Merge pull request #130047 from NixOS/doc-manpage-role

doc: Add support for manpage references

.github/CODEOWNERS

@@ -42,6 +42,12 @@
 # Nixpkgs build-support
 /pkgs/build-support/writers @lassulus @Profpatsch
 
+# Nixpkgs documentation
+/maintainers/scripts/db-to-md.sh @jtojnar @ryantm
+/maintainers/scripts/doc @jtojnar @ryantm
+/doc/build-aux/pandoc-filters @jtojnar
+/doc/contributing/contributing-to-documentation.chapter.md @jtojnar
+
 # NixOS Internals
 /nixos/default.nix          @nbp @infinisil
 /nixos/lib/from-env.nix     @nbp @infinisil

doc/Makefile

@@ -3,12 +3,17 @@ MD_TARGETS=$(addsuffix .xml, $(basename $(shell find . -type f -regex '.*\.md$$'
 PANDOC ?= pandoc
 
 pandoc_media_dir = media
-# NOTE: Keep in sync with NixOS manual (/nixos/doc/manual/md-to-db.sh).
+# NOTE: Keep in sync with NixOS manual (/nixos/doc/manual/md-to-db.sh) and conversion script (/maintainers/scripts/db-to-md.sh).
 # TODO: Remove raw-attribute when we can get rid of DocBook altogether.
 pandoc_commonmark_enabled_extensions = +attributes+fenced_divs+footnotes+bracketed_spans+definition_lists+pipe_tables+raw_attribute
+# Not needed:
+# - docbook-reader/citerefentry-to-rst-role.lua (only relevant for DocBook → MarkDown/rST/MyST)
 pandoc_flags = --extract-media=$(pandoc_media_dir) \
 	--lua-filter=$(PANDOC_LUA_FILTERS_DIR)/diagram-generator.lua \
-	--lua-filter=labelless-link-is-xref.lua \
+	--lua-filter=build-aux/pandoc-filters/myst-reader/roles.lua \
+	--lua-filter=build-aux/pandoc-filters/link-unix-man-references.lua \
+	--lua-filter=build-aux/pandoc-filters/docbook-writer/rst-roles.lua \
+	--lua-filter=build-aux/pandoc-filters/docbook-writer/labelless-link-is-xref.lua \
 	-f commonmark$(pandoc_commonmark_enabled_extensions)+smart
 
 .PHONY: all

doc/build-aux/pandoc-filters/docbook-reader/citerefentry-to-rst-role.lua

@@ -0,0 +1,23 @@
+--[[
+Converts Code AST nodes produced by pandoc’s DocBook reader
+from citerefentry elements into AST for corresponding role
+for reStructuredText.
+
+We use subset of MyST syntax (CommonMark with features from rST)
+so let’s use the rST AST for rST features.
+
+Reference: https://www.sphinx-doc.org/en/master/usage/restructuredtext/roles.html#role-manpage
+]]
+
+function Code(elem)
+  elem.classes = elem.classes:map(function (x)
+    if x == 'citerefentry' then
+      elem.attributes['role'] = 'manpage'
+      return 'interpreted-text'
+    else
+      return x
+    end
+  end)
+
+  return elem
+end

doc/labelless-link-is-xref.lua → doc/build-aux/pandoc-filters/docbook-writer/labelless-link-is-xref.lua

@@ -1,3 +1,13 @@
+--[[
+Converts Link AST nodes with empty label to DocBook xref elements.
+
+This is a temporary script to be able use cross-references conveniently
+using syntax taken from MyST, while we still use docbook-xsl
+for generating the documentation.
+
+Reference: https://myst-parser.readthedocs.io/en/latest/using/syntax.html#targets-and-cross-referencing
+]]
+
 local function starts_with(start, str)
   return str:sub(1, #start) == start
 end

