Example Timings

Robert M Flight

2019-08-27

Purpose

The goal of importedPackageTimings is to help developers determine if any of the R packages their package depends on make loading their package slow.

To accompmlish this, it uses independent R sessions from the future package to time how long it takes to load each of the packages listed in the Imports and Depends fields of the package in question. Although it will take a long time because it only uses a single core at a time (the only way I could get reliable timings), the times seem to be reliable.

Installation

Currently, importedPackageTimings only exists on Github, so install it with:

remotes::install_github("rmflight/importedPackageTimings")

Example

For example, lets look at a Bioconductor package I’ve seen take a long time to load, xcms.

# not run
library(furrr)
plan(multiprocess)
library(importedPackageTimings)
xcms_time = imported_timings("xcms")

The package provides two types of timings, the time required for the import to load (type = pkg), and then the time required for the package to load after the dependency (type = after).

library(importedPackageTimings)
data("xcms_time")
knitr::kable(head(dplyr::select(xcms_time, -timings)))

package	med	min	max	type	which
xcms	4777753784	4540154629	5047299157	pkg	self
xcms	120612	118835	153174	after	self
mzR	695070041	654965228	726937410	pkg	import
mzR	4188562694	4100112954	4393323772	after	import
BiocGenerics	146176624	132330703	160247180	pkg	import
BiocGenerics	4520280675	4330646417	4579202503	after	import

We can use the pkg entries to see which imports actually take a long time to load, possibly contributing to the long load time of our package in question.

library(ggplot2)
ggplot(dplyr::filter(xcms_time, type %in% "pkg"), 
       aes(x = min / 1e9, y = package)) + 
  geom_point()

From this plot, we can see that MSnbase looks like it is taking the longest to load outside of xcms itself.

We can use the after entries to see which imports after loading have the smallest time to load our package in question, which implies they may be the culprit causing long load times.

ggplot(dplyr::filter(xcms_time, type %in% "after", which %in% "import"),
       aes(x = min / 1e9, y = package)) +
  geom_point()