|
| 1 | +""" |
| 2 | +Sphinx extension to enhance LLM discoverability. |
| 3 | +
|
| 4 | +This extension adds: |
| 5 | +1. <script type="text/llms.txt"> tags to HTML pages for inline LLM discovery |
| 6 | + (following Vercel's proposal) |
| 7 | +2. Post-processing of llms.txt to organize pages into logical sections |
| 8 | +""" |
| 9 | +import re |
| 10 | +from pathlib import Path |
| 11 | +from sphinx.application import Sphinx |
| 12 | +from sphinx.util import logging |
| 13 | + |
| 14 | +logger = logging.getLogger(__name__) |
| 15 | + |
| 16 | +# Default section mappings - can be overridden in conf.py |
| 17 | +DEFAULT_SECTION_MAPPINGS = { |
| 18 | + "Getting Started": [ |
| 19 | + r"^index\.html\.md$", |
| 20 | + r"^chipflow-lib/getting-started", |
| 21 | + r"^chipflow-lib/index", |
| 22 | + r"^tutorial", |
| 23 | + r"^examples/getting-started", |
| 24 | + ], |
| 25 | + "User Guide": [ |
| 26 | + r"^chipflow-lib/(?!autoapi)", |
| 27 | + r"^examples/", |
| 28 | + r"^configurator/", |
| 29 | + ], |
| 30 | + "API Reference": [ |
| 31 | + r"^chipflow-lib/autoapi/", |
| 32 | + r"^platform-api", |
| 33 | + ], |
| 34 | + "Digital IP Library": [ |
| 35 | + r"^chipflow-digital-ip/", |
| 36 | + ], |
| 37 | + "Amaranth Language": [ |
| 38 | + r"^amaranth/(?!.*soc)", |
| 39 | + ], |
| 40 | + "Amaranth SoC": [ |
| 41 | + r"^amaranth-soc/", |
| 42 | + ], |
| 43 | + "Optional": [ |
| 44 | + r"^amaranth/changes", |
| 45 | + r"^support", |
| 46 | + ], |
| 47 | +} |
| 48 | + |
| 49 | + |
| 50 | +def add_llms_script_tag(app: Sphinx, pagename: str, templatename: str, |
| 51 | + context: dict, doctree) -> None: |
| 52 | + """Add <script type="text/llms.txt"> to HTML pages.""" |
| 53 | + |
| 54 | + description = getattr(app.config, 'llms_txt_description', '') |
| 55 | + if not description: |
| 56 | + description = "Documentation for this project." |
| 57 | + description = description.strip() |
| 58 | + |
| 59 | + project = app.config.project or "Documentation" |
| 60 | + |
| 61 | + # Create the inline llms.txt content |
| 62 | + llms_script = f'''<script type="text/llms.txt"> |
| 63 | +# {project} |
| 64 | +
|
| 65 | +> {description} |
| 66 | +
|
| 67 | +For complete documentation in LLM-friendly format: |
| 68 | +- [Documentation Index](/llms.txt) - Sitemap of all pages |
| 69 | +- [Full Documentation](/llms-full.txt) - Complete docs in one file |
| 70 | +- [JSON Index](/docs-index.json) - Structured metadata |
| 71 | +</script>''' |
| 72 | + |
| 73 | + # Add to metatags which Sphinx/Furo will include in <head> |
| 74 | + if 'metatags' not in context: |
| 75 | + context['metatags'] = '' |
| 76 | + context['metatags'] = context.get('metatags', '') + llms_script |
| 77 | + |
| 78 | + |
| 79 | +def reorganize_llms_txt(app: Sphinx, exception) -> None: |
| 80 | + """Post-process llms.txt to organize pages into logical sections.""" |
| 81 | + if exception: |
| 82 | + return |
| 83 | + |
| 84 | + outdir = Path(app.outdir) |
| 85 | + llms_txt_path = outdir / "llms.txt" |
| 86 | + |
| 87 | + if not llms_txt_path.exists(): |
| 88 | + logger.warning("llms.txt not found, skipping reorganization") |