doc/build-aux/pandoc-filters/docbook-writer/rst-roles.lua

@@ -0,0 +1,36 @@
+--[[
+Converts AST for reStructuredText roles into corresponding
+DocBook elements.
+
+Currently, only a subset of roles is supported.
+
+Reference:
+  List of roles:
+    https://www.sphinx-doc.org/en/master/usage/restructuredtext/roles.html
+  manpage:
+    https://tdg.docbook.org/tdg/5.1/citerefentry.html
+  file:
+    https://tdg.docbook.org/tdg/5.1/filename.html
+]]
+
+function Code(elem)
+  if elem.classes:includes('interpreted-text') then
+    local tag = nil
+    local content = elem.text
+    if elem.attributes['role'] == 'manpage' then
+      tag = 'citerefentry'
+      local title, volnum = content:match('^(.+)%((%w+)%)$')
+      if title == nil then
+        -- No volnum in parentheses.
+        title = content
+      end
+      content = '<refentrytitle>' .. title .. '</refentrytitle>' .. (volnum ~= nil and ('<manvolnum>' .. volnum .. '</manvolnum>') or '')
+    elseif elem.attributes['role'] == 'file' then
+      tag = 'filename'
+    end
+
+    if tag ~= nil then
+      return pandoc.RawInline('docbook', '<' .. tag .. '>' .. content .. '</' .. tag .. '>')
+    end
+  end
+end

doc/build-aux/pandoc-filters/link-unix-man-references.lua

@@ -0,0 +1,18 @@
+--[[
+Turns a manpage reference into a link, when a mapping is defined
+in the unix-man-urls.lua file.
+]]
+
+local man_urls = {
+  ["tmpfiles.d(5)"] = "https://www.freedesktop.org/software/systemd/man/tmpfiles.d.html",
+  ["nix.conf(5)"] = "https://nixos.org/manual/nix/stable/#sec-conf-file",
+  ["systemd.time(7)"] = "https://www.freedesktop.org/software/systemd/man/systemd.time.html",
+  ["systemd.timer(5)"] = "https://www.freedesktop.org/software/systemd/man/systemd.timer.html",
+}
+
+function Code(elem)
+  local is_man_role = elem.classes:includes('interpreted-text') and elem.attributes['role'] == 'manpage'
+  if is_man_role and man_urls[elem.text] ~= nil then
+    return pandoc.Link(elem, man_urls[elem.text])
+  end
+end

doc/build-aux/pandoc-filters/myst-reader/roles.lua

@@ -0,0 +1,29 @@
+--[[
+Replaces Str AST nodes containing {role}, followed by a Code node
+by a Code node with attrs that would be produced by rST reader
+from the role syntax.
+
+This is to emulate MyST syntax in Pandoc.
+(MyST is a CommonMark flavour with rST features mixed in.)
+
+Reference: https://myst-parser.readthedocs.io/en/latest/syntax/syntax.html#roles-an-in-line-extension-point
+]]
+
+function Inlines(inlines)
+  for i = #inlines-1,1,-1 do
+    local first = inlines[i]
+    local second = inlines[i+1]
+    local correct_tags = first.tag == 'Str' and second.tag == 'Code'
+    if correct_tags then
+      -- docutils supports alphanumeric strings separated by [-._:]
+      -- We are slightly more liberal for simplicity.
+      local role = first.text:match('^{([-._+:%w]+)}$')
+      if role ~= nil then
+        inlines:remove(i)
+        second.attributes['role'] = role
+        second.classes:insert('interpreted-text')
+      end
+    end
+  end
+  return inlines
+end

doc/build-aux/pandoc-filters/myst-writer/roles.lua

@@ -0,0 +1,25 @@
+--[[
+Replaces Code nodes with attrs that would be produced by rST reader
+from the role syntax by a Str AST node containing {role}, followed by a Code node.
+
+This is to emulate MyST syntax in Pandoc.
+(MyST is a CommonMark flavour with rST features mixed in.)
+
+Reference: https://myst-parser.readthedocs.io/en/latest/syntax/syntax.html#roles-an-in-line-extension-point
+]]
+
+function Code(elem)
+  local role = elem.attributes['role']
+
+  if elem.classes:includes('interpreted-text') and role ~= nil then
+    elem.classes = elem.classes:filter(function (c)
+      return c ~= 'interpreted-text'
+    end)
+    elem.attributes['role'] = nil
+
+    return {
+      pandoc.Str('{' .. role .. '}'),
+      elem,
+    }
+  end
+end

doc/contributing/contributing-to-documentation.chapter.md

@@ -52,6 +52,13 @@ Additionally, the following syntax extensions are currently used:
 
   This syntax is taken from [MyST](https://myst-parser.readthedocs.io/en/latest/using/syntax.html#targets-and-cross-referencing).
 
