DocShell extracts documentation into versioned JSON and stops there. It owns generation and the artifact contract; it does not own a renderer, routing, authorization policy, tenancy, or product taxonomy.
Requires Elixir 1.17 or later. Install ash_oaskit, open_api_spex, or plug
when the host uses the corresponding optional integration.
Build and configuration
- Configure only the
:doc_shellapplication. DocShell never reads another application's environment. - Precedence is per-call options to
DocShell.Build.run/1, thenconfig :doc_shell, then package defaults. Defaults live inDocShell.Config, not in a config file, so a host that configures nothing still gets a valid artifact tree. - Use
mix doc_shell.buildfor host builds. It documents every module in the current application. The default task starts the app; pass--no-startto compile and load the app spec without starting the supervision tree. CallDocShell.Build.run/1directly, with an explicit:moduleslist, when you need a different set. - Handle both
{:ok, result}and{:error, reason}. Extraction stops at the first error and names the module or file at fault; it does not skip bad sources. - Pass explicit
:modules,:guide_bases, and:livebook_basevalues when the host layout differs from the defaults ([],["guides"],"notebooks"). - Set
:collectionto a validatedDocShell.Generate.Collectiondescriptor when the output will be imported into a documentation site. It records the caller-supplied source revision and tree digest; DocShell never invokes Git or fetches the descriptor URLs. The public output then includes the envelopedcollection.jsonprovenance artifact. - Treat changelog/release notes as a source adapter. The default
DocShell.Generate.Changelog.Sources.MarkdownFilereadsCHANGELOG.md, but graph/database/CMS hosts should implementDocShell.Generate.Changelog.Sourceand pass:changelog_options. UseDocShell.Generate.Changelog.from_markdown/2when the dynamic source stores Markdown. - Leave
:open_api_adapterunset to emit a valid empty OpenAPI 3.1 document. This is a supported configuration, not a degraded one. - Use
DocShell.Build.run/1's return value to feed a database or knowledge graph, withwrite: falsewhen the files are not wanted. The return value is richer than what is written: entries keep their parsedastand nothing is filtered out. Projector backlinks remain in memory and have no disk artifact. - Set
:presentation_sourceto aDocShell.Presentation.GraphProjectorimplementation to have the build use a host projector.:path_builder,:skip_empty,:search_tokens, and:search_memberspass through to the producer. Member search defaults off. Enable it to append module member names/arities, signatures and parsed docs to page search text, without changing content or overridingskip_empty; malformed member Markdown fails with the module ID. - Set
:openapi_spec_pathwhen external tooling needs a bare OpenAPI file. Put it outside the artifact directories —DocShell.Web.Cacherejects a directory holding an unenveloped.json.
Default API identity
The default OpenAPI 3.1 document includes info.title and the configured
api_version as info.version (default "0.1.0").
Configuration errors
Build.run/1 rejects malformed options and unknown per-call keys before
extraction. Unknown application environment keys remain ignored. Invalid guide
identities, titles, audience, and locale return errors naming the field and file.
Guide IDs and titles accept nonempty strings or numeric/boolean scalars.
Output destinations
Public and private output directories must be disjoint. The optional raw OpenAPI destination must lie outside both. Conflicting paths fail before extraction or writes; use dedicated directories without symlink aliases.
Failed builds and recovery
Builds stage all JSON and back up existing files before publishing. Use the transaction's
delete: [path] option for removals so locks and rollback cover them; duplicate
targets, directories and symlinks are rejected before publication. Returned
publication failures restore earlier files, including deleted artifacts; rollback failures report retained
backup paths. Cooperating builds use .doc-shell-build.lock directories. After
a process or machine crash, recover retained backups and remove stale locks
before rebuilding. Files still publish individually, so cache reloads validate
generation IDs and keep the last complete snapshot. Use dedicated output
directories without external writers or symlink aliases.
Artifact contract
Collection implementation guidance
Follow the collection integrity requirements in
docs/specs/DSH.01-documentation-sites.md and the acceptance status in
docs/plans/documentation-sites.md. Source paths identify files; document IDs
identify records, so separate changelog releases may share one source file.
Include synthetic OpenAPI identity in duplicate checks. Require complete
provenance and a build/load round trip. Reject incompatible custom collection
projections before publication; never reintroduce filtered public bodies implicitly.
Unknown source indexes referenced by provenance use the generic v1 entry shape.
Loading rejects missing or repeated provenance, duplicate document IDs and
malformed source indexes. The ID openapi is reserved in collection mode.
Custom projectors must preserve extracted ASTs and exact JSON numeric types (omitted empty ASTs reconstruct
as []). Dynamic source locators belong in metadata such as source_ref;
an embedded index AST must agree with content, and OpenAPI owns no page AST.
collection source_path values must be relative filesystem paths.
