update for sundials 5.1 #421
Conversation
|
It looks to me like the tests for LAPACK require finding libgfortran when linking with C linkers, |
I can have a look in the next days, unless you figure out yourself. You can try for example to do before running cmake |
|
@giordano thanks I added it but get the same result. |
|
I'm not at the computer to see what happens, I'll be able to get back to this in a couple of days |
|
Are there other Fortran libs being called except blas and lapack? Perhaps those are getting name mangled too but should not be? I am on my phone so unable to dive in on details, which I should be able to in a couple of hours. |
|
@ViralBShah I am not sure but it seems reasonable. To confirm your hypothesis is possible to see that https://dev.azure.com/JuliaPackaging/Yggdrasil/_build/results?buildId=1121&view=logs&j=cff8108e-5e30-5123-99e4-8a9b379302a5&t=5baf66a3-28c5-5694-918f-b66b88316018&l=1722 the name mangling does work. Although LAPACK is still not functional. I uploaded some extra output here LLNL/sundials#23 (comment) for your reference about the LapackTest that Sundials runs. |
|
@giordano is there an easy way you guys test locally with the patches? I haven't found a way to run a local test and pass the patch files. I end up making them manually every time. |
Something like this: https://juliapackaging.github.io/BinaryBuilder.jl/dev/FAQ/#I-love-the-wizard,-but-now-I-want-to-break-free:-can-I-build-the-tarballs-without-it?-1? |
This is exactly what I needed, thanks. I just haven't worked with BinaryBuilder that much. |
|
@giordano I have run out of ideas so far to fix the compile in windows it seems that the problem is in Sundials itself |
|
Yes, I also think this is an issue in Sundials |
|
For PowerPC it looks like |
It's because it doesn't find LAPACK, I need to look into why |
|
Change the last line to build_tarballs(ARGS, name, version, sources, script, platforms, products, dependencies; preferred_gcc_version = v"5")(or click the checkbox to allow maintainers to push to your branch 😉) |
|
@giordano thanks for the fixes for PowerPC. I have been trying to replicate the windows error locally but I get stuck at the C compiler test with the error I don't know if the fact that I am using docker in a Mac might be affecting the exact version of the toolchain that is used. |
It shouldn't, the point of BinaryBuilder is to provide the same environment to everyone 😕 I can reproduce the Windows error locally but I'm mildly confident that this is an upstream error |
|
The macOS failure is a |
|
One downside of doing something like is that probably all libraries will be linked to libgfortran, even if they don't need it, which I think is @ViralBShah's point in a comment he made above. I need to play more with some of the CMake variables to see how to properly do this, yesterday I did some experiments which didn't work very well. |
|
Closing this and opening a new PR - so that sundials v3 and v5 can be both built during the transition. |
* Creating a separate directory for Sundials 5. This is basically the work of @jd-lara and @giordano in #421 Co-authored-by: Jose Daniel Lara <jdlara@berkeley.edu> * [Sundials@5] Use gfortran & Co. only as lapack libraries Co-authored-by: Jose Daniel Lara <jdlara@berkeley.edu> Co-authored-by: Mosè Giordano <giordano@users.noreply.github.com>
@ViralBShah This builder will likely fail because I haven't been able to run the wizard fully because of LLNL/sundials#23