So that R code in a package calling an old version of that package's .dll on Windows can't happen. Almost every time we upgrade, someone somewhere on Windows hits this problem; e.g. #3055.
The worry is that it happens more than we know, but users get wrong results silently. It's quite fortunate when a mismatch is severe enough to cause an error, so at least the user knows something is wrong.
But how to achieve this? It might mean checking the version matches between R package and its .dll on each and every .Call() call, which might be expensive, involve a lot of code changes at every interface point, and might be insufficient if we miss some interface points either now or in the future.
Maybe it's better to fix the issue upstream in R itself (install.packages) or RStudio if it's an RStudio-only problem. Since it must affect other packages too. We need someone with i) the will and ii) Windows, to solve it once and for all, and we need precise information from users on Windows who hit this problem to tell us the conditions under which they upgraded and the problem happens. (My guess is that they upgrade in one R session while another R session has the package loaded using the .dll).
So that R code in a package calling an old version of that package's .dll on Windows can't happen. Almost every time we upgrade, someone somewhere on Windows hits this problem; e.g. #3055.
The worry is that it happens more than we know, but users get wrong results silently. It's quite fortunate when a mismatch is severe enough to cause an error, so at least the user knows something is wrong.
But how to achieve this? It might mean checking the version matches between R package and its .dll on each and every
.Call()call, which might be expensive, involve a lot of code changes at every interface point, and might be insufficient if we miss some interface points either now or in the future.Maybe it's better to fix the issue upstream in R itself (
install.packages) or RStudio if it's an RStudio-only problem. Since it must affect other packages too. We need someone with i) the will and ii) Windows, to solve it once and for all, and we need precise information from users on Windows who hit this problem to tell us the conditions under which they upgraded and the problem happens. (My guess is that they upgrade in one R session while another R session has the package loaded using the .dll).