fix: take the language from the first known class

CHANGELOG.md

@@ -2,6 +2,10 @@
 
 ## Unreleased
 
+### Bug Fixes
+
+- fix: Use the first class that Pandoc can highlight as the language of a block, so `{.terminal .bash}` is labelled and highlighted as `bash`. Line-numbering classes such as `numberLines` are no longer taken for a language. (#84)
+
 ## 1.8.0 (2026-11-04)
 
 ### New Features

_extensions/code-window/_modules/language.lua

@@ -2,7 +2,7 @@
 --- @license MIT
 --- @copyright 2026 Mickaël Canouil
 --- @author Mickaël Canouil
---- @brief Normalise code blocks with no or unknown language class.
+--- @brief Put the language class of a code block first, or add one.
 
 local cell_output = require(
   quarto.utils.resolve_path('_modules/cell-output.lua'):gsub('%.lua$', ''))
@@ -13,12 +13,16 @@ local M = {}
 
 local known_language_cache = {}
 
+--- Classes that turn on line numbers in Pandoc. They are not languages.
+local NUMBERING_CLASSES = { ['numberLines'] = true, ['number-lines'] = true, ['number'] = true }
+
 --- Check if a language is recognised by Pandoc's syntax highlighter.
---- Renders a test CodeBlock to HTML and checks for the sourceCode class.
+--- Renders a test CodeBlock to HTML and checks that Pandoc names a language
+--- on the code element, not only the sourceCode class.
 --- @param lang string Language identifier
 --- @return boolean
 local function is_known_language(lang)
-  if not lang or lang == '' then
+  if not lang or lang == '' or NUMBERING_CLASSES[lang] then
     return false
   end
 
@@ -28,7 +32,7 @@ local function is_known_language(lang)
 
   local test_block = pandoc.CodeBlock('x', pandoc.Attr('', { lang }))
   local html = pandoc.write(pandoc.Pandoc({ test_block }), 'html')
-  local is_known = html:find('sourceCode') ~= nil
+  local is_known = html:find('<code%s+class="sourceCode%s+[^"]') ~= nil
   known_language_cache[lang] = is_known
   return is_known
 end
@@ -54,32 +58,52 @@ local function set_label(block, label)
   end
 end
 
---- Normalise code blocks with no or unknown language class to "default".
---- Both are framed like a highlighted code cell rather than left as a bare
---- block: a no-language block is labelled "default", and an unknown language
---- keeps its original token as the label. The label is carried on the
---- `code-window-auto-label` attribute (not `filename`, which is reserved for
---- author-set filenames) and consumed by the auto-filename windowing path.
---- The output of an executed cell is left as Quarto wrote it.
+--- Find the class that Pandoc highlights the block with: the first class it
+--- knows as a language, which need not be the first class.
+--- @param classes pandoc.List
+--- @return integer|nil index
+local function find_language(classes)
+  for index, class in ipairs(classes) do
+    if is_known_language(class) then
+      return index
+    end
+  end
+  return nil
+end
+
+--- Find the label for a block with no language: the first class that is not a
+--- line-numbering class.
+--- @param classes pandoc.List
+--- @return string|nil label
+local function find_label(classes)
+  for _, class in ipairs(classes) do
+    if not NUMBERING_CLASSES[class] then
+      return class
+    end
+  end
+  return nil
+end
+
+--- Put the language class first, or add "default" in front when there is none.
+--- The window is named after the first class, so it then names the class that
+--- Pandoc highlights with. The label goes on `code-window-auto-label`, not on
+--- `filename`, which is for author-set names. The output of an executed cell
+--- is left as Quarto wrote it.
 --- @param block pandoc.CodeBlock
 --- @return pandoc.CodeBlock|nil
 function M.CodeBlock(block)
   if cell_output.is_marked(block) then
     return nil
   end
 
-  if not block.classes or #block.classes == 0 then
-    block.classes:insert('default')
-    set_label(block, 'default')
+  local index = find_language(block.classes)
+  if index then
+    block.classes:insert(1, block.classes:remove(index))
     return block
   end
 