Canonical digesting rejects duplicate encoded keys, and malformed boundary input
returns tagged errors. Keep deletion inside publication locks and rollback.
Artifact writers validate protocol-generated JSON and return host encoder errors
without replacing files. Changelog extension fields must remain native JSON in
both collection and ordinary builds, including write: false.
ExDoc member metadata is normalized with duplicate-key checking. Keep custom
@doc metadata keys distinct after conversion to strings.
Import assumes a stable caller-owned directory and uses finite resource limits;
hashes and path checks are not source authentication or an OS sandbox.
Use Collection.load/2 or load_many/2 with positive keyword overrides for
max_file_bytes (32 MiB), max_total_bytes (2 GiB including the manifest),
max_artifacts (256), max_sources (100,000), max_json_depth (64), and
max_collections (256). Except collection count, these budgets are per corpus.
No unlimited value is supported. Root symlinks and .. traversal fail, including
symlinks spelled with a trailing slash or /.. Ancestors of the existing working
and system temporary directories are trusted aliases; other symlink components
are rejected. Duplicate JSON object keys also fail before digest validation.
- Treat
DocShell.schema_version/0and thedoc-shell/v1shapes as public API. Never invent fields or change a field's type in place. - Treat the v1 source catalogue as additive. A manifest may list a new
per-source artifact and
kindis an open string; generic consumers ignore unknown files and kinds, while selective consumers may read only their allow-listed artifacts. Removing an existing file or changing an existing field's name or type requires a schema-version change. - A site-ready collection uses the additive
doc-shell-collection/v1payload incollection.json. Its portable descriptor excludes the local artifact directory and includes source identities, source provenance, artifact digests, and a canonical aggregate digest. Load it withDocShell.Generate.Collection.load/1; the loader rejects missing or unlisted files, symlinks, path escapes, mixed generations, descriptor differences, and digest mismatches before qualifying IDs ascollection_id:document_id. - Read the version from
DocShell.schema_version/0rather than writing the literal"doc-shell/v1". - Read and write artifacts through
DocShell.Artifact. Do not bypass the envelope or encode artifact JSON by hand. - Treat
generation_idas an opaque snapshot identity. Every artifact and the manifest from one build must carry the same value; never synthesize or reuse one across builds. - Keep generated content renderer-neutral: no host UI, routing, tenant, or authorization assumptions inside an artifact.
- Produce presentation data with
DocShell.Presentation.NavigationItem,SearchEntry, andBacklinkstructs, not bare maps. - Validate graph-backed output through
DocShell.Presentation.GraphProjector.project/2before exposing it. - Changing a
doc-shell/v1shape is a breaking change to every producer and renderer at once. Adding an optional field is usually safe; renaming, removing, or retyping one is not.
Rendering untrusted content
The AST preserves raw HTML and URL schemes. Renderers must allow-list tags and attributes, reject unsafe URL schemes, and escape text for their output context. Parsing Markdown is not sanitization.
Document identities
Document IDs must be nonempty and unique across all sources, including entries filtered from presentation. Duplicate IDs return an error naming both sources. Overlapping guide directories extract each normalized path once.
Recursive content validation
Host projectors and changelog sources must provide complete recursive AST nodes whose extension fields are native JSON. Presentation IDs must be nonempty and unique within navigation (including descendants) and search; shared IDs use the same path. Local leaves/search results require content. Groups with children and absolute HTTP(S) links may omit content; content may be hidden from both indexes. Backlink targets require content, while origins may belong to the wider host graph provided they do not contradict a known path. Default ordering is kind/title/ID.
Host projectors and changelog sources must provide complete recursive AST nodes
and JSON metadata with string keys. Invalid nested content fails validation
before output is written. DocShell.Ast.valid?/1 checks node lists.
Legacy envelope compatibility
Artifact.read/1 and read_envelope/1 accept legacy v1 envelopes without
generation_id. A present ID must be a nonempty string. Runtime caches require
an ID on every artifact and manifest to verify that they form one generation.
Concurrent artifact writers
Individual artifact writes use exclusively created random temporary files in the destination directory, so independent BEAM instances cannot share a temporary file. A rename publishes each complete file.
Search text
Search content preserves adjacent inline text, including words split by formatting. Block elements and line breaks add separators; image alt text is searchable. Token generation uses this same text.
Document paths
Each document path is calculated once and reused by navigation and search.
Default paths percent-encode kind and ID as individual URL segments. Use a
custom path_builder when IDs intentionally represent a path hierarchy.
