WISH: Auto-installing of non-installed packages about wishlist-for-r

[leeper]

I could imagine that library() could have an ask = interactive() default argument indicating whether to ask the user if they want to install, such that: library("pkg") would fail during non-interactive mode, prompt during interactive mode by default, and not prompt interactively if set explicitly to false (i.e., library("pkg", ask = FALSE).

[leeper]

pacman::p_load() does this.

[jimhester]

This is not a general solution for R or something I would recommend those new to the language to use, but https://github.com/jimhester/autoinst when used as an error handler will automatically install missing packages for you when they throw an error. It also queries GitHub for R packages if the missing package is not on CRAN, so like I said, not a general solution.

[HenrikBengtsson]

There are probably several attempts to work around this out there. Here are the ones I am aware of (please let me know of others):

• pacman::p_load() replaces library() w/ install.packages() iff needed.
• R.utils::use() - replaces library() and loadNamespace() w/ install.packages() iff needed.

Last updated: 2016-04-27

[gmbecker]

@HenrikBengtsson I really don't think we want an option-based approach here. Mostly because I don't want an option-based approach anywhere really (there are some exceptions). Code that does radically different things based on things external to both code and explicit inputs? No thank you.

I'd say adding an argument is more reasonable. I could see library(, install=) working, though we'd need to think hard about what the default to that should be. I suspect you'll want it to be TRUE, but I'm not sure the case for that is as strong as it might appear (network roundtrip and attempted install in the case of typo ...).

[HenrikBengtsson]

Another usecase could be pkg::foo(). In that case there is no explicit function call where we can specify an install argument.

I agree that options should not be over/misused, but in some cases I find them really handy. In this case I'd argue that the behavior is more related to the UI rather than function (with stringsAsFactors being extreme example of what we don't want). I don't see how the end results would differ with and without automatic package installation, expect from a time delay.

I also think it is important not to enforce package installation on people, e.g. poor/expensive internet access. This is why I believe it has to be an option.

[gmbecker]

I would argue fairly strongly that pkg::foo() should not install a package, even if library did. I don't think those are interchangeable enough; :: is (semantically, if not implementation-wise) a strict namespacing operation. :: is also used in package code (whether this is good, bad or mixed is an open question and not the topic of discussion here) so we're talking about unexpected installation. The same goes for the require* functions, which would damage how use of suggested packages works now.

So I'd say if this was done, it really should only be done for library, and as an argument.

I guess my deeper question is whether this is really a major annoyance to people? As far as I know, other interpreted languages (python, julia, etc) don't do this, though I must admit I haven't looked.