Error related to devtools::load_all() in .Rprofile

[lbm364dl]

I decided to first ask here since I'm not sure if this is a bug or I'm just doing something I shouldn't.

Question: is it bad practice to try including devtools::load_all() in .Rprofile?

I had this in .Rprofile:

if (Sys.getenv("USE_RENV", unset = "TRUE") == "TRUE") {
  source("renv/activate.R")
}

if (requireNamespace("devtools", quietly = TRUE)) {
  devtools::load_all()
}

I just installed Positron and tried opening my project and this is what I see:

Error: UI comm is not connected. Can't execute request ExecuteCommand(ExecuteCommandParams { command: "workbench.action.files.saveAllTitled" })
Error while sourcing R profile file at path '/home/usuario/WHEP/.Rprofile':
Unexpected longjump
Likely caused by:
Error: UI comm is not connected. Can't execute request ExecuteCommand(ExecuteCommandParams { command: "workbench.action.files.saveAllTitled" })

If I reduce the .Rprofile to just this:

if (Sys.getenv("USE_RENV", unset = "TRUE") == "TRUE") {
  source("renv/activate.R")
}

Then there's no error anymore. Is there something wrong with loading things in .Rprofile?

PS: The error indeed still happens if the content of .Rprofile is just this:

devtools::load_all()

[jennybc]

Question: is it bad practice to try including devtools::load_all() in .Rprofile?

I guess I wouldn't call it bad but rather sort of pointless/weird? And then, given the way it's interacting with startup, that makes it bad.

In terms of how I work, I don't necessarily do load_all() the very moment I fire up an R package. And I go on to do load_all() on a very frequent basis during any given dev session. So I just don't see the point of having it in .Rprofile, which I think of as doing fiddly stuff I want to invisibly persist throughout my session, like setting my CRAN mirror or warning me about partial matches.

[juliasilge]

I agree with Jenny, for sure!

From a standpoint of what's happening here, the R code in devtools::load_all() uses the rstudioapi package to save untitled files you may have. R is trying to this before all the connecting is fully wired up between R and Positron ("UI comm is not connected"). Your .Rprofile is run pretty early in the process of startup, and there aren't guarantees about Positron being fully ready to hear from R before it is run. If we made some change here, it would make R even slower to be started than it already is.

[jmcphers]

Tangentially: we need a place to put startup code that requires rstudioapi. RStudio has a session init hook that we could pretty easily support in Positron too: #9763

[lbm364dl]

Thank you all. I see @jennybc's point about this being quite pointless. One collaborator suggested including this since they are still not familiar with package development and thought it would be a good idea to have everything loaded the first time. My reply was just OK, why not? I guess it doesn't hurt. Now at least I have a reason why not to add it.

[jennybc]

Yeah I think the easy fix here is to remove load_all() from .Rprofile. (I agree it would be nice if we had a way to defer startup-y things until the session is truly "up". But I don't think particular use case (load_all()) is a super motivating use case for that.)

[jabenninghoff]

Thanks all for the explanation, this was also an issue for me. Personally, I find running load_all() in my .Rprofile convenient as that's typically the first thing I do when starting package development. Do I understand correctly that #9763 will fix the "UI comm is not connected" error?