Use Rstudio editor selection as fallback when clipboard not available

[gadenbuie]

Following up on the conversation we had last night (your morning I guess) on Twitter. I was going to open a long detailed issue but then figured it'd just be easier to write out the code. Of course, I was probably wrong; I ran into a couple twists that I hadn't expected but I think I have them sorted out.

In short, this PR refactors the clipboard reading step into a single function that will attempt to fall back to the current selection in the RStudio editor if the clipboard isn't available. This makes datapasta seamless on RStudio Server.

The overall logic is to test if clipr::clip_available() and if it is then to read from the clipboard. If not, then try reading using the rstudioapi. If that's not available (not in RStudio) then throw the typical clipboard not available error. If we're in RStudio but nothing is selected, then inform the user that they can paste into the editor, select the pasted text, and re-run the addin.

The RStudio API gives a different format than clipr::read_clip() so I wrote the selection to a temp file that I read in with scan(), using the same line as in clipr:::osx_read_clip() and clipr:::X11_read_clip(). The temp file is deleted immediately so I don't think this will cause problems with CRAN.

To test the PR, at least interactively, I discovered that you can set

Sys.setenv("CLIPR_ALLOW" = FALSE)

to temporarily disable clipr and force datapasta to fall back on RStudio. I didn't add any tests, but I did make sure that this PR doesn't break any (more) tests (than are currently failing).

cc @jonthegeek the above gif is from an RStudio Server 😃

[jonthegeek]

Ooooh, that's awesome! That will definitely cover my use case!

It looks like there's something causing an error in the documentation, though...

[gadenbuie]

Awesome!

The checks are failing because I didn't re-render the documentation (didn't want to clutter up the PR)

[gadenbuie]

Spurred by my desire to help @jonthegeek avoid carpal tunnel with crazy keyboard shortcuts, I added another final step that opens an editor with utils::file.edit() for a temp file where the user can just paste their text. The overall chain is then clipboard > selection > text prompt.

[gadenbuie]

I added an option to disable the instructions prompt by setting

options(datapasta.quiet_manual_paste = TRUE)

and I also set it to muffle the instructions after the first prompt per session. (Note I haven't documented this anywhere, not sure where best to put that information.)

[jonthegeek]

and I also set it to muffle the instructions after the first prompt per session. (Note I haven't documented this anywhere, not sure where best to put that information.)

Hmm. I'm not sure about addin documentation in general (definitely something I should learn if it's a thing), but I can imagine a vignette along the lines of "Using datapasta in a browser". Maybe THAT will be left to do on Monday? 😁

[gadenbuie]

It occurred to me that utils::file.edit() doesn't need to be in RStudio, so we could also use the "paste into editor" workflow on terminals where the clipboard isn't available, like a remote server via SSH.

The new "paste" flow looks like this: clipboard > (RStudio? try selection) > text prompt. Here's a short asciicast demo.

[MilesMcBain]

Thanks for this @gadenbuie! I just noticed your fork is quite behind, which explains the test failures. There are none on master currently.

Any advice on how to proceed here? I could change the target branch to a copy of master and pull it there for testing... or you could resync your fork and make a new PR?

[gadenbuie]

ugh. Sorry about that! I use dev in my projects for pre-release code; I guess I just ran with that.

It looks like conflicts are limited primarily to df_paste.R ... it's hard to see on GitHub but I'm pretty sure I can resolve manually. If not I'll rebase my branch on master and open a new PR.

[MilesMcBain]

Ohhhh I see. Easy misake to make. I've got a deadwood cleanup to do.

[gadenbuie]

Okay, so that wasn't too bad! I'm shocked that the PR is still readable. I merged in master and fixed the few minor conflicts so this should be up to date now.

Most of what I wrote is hard to test, but I did extract the function that reads the text gathered by the fallback methods so that we can test that clipr::read_clip() and this method give us the same answer.

[MilesMcBain]

This is really nice work, thanks for the crisp PR @gadenbuie!