Passing an id to a code cell

[andrewpbray]

Description

Question

Is there a way to pass a pandoc id to the top level block that is created by a code cell (the Div with a class cell)?

Examples

When doing a crossref, you can the label key in a code along with the fig- prefix. That passes the value for label as the id of the Image that is created by the cell.

---
format: native
---

```{r}
#| label: fig-histplot

hist(rnorm(5))
```

Pandoc
  Meta { unMeta = fromList [] }
  [ Div
      ( "" , [ "cell" ] , [] )
      [ CodeBlock
          ( "" , [ "r" , "cell-code" ] , [] ) "hist(rnorm(5))"
      , Div
          ( "" , [ "cell-output-display" ] , [] )
          [ Plain
              [ Image
                  ( "fig-histplot" , [] , [] )
                  []
                  ( "Untitled_files/figure-native/fig-histplot-1.png"
                  , ""
                  )
              ]
          ]
      ]
  ]

However, if you try to use that logic to generically assign an id to the enclosing Div with class cell, no luck:

---
format: native
---

```{r}
#| label: histplot

hist(rnorm(5))
```

Pandoc
  Meta { unMeta = fromList [] }
  [ Div
      ( "" , [ "cell" ] , [] )
      [ CodeBlock
          ( "" , [ "r" , "cell-code" ] , [] ) "hist(rnorm(5))"
      , Div
          ( "" , [ "cell-output-display" ] , [] )
          [ Para
              [ Image
                  ( "" , [] , [] )
                  []
                  ( "Untitled_files/figure-native/histplot-1.png" , "" )
              ]
          ]
      ]
  ]

Note that not only is a non-prefixed label not propagated into the AST, there is a filter that scrubs that attribute from the block.

The behavior that I'd like to see

Fundamentally, I'd like to be able to create an AST that looks like

Pandoc
  Meta { unMeta = fromList [] }
  [ Div
      ( "histplot" , [ "cell" ] , [] )
      [ CodeBlock
          ( "" , [ "r" , "cell-code" ] , [] ) "hist(rnorm(5))"
      , Div
          ( "" , [ "cell-output-display" ] , [] )
          [ Para
              [ Image
                  ( "" , [] , [] )
                  []
                  ( "Untitled_files/figure-native/histplot-1.png" , "" )
              ]
          ]
      ]
  ]

where the enclosing Div of class cell that is generated by a code cell can be passed an id.

One approach to implementation would be to assign the value of label in this manner if it doesn't use one of the protected prefixes. Another second would be to create a separate key for this purpose. A third would be to pass label: histplot as an attribute on that enclosing Div so that I can move it from an attribute to an id using a filter.

Thanks!

$ quarto --version
1.5.47

[jimjam-slam]

I also note that id seems to work the same way as label, but classes targets the cell:

---
format: native
---

```{r}
#| id: fig-histplot
#| classes: my-histplot

hist(rnorm(5))
```

Pandoc
  Meta { unMeta = fromList [] }
  [ Div
      ( "" , [ "cell" , "my-histplot" ] , [] )
      [ CodeBlock
          ( "" , [ "r" , "cell-code" ] , [] ) "hist(rnorm(5))"
      , Div
          ( "" , [ "cell-output-display" ] , [] )
          [ Plain
              [ Image
                  ( "fig-histplot" , [] , [] )
                  []
                  ( "id_files/figure-native/fig-histplot-1.png" , "" )
              ]
          ]
      ]
  ]

I'm assuming that the ID moving to the output is part of making crossrefs work in Quarto, but it does complicate our use case, @andrewpbray!

[cderv]

quite a complex topic today - we would need to think of the best way to allow such addition.

The Div of class cell is really something from quarto, with quite some internal usage.

The two mechanism in Quarto are the following:

• classes allows to set classes to Div of class cell.
• Any unknown yaml option will be set as attributes to this Div too. (see Add attributes as code cells option (alongside classes) #6253)

an id of the div is quite specific, and there is no mechanism for id.

This means you could do

---
format: native
---

```{r}
#| custom-id: '#histplot'

hist(rnorm(5))
```

Pandoc
  Meta { unMeta = fromList [] }
  [ Div
      ( "" , [ "cell" ] , [ ( "custom-id" , "#histplot" ) ] )
      [ CodeBlock
          ( "" , [ "r" , "cell-code" ] , [] ) "hist(rnorm(5))"
      , Div
          ( "" , [ "cell-output-display" ] , [] )
          [ Para
              [ Image
                  ( "" , [] , [] )
                  []
                  ( "test_files/figure-native/unnamed-chunk-1-1.png"
                  , ""
                  )
              ]
          ]
      ]
  ]

Then a Lua filter to catch custom-id attributes and move it as id of the block would work.

