fix: keep a block's own language where no chrome is drawn

CHANGELOG.md

@@ -8,6 +8,7 @@
 
 ### Bug Fixes
 
+- fix: Keep a block's own language where the filter derives no name for it, which means the filter is off, or `auto-filename` is off, or the output format is neither HTML nor Typst. The block was relabelled as `default` before, and the language the author wrote was lost. (#66)
 - fix: Keep inline code in a Typst document as a code element, so Pandoc writes the syntax highlighting definitions for it. A document with inline code and no code block failed to compile before. (#65)
 - fix: Remove the code-window attributes from a block that sets code-window-enabled="false", and from a block that sets code-window-lines when lines-label is off. They reached the HTML output as data-code-window-* attributes before. (#63)
 - fix: Remove the extension's own label attribute when the output format gets no window chrome, so a writer that keeps attributes no longer prints it. (#63)

_extensions/code-window/code-window.lua

@@ -72,6 +72,23 @@ local ANNOTATION_BLOCK_COUNTER = 0
 -- CELL OUTPUT
 -- ============================================================================
 
+--- Check whether the extension acts on the format being rendered. It draws
+--- chrome for html, which covers Reveal.js, and for typst, and leaves every
+--- other format as Quarto writes it.
+--- @return boolean
+local function acts_on_format()
+  return CURRENT_FORMAT == 'html' or CURRENT_FORMAT == 'typst'
+end
+
+--- Check whether this render draws chrome at all: the extension is on, and the
+--- format is one it acts on. Every pass that exists only to serve the chrome
+--- asks this before it does any work, so none of them has to carry its own
+--- copy of the two conditions.
+--- @return boolean
+local function draws_chrome()
+  return CONFIG ~= nil and CONFIG.enabled and acts_on_format()
+end
+
 --- Check whether a block holds the output of an executed cell that the engine
 --- did not name. Such a block keeps the shape Quarto gave it.
 --- @param block pandoc.CodeBlock Code block element
@@ -749,9 +766,9 @@ function Meta(meta)
   -- This is the pass that reads the configuration, so the check runs here,
   -- before the first option is read. An option the check rejects is still
   -- read below, because the report says what the extension cannot use and the
-  -- document renders either way. The extension only acts on html and typst,
-  -- so the check is gated on the same union those formats already use below.
-  if CURRENT_FORMAT == 'html' or CURRENT_FORMAT == 'typst' then
+  -- document renders either way. Only a format the extension acts on reports,
+  -- because nothing it could say applies anywhere else.
+  if acts_on_format() then
     checker:options(meta)
   end
 
@@ -902,9 +919,12 @@ function CodeBlock(block)
   -- A filter that draws nothing changes nothing an author wrote. The
   -- attributes it would read stay on the block and reach the output, which is
   -- also what a document with this extension not installed produces. Only
-  -- code-window-auto-label goes, because the language module wrote it and no
-  -- author did. This holds for a filter switched off, below, and for a format
-  -- the extension does not act on, at the end of this function.
+  -- code-window-auto-label goes. The language module no longer writes it in
+  -- either of the two branches that clear it, since it asks the same question
+  -- before it runs, so what is left to clear is a document that wrote the
+  -- extension's own attribute name on a fence by hand. This holds for a filter
+  -- switched off, below, and for a format the extension does not act on, at
+  -- the end of this function.
   if not CURRENT_FORMAT or not CONFIG or not CONFIG.enabled then
     checker:attributes(block.attributes, 'CodeBlock')
     block.attributes['code-window-auto-label'] = nil
@@ -921,8 +941,9 @@ function CodeBlock(block)
 
   -- Typst is finished by the Pandoc filter ahead of this one, which takes the
   -- attributes off there. Every other format draws no chrome, so the block
-  -- keeps what its author wrote and loses only the language module's label,
-  -- which a writer that preserves attributes would otherwise print.
+  -- keeps what its author wrote and loses only the label, which a writer that
+  -- preserves attributes would otherwise print. Nothing writes that label here
+  -- any more, for the reason given above, so this guards a hand-written one.
   block.attributes['code-window-auto-label'] = nil
   return block
 end
@@ -1308,4 +1329,5 @@ return {
   Pandoc = Pandoc,
   CodeBlock = CodeBlock,
   CONFIG = function() return CONFIG end,
+  draws_chrome = draws_chrome,
 }

_extensions/code-window/main.lua

@@ -42,21 +42,52 @@ code_window.set_checker(checker)
 -- ============================================================================
 
