[R-pkg-devel] Troubleshooting Winbuilder Run Timeouts

[Andrew Robbins]

Hi Ivan,

Thank you for giving my package check a run-through. You have it 
correct-every win-builder log hangs/halts in that exact example. The 
only other anomaly I notice on win-builder is that the elapsed time 
reported to Rcout is 0-which is clearly incorrect. DNDEBUG removing the 
assert calls would certainly make my life quite a bit easier when it 
comes time to maintain the package, so I'll definitely give that a try. 
Re: declared requirements, WindowsOverrides contains slightly modified 
copies of cmake locator files with patches that will likely be included 
in a later Rtools43 release. I'm fairly sure pkgbuild in rtools can be 
assumed present, but that's more of a Tomas question (and it may be 
worth declaring either way). Regardless, those files aren't intended to 
stick around for any longer than they need to. Any other suggestions are 
always welcome, but I do get the feeling that something is fishy on 
win-builder's end.


Best,

Andrew

On 11/30/2023 7:40 AM, Ivan Krylov wrote:
> Hi Andrew Robbins,
>
> В Mon, 27 Nov 2023 12:22:44 -0500
> Andrew Robbins via R-package-devel <r-package-devel using r-project.org>
> пишет:
>
>> I'm currently attempting to submit a package to CRAN and am getting
>> timeouts during the "running examples" phase of the winbuilder tests
>> that I cannot reproduce locally or on the r-hub windows runner
> Do I understand it correctly that in multiple win-builder tests, you
> always get a timeout in the same inmf example? Or is it different
> examples?
>
> There doesn't seem to be a lot of code between Rcpp::Rcerr << "INMF
> started, niter=" << niter << std::endl; and Progress p(niter, verbose);
> to crash or hang in. I've tried to reproduce the problem myself, but it
> doesn't happen to me either.
>
> So far I only noticed minor things, like the build-time dependency on
> pkgbuild (from tools/WindowsOverrides/hdf5/hdf5-config.cmake) that may
> be worth declaring in your DESCRIPTION. I also think that your patch to
> hwloc can be reduced by adding a -DNDEBUG somewhere instead of patching
> out calls to assert(). I think that the C standard (e.g. C99, WG14 draft
> version N1256, 7.4) requires assert() to expand to ((void)0) if NDEBUG
> is defined.
>
> The amount of systems engineering effort to do the right thing that
> went into this package is really impressive; I'd hate to see it fail.
>
-- 
Andrew Robbins
Systems Analyst, Welch Lab<https://welch-lab.github.io>
University of Michigan
Department of Computational Medicine and Bioinformatics


-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature.asc
Type: application/pgp-signature
Size: 840 bytes
Desc: OpenPGP digital signature
URL: <https://stat.ethz.ch/pipermail/r-package-devel/attachments/20231130/4f675e99/attachment.sig>