Closed patrickmdnet closed 4 years ago
Travis works. Appveyor fails with this error:
Error: (converted from warning) unable to access index for repository https://cloud.r-project.org/bin/windows/contrib/4.1:
cannot open URL 'https://cloud.r-project.org/bin/windows/contrib/4.1/PACKAGES'
R-devel recently bumped its version number to 4.1. However, CRAN does not yet have windows binaries for 4.1, so the directory doesn't exist. I could change PKGTYPE=source and that might help, or we can just make R-devel builds optional until CRAN catches up in a few weeks.
Patrick - thanks for all this work. CRAN, as ever, a moving target. The original travis builds were so complicated because the dependency load was huge, build/test times timed-out, and there were platform specific compiler flags required, partly relating to the (now default) C++11 requirement, OpenMP, etc.. I can accept R-devel failing sometimes, but since I need to submit to CRAN and they require an affirmation that I have tested against the very latest R-devel, this is a problem. Will plough through your pull requests today. I'm really keen to get this bug-fix release out.
I tried to stop travis adding the --as-cran flag to eliminate some checks which inevitably failed, or hugely extended the dependency burden and build time. E.g. I don't want to check building the PDFs, etc on each platform, and this requires huge load of tidyverse, pandoc installation, etc., which takes half an hour or more. There were also some checks which always errored (compiler flags is an example you helped with), or gave notes for (UTF-8 strings in the diagnosis descriptions), lack of qpdf to check the pdf (which i don't even want to build in most cases), etc., etc.. I really want to keep it simple, but the checks have to be both fast enough for dev cycle, and thorough enough for CRAN. I'd rather struggle with travis builds now, than similar errors on the CRAN side, where the feedback loop is much longer, and consumes CRAN resources.
Am also using r-hub. Each platform has own quirks, which is probably good for testing, but the ratio of CI problems to icd code work is not good right now!
Really appreciate your help.
FYI I'm starting to work with @ellessenne to take fit the best bits of his comorbidity package with my matrix-based comorbidity engine, and keeping all the ICD processing, validation in this package.
Fixed with f4bca68c362d8acf6dc622d0f5403a0c67a8518a again with thanks and help from @patrickmdnet
As described in issue #197, this pull has changes designed to fix travis. Work in progress.