Make stringr / stringi a soft dependency

[HughParsonage]

knitr's dependency graph includes evaluate and stringr, and thus stringi. Package stringi is a heavy dependency: it takes around a minute to install from source (or several on OSX or Windows) -- a typical use-case for things like Travis-CI. Since knitr (to its immense credit) is required for testing by an enormous number of packages, shaving a minute from each test could have a large payoff. (I also use Travis-CI to deploy knitr+LaTeX documents and installing stringi is a significant component of the build time.)

I'd propose to move stringr to Suggests both here and in evaluate. I'd keep the same verbs as stringr uses, but switch to base R if stringi is not installed. See https://github.com/HughParsonage/TeXCheckR/blob/master/R/suggested_speedups.R for an example of softening the dependencies for readr and stringi.

A glance at the parts of the source currently importing stringr makes me think this would not be too hard. I'd be happy to do the work, but your contribution guidelines say to raise an issue before starting work on something like this.

[yihui]

Thanks for asking first! I'd love to get rid of the stringr/stringi dependencies if possible. Installing stringi also bothers myself a bit. Please feel free to work on it and I'll be happy to review the PR. Thanks!

[jangorecki]

Hi guys, any progress on that issue? knitr is so crucial package that it should be as lightweight as possible. stringr is definitely first candidate to get rid off from mandatory deps.

[yihui]

@jangorecki Sorry, but no yet. As I said above, I'd love to get rid of the stringr dependency. @HughParsonage has kindly contributed a big PR #1549 that I haven't had time to digest yet. I feel it may be easier to achieve the goal if we replace one stringr function at one step (i.e., replace each function in a separate PR). We could start with the easier ones, and decide what to do with the hard bones later.

[cderv]

For later reference, knitr is currently using 13 functions for stringr that would need to be rewrite without stringr.

library(dplyr)
itdepends::dep_usage_pkg("knitr") %>% filter(pkg == "stringr") %>% distinct()
#> Warning: `as.tibble()` is deprecated as of tibble 2.0.0.
#> Please use `as_tibble()` instead.
#> The signature and semantics have changed, see `?as_tibble`.
#> This warning is displayed once every 8 hours.
#> Call `lifecycle::last_warnings()` to see where this warning was generated.
#> # A tibble: 13 x 2
#>    pkg     fun            
#>    <chr>   <chr>          
#>  1 stringr str_detect     
#>  2 stringr str_count      
#>  3 stringr str_locate     
#>  4 stringr str_sub        
#>  5 stringr str_locate_all 
#>  6 stringr str_extract_all
#>  7 stringr str_wrap       
#>  8 stringr str_trim       
#>  9 stringr str_pad        
#> 10 stringr str_dup        
#> 11 stringr str_split      
#> 12 stringr str_match_all  
#> 13 stringr str_replace_all

One thing to watch out for is the performance if we remove stringi as we don't want to lose performance.

[cderv]

For reference, this could be of interest
https://github.com/hadley/stringb

[yihui]

We have finally finished removing stringr from the dependency. We will do more testing and also benchmarking later. Everyone is welcome to help us test and benchmark the dev version! Thanks!

P.S. The benchmark should be done against knitr v1.40 but we did the stringr removal in several commits since September (9c92eff 8fa7d17 1ce8286 67b973d 5a2cb72 1a0f2cc cc3b92a b1ce0c7), so benchmarking can be tricky.

[github-actions]

This old thread has been automatically locked. If you think you have found something related to this, please open a new issue by following the issue guide (https://yihui.org/issue/), and link to this old issue if necessary.