From 4d6e55fd770eb1c81b130ca3b7af5840c17c8c68 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Micka=C3=ABl=20Canouil?=
 <8896044+mcanouil@users.noreply.github.com>
Date: Thu, 5 Nov 2026 10:13:13 +0100
Subject: [PATCH 1/6] fix: take the language from the first known class

Pandoc highlights a block with the first class it knows as a language, but
the language pass read only the first class. A block such as {.terminal .bash}
lost its language to default and was named after terminal. The probe also
took numberLines, number-lines and number for languages, because Pandoc
writes the sourceCode class on any numbered block.

The pass now moves the first class Pandoc can highlight to the front. A block
with none gets default in front of its own classes, so line numbers survive.
---
 _extensions/code-window/_modules/language.lua | 72 ++++++++++++++-----
 docs/examples.qmd                             |  3 +
 tests/fixtures/language-not-first.qmd         | 26 +++++++
 tests/fixtures/language-relabelled.qmd        |  4 +-
 tests/run.sh                                  | 33 +++++++++
 5 files changed, 117 insertions(+), 21 deletions(-)
 create mode 100644 tests/fixtures/language-not-first.qmd

diff --git a/_extensions/code-window/_modules/language.lua b/_extensions/code-window/_modules/language.lua
index d4763a7..3624008 100644
--- a/_extensions/code-window/_modules/language.lua
+++ b/_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$', ''))
@@ -13,8 +13,14 @@ local M = {}
 
 local known_language_cache = {}
 
+--- Classes that turn on line numbers in Pandoc. They are not languages, and a
+--- block that has only these classes has no language.
+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. The sourceCode class alone is not enough, because
+--- Pandoc also writes it for a numbered block that it does not highlight.
 --- @param lang string Language identifier
 --- @return boolean
 local function is_known_language(lang)
@@ -28,7 +34,7 @@ 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 class="sourceCode ', 1, true) ~= nil
   known_language_cache[lang] = is_known
   return is_known
 end
@@ -54,13 +60,43 @@ 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" when the block has none.
+--- The window is named after the first class, and Pandoc highlights with the
+--- first class it knows, so the language class is moved to the front and both
+--- read the same class. A block with no language is framed like a highlighted
+--- code cell rather than left as a bare block: "default" is added in front of
+--- its classes, so that the other classes, line numbers among them, still
+--- apply. The block is labelled with its first class that is not a
+--- line-numbering class, or with "default" when it has none. 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.
 --- @param block pandoc.CodeBlock
 --- @return pandoc.CodeBlock|nil
 function M.CodeBlock(block)
@@ -68,18 +104,16 @@ function M.CodeBlock(block)
     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
+    if index > 1 then
+      block.classes:insert(1, block.classes:remove(index))
+    end
     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
 
diff --git a/docs/examples.qmd b/docs/examples.qmd
index a0a71b2..21dfe6f 100644
--- a/docs/examples.qmd
+++ b/docs/examples.qmd
@@ -46,6 +46,9 @@ 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, which is also the class Pandoc highlights the block with.
+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.
diff --git a/tests/fixtures/language-not-first.qmd b/tests/fixtures/language-not-first.qmd
new file mode 100644
index 0000000..3a7ffe0
--- /dev/null
+++ b/tests/fixtures/language-not-first.qmd
@@ -0,0 +1,26 @@
+---
+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.
+
+```{.numberLines}
+x = 1
+```
diff --git a/tests/fixtures/language-relabelled.qmd b/tests/fixtures/language-relabelled.qmd
index 84d77a6..531cc92 100644
--- a/tests/fixtures/language-relabelled.qmd
+++ b/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
diff --git a/tests/run.sh b/tests/run.sh
index 15b07c4..40f1c8e 100755
--- a/tests/run.sh
+++ b/tests/run.sh
@@ -504,6 +504,39 @@ 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"' &&
+	[ "$(block_classes "${not_first_html}" | grep -c 'numberLines')" -eq 2 ]; 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 numberLines class on both numbered blocks, and data-filename=\"default\" on one"
+fi
+
+render language-not-first typst
+not_first_typ="${work_dir}/language-not-first.typ"
+if [ "$(grep -c '^#code-window(filename: "bash"' "${not_first_typ}" || true)" -eq 2 ] &&
+	[ "$(grep -c '^#code-window(filename: "default"' "${not_first_typ}" || true)" -eq 1 ]; 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" \
+		"two windows named bash and one named default in language-not-first.typ"
+fi
+
 # ============================================================================
 
 printf '\n%s passed, %s failed\n' "${passed}" "${failed}"

