Bookdown knitting slow when there are gganimate chunks (caching does not help as much as expected)

[nuskcalb]

The time it takes to knit a book (bookdown) that has a gganimate animated plot in it is quite long, even then re-knitting and having cache=TRUE in the code chunks. The expected behavior would be that the first time knitting it takes a while, the second time almost nothing.

Below there are a few more benchmarks, but, in brief (using CRAN gganimate):

• The time for a clean knit is ~50s.
• The second knit with cache=TRUE is ~15s (99% time spent in the final pandoc command)
• Commenting out the animate() part, cached knit time goes down to ~2s
• If I add a second gganimated chunk, clean knit and cached knit times are 2x (~30s when cached; 99% time spent in the final pandoc command)

Weird enough, if I use the Github dev version of gganimate, things are MUCH worse (~170 1st / 110s 2nd knit).

All the code for these tests is in: https://github.com/gorkang/gganimate-bookdown

Posting here before reporting it as a bug, in case I am doing something wrong.

Detailed benchmarks

• A1. CLEAN

  • 48, 50, 55s CRAN version gganimate
  • 89, 91s dev version gganimate

• A2. 2nd run with CACHE

  • 14.5 (99% in pandoc command) CRAN version gganimate
  • 57, 58s dev version gganimate

• B1. CLEAN commenting animate()

  • 2s CRAN version gganimate
  • 3.2 dev version gganimate

• B2. 2nd run with CACHE commenting animate()

  • 1.3s CRAN version gganimate
  • 2.5 dev version gganimate

• C1. CLEAN 2 anime chunks

  • 111s, 108 CRAN version gganimate
  • 173s dev version gganimate

• C2. 2nd run with CACHE

  • 27, 29s CRAN version gganimate
  • 110s dev version gganimate