 --- Mark the code blocks that hold the output of an executed cell, so the later
---- passes leave them as Quarto wrote them. Reads the configuration once and
---- walks the document only when the output has to stay unframed. The language
---- pass runs whether the extension is on or off, so the mark is set in both
---- cases; the window passes remove it either way.
+--- passes leave them as Quarto wrote them. Walks the document only when there
+--- is a pass to hold back, which means this render draws chrome and the output
+--- has to stay unframed. Every reader that acts on the mark asks draws_chrome
+--- first: the language pass below, and the two window paths. CodeBlock reads it
+--- too, but only to remove it, and a mark that was never set costs nothing
+--- there. So a render that draws no chrome would walk the whole document to set
+--- an attribute nothing goes on to act on. draws_chrome answers false when
+--- there is no configuration yet, so the second test below always has one in
+--- hand.
 --- @param doc pandoc.Pandoc
 --- @return pandoc.Pandoc|nil Marked document, or nil when the pass is skipped
 local function mark_cell_output(doc)
-  local config = code_window.CONFIG()
-  if not config or (config.enabled and config.cell_output) then
+  if not code_window.draws_chrome() or code_window.CONFIG().cell_output then
     return nil
   end
   doc.blocks = doc.blocks:walk({ Div = cell_output.Div })
   return doc
 end
 
+-- ============================================================================
+-- LANGUAGE
+-- ============================================================================
+
+--- Normalise a block's language where the render draws chrome.
+--- The pass labels a block whose language Pandoc cannot highlight, and the
+--- derived filename is the only reader of that label. Nothing derives a
+--- filename in a render that draws no chrome, so the pass would rewrite a
+--- class for nobody and hand the author back a language they did not write.
+--- "auto-filename" belongs in the same question, because it is the reader
+--- itself: with no derived name to build, both window paths return before they
+--- read the label, so the pass would rewrite a class for nobody again.
+--- Every question this asks is about the render, not about one block. A block
+--- can still draw no chrome inside a render that does, through
+--- "code-window-no-auto-filename" or "code-window-enabled", and its class is
+--- rewritten with no reader either. Answering that per block means relabelling
+--- where the name is built, which is a change to the two window paths rather
+--- than to this gate.
+--- @param block pandoc.CodeBlock
+--- @return pandoc.CodeBlock|nil Relabelled block, or nil when the pass is skipped
+local function normalise_language(block)
+  if not code_window.draws_chrome() or not code_window.CONFIG().auto_filename then
+    return nil
+  end
+  return language.CodeBlock(block)
+end
+
 -- ============================================================================
 -- SKYLIGHTING HOT-FIX
 -- ============================================================================
@@ -88,7 +119,7 @@ end
 local filters = {
   { Meta = code_window.Meta },
   { Pandoc = mark_cell_output },
-  { CodeBlock = language.CodeBlock },
+  { CodeBlock = normalise_language },
   { Pandoc = code_window.Pandoc },
   { CodeBlock = code_window.CodeBlock },
 }

docs/examples.qmd

@@ -165,8 +165,9 @@ It then adds no chrome to any block, and loads neither the stylesheet nor the sc
 So `code-window-enabled="true"` on a block cannot switch the chrome back on.
 `code-window-enabled` can only turn a block off, never on.
 
-Switching the filter off does not put the document back exactly as Quarto would write it.
-The pass that normalises a block's language runs either way, so a block with no language still gains the `default` class.
+With the filter off, every code block keeps the classes Quarto gives it.
+A block whose language Pandoc cannot highlight keeps that language, and a block with no language gains none.
+The attributes the filter reads stay on the block too, and reach the HTML as `data-code-window-*`, which is what a document without the extension also produces.
 
 A block that names a file still gets the plain title bar Quarto builds for it.
 That bar carries no traffic lights, no fold, and no line chip.

tests/fixtures/auto-filename-off.qmd

@@ -0,0 +1,16 @@
+---
+title: "No derived names anywhere"
+filters:
+  - code-window
+extensions:
+  code-window:
+    auto-filename: false
+---
+
+With no derived name to build, nothing reads the label, so the block below keeps
+the language its author wrote even though the filter is on and the format gets
+chrome.
+
+```foo
+x = 1
+```

tests/fixtures/filter-disabled-no-language.qmd

@@ -0,0 +1,13 @@
+---
+title: "The filter turned off, and a block with no language"
+filters:
+  - code-window
+extensions:
+  code-window:
+    enabled: false
+---
+
+A block with no language gains none while the filter is off. The block below is
+indented rather than fenced, which gives the same code block with no language.
+
+    x = 1

