Dear R-package-developers, On 'r-oldrel-windows-ix86_x86_64' I get in 'check.log' the error message "Error: package 'highs' is not installed for 'arch = i386'" the statement that for arch = i386 the package is not installed is correct and expected, since in 'install.out' I see the warning "Warning: this package has a non-empty 'configure.win' file, so building only the main architecture" Looking at r-package-devel archive, I found the suggestion to set Biarch TRUE in the DESCRIPTION file. I don't want to force the build of "i386" by setting Biarch: TRUE, since as far as I see from the source I link to this should not work and I don't want to alter the library I link to as little as possible. Therefore, I want skip the step "loading checks for arch 'i386'". Is this possible? I know locally this can be resolved by adding the option '--no-multiarch', but can a similar effect be accomplished at the CRAN checks? Furthermore, I know I can could this issue by adding R >= 4.2 to the depends since Rtools 4.2 does not build for i386. But this does not seem like a nice fix. Is there a better option and how would it work? Best regards, Florian
[R-pkg-devel] windows i386
4 messages · Florian Schwendinger, Simon Urbanek, Uwe Ligges
1 day later
Florian, since there was no direct response and given the earlier discussion I figured I chime. The main problem seems to be your build system that doesn't work. Since you didn't post the actual version of the package, I can only see the CRAN version which still don't set any of the necessary flags from R so it won't work (the error is that it can't even use the C compiler). Your package has to work both in 32-bit and 64-bit, so you need to make sure you pass the correct flags for cmake for each architecture - it's really the same concept as discussed earlier. As for multi-arch, if you have configure.win then you may need to use --merge-multiarch such that each architecture is built separately and the merged into one binary. I think an easier approach would be if you simply dropped configure* and just used Makevars to build the dependent library as part of the build process since you are building in a sub-directory, so you don't really need the configure and could build both archs in one R CMD INSTALL run. Cheers, Simon
On 13/09/2022, at 11:14 PM, Florian Schwendinger <FlorianSchwendinger at gmx.at> wrote: Dear R-package-developers, On 'r-oldrel-windows-ix86_x86_64' I get in 'check.log' the error message "Error: package 'highs' is not installed for 'arch = i386'" the statement that for arch = i386 the package is not installed is correct and expected, since in 'install.out' I see the warning "Warning: this package has a non-empty 'configure.win' file, so building only the main architecture" Looking at r-package-devel archive, I found the suggestion to set Biarch TRUE in the DESCRIPTION file. I don't want to force the build of "i386" by setting Biarch: TRUE, since as far as I see from the source I link to this should not work and I don't want to alter the library I link to as little as possible. Therefore, I want skip the step "loading checks for arch 'i386'". Is this possible? I know locally this can be resolved by adding the option '--no-multiarch', but can a similar effect be accomplished at the CRAN checks? Furthermore, I know I can could this issue by adding R >= 4.2 to the depends since Rtools 4.2 does not build for i386. But this does not seem like a nice fix. Is there a better option and how would it work? Best regards, Florian
______________________________________________ R-package-devel at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-package-devel
On 13.09.2022 13:14, Florian Schwendinger wrote:
Dear R-package-developers, On 'r-oldrel-windows-ix86_x86_64' I get in 'check.log' the error message "Error: package 'highs' is not installed for 'arch = i386'" the statement that for arch = i386 the package is not installed is correct and expected, since in 'install.out' I see the warning "Warning: this package has a non-empty 'configure.win' file, so building only the main architecture" Looking at r-package-devel archive, I found the suggestion to set Biarch TRUE in the DESCRIPTION file. I don't want to force the build of "i386" by setting Biarch: TRUE, since as far as I see from the source I link to this should not work and I don't want to alter the library I link to as little as possible. Therefore, I want skip the step "loading checks for arch 'i386'". Is this possible?
No, up to the R-4.1.x series we provide only bi-arch packages. Not packages for single architectures.
I know locally this can be resolved by adding the option '--no-multiarch', but can a similar effect be accomplished at the CRAN checks? Furthermore, I know I can could this issue by adding R >= 4.2 to the depends since Rtools 4.2 does not build for i386. But this does not seem like a nice fix. Is there a better option and how would it work?
If it is not too important for you that CRAN provides Windows binaries for the R-4.1.x series, simply keep it as is. Users can install for x64 from source or upgrade R. Best, Uwe
Best regards, Florian
______________________________________________ R-package-devel at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-package-devel
3 days later
Dear Simon, The repo is at https://gitlab.com/roigrp/solver/highs . Thank you. If I tried the Makevars approach but this leads to other issues with the 32-bit version. I resubmittet the the version with the configure file. Cheers, Florian ? Gesendet:?Donnerstag, 15. September 2022 um 01:39 Uhr Von:?"Simon Urbanek" <simon.urbanek at R-project.org> An:?"Florian Schwendinger" <FlorianSchwendinger at gmx.at> Cc:?r-package-devel at r-project.org Betreff:?Re: [R-pkg-devel] windows i386 Florian, since there was no direct response and given the earlier discussion I figured I chime. The main problem seems to be your build system that doesn't work. Since you didn't post the actual version of the package, I can only see the CRAN version which still don't set any of the necessary flags from R so it won't work (the error is that it can't even use the C compiler). Your package has to work both in 32-bit and 64-bit, so you need to make sure you pass the correct flags for cmake for each architecture - it's really the same concept as discussed earlier. As for multi-arch, if you have configure.win then you may need to use --merge-multiarch such that each architecture is built separately and the merged into one binary. I think an easier approach would be if you simply dropped configure* and just used Makevars to build the dependent library as part of the build process since you are building in a sub-directory, so you don't really need the configure and could build both archs in one R CMD INSTALL run. Cheers, Simon
On 13/09/2022, at 11:14 PM, Florian Schwendinger <FlorianSchwendinger at gmx.at> wrote: Dear R-package-developers, On 'r-oldrel-windows-ix86_x86_64' I get in 'check.log' the error message "Error: package 'highs' is not installed for 'arch = i386'" the statement that for arch = i386 the package is not installed is correct and expected, since in 'install.out' I see the warning "Warning: this package has a non-empty 'configure.win' file, so building only the main architecture" Looking at r-package-devel archive, I found the suggestion to set Biarch TRUE in the DESCRIPTION file. I don't want to force the build of "i386" by setting Biarch: TRUE, since as far as I see from the source I link to this should not work and I don't want to alter the library I link to as little as possible. Therefore, I want skip the step "loading checks for arch 'i386'". Is this possible? I know locally this can be resolved by adding the option '--no-multiarch', but can a similar effect be accomplished at the CRAN checks? Furthermore, I know I can could this issue by adding R >= 4.2 to the depends since Rtools 4.2 does not build for i386. But this does not seem like a nice fix. Is there a better option and how would it work? Best regards, Florian
______________________________________________ R-package-devel at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-package-devel
?