Detecting text changes in Rmd

[pat-s]

Related #23 workflowr/workflowr#238

I was reading the issues related to {workflowr} + {targets} and currently using the following approach

  tar_target(report_01,
    command = {
      !!tar_knitr_deps_expr("analysis/rep-01.Rmd")
      rmarkdown::render_site("analysis/rep-01.Rmd", quiet = TRUE)
      "docs/rep-01.html"
    },
    format = "file"
  ),

However, this approach does not watch for text-based changes in rep-01.Rmd as it only watches variables (which are extracted via tar_knitr_deps_expr).

IIRC in {drake} knitr_in() was able to also watch for text-based changes in Rmds. Is this somehow possible in {targets}?

[pat-s]

The following seems to work if one wants to watch for dependency and text based changes:

  tar_target(report_01,
    command = {
      !!tar_knitr_deps_expr("analysis/rep-01.Rmd")
      suppressMessages(workflowr::wflow_build("analysis/rep-01.Rmd", verbose = FALSE))
      "analysis/rep-01.Rmd"
    },
    format = "file"
  ),

[petrbouchal]

Just to add one more to the list (sorry!): Rmarkdown websites. They are somewhat more complex (there is a setup file; much of the working dir is automatically copied to the output dir unless excluded) so handling them using the above approach gets quite verbose. But the relevant target factory would not introduce another dependency.

The main value of such a target factory would be in

• detecting all the Rmd files in scope for the website
• watching other _site.yml and files references in _site.yml, e.g. CSS and HTML includes
• watching the output directory as specified in _site.yml.

[wlandau]

R Markdown websites are a neat idea if we can reliably identify all the input and output files.

[ginolhac]

I have this very application working but am tracking things manually with targets. Would be awesome to have some helpers for those websites too!

[wlandau]

I had a look at what render_site() is doing, and I no longer think I have enough information to reliably create a target factory for the general case. render_site() is not always explicit about the files it depends on, and do not think I can infer or replicate all of that faithfully. #49 may turn out to be an alternative, and I would like to see how that goes first.

[rgayler]

FWIW I have just posted workflowr/workflowr#238 (comment) over on {workflowr} showing how I am using {workflowr} + {targets}.

The key point is that I am only using {workflowr} notebooks to document the design of the core computational pipeline, which is managed by {targets}. So it is primarily a {targets} project, with {workflowr} only used to render the design documentation as a website.

The targets for the {workflowr} notebooks use workflowr::wlfow_build() to render the notebooks, but the decision to publish the notebooks strikes me as essentially manual, so I don't manage that aspect with {targets}. When I want to publish I manually invoke workflowr::wflow_publish().

The target for one of these {workflowr} notebooks looks like:

tar_target(
        meta_notebook_1,
        command = {
            !! tar_knitr_deps_expr("meta_notebook_1.Rmd")
            list(core_fn_1, core_fn_2, core_fn_3)
            workflowr::wflow_build("meta_notebook_1.Rmd")
            "meta_notebook_1.Rmd""
        },
        format = "file"
    )

It checks for changes in the targets accessed by tar_read(), the core pipeline functions used in the Rmd file, and changes in the Rmd file itself. However, it does not check for changes in the rendered HTML file, because publishing the notebook would update the rendered file and invalidate the target.

core_fn_1 ... are functions to be used in the core pipeline that are developed and tested in the meta notebooks. So the function is a dependency for the relevant target in the core pipeline and also the meta notebook notebook where it is developed/used.