[R-pkg-devel] Advice on R-forge to Github migration

[Duncan Murdoch]

Thanks to everyone who commented.  A few replies inline:

On 31/01/2021 1:23 p.m., Ben Bolker wrote:
>     I think it should be fine to have your package in a subdirectory:
> this works as expected.
> 
> remotes::install_github("dmurdoch/rgl/pkg/rgl")
> 
>     The only disadvantages to this structure are (1) it's a little bit
> surprising to users (who might try install_github("r-forge/rgl" and
> wonder why it didn't work and (2) setting up Travis machinery takes a
> little more effort to get all processes to work in the proper
> subdirectories (see the glmmTMB package for examples).

I'm setting up Github actions instead of Travis; about a third of my 22 
failed attempts were caused by not dealing with the non-standard 
structure properly :-).

It's true that remotes::install_github("dmurdoch/rgl/pkg/rgl") works, 
but the RStudio "New Project | Version Control | Git" menu isn't so 
forgiving, presumably because git isn't.

On 31/01/2021 1:35 p.m., Jeff Newmiller wrote:
 > Technically, git does not record file history... it records commit
 > history, and reconstructs file history as needed. The contents of a
 > file identify it uniquely among all commits ever made regardless of
 > which directory it was ever in. IMO this is the main reason why git 
is > superior and I put up wth the inconsistent command line interface
 > because of it.
 >
 > git mv is very simple [1].

Given how long I've been using svn it may take me a while to recognize 
the benefits of this mindset, but I'll give it a try.

On 31/01/2021 1:21 p.m., Joshua Ulrich wrote:
 > I've moved history and issues from R-Forge to GitHub for half a dozen
 > R packages. I might be able to do this rgl. At minimum, I could help
 > you do it.

Thanks for the offer (and the more detailed one offline, where you 
recommend cloning the repository rather than forking it).  Just for 
future reference, could you tell me the benefits of cloning over forking?

Duncan Murdoch