I seem to have hit an S4 method caching issue on R 3.5.1 / Bioconductor 3.8 that appears to be specific to Linux. I can reproduce this issue on the Bioc 3.8 release AMI as well as a VM running Red Hat Linux. Running the same version of R/Bioconductor on macOS with the same versions of my toolchain doesn't reproduce this issue described below:
My basejump package (see develop branch for latest code) defines an S4 generic plotHeatmap(), along with corresponding method support for SummarizedExperiment. My bcbioRNASeq package depends on basejump, and therefore inherits the SummarizedExperiment method. When I attach basejump and call plotHeatmap() on a SummarizedExperiment, the SE method is current and works as expected. However, when I attach bcbioRNASeq, a cached version of the inherited S4 method for SummarizedExperiment (exported from basejump), is incorrectly loaded, and my function doesn't work as expected.
R on Linux doesn't seem to be updating the method cache as expected, whereas this works fine on macOS. I've tried digging around in the documentation for the methods package, specifically Methods_Details, which suggests resetGeneric() as a potential solution, but I can't seem to figure out how to clear the method cache.
As a follow-up, I've found some additional information in How Methods Work regarding how S4 generics cache methods, but there's very little information in the R documentation on how to troubleshoot S4 caching issues.
I'm working on trying to generate a reprex for this issue, but it does seem to be isolated to RStudio on Linux, and potentially only affects packages loaded inside an R Project.
I think the cache is created when a package imports a method, so the problem is that one or more of your packages was installed before the method was updated. The solution is to re-install the package(s) with stale cached methods; the challenge is to find out what those are...
Indeed, I think that's what I'm seeing in my library. I'm working on reinstalling the dependencies from scratch to see if that fixes the issue. Thanks Martin!