| 89 | + return |
| 90 | + |
| 91 | + # Read current llms.txt |
| 92 | + content = llms_txt_path.read_text(encoding='utf-8') |
| 93 | + |
| 94 | + # Parse the header (everything before ## Pages) |
| 95 | + header_match = re.match(r'^(.*?)(?=^## Pages|\Z)', content, re.MULTILINE | re.DOTALL) |
| 96 | + if not header_match: |
| 97 | + logger.warning("Could not parse llms.txt header") |
| 98 | + return |
| 99 | + |
| 100 | + header = header_match.group(1).strip() |
| 101 | + |
| 102 | + # Extract all page links |
| 103 | + page_pattern = re.compile(r'^- \[([^\]]+)\]\(([^)]+)\)(?::\s*(.*))?$', re.MULTILINE) |
| 104 | + pages = [(m.group(1), m.group(2), m.group(3) or '') for m in page_pattern.finditer(content)] |
| 105 | + |
| 106 | + if not pages: |
| 107 | + logger.warning("No pages found in llms.txt") |
| 108 | + return |
| 109 | + |
| 110 | + # Get section mappings from config or use defaults |
| 111 | + section_mappings = getattr(app.config, 'llms_sections', DEFAULT_SECTION_MAPPINGS) |
| 112 | + |
| 113 | + # Categorize pages into sections |
| 114 | + sections = {name: [] for name in section_mappings.keys()} |
| 115 | + sections["Other"] = [] # Catch-all |
| 116 | + |
| 117 | + for title, path, description in pages: |
| 118 | + categorized = False |
| 119 | + for section_name, patterns in section_mappings.items(): |
| 120 | + for pattern in patterns: |
| 121 | + if re.search(pattern, path): |
| 122 | + sections[section_name].append((title, path, description)) |
| 123 | + categorized = True |
| 124 | + break |
| 125 | + if categorized: |
| 126 | + break |
| 127 | + if not categorized: |
| 128 | + sections["Other"].append((title, path, description)) |
| 129 | + |
| 130 | + # Build new llms.txt with sections |
| 131 | + new_content = header + "\n\n" |
| 132 | + |
| 133 | + for section_name, section_pages in sections.items(): |
| 134 | + if not section_pages: |
| 135 | + continue |
| 136 | + |
| 137 | + new_content += f"## {section_name}\n\n" |
| 138 | + for title, path, description in section_pages: |
| 139 | + if description: |
| 140 | + new_content += f"- [{title}]({path}): {description}\n" |
| 141 | + else: |
| 142 | + new_content += f"- [{title}]({path})\n" |
| 143 | + new_content += "\n" |
| 144 | + |
| 145 | + # Write reorganized llms.txt |
| 146 | + llms_txt_path.write_text(new_content.strip() + "\n", encoding='utf-8') |
| 147 | + logger.info(f"Reorganized llms.txt with {len(sections)} sections") |
| 148 | + |
| 149 | + |
| 150 | +def setup(app: Sphinx): |
| 151 | + # Config value for custom section mappings |
| 152 | + app.add_config_value('llms_sections', DEFAULT_SECTION_MAPPINGS, 'html') |
| 153 | + |
| 154 | + # Add script tag to each HTML page |
| 155 | + app.connect('html-page-context', add_llms_script_tag) |
| 156 | + |
| 157 | + # Reorganize llms.txt after build (run after sphinx-llm) |
| 158 | + app.connect('build-finished', reorganize_llms_txt, priority=900) |
| 159 | + |
| 160 | + return { |
| 161 | + 'version': '0.1', |
| 162 | + 'parallel_read_safe': True, |
| 163 | + 'parallel_write_safe': True, |
| 164 | + } |
0 commit comments