[R-package][ci] Added more R linters (fixes #2477)

[jameslamb]

lintr released 2.0.0 to CRAN (and r-lintr to conda-forge).

In this PR, I propose adding additional linters on the R code in the project, to enforce particular style. See #2477 for background.

Start with the diff to .ci/lint_r_code.R to see the new linters being added.

One change you will see a LOT of is something like replacing i + 1 with i +1L to satisfy the implicit_integer_linter. This one caught a lot of cases where we were relying on R sometimes treating 1 as double precision and sometimes as integer. This is basically always fine in R, but it can mean you're using more memory than necessary. So think of the addition of the L as a tiny tiny memory savor and, more importantly, documentation for people reading the code.

[jameslamb]

Just rebased to most recent master. I think this is ready for review

[StrikerRUS]

Below are some comments about a CI part.

[StrikerRUS]

It takes soo much time! Is it the only available fix? Maybe it will be faster to update lintr from CRAN?

[jameslamb]

I can try setting up an R environment like I'm doing in #2530 so that we can just use install.packages() directly. Let's see if that speeds it up.

But in general the install process for R packages is notoriously time-consuming :/

[StrikerRUS]

Oh no! I think we should stick to conda's environment, since it's significantly faster than ordinary one and suits to our purpose (just linting) very well.
Sorry for my ambiguous comment above! I just meant to try Rscript -e "install.packages('lintr', repos = 'http://cran.rstudio.com')" to see whether it's faster than reinstalling stringi or not.

But in general the install process for R packages is notoriously time-consuming :/

But not from conda 😉

[jameslamb]

ah I see, ok. Yes I think I misinterpreted what you were saying. I'll change it back and try.

[jameslamb]

Looks like linting task took almost 7 minutes to run still with this approach (using alternative R setup)

[jameslamb]

creating an R environment from conda and then install.packages('lintr') from CRAN, the lint task ran for 8m38s and then failed with

ERROR: dependency ‘curl’ is not available for package ‘httr’

I could add an install.packages('curl') but that would make it run even longer than 8 minutes.

I decided to summarize all the attempts here (and add a few new ideas):

1. (-failure-) R env from conda, lintr from conda-forge
2. (success) R env from conda, lintr from conda-forge, install.packages('stringi'): 6m9s
3. (success) R env from CRAN like [R-package][ci] added CI stage for R package (fixes #2335, fixes #2569) #2530, install.packages('lintr'): 6m48s
4. (-failure-) R env from conda, install.packages('lintr'): 8+ minutes
5. (-failure-) R env from conda, lintr from conda-forge, re-installing stringi from default conda channels: failed
6. (success) R env from conda, installing stringi from conda first, then lintr from conda-forge: 1m57s

So I think we should go with installing stringi explicitly, THEN lintr! This works and runs in under 2 minutes 🎉🎉 🎉

[StrikerRUS]

@jameslamb Thanks a lot for making it work! Please see my comments below.

[jameslamb]

ping @Laurae2 for review on changes to the R code

[StrikerRUS]

I don't think we need progress bars for downloading packages, which only litter logs. Moreover, sometimes progress bar causes errors during the installation process 😮 .

[StrikerRUS]

LGTM for CI scripting part! Thank you very much!

[jameslamb]

just rebased this to most recent master