Project cache error for notebooks in manuscript project

[cderv]

I cannot reproduce your issue using 1.6.42 or 1.7.* but I get another error using latest development version (2546027):

❯ quarto render
[1/2] notebooks/notebook.qmd
ℹ R version 4.4.3 (2025-02-28)
! Config '~/.Rprofile' was loaded!


processing file: notebook.qmd
1/3                  
2/3 [unnamed-chunk-1]
3/3                  
output file: notebook.knit.md

Error adding css vars block TypeError: Cannot read private member #rid from an object whose class did not declare it
    at Kv.get (ext:deno_kv/01_db.ts:57:59)
    at ProjectCacheImpl.get (file:///Users/mcanouil/Projects/quarto/quarto-cli/src/core/cache/cache.ts:137:39)
    at ProjectCacheImpl._getAs (file:///Users/mcanouil/Projects/quarto/quarto-cli/src/core/cache/cache.ts:149:31)
    at ProjectCacheImpl.getString (file:///Users/mcanouil/Projects/quarto/quarto-cli/src/core/cache/cache.ts:192:17)
    at file:///Users/mcanouil/Projects/quarto/quarto-cli/src/core/cache/cache.ts:208:33
    at eventLoopTick (ext:core/01_core.js:175:7)
    at async handleVarsBlock (file:///Users/mcanouil/Projects/quarto/quarto-cli/src/core/sass.ts:384:22)
    at async file:///Users/mcanouil/Projects/quarto/quarto-cli/src/core/sass.ts:415:17
    at async SassCache.setFromHash (file:///Users/mcanouil/Projects/quarto/quarto-cli/src/core/sass/cache.ts:85:7)
    at async compileSass (file:///Users/mcanouil/Projects/quarto/quarto-cli/src/core/sass.ts:161:18)
The resulting CSS file will not have SCSS color variables exported as CSS.
This is likely a Quarto bug.
Please consider reporting it at https://github.com/quarto-dev/quarto-cli,
along with the _quarto_internal_scss_error.scss file that can be found in the current working directory.
ℹ R version 4.4.3 (2025-02-28)
! Config '~/.Rprofile' was loaded!


processing file: notebook.qmd
1/3                  
2/3 [unnamed-chunk-1]
3/3                  
output file: notebook.knit.md

[2/2] index.qmd

Creating MECA Bundle
  3 article files
  5 source files
  1 notebook
  1 other file

Output created: _manuscript/index.html

Originally posted by @mcanouil in #12360

This happens rendenring manuscript project with a simple notebooks/notebooks.qmd

---
title: "Notebook"
---

## A test

```{r}
x <- 1:4
y <- 1:4
plot(x,y)
```

I am surprised our test does not catch that. 🤔

[mcanouil]

I forgot to link with the Git repository/directory: https://github.com/mcanouil/quarto-issues-experiments/tree/main/quarto-cli-12360

[cderv]

Something is not working somehow with the ProjectCacheImpl - it is used for memoization
and this is where it throws issue

quarto-cli/src/core/sass.ts

Lines 379 to 387 in c2d7723

	const handleVarsBlock = async (input: string) => {
	if (!addVarsBlock) {
	return input;
	}
	try {
	const result = await memoizedGetVarsBlock(project, input);
	return input + "\n" + result;
	} catch (e) {
	console.warn("Error adding css vars block", e);

I am currently looking into it.

[cscheid]

We cannot call cloneDeep on an object that holds (or [an object that holds]*) a Deno.KV object. That's the cause of the bug here.

[cderv]

I believe these are the one we do for Notebook resolution

quarto-cli/src/render/notebook/notebook-contributor-qmd.ts

Lines 83 to 112 in c2d7723

	function resolveOutputNotebook(
	nbAbsPath: string,
	_token: string,
	executedFile: ExecutedFile,
	_notebookMetadata?: NotebookMetadata,
	) {
	const resolved = ld.cloneDeep(executedFile);
	resolved.recipe.format.pandoc[kOutputFile] = ipynbOutputFile(nbAbsPath);
	resolved.recipe.output = resolved.recipe.format.pandoc[kOutputFile];
	resolved.recipe.format.pandoc.to = "ipynb";
	// TODO: Allow YAML to pass through as raw or markdown block
	const template = ipynbTitleTemplatePath();
	// Configure echo for this rendering
	resolved.recipe.format.execute.echo = false;
	resolved.recipe.format.execute.warning = false;
	resolved.recipe.format.render[kKeepHidden] = true;
	resolved.recipe.format.render[kNotebookPreserveCells] = true;
	resolved.recipe.format.metadata[kClearHiddenClasses] = "all";
	resolved.recipe.format.metadata[kRemoveHidden] = "none";
	resolved.recipe.format.metadata[kIPynbTitleBlockTemplate] = template;
	resolved.recipe.format.render[kIpynbProduceSourceNotebook] = true;
	resolved.recipe.format.pandoc.citeproc = false;
	// Configure markdown behavior for this rendering
	resolved.recipe.format.metadata[kUnrollMarkdownCells] = false;
	return resolved;
	}

We do that for each contributor type. I'll verify and suggest a change; We do set a lot of value, so we need to project against mutation I guee.

[cscheid]

Great! Maybe the easiest fix is to call cloneDeep and then replace the sasscache in the clone with the original.

[cderv]

yeah - I am looking at making a function to deepClone safely a ProjectContext.
Or maybe deepCloneWithCache for object having DenoKV instance ?

Something like

export function deepCloneWithCache<T>(obj: T): T {
  if (obj === null || typeof obj !== "object") {
    return obj;
  }

  // Skip cloning Deno.Kv instances
  if (obj instanceof Deno.Kv) {
    // Return the original reference
    return obj;
  }

  // Handle arrays
  if (Array.isArray(obj)) {
    return obj.map((item) => deepCloneWithCache(item)) as T;
  }

  // Handle regular objects
  const result = {} as T;
  for (const key in obj) {
    if (Object.prototype.hasOwnProperty.call(obj, key)) {
      result[key] = deepCloneWithCache(obj[key]);
    }
  }

  return result;
}

Good idea ? too much ?

We do shallow copy in some places, but this starts to be hard to follow and not miss anything when handling nested object. If we don't shallow copy every nested object, it seems there is a risk that later processing using the returned value do mutation.

Anyhow, it seems you are suggesting a more scoped version for ProjectContext

export function cloneProjectContext(context: ProjectContext): ProjectContext {
  const cloned = ld.cloneDeep(context);
  cloned.diskCache = context.diskCache;
  
  return cloned;
}

it seems probably safer as more scoped to our class or issues. 🤔

then replace the sasscache in the clone with the original.

It is the diskCache that cause problem - but anyhow, doesn't change the above

[cscheid]

It's a really good idea! I would go even further. Instead of checking for Deno.KV explicitly, we could have an interface for "do-not-clone" objects, and then check for those directly in a new "safeDeepClone" method, and then:

• always use it,
• and always protect uncloneable objects with a wrapper object that implements this interface