopt_jll", uuid="079eb43e-fd8e-5478-9966-2cf3e3edb778"); compat="2.8.0") | |||
There was a problem hiding this comment.
Fixed NLopt version to 2.8.0 as more recent version leads to large variation in calculation results between platforms
This compat follows Pkg compat bounds syntax. 2.8.0 means [2.8.0, 3), which I'm not sure is what you want, based on your comment.
There was a problem hiding this comment.
Ah sorry. You are right. I want to fix the specific version not put a lower bound. Thank you for providing the documentation. I fix this after lunch
There was a problem hiding this comment.
Might also be beneficial for you to work with upstream to get whatever issue you see fixed, not to get stuck with an outdated version forever.
There was a problem hiding this comment.
Yes, absolutely. This is also what I want to do. I am teaching with my code next week, and my plan was to contact NLopt author afterward.
|
@giordano Sorry to annoy you. But I am not sure what to do here. It seems "aarch64-apple-darwin-libgfortran5-mpi+mpich" has a problem with clang:
It compiled correctly with the previous version 1.6.8 :( Do you think this is because I fixed NLopt version? |
|
@giordano Well thank you... my apologies for the hassle. |
