Improve VisionEval Developer Environment
Closed this issue · 2 comments
The current ve.build() process is being augmented in Next Generation Beta 0.7 to add a ve.test() function, with the basic workflow being to build, test and then run (launching a VisionEval runtime). The following additions are desirable (and will eventually be implemented - this issue is a reminder):
- If VE_RUNTIME is not set in the user's environment, ve.run() will currently go the built runtime folder. Any work done there is, however, at risk since that folder can be wiped with instructions like
ve.build('clean')(something that should happen before building an installer so your local models don't end up being packaged).
So we're going to change it such that the first time you do ve.run(), it prompts for a directory (ideally outside your Git tree), creates it if it doesn't exist, and optionally copies the built runtime recursively into it. Then it (re)writes the .Renviron at the root of the VisionEval-dev directory hierarchy to define VE_RUNTIME as that directory and also to set up R_LIBS_USER to point into the libraries that were set up during the build. You can (and should) currently do all that manually. Also change ve.build so that if VE_RUNTIME is set, the tools and walkthrough will get copied over to the folder listed in VE_RUNTIME after the successful build (so the local runtime supporting files stay up to date. Also create a suitable .Renviron in VE_RUNTIME so VisionEval can be started from that location (either in Rstudio by opening the project, or in RGui by running launch.bat).
- Since RStudio has a nifty "add ins" function, it is also desirable to push the ve.build and related stuff into a VE-Developer package.
So rather than run build\VisonEval-dev.R in .Rprofile (current method), users should be able to download and install package VEDeveloper in RStudio and have access to functions for cloning from Git (using git2r) and then doing all the build stuff. With that workflow, there's no need for bringing in Git for Windows (or Linux). Just install R, RStudio (and RTools40 within it - VisionEval needs the make program) and then install the VEDeveloper package as an RStudio add-in. That will offer up a menu of functions for cloning a branch of VisionEval, pulling updates, building the source via ve.build etc, setting up a local runtime, etc. In that workflow, starting a developer environment of VisionEval would just entail installing the package in a fully-provisioned (i.e. with RTools40) RStudio environment and then working the menus.
Fixes are in the pipeline
Simplifying changes to development environment are in the development-next push 2023-03-24