Working interactively on Quarto document inside an R project with relative paths

[Art-Ink]

System details:

Positron and OS details:

Positron Version: 2025.01.0 (user setup) build 159
Code - OSS Version: 1.95.0
Commit: bc8820b
Date: 2025-01-09T23:52:00.440Z
Electron: 32.2.1
Chromium: 128.0.6613.186
Node.js: 20.18.0
V8: 12.8.374.38-electron.0
OS: Windows_NT x64 10.0.22631

Interpreter details:

R 4.4.2

Describe the issue:

I keep Quarto documents in a documents folder inside my R project folder and source certain functions from R files in a spearate R folder on the same level. I also maintain the project with renv.

When I use relative paths for the source command (../R/some_function.r) I can render the document via the cli without issues from the project directory. Hoewever, when I interactively work with the Quarto document all chunks are evaluated using the project directory as WD, not the project/documents directory. causing the source path fail to resolve.

Adjusting the working directory fixes the issue but causes the R session to not detect the project associated renv leading to a different mode of failure.

This is not the case in RStudio where I also work on the project level directory but the Quarto file automatically uses its own position to resolve relative paths when executed interactively.

Expected or desired behavior:

A Quarto document executed interactively inside a subdirectory of a R project should resolve relative paths relative to its own location, not the project WD as it is done in RStudio.

[juliasilge]

• Related to Provide options to specify working directory for Quarto, .Rmd, notebooks #4692 and especially see the notes I made in Provide options to specify working directory for Quarto, .Rmd, notebooks #4692 (comment)
• Also related to How to activate renv in notebook #4695

[jmcphers]

Hoewever, when I interactively work with the Quarto document all chunks are evaluated using the project directory as WD

What specific gesture are you using to execute code in this case? (Run Cell? Cmd + Enter? something else?)

RStudio has special behavior when you run Quarto or R Markdown chunks that causes it to temporarily switch the working directory to the document's directory, evaluate the code, then switch it back. This mostly does what people expect but can also be confusing since the directory shown in the Console is not always the directory in which code is actually run. It can also lead to confusion if you make further changes to the working directory from within a Quarto cell (now what do we change it to?)

Positron (currently) sends code to the console more or less as though you'd typed it, so you get whatever the Console shows (always). You can switch to the document's directory by right-clicking on it in the Explorer and running Set as Working Directory in Active Console.

Just spitballing, what would you think about having a separate R console per quarto doc/folder that has its working directory pre-set to the document directory? We have multi-console landing shortly and I wonder if that'd be easier to reason about than swapping the directory back and forth in a single active session.

[Art-Ink]

What specific gesture are you using to execute code in this case? (Run Cell? Cmd + Enter? something else?)

I use both run cell und Cmd + Enter to work interactively, in both cases the path does not resolve since the WD is not at the .qmd location. As juliasilge mentioned in one related topic I might have to work around that with here somehow.

RStudio has special behavior when you run Quarto or R Markdown chunks that causes it to temporarily switch the working directory to the document's directory, evaluate the code, then switch it back. This mostly does what people expect but can also be confusing since the directory shown in the Console is not always the directory in which code is actually run. It can also lead to confusion if you make further changes to the working directory from within a Quarto cell (now what do we change it to?)

Yes I have learned since yesterday that what I took as a natural way to handle things is in fact not so simple :) I would assume an option to choose the desired behavior as is possible in RStudio might be useful here.

Positron (currently) sends code to the console more or less as though you'd typed it, so you get whatever the Console shows (always). You can switch to the document's directory by right-clicking on it in the Explorer and running Set as Working Directory in Active Console.

This fails a couple lines later when I use project specific packages and do not find the renv environment. It is a workaround but a tedious one.

Just spitballing, what would you think about having a separate R console per quarto doc/folder that has its working directory pre-set to the document directory? We have multi-console landing shortly and I wonder if that'd be easier to reason about than swapping the directory back and forth in a single active session.

As long as this R console is forwarded the renv lib paths that might work.

Thank you for the great work!

[teecrow]

I echo the comments from @Art-Ink - in Positron, I would prefer to have an option for the same behavior as Rstudio, where running getwd() interactively within a code chunk in a QMD file (using e.g., CTRL+Enter) returns the working directory of the document (in my case Projects/Project1/Reports/) whereas running getwd() interactively in the console may return a different working directory (in my case Projects/Project1/). Meanwhile, running getwd() interactively within a .R script in the same session returns Projects/Project1/, as it would if I ran it from the console, as expected. (FWIW I always found the Rstudio default behavior a bit strange, although I've gotten used to QMD files and .R files having a different working directory when running lines interactively.)

Switching to Positron has me uncertain how best to handle this: should I change some option in Rstudio to get the same behavior as in Positron, and then change all my QMD files? Or perhaps explicitly set my desired folders with the {here} package in the setup chunk (or the YAML?) of my QMD files? Something else?

[juliasilge]

I'll paste in here the discussion I outlined in #4692 (comment) to summarize current options for folks:

OK, I am going to sum up the current situation:

• The defaults right now are similar to RStudio's behavior; we use the file's directory as the working directory when rendering, and there are no changes to the console's working directory for code coming from .Rmd or .qmd. All of this can be set/changed by the user, using here::here() or the knit_root_dir argument to rmarkdown::render() or by explicitly changing the working directory.
• Positron does not have inline output, which is I bet what the original discussion was getting at. Setting the working directory in inline output makes sense, but it doesn't really for send-to-console execution.
• We could address this use case in the future with a treatment similar to what we land on for Python cells are evaluated in the project or current working directory #3156, for example providing a way to spawn a new, quasi-notebook-like console session for .Rmd and .qmd files.