pak vs .Rbuildignore in Package Building - Am I Missing Something?

[alexmundayso]

Question: pak vs .Rbuildignore in Package Building - Am I Missing Something?

Observation: I've noticed that Positron's default package build command (Ctrl+Shift+B) seems to use pak::local_install(), and I'm having issues where files that I expect to be excluded via .Rbuildignore appear to be included in my package. However, I'm not entirely sure if this is actually a pak issue or if I'm doing something wrong.

What I Think I'm Seeing:

• My .Rbuildignore patterns don't seem to be respected when using Ctrl+Shift+B
• Files/directories I expect to be excluded are still being processed
• I can't find a setting to change which package manager Positron uses for building

Specific Example:
I discovered a large file in my .Rproj.user directory that appeared to be getting built into my package. My .Rbuildignore contains the line:

^\.Rproj\.user$

which I expected would exclude the entire .Rproj.user directory, but it seems like pak might not be respecting this pattern.

Questions:

• Is there a setting I'm missing to configure the package build method?
• Should pak::local_install() respect .Rbuildignore by default, or is this expected behavior?
• Am I structuring my .Rbuildignore incorrectly for pak?
• Is there something else I am missing?

Context: Coming from RStudio where package building typically used devtools/R CMD build, so I might be expecting behavior that isn't intended in Positron.

Would appreciate any guidance on whether this is user error on my part or if others have seen similar behavior!

[jennybc]

Just so we understand each other: can you find .Rbuildignored files in the package built by pak::local_install()? And if so, can you point to package source on GitHub?

Or are you seeing evidence of .Rbuildignored files being copied during "build and install"? Because if it's this, it's actually a bug (or infelicity, at least) in R itself. Strange though it may sound, .Rbuildignore is currently enforced after the entire source folder is copied to a temp directory. This is a base R problem. Now pak might have some (attempted?) workaround for this that causes differences in behaviour. I have some vague memory of that, but I can't summon any details at the moment.

So before we go any further, I'd like to know exactly what it is that you're noticing around unhappy handling of .Rbuildignored files.

[alexmundayso]

Hi, thanks for the quick response!

It turns out the issue isn’t actually with .Rbuildignore as initially thought — the package build itself works correctly and ignored files are not included in the final tarball.

The real problem is with pak: during local_install(), it copies the entire package source into a temporary staging folder, and this copy is never cleaned up.

In some of my local packages, I have large .git and .Rproj.user directories, and as a result, my tmp folder becomes very large.

[jennybc]

it copies the entire package source into a temporary staging folder

Just to clarify, what I'm saying above is that this is base R behaviour, not something coming from pak. Here's is the bug on R itself about when .Rbuildignore is enforced after "the big copy" and not before, which causes various problems:

https://bugs.r-project.org/show_bug.cgi?id=18186

... and this copy is never cleaned up.

So, do you think that this is where the pak-specific behaviour lies? I can't see any issues on pak re: having different cleanup behaviour than base R:

https://github.com/r-lib/pak

What I'm really trying to get at is whether pak is doing something different re: copying files, because it's certainly not one of the intended features (i.e. an intended departure from using just the base R tooling).

Do you think it's possible that you're just noticing this infelicity around local package building for the first time (because your current project has a large file in .Rproj.user directory, which is itself an RStudio matter) and you happen to be using Positron at the moment?

[alexmundayso]

I’m new to this and by no means an expert, but here’s what I’ve noticed: when I was developing in RStudio, I never ran into this issue.

Since switching to Positron, my /tmp folder has filled up completely twice in two weeks (I cleared it out after the first time).

From what I can see, RStudio doesn’t create a full copy of the project in /tmp during builds, but Positron does — and I’m not sure why. That seems to be what’s causing the disk space problems on my end.

[jennybc]

Ah, I have another 💡 I can shed: (I think) RStudio just installs the package, whereas what Positron is currently doing is building then installing (because of how pak does things). So that is probably the difference that you're noticing. That is, the both IDEs ultimately are using base R tools, but in 2 different ways.

For reference, in the ?INSTALL help, what we're seeing is two different journeys to package installation:

Both lib and the elements of pkgs may be absolute or relative path names of directories. pkgs may also contain names of package archive files: these are then extracted to a temporary directory.

[juliasilge]

Jenny's questions for you are very relevant and important, to get unstuck.

I also want to highlight that we have heard from multiple folks that our current use of pak for this command is causing problems in #2955. Feel free to add a 👍 to that issue if you think it's relevant for you!

[alexmundayso]

That issue does look like it may be relevant, however, I cannot access the links in its description so it is hard to say.