38 KiB
LDoc, a Lua Documentation Tool
Introduction
LDoc is a second-generation documentation tool that can be used as a replacement for LuaDoc. It arose out of my need to document my own projects and only depends on the Penlight libraries.
It is mostly compatible with LuaDoc, except that certain workarounds are no longer needed. For instance, it is not so married to the idea that Lua modules should be defined using the module
function; this is not only a matter of taste since this has been deprecated in Lua 5.2.
Otherwise, the output is very similar, which is no accident since the HTML templates are based directly on LuaDoc. You can ship your own customized templates and style sheets with your own project, however. You have an option to use Markdown to process the documentation, which means no ugly HTML is needed in doc comments. C/C++ extension modules may be documented in a similar way, although function names cannot be inferred from the code itself.
LDoc can provide integrated documentation, with traditional function comments, any documents in Markdown format, and specified source examples. Lua source in examples and the documents will be prettified.
Although there are a fair number of command-line options, the preferred route is to write a config.ld
configuration file in Lua format. By convention, if LDoc is simply invoked as ldoc .
it will read this file first. In this way, the aim is to make it very easy for end-users to build your documentation using this simple command.
Commenting Conventions
LDoc follows the conventions established by Javadoc and later by LuaDoc.
Only 'doc comments' are parsed; these can be started with at least 3 hyphens, or by a empty comment line with at least 3 hypens:
--- summary.
-- Description; this can extend over
-- several lines
-----------------
-- This will also do.
You can also use Lua block comments:
--[[--
Summary. A description
...;
]]
Any module or script must start with a doc comment; any other files are ignored and a warning issued. The only exception is if the module starts with an explicit module
statement.
All doc comments start with a summary sentence, that ends with a period or a question mark. An optional description may follow. Normally the summary sentence will appear in the module contents.
After this descriptive text, there will typically be tags. These follow the convention established by Javadoc and widely used in tools for other languages.
--- foo explodes text.
-- It is a specialized splitting operation on a string.
-- @param text the string
-- @return a table of substrings
function foo (text)
....
end
There are also 'tparam' and 'treturn' which let you specify a type:
-- @tparam string text the string
-- @treturn {string,...} a table of substrings
There may be multiple 'param' tags, which should document each formal parameter of the function. For Lua, there can also be multiple 'return' tags
--- solvers for common equations.
module("solvers", package.seeall)
--- solve a quadratic equation.
-- @param a first coeff
-- @param b second coeff
-- @param c third coeff
-- @return first root, or nil
-- @return second root, or imaginary root error
function solve (a,b,c)
local disc = b^2 - 4*a*c
if disc < 0 then
return nil,"imaginary roots"
else
disc = math.sqrt(disc)
return (-b + disc)/2*a,
(-b - disc)/2*a
end
end
...
This is the common module style used in Lua 5.1, but it's increasingly common to see less 'magic' ways of creating modules in Lua. Since module
is deprecated in Lua 5.2, any future-proof documentation tool needs to handle these styles gracefully:
--- a test module
-- @module test
local test = {}
--- first test.
function test.one()
...
end
...
return test
Here the name of the module is explicitly given using the 'module' tag. If you leave this out, then LDoc will infer the name of the module from the name of the file and its relative location in the filesystem; this logic is also used for the module(...)
idiom. (How this works and when you need to provide extra information is discussed later.)
It is common to use a local name for a module when declaring its contents. In this case the 'alias' tag can tell LDoc that these functions do belong to the module:
--- another test.
-- @module test2
-- @alias M
local M = {}
-- first test.
function M.one()
..
end
return M
M
and _M
are used commonly enough that LDoc will recognize them as aliases automatically, but 'alias' allows you to use any identifier.
LDoc tries to deduce the function name and the formal parameter names from examining the code after the doc comment. It also recognizes the 'unsugared' way of defining functions as explicit assignment to a variable:
--- second test.
M.two = function(...) ... end
Apart from exported functions, a module usually contains local functions. By default, LDoc does not include these in the documentation, but they can be enabled using the --all
flag. They can be documented just like 'public' functions:
--- it's clear that boo is local from context.
local function boo(...) .. end
local foo
--- we need to give a hint here for foo
-- @local here
function foo(...) .. end
Modules can of course export tables and other values. The classic way to document a table looks like this:
--- a useful table of constants
-- @field alpha first correction
-- @field beta second correction
-- @field gamma fudge factor
-- @table constants
Here the kind of item is made explicit by the 'table' tag; tables have 'fields' in the same way as functions have parameters.
This can get tedious, so LDoc will attempt to extract table documentation from code:
--- a useful table of constants
M.constants = {
alpha = 0.23, -- first correction
beta = 0.443, -- second correction
gamma = 0.01 -- fudge factor
}
The rule followed here is NAME = <table-constructor>
. If LDoc can't work out the name and type from the following code, then a warning will be issued, pointing to the file and location.