From d3d0275d61e71287966ba3af4741c14a21f6f598 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Micka=C3=ABl=20Canouil?=
 <8896044+mcanouil@users.noreply.github.com>
Date: Thu, 5 Nov 2026 10:15:40 +0100
Subject: [PATCH 2/6] refactor: tighten the language pass and its tests

---
 _extensions/code-window/_modules/language.lua | 28 ++++++-------------
 docs/examples.qmd                             |  3 +-
 tests/run.sh                                  |  6 ++--
 3 files changed, 13 insertions(+), 24 deletions(-)

diff --git a/_extensions/code-window/_modules/language.lua b/_extensions/code-window/_modules/language.lua
index 3624008..2611859 100644
--- a/_extensions/code-window/_modules/language.lua
+++ b/_extensions/code-window/_modules/language.lua
@@ -13,18 +13,16 @@ local M = {}
 
 local known_language_cache = {}
 
---- Classes that turn on line numbers in Pandoc. They are not languages, and a
---- block that has only these classes has no language.
+--- 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 that Pandoc names a language
---- on the code element. The sourceCode class alone is not enough, because
---- Pandoc also writes it for a numbered block that it does not highlight.
+--- 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
 
@@ -86,17 +84,11 @@ local function find_label(classes)
   return nil
 end
 
---- Put the language class first, or add "default" when the block has none.
---- The window is named after the first class, and Pandoc highlights with the
---- first class it knows, so the language class is moved to the front and both
---- read the same class. A block with no language is framed like a highlighted
---- code cell rather than left as a bare block: "default" is added in front of
---- its classes, so that the other classes, line numbers among them, still
---- apply. The block is labelled with its first class that is not a
---- line-numbering class, or with "default" when it has none. 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.
+--- 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)
@@ -106,9 +98,7 @@ function M.CodeBlock(block)
 
   local index = find_language(block.classes)
   if index then
-    if index > 1 then
-      block.classes:insert(1, block.classes:remove(index))
-    end
+    block.classes:insert(1, block.classes:remove(index))
     return block
   end
 
diff --git a/docs/examples.qmd b/docs/examples.qmd
index 21dfe6f..411de5d 100644
--- a/docs/examples.qmd
+++ b/docs/examples.qmd
@@ -46,8 +46,7 @@ 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, which is also the class Pandoc highlights the block with.
-So `{.terminal .bash}` is labelled `BASH`, the same as `{.bash .terminal}`.
+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.
diff --git a/tests/run.sh b/tests/run.sh
index 40f1c8e..fd85f14 100755
--- a/tests/run.sh
+++ b/tests/run.sh
@@ -529,12 +529,12 @@ fi
 
 render language-not-first typst
 not_first_typ="${work_dir}/language-not-first.typ"
-if [ "$(grep -c '^#code-window(filename: "bash"' "${not_first_typ}" || true)" -eq 2 ] &&
-	[ "$(grep -c '^#code-window(filename: "default"' "${not_first_typ}" || true)" -eq 1 ]; then
+if [ "$(count_typst_windows "${not_first_typ}")" -eq 3 ] &&
+	[ "$(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" \
-		"two windows named bash and one named default in language-not-first.typ"
+		"three windows, two of them named bash, in language-not-first.typ"
 fi
 
 # ============================================================================

From 40805afca11b6ee241e97d0a2793c3cd122f8fe2 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Micka=C3=ABl=20Canouil?=
 <8896044+mcanouil@users.noreply.github.com>
Date: Thu, 5 Nov 2026 10:23:23 +0100
Subject: [PATCH 3/6] fix: let the language probe match a wrapped tag

