Build broken on Windows

[wmay]

Build broken on Windows

[wmay]

After that commit you need BLAS and LAPACK to compile it. They should be included in Rtools. Let me know if that helps.

[knutwannheden]

I do have Rtools 4.0 and R.Project 4.0.4 installed. BLAS AND LAPACK appear to be part of R.Project and available as Rblas.dll and Rlapack.dll. And I assume the older versions of this library also use LAPACK and / or BLAS. Since I am able to install the oder versions just fine, I assume my setup is basically fine. I suspect this is some platform specific issue. I don't have much experience with R, so I am afraid I won't be able to help much. Possibly someone else could try to confirm these problems on Windows?

[wmay]

Not quite. You need the BLAS and LAPACK development headers to do compiling. R itself only comes with the shared libraries (the dll's). That's why there's a separate Rtools package for compiling packages. And that's why gcc complains about sgesvd, which is a LAPACK function.

Older versions of this package didn't use LAPACK. They used IMSL and EISPACK code included in the fortran source file. I replaced it with LAPACK in this commit because the IMSL code is proprietary.

So this almost certainly suggests that Rtools somehow isn't set up correctly. Your compiler just needs to be pointed to the LAPACK development headers.

[knutwannheden]

My R on Windows does also have LAPACK and BLAS C header files. They are under include/R_ext and you can inspect them here: https://svn.r-project.org/R/trunk/src/include/R_ext/Lapack.h and https://svn.r-project.org/R/trunk/src/include/R_ext/BLAS.h.

I have no idea what the "R flavor" of these two libraries is or how it differs from the standard libraries. But I think I have my R and Rtools installed properly and suspect that only installing Rtools is not quite enough on Windows.

[wmay]

You're linking to the R source code, but if you install R on windows you would typically only have the binary. I'm not working on a windows machine at the moment but my understanding is that you won't end up with those files on your system. They may not be found by the compiler even if you do have the files.

The Rtools source code shows that it includes lapack, which is why I expect it to work.

But I also haven't used this new version of Rtools before, and I'm a bit confused about what they're doing. Let me try this out with my windows machine and get back to you.

[wmay]

Can you try following the Rtools instructions here?

It looks like you would need to run something along the lines of:

pacman -S  mingw-w64-{i686,x86_64}-lapack

And that should provide the development headers for you.

[wmay]

Wow, I got that wrong-- the lapack headers really are installed with R. The issue seems to be that the functions I'm calling aren't included in R's lapack. I'll have to ask around on some mailing lists, because I actually don't know how to make this work right now.

[knutwannheden]

Thanks a lot for looking into this. If you figure out a way to solve this with using BLAS and LAPACK bundled with R, then great. Otherwise it would maybe be possible to publish a compiled package to CRAN to make it easier to use for Windows users.

[wmay]

Unfortunately it's not possible, because I literally can't compile it on windows right now (and putting the package on CRAN is not an option at the moment). After I get this sorted out though, if the compilation process is too complicated, I will definitely try to provide a binary version.

[knutwannheden]

I am happy to help you test, if that is of any help.

[wmay]

I think I've got it. Try running

remotes::install_github('wmay/dwnominate', ref = 'lapack')

[knutwannheden]

Yes, indeed. Like this it now works!