A light-weight tool for tracking R packages. It sits somewhere between reprex and packrat and revolves around CSV files to store key package information.

The main goal of sesh, is to make it simpler to reproduce R code. The ability to restore specific versions/commits of packages, is security for rapid development. And sesh doesn't require anything beyond an Rscript, no Docker, external managers or RStudio projects.



The concept behind sesh is being able to record information about specific package versions for sharing with others, including your future self. Using sesh as part of your workflow gives you extra reprodu-security for individual scripts.

It is very similar in aim to packrat, but doesn't tie you into a RStudio Project. It is also similar to docker but doesn't try to track everything, just the R packages you are using.


Lets pretend we are using two tidyverse pacakges, forcats and tibble, over the course of a two year project. Today we are using the current CRAN versions, but in the past we have used various version and we prefer updating to keep up with the latest features.

# today
library(forcats) # v0.3.0
library(tidyr) # v0.8.1

Saving critial packages for your script

Recording the package versions we use for analysis is important for reproducibily. But part of the open-source experience is being able play around with lots of rapidly changing tools.

Using devtools::session_info() or sessionInfo() to look at this information is great and sesh makes it easier to manage this data. The function sesh() gives an overview of the attached packages, their version and their source.

##   package version                           source
## 1    base   3.5.0                            local
## 2 forcats   0.3.0                   CRAN (R 3.5.0)
## 3    sesh   0.0.1 Github (nathancday/[email protected])
## 4   tidyr   0.8.1                   CRAN (R 3.5.0)

This dataframe is the essense of sesh, a light, easy to share, record of your R session's essential information.

Saving your sesh

To save your current sesh(), use save_sesh(). It will record the vital information to re-load all of your attached [email protected]s and write it to disk as a CSV.

By recording the attached packages and saving it with your analysis, you can pick up scripts in the future with no worries about breaking package changes.

## Saved sesh as: sesh_2018-08-20.csv

The default path, is set up to name the output as "sesh_$SYS-DATE.csv". But it uses the glue package to paste together R variables, so you could include custom gloabls to fit your tastes.

A simple case

The function check_sesh() will compare the currently loaded/installed package versions against a sesh record. It will report which packages are already loaded, installed (but not loaded) or require installation.

check_sesh("sesh_2018-08-20.csv") # check against currently installed versions
## Loaded versions match sesh!

This just confirms our current session info matches the session info we saved two seconds ago, duh. Not a very cool use case

A messier case

We need to reproduce a script we wrote back on October 1st 2017.

To re-create this scenario, we will re-install two common cases:

devtools::install_version("forcats", "0.1.0", reload = F)
devtools::install_github("tidyverse/[email protected]", reload = F)

It is a good idea to restart your R session whenever you install an already attached package, like now.



Ending the time warp

And bringing us back to back to the present



Picking up old code

First thing we do if we are picking up that year-old, dusty script, is look at the packages we used.

## # A tibble: 3 x 3
##   package v     s                               
##   <chr>   <chr> <chr>                           
## 1 base    3.5.0 local                           
## 2 forcats 0.1.0 url                             
## 3 tidyr   0.7.1 Github (tidyverse/[email protected])

Sure, we are a little nervous because perhaps by upgrading our package versions we have inadvertantly broken something. But our new found seshabilities make it easy to replicate specific versions.

Checking diffs

In order to see what pacakges are differnt between the "old script" and our library today, we would use check_sesh to learn the next steps.

## These sesh_version / installed_version do not match:
## Call install_sesh() to safely install.

We see we will need to install the two older copies fo the libraries, so sesh makes a temporary library in ~/.Trash/.

The function install_sesh() will handle that and re-install matching versions. By looking at source and version, it will try to track the right version from CRAN or GitHub and tell you if it was succesful. By keeping a temporary library in ~/.Trash/, sesh contains conflicts and doesn't take up disk space long term, because macOS deletes any files in there after 30 days.

sesh does not touch .libPaths(), so it will not interfer with your globally installed package versions.

## Installing sesh lib in: ~/.Trash/sesh_2017-10-01/
## Using GitHub PAT from envvar GITHUB_PAT
## Downloading GitHub repo tidyverse/[email protected]
## from URL
## Installing tidyr
## Installing rlang
## '/Library/Frameworks/R.framework/Resources/bin/R' --no-site-file  \
##   --no-environ --no-save --no-restore --quiet CMD INSTALL  \
##   '/private/var/folders/09/fpsmqczx5sb6t4v1b66tv7tm0000gp/T/RtmpbNvb69/devtools1de252f7057d/rlang'  \
##   --library='/Users/nathanday/.Trash/sesh_2017-10-01' --install-tests
## '/Library/Frameworks/R.framework/Resources/bin/R' --no-site-file  \
##   --no-environ --no-save --no-restore --quiet CMD INSTALL  \
##   '/private/var/folders/09/fpsmqczx5sb6t4v1b66tv7tm0000gp/T/RtmpbNvb69/devtools1de2148c4d14/tidyverse-tidyr-bd0c6b0'  \
##   --library='/Users/nathanday/.Trash/sesh_2017-10-01' --install-tests
## Downloading package from url:
## Installing forcats
## '/Library/Frameworks/R.framework/Resources/bin/R' --no-site-file  \
##   --no-environ --no-save --no-restore --quiet CMD INSTALL  \
##   '/private/var/folders/09/fpsmqczx5sb6t4v1b66tv7tm0000gp/T/RtmpbNvb69/devtools1de2a370a2f/forcats'  \
##   --library='/Users/nathanday/.Trash/sesh_2017-10-01' --install-tests
## 2 succesful install of 2 needed package.
## Use `sesh_load()` to attach sesh versions.

Great, now we have the the right package versions installed to rerun our "old" script, but as it sits right now the matching versions are not loaded.

We need to call sesh_load() to attach them. We should also restart the R session again because we re-installed loaded packages.

## Joining, by = c("package", "v", "s", "version", "source")
## Loading from sesh library: forcats, tidyr

And re-checking we see...

## Loaded versions match sesh!

All that is left to do now is source that old script!


When you are done working with the past sesh versions, you can go back to your current global versions immediately. Just restart your R session and attach your libraries like normal.

There is also the helper function unload_sesh() to do this without restarting, but it's vunerable to dependencies,



##    package version                           source
## 1     base   3.5.0                            local
## 2 bindrcpp   0.2.2                   CRAN (R 3.5.0)
## 3  forcats   0.3.0                   CRAN (R 3.5.0)
## 4     sesh   0.0.1 Github (nathancday/[email protected])
## 5    tidyr   0.8.1                   CRAN (R 3.5.0)

nathancday/sesh documentation built on Aug. 22, 2018, 11:13 p.m.