---
 _extensions/code-window/_modules/language.lua | 2 +-
 tests/fixtures/language-not-first.qmd         | 7 ++++++-
 tests/run.sh                                  | 9 +++++----
 3 files changed, 12 insertions(+), 6 deletions(-)

diff --git a/_extensions/code-window/_modules/language.lua b/_extensions/code-window/_modules/language.lua
index 2611859..20f09e4 100644
--- a/_extensions/code-window/_modules/language.lua
+++ b/_extensions/code-window/_modules/language.lua
@@ -32,7 +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('<code class="sourceCode ', 1, true) ~= nil
+  local is_known = html:find('<code%s+class="sourceCode%s+[^"]') ~= nil
   known_language_cache[lang] = is_known
   return is_known
 end
diff --git a/tests/fixtures/language-not-first.qmd b/tests/fixtures/language-not-first.qmd
index 3a7ffe0..f1385ed 100644
--- a/tests/fixtures/language-not-first.qmd
+++ b/tests/fixtures/language-not-first.qmd
@@ -19,8 +19,13 @@ echo terminal
 echo numbered
 ```
 
-A block with line numbers and no language keeps its line numbers.
+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
+```
diff --git a/tests/run.sh b/tests/run.sh
index fd85f14..42b89e1 100755
--- a/tests/run.sh
+++ b/tests/run.sh
@@ -520,21 +520,22 @@ else
 fi
 
 if block_wrappers "${not_first_html}" | grep -q 'data-filename="default"' &&
-	[ "$(block_classes "${not_first_html}" | grep -c 'numberLines')" -eq 2 ]; then
+	block_wrappers "${not_first_html}" | grep -q 'data-filename="foo"' &&
+	[ "$(block_classes "${not_first_html}" | grep -c 'numberLines')" -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 numberLines class on both numbered blocks, and data-filename=\"default\" on one"
+		"a numberLines class 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 "${not_first_typ}")" -eq 3 ] &&
+if [ "$(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" \
-		"three windows, two of them named bash, in language-not-first.typ"
+		"four windows, two of them named bash, in language-not-first.typ"
 fi
 
 # ============================================================================

From 6bf5ce14270fbe2b94afd1708136e865a7409430 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Micka=C3=ABl=20Canouil?=
 <8896044+mcanouil@users.noreply.github.com>
Date: Thu, 5 Nov 2026 10:30:33 +0100
Subject: [PATCH 4/6] test: check line numbers through numberSource

---
 tests/run.sh | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/tests/run.sh b/tests/run.sh
index 42b89e1..e2e5ba8 100755
--- a/tests/run.sh
+++ b/tests/run.sh
@@ -512,7 +512,7 @@ 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 "${not_first_html}" | grep -q 'default bash'; then
+	! block_classes "${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" \
@@ -521,11 +521,11 @@ 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 'numberLines')" -eq 3 ]; then
+	[ "$(block_classes "${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 numberLines class on the three numbered blocks, data-filename=\"default\" on one and data-filename=\"foo\" on another"
+		"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

From 15a5be02723ae095b94410bcf9286434daaeb83f Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Micka=C3=ABl=20Canouil?=
 <8896044+mcanouil@users.noreply.github.com>
Date: Thu, 5 Nov 2026 10:39:50 +0100
Subject: [PATCH 5/6] test: pin the class kept for an unknown language

---
 tests/run.sh | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/tests/run.sh b/tests/run.sh
index e2e5ba8..d6d17ed 100755
--- a/tests/run.sh
+++ b/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
 
 # ============================================================================

From e6f6aa4310d61620f56be8cb9815509d0ab14625 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Micka=C3=ABl=20Canouil?=
 <8896044+mcanouil@users.noreply.github.com>
Date: Thu, 5 Nov 2026 10:40:03 +0100
Subject: [PATCH 6/6] docs: add changelog entry for the language fix

---
 CHANGELOG.md | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index efe2b0a..24ee76d 100644
--- a/CHANGELOG.md
+++ b/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