Use private event loop for chrome_print(async=FALSE)

[jcheng5]

Note: The diff is pretty minimal if you ignore whitespace.

This PR takes a dependency on the GitHub master version of later. We're hoping to do a release of later soon, so I'm guessing you'll want to wait for that before merging, but I thought I'd create the PR now in case you have questions.

I have not had a chance to fully document the reasons why blocking-while-pumping-events is dangerous unless you use a private event loop, but it looks like @yihui was already aware?

In Chromote we found it necessary to do a lot more work to get user interrupts (i.e. Esc or Ctrl-C) to be respected quickly and to not leak resources. That might not be too big of a concern in chrome_print; the promises::finally() will likely not execute, but in this case the finally doesn't do anything that isn't also being done in on.exit.

I was going to add unit tests, but given the dependency on Chrome/Chromium, I didn't think they would pass on CRAN (or Travis CI for that matter). These are the tests I ran manually:

# sync
{
  later::later(~message("This should print third"))
  pagedown::chrome_print("https://www.google.com") %>%
    { message("This should print first") }
  message("This should print second")
}

# async
library(promises)

{
  later::later(~message("This should print second"))
  pagedown::chrome_print("https://www.google.com", async = TRUE) %...>%
    { message("This should print third") }
  message("This should print first")
}

[cderv]

I was going to add unit tests, but given the dependency on Chrome/Chromium, I didn't think they would pass on CRAN (or Travis CI for that matter)

FWIW we have some tests in crrri that are using chrome on Travis so I guess we could add some tests in pagedown about this. Chrome on Travis in not an issue. I let you see what we have done in crrri and tell us if you think it could be useful here. (and maybe in chromote). I can contribute on that if required.

[yihui]

@jcheng5 If I understand it correctly, the essential change is that you wrapped the code from line 89 in with_temp_loop_maybe() and indented these lines by two spaces. That sounds good to me. Thank you!

@cderv Please feel free to add tests for Travis CI only. I have done such tests in other packages, e.g. https://github.com/yihui/tinytex/blob/master/tests/test-travis.R Thanks!