Default map of file extensions to MIME types.

    ".txt" => MIME("text/x-intersphinx"),
    ".inv" => MIME("application/x-intersphinx"),
    ".toml" => MIME("application/toml"),
    ".txt.gz" => MIME("text/x-intersphinx+gzip"),
    ".toml.gz" => MIME("application/toml+gzip"),

See Inventory File Formats for details.


An inventory of link targets in a project documentation.

inventory = Inventory(

loads an inventory file from the given source, which can be a URL or the path to a local file. If it is a URL, the options timeout (seconds to wait for network connections), retries (number of times to retry) and wait_time (seconds longer to wait between each retry) may be given. The source must contain data in the given mime type. By default, the mime type is derived from the file extension, via auto_mime.

The Inventory acts as a collection of InventoryItems, representing all the objects, sections, or other linkable items in the online documentation of a project.


inventory = Inventory(; project, version="", root_url="", items=[])

with a mandatory project argument instantiates an inventory with the InventoryItems in items. If items is not given, the resulting empty inventory can have InventoryItems added afterwards via push!.


  • project: The name of the project
  • version: The version of the project (e.g., "1.0.0")
  • root_url: The root URL to which the item.uri of any InventoryItem is relative. If not empty, should start with "https://" and end with a slash.
  • source: The URL or filename from which the inventory was loaded.
  • sorted: A boolean to indicate whether the items are sorted by their name attribute, allowing for efficient lookup. This is true for all inventories loaded from a URL or file and false for manually instantiated inventories.

Note that source and sorted are informational attributes only and are ignored when comparing two Inventory objects.

Item access

Items can be accessed via iteration (for item in inventory), by numeric index (inventory[1], inventory[2], … inventory[end]), or by lookup: inventory[name] or inventory[spec], where spec is a string of the form ":[domain:]role:`name`", see the discussion of spec in InventoryItem. The lookup delegates to find_in_inventory with quiet=true and takes into account item.priority.


The inventory can be searched by calling inventory(search; include_hidden_priority=true). This returns a list of all items that contain search in spec(item) or repr(item; context=(:full => true)). Typically, search would be a string or a Regex. Some examples for common searches:

  • A spec of the form ":domain:role:`name`", in full, partially, or as a regex.
  • Part of a url of a page in the project's documentation, as a string
  • The title of a section as it appears somewhere in the project's documentation.

The search results are sorted by abs(item.priority). If include_hidden_priority=false, negative item.priority values are omitted.



An item inside an Inventory.

item = InventoryItem(; name, role, uri, priority=1, domain="jl", dispname="-")

represents a linkable item inside a project documentation, referenced by name. The domain and role take their semantics from the Sphinx project, see Attributes for details on these parameters, as well as priority and dispname. The uri is relative to a project root, which should be the Inventory.root_url of the inventory containing the InventoryItem.

For convenience, an InventoryItem can also be instantiated from a mapping spec => uri, where spec=":domain:role:`name`" borrows from Sphinx' cross-referencing syntax:

item = InventoryItem(
    ":domain:role:`name`" => uri;
    priority=(<domain == "std" ? -1 : 1>)

The domain is optional: if spec=":role:`name`", the domain is "std" for role="label" or role="doc", and "jl" otherwise. The role is mandatory for code objects. For non-code objects,

item = InventoryItem(
    "title" => uri;

indicates a link to a section header in the documentation of a project. The name will be a sluggified version of the title, making the item equivalent to item = InventoryItem(":std:label:`name`" => uri; dispname=title, priority=-1).


  • name: The object name for referencing. For code objects, this should be the fully qualified name. For section names, it may be a slugified version of the section title. It must have non-zero length.

  • domain: The name of a Sphinx domain. Should be "jl" for Julia code objects (default), "py" for Python code objects, and "std" for text objects such as section names. Must have non-zero length, and must not contain whitespace or a colon.

  • role: A domain-specific role (type). Must have nonzero length and not contain whitespace.