-  local lang = block.classes[1]
-  if not is_known_language(lang) then
-    block.classes[1] = 'default'
-    set_label(block, lang)
-  end
-
+  set_label(block, find_label(block.classes) or 'default')
+  block.classes:insert(1, 'default')
   return block
 end
 

docs/examples.qmd

@@ -46,6 +46,8 @@ print("hello")
 The title bar then reads `PYTHON`.
 A derived name is drawn smaller and in capitals, so a reader can tell it from a file that exists on disk.
 A block with no language at all is labelled `DEFAULT`, and a block whose language Pandoc cannot highlight keeps its own token as the label.
+The language is the first class that Pandoc can highlight, so `{.terminal .bash}` is labelled `BASH`, the same as `{.bash .terminal}`.
+Line-numbering classes such as `numberLines` are never the language.
 
 This section shows source only.
 The site sets `auto-filename: false`, so it frames no block that leaves `filename` out.

tests/fixtures/language-not-first.qmd

@@ -0,0 +1,31 @@
+---
+title: "A language class that is not the first class"
+filters:
+  - code-window
+format:
+  html: default
+  typst:
+    keep-typ: true
+---
+
+Pandoc highlights a block with the first class it knows as a language, which
+need not be the first class. The window is named after that same class.
+
+```{.terminal .bash}
+echo terminal
+```
+
+```{.numberLines .bash}
+echo numbered
+```
+
+A block with line numbers and no language keeps its line numbers. Its window is
+named after its first class that does not turn on line numbers.
+
+```{.numberLines}
+x = 1
+```
+
+```{.numberLines .foo}
+x = 2
+```

tests/fixtures/language-relabelled.qmd

@@ -5,8 +5,8 @@ filters:
 ---
 
 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.
+like any other. The label keeps the language the author wrote, and the block
+gains a `default` class in front, which is the one Pandoc has a theme for.
 
 ```foo
 x = 1

tests/run.sh

@@ -237,13 +237,13 @@ fi
 # 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 "${work_dir}/language-relabelled.html" | grep -q 'default' &&
+if block_classes "${work_dir}/language-relabelled.html" | grep -q 'default foo' &&
 	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"
+		"a default class before the foo class, a cw-auto class, and data-filename=\"foo\" on the block"
 fi
 
 # ============================================================================
@@ -504,6 +504,40 @@ else
 		"two blocks with a foo class in auto-label-forced.html, got ${opted_out}"
 fi
 
+# ============================================================================
+# The language is the first class Pandoc can highlight
+# ============================================================================
+
+render language-not-first html
+not_first_html="${work_dir}/language-not-first.html"
+named_bash="$(block_wrappers "${not_first_html}" | grep -c 'data-filename="bash"' || true)"
+if [ "${named_bash}" -eq 2 ] &&
+	! block_classes "${not_first_html}" | grep -q 'default.*bash'; then
+	report pass "language-not-first: a later language class names and highlights the block"
+else
+	report fail "language-not-first: a later language class names and highlights the block" \
+		"two wrappers with data-filename=\"bash\" and no default class before bash, got ${named_bash}"
+fi
+
+if block_wrappers "${not_first_html}" | grep -q 'data-filename="default"' &&
+	block_wrappers "${not_first_html}" | grep -q 'data-filename="foo"' &&
+	[ "$(block_classes "${not_first_html}" | grep -c 'numberSource')" -eq 3 ]; then
+	report pass "language-not-first: a numbered block with no language keeps its numbers"
+else
+	report fail "language-not-first: a numbered block with no language keeps its numbers" \
+		"a numberSource class, which Pandoc writes on a numbered block, on the three numbered blocks, data-filename=\"default\" on one and data-filename=\"foo\" on another"
+fi
+
+render language-not-first typst
+not_first_typ="${work_dir}/language-not-first.typ"
+if [ "$(count_typst_windows "${not_first_typ}")" -eq 4 ] &&
+	[ "$(grep -c '^#code-window(filename: "bash"' "${not_first_typ}" || true)" -eq 2 ]; then
+	report pass "language-not-first: the Typst windows are named after the language"
+else
+	report fail "language-not-first: the Typst windows are named after the language" \
+		"four windows, two of them named bash, in language-not-first.typ"
+fi
+
 # ============================================================================
 
 printf '\n%s passed, %s failed\n' "${passed}" "${failed}"