Gábor Csárdi

Note: tokens are here: https://github.com/wch/r-source/blob/abb550c99b3927e5fc03d12f1a8e7593fddc04d2/src/main/gram.c#L511

The main issue for local packages is that it is harder to decide if the cache is valid or not.

> What about something like this: one silently builds the local package, takes a checksum and checks it agains the cache. If building the package has failed, we throw an...

The tricky stuff is that if you build-time depend on some package, and that has changed, that should ideally trigger a re-install. Similarly, if the system dependencies have changed, that...

In theory we could also try to use the mtime of the files/directories, but that is unreliable on some systems, so better to stay away from it, I think. I...

Oh, I do want to have caching support for local packages, so let's keep this open. :) I think more people would use this than git remotes, actually. But understand...

Btw. this is a good summary of the issues with mtimes: https://apenwarr.ca/log/20181113

Avoiding the install is less important, I think. Installing the binary package is fast, basically an unzip/untargz. I don't want to be paranoid, hence my opinion re checksum. I would...

AFAIR `R CMD INSTALL` does that as well by default when installing a package from source. Is that OK, and if yes, why?

Unfortunately this is not an easy fix, because Windows locks the loaded DLL file which leads to complications. Typically what happens is that users cannot update packages because some process...