I also note that id seems to work the same way as label

I can't find it, but I wonder if this is not an alias... 🤔

Specificity of knitr engine.

knitr has a concept of class.* and attr.* options (knitr options)

For Quarto, they will work to pass to Codeblock part.

If you try for example

---
format: native
---

```{r}
#| attr-source: '#histplot'

hist(rnorm(5))
```

Pandoc
  Meta { unMeta = fromList [] }
  [ Div
      ( "" , [ "cell" ] , [] )
      [ CodeBlock
          ( "histplot" , [ "r" , "cell-code" ] , [] ) "hist(rnorm(5))"
      , Div
          ( "" , [ "cell-output-display" ] , [] )
          [ Para
              [ Image
                  ( "" , [] , [] )
                  []
                  ( "test_files/figure-native/unnamed-chunk-1-1.png"
                  , ""
                  )
              ]
          ]
      ]
  ]

attr.output does work in some cases with evaluated content for the fenced code block created for outputs

---
format: native
---

```{r}
#| attr-output: '#histplot'

1 + 1
```

Pandoc
  Meta { unMeta = fromList [] }
  [ Div
      ( "" , [ "cell" ] , [] )
      [ CodeBlock ( "" , [ "r" , "cell-code" ] , [] ) "1 + 1"
      , Div
          ( "" , [ "cell-output" , "cell-output-stdout" ] , [] )
          [ CodeBlock ( "" , [ "#histplot" ] , [] ) "[1] 2" ]
      ]
  ]

Like in knitr directly. No application for plot output for example.

So as you see, several layer of things going on. If this needs improvement, we need to think about it carefully

[andrewpbray]

Thanks @cderv this is helpful!

For now I'm thinking the easiest route will be to write a Lua filter to move #| id: from the output to the Div of class cell. Since all of the docs around crossrefs use #| label:, hopefully we'll be able to avoid situations where we're breaking or changing that functionality. And if it is too messy, we can switch to using a custom attribute name.

[andrewpbray]

Looks like that approach is more brittle than I expected :/

Here is @jimjam-slam 's code block using id: fig-histplot:

```{r}
#| id: fig-histplot
#| classes: my-histplot

hist(rnorm(5))
```

      Div
          ( "" , [ "cell" , "my-histplot" ] , [] )
          [ CodeBlock
              ( "" , [ "r" , "cell-code" ] , [] ) "hist(rnorm(5))"
          , Div
              ( "" , [ "cell-output-display" ] , [] )
              [ Plain
                  [ Image
                      ( "fig-histplot" , [] , [] )
                      []
                      ( "demo_files/figure-native/fig-histplot-1.png" , "")

Here it is if you don't use a protected prefix:

```{r}
#| id: cr-histplot
#| classes: my-histplot

hist(rnorm(5))
```

      Div
          ( "" , [ "cell" , "my-histplot" ] , [] )
          [ CodeBlock
              ( "" , [ "r" , "cell-code" ] , [] ) "hist(rnorm(5))"
          , Div
              ( "" , [ "cell-output-display" ] , [] )
              [ Para                    <<<<<< Changes from Plain to Para >>>>>>>
                  [ Image
                      ( "" , [] , [] ). <<<<<< id disappears >>>>>>>
                      []
                      ( "demo_files/figure-native/fig-histplot-1.png" , "")

It seems like id is aliased with label and also checks for the prefix before it propagates through to the AST as an identifier. Bummer!

So it seems like, barring any changes to how id is handled, our best bet is to use a custom-id?

[cderv]

That is an interesting difference !

id and label are alias for sure. And we trigger different processing if we find a know prefix in the id. Though I don't know if the Para vs Plain is on purpose.

However, we do keep the label in our cell processing when it is a known one. otherwise, it is thrown away. This happens in our engine processing - label has special meaning and so id

This could be in part related to

• Teach engine about new custom crossref #7665

Though I don't think it will help here as id as really a meaning like label. Unless we revert that, or we consider that any label processing should happen in lua and no more at engine level 🤔

[andrewpbray]

I had a chat with @cscheid about this and we tinkered with a couple potential fixes - none of them straightforward! It brought up this thread: #7062 . Even if we solved it for knitr-run cells, Jupyter handles the cell ids in a different manner (in this case, one that'd work better for our use-case).

Standardizing the way the AST looks across engines sounds very tricky, so for now I think we'll move ahead with wrapping the code cells in a div and labeling it using pandoc.

:::{#cr-histplot}
```{r}
hist(rnorm(5))
```
:::