R CMD check didn't like Cairo::CairoPNG. With this method, check wants
Cairo to be imported in NAMESPACE, but it shouldn't be - Cairo should
be optional.
The `content` function now takes a file path, not writable connection, as an argument.
This makes it much easier to work with APIs that only write to file paths, not
connections.
Some of our examples omit library(shiny) from the top of ui.R and server.R,
which worked fine before but not with the namespace fix from yesterday.
Requiring shiny at startup fixes the problem.
See in particular:
http://stackoverflow.com/questions/13575353/how-does-the-shiny-r-package-deal-with-data-frames
Also reported at different times by Dirk Eddelbuettel and Jay Emerson.
The observed behavior is that S3/S4 method dispatch does not always seem to
work; the desired methods are not invoked despite appearing to be in the
search path.
The problem was that sourcing files with local=TRUE creates a new environment
based on the parent frame, which in our case is Shiny's package environment.
What we really want is to read from the global environment but write to a
throwaway environment. The correct way to do that is to make a new environment
with .GlobalEnv as the parent.
Very simple tweak to R's untar2 code was all that was
required to fix the built-in untar's problems with
gists. Seemed best to just fork it and start using
the forked version directly, regardless of what is
installed on your machine.
There is a deeper problem here, that reactiveUI output that renders stuff to the <head> will generally not work. We're not in a position to fix that yet and this problem has been reported twice, so we'll just fix this instance by making the slider dependencies built into the framework.
