Leaflet not rendering based on value of new_session

[gregmacfarlane]

I have a book written in Bookdown, in which I would like to include Leaflet maps. I have noticed that my ability to print these maps in the interactive html is a function of

1. the value of the new_session bookdown parameter, which determines whether each Rmd file should be rendered in a new R session.
2. the use of the book_filename bookdown parameter.

Perhaps these are two completely separate issues, but I discovered one while looking for the other and I don't know how intertwined they are.

MWE

---
title: "A Book"
author: "Frida Gomam"
site: bookdown::bookdown_site
documentclass: book
output:
  bookdown::gitbook: default
  #bookdown::pdf_book: default
---

# Test

```{r setup}
library(leaflet)
```


Don't miss Figure \@ref(fig:nice-tab).

```{r nice-tab, fig.cap="Figure"}
leaflet() %>%
  addProviderTiles(providers$CartoDB) %>%
  addMarkers(lng=174.768, lat=-36.852, popup="The birthplace of R")
```

As posted, this renders correctly (as expected) using bookdown::render_book().

Issue 1: HTML dependency based on background layer

If I render as bookdown::render_book(new_session = TRUE) however, I get the following error:

Error: path for html_dependency not found: /var/folders/0f/f4v60fbn6psdh5r_wlqz0qvm0000gn/T//RtmpsCf5pj
Please delete _main.md after you finish debugging the error.

This persists even after clearing the knitr cache and restarting R, as suggested on StackOverflow

I can avoid this error, however, by using a different map tiles layer. I feel like this is a bookdown issue and not a leaflet issue because the performance changes based on a bookdown flag.

Issue 2: Not printing leaflet with named file

If I call the build using bookdown::render_book(config_file = "_bookdown.yml") and alter the name following _bookdown.yml config file:

new_session: yes
book_filename: "mydoc.Rmd"

then the leaflet map is completely empty, regardless of the background map used.

Viewing the page in the debugger shows that there is an htmlwidget in the proper place, but I don't know enough to tell where exactly this is a problem.

session info

R version 4.2.0 (2022-04-22)
Platform: aarch64-apple-darwin20 (64-bit)
Running under: macOS Monterey 12.3.1, RStudio 2022.2.1.461

Locale: en_US.UTF-8 / en_US.UTF-8 / en_US.UTF-8 / C / en_US.UTF-8 / en_US.UTF-8

Package version:
  base64enc_0.1.3 bookdown_0.26.3 bslib_0.3.1     digest_0.6.29   evaluate_0.15  
  fastmap_1.1.0   fs_1.5.2        glue_1.6.2      graphics_4.2.0  grDevices_4.2.0
  highr_0.9       htmltools_0.5.2 jquerylib_0.1.4 jsonlite_1.8.0  knitr_1.39     
  magrittr_2.0.3  methods_4.2.0   R6_2.5.1        rappdirs_0.3.3  rlang_1.0.2    
  rmarkdown_2.14  sass_0.4.1      stats_4.2.0     stringi_1.7.6   stringr_1.4.0  
  tinytex_0.39    tools_4.2.0     utils_4.2.0     xfun_0.31       yaml_2.3.5  

[cderv]

Thanks for the report.

Regarding issue 1

Error: path for html_dependency not found: /var/folders/0f/f4v60fbn6psdh5r_wlqz0qvm0000gn/T//RtmpsCf5pj

It seems the html dependency is downloaded / put into a temp folder for the R session. Using new_session: true will run each file in a different R session, so this can cause issue. I am surprised that it happens only now. I'll see if there were some changes somewhere.

Regarding issue2, we'll see after understanding more about issue 1, unless it happens also when new_session: false

[yihui]

I am surprised that it happens only now.

I'm surprised, too. I did save and restore the HTML dependencies:

bookdown/R/render.R

Line 225 in 169c43b

knit_meta = unlist(lapply(meta, attr, 'knit_meta', exact = TRUE), recursive = FALSE)

Not sure why or when that stopped working.

[cderv]

So after debugging, here is the story:

Traceback for the issue is this one:

<error/rlang_error>
Error:
! path for html_dependency not found: C:/Users/chris/AppData/Local/Temp/Rtmpk1vYTY
---
Backtrace:
  1. bookdown::render_book()
  2. bookdown:::render_new_session(...)
  3. rmarkdown::render(...)
  4. output_format$pre_processor(...)
  5. rmarkdown base(...)
  6. rmarkdown:::html_extras_for_document(...)
  7. rmarkdown dependency_resolver(all_dependencies)
  8. base::lapply(dependencies, validate_html_dependency)
  9. rmarkdown FUN(X[[i]], ...)
 10. rmarkdown:::stop2("path for html_dependency not found: ", file)
Run `rlang::last_trace()` to see the full context.

So issue happens during validation of dependencies in rmarkdown where we check for file existence when a src$file dependency is provided
https://github.com/rstudio/rmarkdown/blob/1ca8a87cc9a6bd009543e636cbd1f253c8d5539f/R/html_dependencies.R#L374

When using leaflet::addProvidersTiles(), it will load a specific HTML dependencies that leaflet creates in a R temp folder
https://github.com/rstudio/leaflet/blob/main/R/plugin-providers.R#L92-L107
This lead to this html dependencies that will be validated by R Markdown:

List of 10
 $ name      : chr "leaflet-providers"
 $ version   : chr "1.9.0"
 $ src       :List of 1
  ..$ file: chr "C:/Users/chris/AppData/Local/Temp/Rtmp618RtY"
 $ meta      : NULL
 $ script    : chr "leaflet-providers_1.9.0.js"
 $ stylesheet: NULL
 $ head      : NULL
 $ attachment: NULL
 $ package   : NULL
 $ all_files : logi FALSE
 - attr(*, "class")= chr "html_dependency"

This temp folder is the one that was set from the R session where the Rmd file containing the leaflet code was knitted. So at the validation stage, it does not exist anymore.

I don't really know if this is expected that leaflet is setting a dependency in an absolute path which is temporary to a R session.
Doesn't seem so right, but could be understandable when used in an R session. Probably we should handle this better on our side.

Regarding why not before: We have touch this validate_html_dependency function in rstudio/rmarkdown#2151 so maybe we missed a edge case.

@yihui do you have insights on this ?

[gregmacfarlane]

Thanks for looking into this. As far as I can tell, the issue with leaflet maps not being included in the output HTML only occurs when the output document is renamed and new_session: yes

[yihui]

@cderv Oh, this reminded me that I had investigated this issue two years ago, which I have totally forgotten: #15 (comment) I'm afraid that this won't be easy to fix for new_session: yes...

[cderv]

Oh thanks. Did not thought to look into old issues.
I think our best shots is to work with our colleague on leaflet to stabilize the dependency. I'll see what we can do on leaflet side.

@gregmacfarlane unfortunately current state is that the tile provides feature of leaflet won't work with bookdown knit and merge approach. You'll have to use new_session: false.