tests/fixtures/filter-disabled.qmd

@@ -0,0 +1,16 @@
+---
+title: "The filter turned off"
+filters:
+  - code-window
+extensions:
+  code-window:
+    enabled: false
+---
+
+A filter that is off leaves every block as Quarto writes it. The block below
+keeps the language its author gave it, even though Pandoc cannot highlight it
+and the filter would relabel it when it is on.
+
+```foo
+x = 1
+```

tests/fixtures/language-relabelled.qmd

@@ -0,0 +1,13 @@
+---
+title: "A language Pandoc cannot highlight"
+filters:
+  - code-window
+---
+
+The filter labels a block whose language Pandoc cannot highlight, and frames it
+like any other. The label keeps the language the author wrote, and the class
+becomes `default`, which is the one Pandoc has a theme for.
+
+```foo
+x = 1
+```

tests/fixtures/unsupported-format.qmd

@@ -4,9 +4,9 @@ filters:
   - code-window
 ---
 
-The filter draws no chrome outside HTML and Typst. The language module still
-labels the block below, and that label is the filter's own, so it does not
-reach the output.
+The filter draws no chrome outside HTML and Typst. Nothing reads a label here,
+so the block below keeps the language its author wrote, and the filter's own
+label for it is never written at all.
 
 ```foo
 x = 1

tests/run.sh

@@ -168,8 +168,9 @@ for fixture in disabled-block lines-label-off; do
 	fi
 done
 
-# The label the language module writes is the filter's own, so a format that
-# draws no chrome drops it rather than printing it.
+# The label is the filter's own name for a block, and no reader of it exists on
+# a format that draws no chrome, so neither the label nor the pass that writes
+# it reaches the output.
 render unsupported-format markdown
 if grep -q 'code-window-auto-label' "${work_dir}/unsupported-format.md"; then
 	report fail "unsupported-format: the internal label stays out of the output" \
@@ -178,6 +179,62 @@ else
 	report pass "unsupported-format: the internal label stays out of the output"
 fi
 
+# ============================================================================
+# A filter that draws no chrome leaves a block's language alone
+# ============================================================================
+
+# The pass that relabels a language serves the derived filename, and nothing
+# derives a filename with the filter off.
+render filter-disabled html
+if block_classes "${work_dir}/filter-disabled.html" | grep -q 'foo'; then
+	report pass "filter-disabled: the block keeps its own language"
+else
+	report fail "filter-disabled: the block keeps its own language" \
+		"a foo class on the block"
+fi
+
+# The other branch of the same pass inserts a class where the block had none,
+# which turns a bare block into a highlighted one.
+render filter-disabled-no-language html
+if block_classes "${work_dir}/filter-disabled-no-language.html" | grep -q 'default'; then
+	report fail "filter-disabled-no-language: the block gains no class" \
+		"no default class on the block"
+else
+	report pass "filter-disabled-no-language: the block gains no class"
+fi
+
+# A render with no derived name to build reads no label either, whatever the
+# format, so the pass has no reader there.
+render auto-filename-off html
+if block_classes "${work_dir}/auto-filename-off.html" | grep -q 'foo'; then
+	report pass "auto-filename-off: the block keeps its own language"
+else
+	report fail "auto-filename-off: the block keeps its own language" \
+		"a foo class on the block"
+fi
+
+# The same pass serves no reader on a format that gets no chrome either.
+if grep -q '^``` foo' "${work_dir}/unsupported-format.md"; then
+	report pass "unsupported-format: the block keeps its own language"
+else
+	report fail "unsupported-format: the block keeps its own language" \
+		"a fence reading \`\`\` foo in unsupported-format.md"
+fi
+
+# Where the chrome is drawn, the pass has work to do: the class becomes the one
+# Pandoc has a theme for, the block is framed, and the title bar keeps the
+# language the author wrote. Without this, the two tests above would stay green
+# if the gate ever closed on a render it should let through.
+render language-relabelled html
+if block_classes "${work_dir}/language-relabelled.html" | grep -q 'default' &&
+	block_classes "${work_dir}/language-relabelled.html" | grep -q 'cw-auto' &&
+	block_wrappers "${work_dir}/language-relabelled.html" | grep -q 'data-filename="foo"'; then
+	report pass "language-relabelled: the block is relabelled and framed"
+else
+	report fail "language-relabelled: the block is relabelled and framed" \
+		"a default class, a cw-auto class, and data-filename=\"foo\" on the block"
+fi
+
 # ============================================================================
 
 printf '\n%s passed, %s failed\n' "${passed}" "${failed}"