More with_* function ideas

[r-lib]

More with_* function ideas

[krlmlr]

Does with_png <- with_(png, dev.off) work for you?

The other two suggestions seem to require some global stack that stores the old values.

[richierocks]

@krlmlr
This definition of with_png gives me an error:

with_png <- with_(png, dev.off)
with_png("test.png", plot(1:10))
## Error in match.arg(antialias, aa.win) : 'arg' must be of length 1

The problem occurs on the line list(bitmap.aa.win = match.arg(antialias, aa.win)), where aa.win is a constant in the grDevices package. The problem is that if the first argument to match.arg isn't a formal argument of a function, then match.arg wants it to be a single string rather than a character vector.

---

A similar function for PDFs gives a different error.

with_pdf <- with_(pdf, dev.off)
with_pdf("test.pdf", plot(1:10))
## Error in if (which == 1) stop("cannot shut down device 1 (the null device)") : 
##   argument is of length zero

The plot does get written to file, but R is trying to shut down the wrong device. If I call dev.off() after the error, then the plot file is correctly generated.

[jimhester]

There was some previous discussion about the device functions at #16 (comment)

[krlmlr]

I forgot. Perhaps:

withr_ <- function(set, reset = set, transform = identity, envir = parent.frame()) { ... }

?

Then:

with_png <- with_(png, dev.off, function(x) dev.cur())

[jimhester]

Yeah I like that approach 👍

[krlmlr]

@richierocks: Once #27 is merged, devices will be easy, e.g., for the pdf() device:

pdf_dev <- wrap(pdf, NULL, dev.cur())
with_pdf <- with_(pdf_dev, dev.off)

Would you like to contribute?

A similar definition of with_png() doesn't show the error you mentioned before (#29 (comment)), but I'm on Ubuntu, and things might be different on your system.

[richierocks]

@krlmlr @jimhester OK, I'm happy with the with_device functions now, so feel free to merge at your leisure.

Still not quite sure how to go about the with_connection/with_db_connection functions.

[richierocks]

@krlmlr @jimhester

Feedback to your queries

Could you please update your roxygen2 installation?

Done.

The names "with_cairo_ps()" and "with_postscript()" look inconsistent to me.

Agreed, but there is a reason. cairo_ps and postscript are the names of the devices in the grDevices package, so I'm consistently following that package's inconsistencies.

Please double-check documentation of this argument.

I'd copied and pasted from the with_locale function. Now updated to use the template suggested by Jim.

antialias <- match.arg(antialias) for jpeg_dev() and png_dev(), why is this necessary?

There are different implementations of grDevices::png and grDevices::jpeg depending upon your OS. On Windows, the functions use a non-standard way of matching the possible values of antialias, and jpeg_dev and png_dev throw an error unless you call match.arg first.

grDevices has man/unix and man/windows folders, it is possible for
external packages to do the same thing?

Not as far as I know, but I've asked the roxygen people.
See Issue r-lib/roxygen2#438

we have been avoiding @importFrom declarations

Fixed.

Another problem

The documentation for the useDingbats argument to grDevices::pdf differs between Unix and Windows: the Unix version has an extra sentence

See the ‘Note’ for a possible fix for some viewers.

When I build withr on Windows, instead of that sentence I get

#ifdef unix not active

and lots of warnings because R can't properly parse the rest of the Rd file.

The same issue occurs with the command argument to grDevices::postscript.

[jimhester]

I think we have included all of these ideas now, thanks again!