+- []{#ssec-contributing-markup-inline-roles}
+  If you want to link to a man page, you can use `` {manpage}`nix.conf(5)` ``, which will turn into {manpage}`nix.conf(5)`.
+
+  The references will turn into links when a mapping exists in {file}`doc/build-aux/pandoc-filters/unix-man-urls.lua`.
+
+  This syntax is taken from [MyST](https://myst-parser.readthedocs.io/en/latest/syntax/syntax.html#roles-an-in-line-extension-point). Though, the feature originates from [reStructuredText](https://www.sphinx-doc.org/en/master/usage/restructuredtext/roles.html#role-manpage) with slightly different syntax.
+
 - []{#ssec-contributing-markup-admonitions}
   **Admonitions**, set off from the text to bring attention to something.
 

maintainers/scripts/db-to-md.sh

@@ -0,0 +1,88 @@
+#! /usr/bin/env nix-shell
+#! nix-shell -I nixpkgs=. -i bash -p pandoc
+
+# This script is temporarily needed while we transition the manual to
+# CommonMark. It converts DocBook files into our CommonMark flavour.
+
+debug=
+files=()
+
+while [ "$#" -gt 0 ]; do
+    i="$1"; shift 1
+    case "$i" in
+      --debug)
+        debug=1
+        ;;
+      *)
+        files+=("$i")
+        ;;
+    esac
+done
+
+echo "WARNING: This is an experimental script and might not preserve all formatting." > /dev/stderr
+echo "Please report any issues you discover." > /dev/stderr
+
+outExtension="md"
+if [[ $debug ]]; then
+    outExtension="json"
+fi
+
+DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" >/dev/null 2>&1 && pwd )"
+
+# NOTE: Keep in sync with Nixpkgs manual (/doc/Makefile).
+# TODO: Remove raw-attribute when we can get rid of DocBook altogether.
+pandoc_commonmark_enabled_extensions=+attributes+fenced_divs+footnotes+bracketed_spans+definition_lists+pipe_tables+raw_attribute
+targetLang="commonmark${pandoc_commonmark_enabled_extensions}+smart"
+if [[ $debug ]]; then
+    targetLang=json
+fi
+pandoc_flags=(
+    # Not needed:
+    # - diagram-generator.lua (we do not support that in NixOS manual to limit dependencies)
+    # - media extraction (was only required for diagram generator)
+    # - myst-reader/roles.lua (only relevant for MyST → DocBook)
+    # - link-unix-man-references.lua (links should only be added to display output)
+    # - docbook-writer/rst-roles.lua (only relevant for → DocBook)
+    # - docbook-writer/labelless-link-is-xref.lua (only relevant for → DocBook)
+    "--lua-filter=$DIR/../../doc/build-aux/pandoc-filters/docbook-reader/citerefentry-to-rst-role.lua"
+    "--lua-filter=$DIR/../../doc/build-aux/pandoc-filters/myst-writer/roles.lua"
+    "--lua-filter=$DIR/doc/unknown-code-language.lua"
+    -f docbook
+    -t "$targetLang"
+    --tab-stop=2
+    --wrap=none
+)
+
+for file in "${files[@]}"; do
+    if [[ ! -f "$file" ]]; then
+        echo "db-to-md.sh: $file does not exist" > /dev/stderr
+        exit 1
+    else
+    rootElement=$(xmllint --xpath 'name(//*)' "$file")
+
+    if [[ $rootElement = chapter ]]; then
+        extension=".chapter.$outExtension"
+    elif [[ $rootElement = section ]]; then
+        extension=".section.$outExtension"
+    else
+        echo "db-to-md.sh: $file contains an unsupported root element $rootElement" > /dev/stderr
+        exit 1
+    fi
+
+    outFile="${file%".section.xml"}"
+    outFile="${outFile%".chapter.xml"}"
+    outFile="${outFile%".xml"}$extension"
+    temp1=$(mktemp)
+    $DIR/doc/escape-code-markup.py "$file" "$temp1"
+    if [[ $debug ]]; then
+        echo "Converted $file to $temp1" > /dev/stderr
+    fi
+    temp2=$(mktemp)
+    $DIR/doc/replace-xrefs-by-empty-links.py "$temp1" "$temp2"
+    if [[ $debug ]]; then
+        echo "Converted $temp1 to $temp2" > /dev/stderr
+    fi
+    pandoc "$temp2" -o "$outFile" "${pandoc_flags[@]}"
+    echo "Converted $file to $outFile" > /dev/stderr
+  fi
+done

