FAQ: updating certain packages while a kernel is running is going to break the installation/fail #32
Description
If update.packages touches a native package which is loaded, it will fail and leave the package in a weired state (at least if I use R CMD build in my local git).
This is not a problem unless that that package is jsonlite, which the kernel needs... In that case you need to shutdown all kernels and then reinstall jsonlite...
Activity
takluyver commentedon May 17, 2016
Is this a Windows issue with multiple processes trying to open the same file?
jankatins commentedon May 17, 2016
No, I think it is a problem with native code (=
*.soon linux) trying to be replaced while in use.jankatins commentedon May 17, 2016
Maybe it would be interesting to replace update.packages with a version which checks if one of the to-be-updated packages are in a "blacklist" and if so errors instead of crashing...?
takluyver commentedon May 17, 2016
I feel like we don't understand the problem well enough, and I'm wary of attempting to paper over bugs in R package installation - that's the kind of fix that can end up causing as much aggravation as the problem it aims to resolve.
jankatins commentedon May 17, 2016
true...
BTW: I looked into the Rsession process of rstudio and they use rcp messages to communicate between the UI and the session... Not sure how easy it would be to translate the jupyter messages to rstudio messages and then use their rsession as the backend for the kernel... would remove a lot of headache from our usage of
evaluate...flying-sheep commentedon May 18, 2016
well, i guess that would be a separate project then.
i think a hard dependency on RStudio is out of scope for this
jankatins commentedon May 18, 2016
Yep, the last would a replacement for our current evaluate based implementation...
flying-sheep commentedon May 20, 2016
i’ve added most FAQ entries, but it’s not clear if we should really add this