Restore opts_current after each code chunk, and also after knit() exits

[yihui]

Will this fix #1988? (except for the support of inline labels)

[cderv]

Yes I believe this is what is needed. This means that Inline code won't have any label.
We can think of something later; but this will probably make some existing code fails.

See related magick issue: ropensci/magick#310
It is using inline code to insert an image and rely on the fact that knitr will provide a name to a file when using plot_counter(). Previously it was using previous chunk name and overriding the previous image, but now it will have no name.

When running the example there, I see for the inline code

• options$fig.path being figure/ instead of test_files/figure-gfm/. Not sure why this options is different.. 🤔
• No label so valid_path(options$fig.path, options$label) is only figure/. Then suffix -1 will apply

It seems opts_chunk$get() and opts_current$get() are not the same.

Should opts_current in inline code be set to opts_chunk maybe ? Like in chunk ?
Inline have no params, but they should probably default to the global options that could have been modified ?

diff --git a/R/block.R b/R/block.R
index 35bbed87..12c486dc 100644
--- a/R/block.R
+++ b/R/block.R
@@ -547,6 +547,8 @@ merge_character = function(res) {
 }

 call_inline = function(block) {
+  optc = opts_current$get(); on.exit(opts_current$restore(optc), add = TRUE)
+  opts_current$restore(opts_chunk$get())
   if (opts_knit$get('progress')) print(block)
   in_input_dir(inline_exec(block))
 }

We could probably also deal with adding a label this way IMO

diff --git a/R/block.R b/R/block.R
index 35bbed87..4e1c5c80 100644
--- a/R/block.R
+++ b/R/block.R
@@ -547,6 +547,9 @@ merge_character = function(res) {
 }

 call_inline = function(block) {
+  optc = opts_current$get(); on.exit(opts_current$restore(optc), add = TRUE)
+  params = opts_chunk$merge(list(label = unnamed_chunk("inline", inline_counter())))
+  opts_current$restore(params)
   if (opts_knit$get('progress')) print(block)
   in_input_dir(inline_exec(block))
 }
diff --git a/R/utils.R b/R/utils.R
index 404dd768..271dae05 100644
--- a/R/utils.R
+++ b/R/utils.R
@@ -17,6 +17,7 @@ knit_counter = function(init = 0L) {
 plot_counter = knit_counter(1L)
 shot_counter = knit_counter(1L)
 chunk_counter = knit_counter(1L)
+inline_counter = knit_counter(1L)

 # a vectorized and better version than evaluate:::line_prompt
 line_prompt = function(x, prompt = getOption('prompt'), continue = getOption('continue')) {

I know using Images in inline code is not the best usage - problaby magick could be fixed to correctly handle inlines.

However, just ideas that you can tweak if you are interested,a s it seems setting label and global options as default for inline seems what should be to me.

@rundel this will partially solve #1988 as discussed.

@gadenbuie this should avoid the need to do this yourself (#1798 (comment))

[yihui]

I've applied your patch with a small change. Thanks!

Now inline code will have access to a unique label in opts_current$get('label').

[cderv]

Cool thanks !

I wonder how many side effect this unamed chunk scheme could cause 😅 it is possible that some user may rely on it too much. Would it invalidate existing cache too ?

Not saying it is bad, but we could probably communicate ahead of time if this could help

[yihui]

Yes, this could be a breaking change, so I've started the revdep check: https://github.com/yihui/crandalf/actions/runs/6644880330/job/18054982641 I'll see how bad this change could be, and write a blog post accordingly.

[yihui]

I've reverted the change in the naming scheme, since the revdep check showed that a few packages would be broken.