  • priority: An integer flag for placement in search results. Used when searching in an Inventory, for item access in an Inventory, and with find_in_inventory. The following flag values are supported:

    • 1: the "default" priority. Used by default for all objects not in the "std" domain (that is, all "code" objects such as those in the "jl" domain).
    • 0: object is important
    • 2 (or higher): object is unimportant
    • -1 (or lower): object is "hidden" (may be omitted from search). Used by default for all objects in the std domain (section titles)

    See find_in_inventory for details. The above semantics match those used by Sphinx.

  • uri: A URI for the location of the object's documentation, relative to the location of the inventory file containing the item. Must not contain whitespace. May end with "$" to indicate a placeholder for name (usually as "#$", for an HTML anchor matching name).

  • dispname: A full plain text representation of the object. May be "-" if the display name is identical to name (which it should be for code objects). For section titles, this should be the plain text of the title, without formatting, but not slugified.


  • uri – Extract the full URI, resolving the $ placeholder and prepending a root_url, if applicable.
  • dispname – Extract the dispname, resolving the "-" shorthand, if applicable.
  • spec – Return the specification string ":domain:role:`name`" associated with the item

Determine the MIME type of the given file path or URL from the file extension.

mime = auto_mime(source)

returns a MIME type from the extension of source. The default mapping is in MIME_TYPES.

Unknown or unsupported extensions throw an ArgumentError.


Convert an inventory file.

DocInventories.convert("objects.inv", "inventory.toml")

converts the input file "objects.inv" in the Sphinx Inventory Format to the TOML Format "inventory.toml".

This is a convenience function to simply load an Inventory from the input file and write it to the output file. Both the input and output file must have known file extensions. The project and version metadata may be given as additional keyword arguments to be written to the output file, see set_metadata.


Find an item in the inventory.

item = find_in_inventory(

returns the top priority InventoryItem matching the given name. If the inventory contains no matching item, returns nothing.


  • inventory: The Inventory to search.
  • name: The value of the name attribute of the InventoryItem to find. Must match exactly.
  • domain: If not empty, restrict search to items with a matching domain attribute.
  • role: If not empty, restrict search to items with a matching role attribute.
  • quiet: If false (default), log a warning if the item specification is ambiguous (the top priority item of multiple candidates is returned). If no matching item can be found, an error will be logged in addition to returning nothing.
  • include_hidden_priority: Whether or not to consider items with a negative priority attribute. If "hidden" items are included (default), they are sorted by the absolute value of the priority. That is, items with priority=-1 and priority=1 are considered to be equivalent.

Note that direct item lookup as inventory[spec] where spec is a string of the form "[:[domain:]role:]`name`" is available as a simplified way to call find_in_inventory with quiet=true.


Obtain the root url from an inventory source.

url = root_url(source; warn=true)

returns the root url as determined by split_url if source starts with "https://" or "http://", or an empty string otherwise (if source is a local file path). An empty root url will emit a warning unless warn=false.


Write the Inventory to file in the specified format., inventory; mime=auto_mime(filename))

writes inventory to filename in the specified MIME type. By default, the MIME type is derived from the file extension of filename via auto_mime.


Modify the project and version metadata of an inventory or inventory file.

new_inventory = set_metadata(

returns a new Inventory with a modified version and/or project attribute.


modifies the project and/or version in the given inventory file (objects.inv, inventory.toml, etc.)

show_full(item)  # io=stdout
show_full(io, item)

is equivalent to

show(IOContext(io, :full => true), "text/plain", item)

and shows the InventoryItem with all attributes.

show_full(inventory)  # io=stdout
show_full(io, inventory)

is equivalent to

show(IOContext(io, :limit => false), "text/plain", inventory)

and shows the entire inventory without truncating the list of items. This may produce large output, so you may want to make use of the TerminalPager package.


Return the specification string of an InventoryItem.

item_spec = spec(item)

returns a string of the form ":domain:role:`name`" using the attributes of the given item.


Split a URL into a root URL and a filename.

root_url, filename = split_url(url)

splits url at the last slash. This behaves like splitdir, but operates on URLs instead of file paths. The URL must start with "https://" or "http://".

uri_str = uri(inventory, key)

is equivalent to uri(inventory[key]; root_url=inventory.root_url).

uri_str = uri(item; root_url="")

fully expands item.uri and prepends root_url.