<root_doc>.longmd.mdOne Markdown file containing the entire project in reading order.
<root_doc>.longmd.map.jsonA machine-readable sidecar with provenance and build metadata.
assets/Copied image assets with collision-safe output names.
The emitted Markdown currently uses these patterns:
ATX headings
HTML anchors such as <a id="..."></a> for stable targets
HTML comments to mark per-document boundaries
standard Markdown for prose, lists, code blocks, links, and images
MyST colon fences for semantics that do not map cleanly to plain Markdown
fallback sphinx-node blocks for unknown custom nodes in non-strict mode
Based on the current code and tests, the builder is designed to preserve or represent:
toctree reading order
section and target anchors
glossary terms
Python domain object descriptions
admonitions and common block-level Sphinx constructs
figures, tables, footnotes, citations, and field lists
copied image assets
warnings, losses, timing, and span metadata in the sidecar
The current sidecar includes top-level keys such as:
version
profile
root_doc
output_file
document_order
anchors
aliases
objects
assets
spans
warnings
losses
stats
The builder pre-registers anchors before emission and resolves cross-document heading collisions with docname-prefixed IDs instead of numeric suffixes. That makes emitted IDs more stable when the toctree order changes.
Raw HTML can pass through to the Markdown body when longmd_raw_html = True. Non-HTML raw blocks are omitted from the Markdown body and recorded in the sidecar instead.