Source integrations
- Implement
DocShell.Generate.OpenApi.Adapter.load/1for a new OpenAPI source. Return{:ok, map}with anopenapikey of"3.0.x","3.1.x", or"3.2.x", or{:error, reason}with a reason worth reading in a failed build. - Implement
DocShell.Generate.Changelog.Source.load/1for a new release-note source. Return validated DocShell changelog entries, or fetch Markdown dynamically and pass it throughDocShell.Generate.Changelog.from_markdown/2. - The bundled Markdown parser accepts git_ops headings and Keep a Changelog headings, including parenthesised or hyphenated dates, inline or reference-style links, unlinked releases, and SemVer prereleases.
- Resolve optional libraries at runtime with
Code.ensure_loaded?/1. A compile-time reference breaks every host that does not install the library. - Preserve Markdown as the renderer-neutral AST from
DocShell.Ast. Never emit HTML from an extractor. - Surface malformed configured sources as errors; do not silently discard them.
- Normalize input metadata with
DocShell.Json.normalize/1and handle converted-key collisions as errors. Structs become theirString.Charstext where they have one and theirinspect/1form otherwise. - Read
meta["moduledoc"]("present","hidden","none") rather than inferring documentation coverage from an emptyast.
Guide line endings
YAML frontmatter accepts LF, CRLF, and CR line endings, including a closing
--- delimiter at end of file.
Markdown titles
Guide and notebook titles come from the first top-level parsed H1, including Setext headings. Inline formatting is flattened, and headings inside code examples are ignored. Explicit guide frontmatter titles still take precedence. Explicit titles avoid a title-only parse. Unfenced bodies reuse their AST; fenced input retains defensive title parsing so malformed closings cannot expose code comments as titles.
Changelog source validation
Every changelog source entry must be a valid entry map; nil and other invalid
entries return {:error, {:invalid_changelog_entry, entry}}.
Changelog Markdown context
Changelogs are parsed as complete Markdown documents before splitting on top-level release headings. Code examples remain within their release, and reference links resolve across the whole document. Parse errors in any part of the source return a source-tagged error.
JSON metadata normalization
Use DocShell.Json.Canonical.encode/1 or digest/1 at serialization boundaries;
both return tagged errors and reject duplicate encoded keys. Collection.digest/1
requires encodable input and raises ArgumentError otherwise. Invalid UTF-8
source files fail with :invalid_utf8, tagged with the file by extraction/build.
Metadata preserves JSON scalars and uses UTF-8 string keys. Unsupported terms
become inspected text; improper list tails become a final array value.
DocShell.Json.normalize/1 rejects converted-key collisions. The legacy
stringify/1 keeps string keys when a collision occurs. Guides use normalize/1.
OpenAPI version support
Raw and custom adapters accept OpenAPI 3.0, 3.1, and 3.2 documents without rewriting their fields. Validation checks the version and unambiguous JSON encoding; source libraries own schema validation. The default document remains OpenAPI 3.1.
Optional integration dependencies
Plug is an optional package dependency because web modules compile against it. AshOaskit is a development/test fixture; hosts using its runtime adapter install AshOaskit themselves. Core consumers do not resolve its dependency tree.
Optional web serving
- Add
DocShell.Web.Cacheto a supervision tree before serving artifacts, and callDocShell.Web.Cache.reload/1after a rebuild. - Keep
manifest.jsonbeside the artifacts it lists. Cache startup and reload reject missing manifests, unlisted files, and mixed generation identifiers. - Use
DocShell.Web.Plugonly when Plug is installed. The Plug, controller, and response modules compile conditionally; the cache can be used without Plug. - Supply host authorization through the plug's
:gateoption — a unary function or an MFA tuple, returning:okortrueto allow. Omitting it serves everything to everyone. - Use
DocShell.Web.Controller.show/2instead when the host wants its own pipeline in front; put authorization in a plug there. - DocShell must not implement application-specific access policy.
Supervising named caches
A cache child specification uses its registered name as its child ID. Multiple named caches can be listed directly in one supervision tree.
Cache ownership
Cache ETS tables permit direct concurrent reads, but only the cache process
may write them. Individual fetches are consistent; separate fetches may cross a
reload. Use Cache.snapshot/2 for a caller-owned copy of all envelopes from one
generation, including the manifest. It returns {:ok, %{generation_id: id, artifacts: envelopes}} and remains valid after reloads, at the cost of retaining
that copy in caller memory. Snapshot calls queue behind reloads.
Cache.reload(server, timeout) and snapshot(server, timeout) default to 5,000
milliseconds. They follow GenServer.call/3: timeouts/unavailable processes exit
the caller, and timing out does not cancel a queued or running reload.
Reusing an active generation ID with changed envelopes returns
{:error, {:generation_id_reused, id}} and preserves the prior snapshot.
HTTP response caching
HTTP serving caches encoded JSON and an ETag per generation. GET and HEAD
share headers; matching If-None-Match requests return 304 after authorization.
Other methods return 405 with Allow: GET, HEAD. The host retains control of
Cache-Control and Vary. Encoding happens during cache publication, not requests.