maintainers/scripts/doc/escape-code-markup.py

@@ -0,0 +1,97 @@
+#! /usr/bin/env nix-shell
+#! nix-shell -I nixpkgs=channel:nixos-unstable -i python3 -p python3 -p python3.pkgs.lxml
+
+"""
+Pandoc will strip any markup within code elements so
+let’s escape them so that they can be handled manually.
+"""
+
+import lxml.etree as ET
+import re
+import sys
+
+def replace_element_by_text(el: ET.Element, text: str) -> None:
+    """
+    Author: bernulf
+    Source: https://stackoverflow.com/a/10520552/160386
+    SPDX-License-Identifier: CC-BY-SA-3.0
+    """
+    text = text + (el.tail or "")
+    parent = el.getparent()
+    if parent is not None:
+        previous = el.getprevious()
+        if previous is not None:
+            previous.tail = (previous.tail or "") + text
+        else:
+            parent.text = (parent.text or "") + text
+        parent.remove(el)
+
+DOCBOOK_NS = "http://docbook.org/ns/docbook"
+
+# List of elements that pandoc’s DocBook reader strips markup from.
+# https://github.com/jgm/pandoc/blob/master/src/Text/Pandoc/Readers/DocBook.hs
+code_elements = [
+    # CodeBlock
+    "literallayout",
+    "screen",
+    "programlisting",
+    # Code (inline)
+    "classname",
+    "code",
+    "filename",
+    "envar",
+    "literal",
+    "computeroutput",
+    "prompt",
+    "parameter",
+    "option",
+    "markup",
+    "wordasword",
+    "command",
+    "varname",
+    "function",
+    "type",
+    "symbol",
+    "constant",
+    "userinput",
+    "systemitem",
+]
+
+XMLNS_REGEX = re.compile(r'\s+xmlns(?::[^=]+)?="[^"]*"')
+ROOT_ELEMENT_REGEX = re.compile(r'^\s*<[^>]+>')
+
+def remove_xmlns(match: re.Match) -> str:
+    """
+    Removes xmlns attributes.
+
+    Expects a match containing an opening tag.
+    """
+    return XMLNS_REGEX.sub('', match.group(0))
+
+if __name__ == '__main__':
+    assert len(sys.argv) >= 3, "usage: escape-code-markup.py <input> <output>"
+
+    tree = ET.parse(sys.argv[1])
+    name_predicate = " or ".join([f"local-name()='{el}'" for el in code_elements])
+
+    for markup in tree.xpath(f"//*[({name_predicate}) and namespace-uri()='{DOCBOOK_NS}']/*"):
+        text = ET.tostring(markup, encoding=str)
+
+        # tostring adds xmlns attributes to the element we want to stringify
+        # as if it was supposed to be usable standalone.
+        # We are just converting it to CDATA so we do not care.
+        # Let’s strip the namespace declarations to keep the code clean.
+        #
+        # Note that this removes even namespaces that were potentially
+        # in the original file. Though, that should be very rare –
+        # most of the time, we will stringify empty DocBook elements
+        # like <xref> or <co> or, at worst, <link> with xlink:href attribute.
+        #
+        # Also note that the regex expects the root element to be first
+        # thing in the string. But that should be fine, the tostring method
+        # does not produce XML declaration or doctype by default.
+        text = ROOT_ELEMENT_REGEX.sub(remove_xmlns, text)
+
+        replace_element_by_text(markup, text)
+
+    tree.write(sys.argv[2])