Another kind of module-level type is 'field', such as follows:
--- module version.
M._VERSION = '0.5'
That is, a module may contain exported functions, local functions, tables and fields.
When the code analysis would lead to the wrong type, you can always be explicit.
--- module contents.
-- @field _CONTENTS
M._CONTENTS = {constants=true,one=true,...}
The order of tags is not important, but as always, consistency is useful. Tags like 'param' and 'return' can be specified multiple times, whereas a type tag like 'function' can only occur once in a comment. The basic rule is that a single doc comment can only document one entity.
By default, LDoc will process any file ending in '.lua' or '.luadoc' in a specified directory; you may point it to a single file as well. A 'project' usually consists of many modules in one or more packages. The generated index.html
will point to the generated documentation for each of these modules.
If only one module or script is documented for a project, then the index.html
generated contains the documentation for that module, since an index pointing to one module would be redundant.
(If you want to document a script, there is a project-level type 'script' for that.)
See References
The tag 'see' is used to reference other parts of the documentation, and 'usage' can provide examples of use:
---------
-- split a string in two.
-- @param s the string
-- @param delim the delimiter (default space)
-- @return first part
-- @return second part
-- @usage local hello,world = split2("hello world")
-- @see split
funtion split2(s,delim) .. end
Here it's assumed that 'split' is a function defined in the same module. If you wish to link to a function in another module, then the reference has to be qualified.
References to methods use a colon: myclass:method
; this is for instance how you would refer to members of a @type
section.
The example at tests/complex
shows how @see references are interpreted:
complex.util.parse
complex.convert.basic
complex.util
complex.display
complex
You may of course use the full name of a module or function, but can omit the top-level namespace - e.g. can refer to the module util
and the function display.display_that
directly. Within a module, you can directly use a function name, e.g. in display
you can say display_this
.
What applies to functions also applies to any module-level item like tables. New module-level items can be defined and they will work according to these rules.
If a reference is not found within the project, LDoc checks to see if it is a reference to a Lua standard function or table, and links to the online Lua manual. So references like 'table.concat' are handled sensibly.
References may be made inline using the @{ref} syntax. This may appear anywhere in the text, and is more flexible than @see. In particular, it provides one way to document the type of a parameter or return value when that type has a particular structure:
------
-- extract standard variables.
-- @param s the string
-- @return @\{stdvars}
function extract_std(s) ... end
------
-- standard variables.
-- Use @\{extract_std} to parse a string containing variables,
-- and @\{pack_std} to make such a string.
-- @field length
-- @field duration
-- @field viscosity
-- @table stdvars
@{ref} is very useful for referencing your API from code samples and readme text. (I've had to throw in a spurious backspace to stop expansion in this example.)
The link text can be changed from the default by the extended syntax @{ref|text}.
You can also put references in backticks, like \
stdvars`. This is commonly used in Markdown to indicate code, so it comes naturally when writing documents. It is controlled by the configuration variable
backtick_references; the default is
trueif you use Markdown in your project, but can be specified explicitly in your
config.ld`.
Sections
LDoc supports explicit sections. By default, the sections correspond to the pre-existing types in a module: 'Functions', 'Tables' and 'Fields' (There is another default section 'Local Functions' which only appears if LDoc is invoked with the --all
flag.) But new sections can be added; the first mechanism is when you define a new type (say 'macro') a new section ('Macros') is created to contain these types. There is also a way to declare ad-hoc sections using the @section
tag.
The need occurs when a module has a lot of functions that need to be put into logical sections.
--- File functions.
-- Useful utilities for opening foobar format files.
-- @section file
--- open a file
...
--- read a file
...
--- Encoding operations.
-- Encoding foobar output in different ways.
-- @section encoding
...
A section doc-comment has the same structure as a normal doc-comment; the summary is used as the new section title, and the description will be output at the start of the function details for that section.
In any case, sections appear under 'Contents' on the left-hand side. See the winapi documentation for an example of how this looks.
Arguably a module writer should not write such very long modules, but it is not the job of the documentation tool to limit the programmer!
A specialized kind of section is type
: it is used for documenting classes. The functions (or fields) within a type section are considered to be the methods of that class.
--- A File class.
-- @type File
....
--- get the modification time.
-- @return standard time since epoch
function File:mtime()
...
end
(In an ideal world, we would use the word 'class' instead of 'type', but this would conflict with the LuaDoc usage.)
A section continues until the next section is found, @section end
, or end of file.
Differences from LuaDoc
LDoc only does 'module' documentation, so the idea of 'files' is redundant.
One added convenience is that it is easier to name entities:
------------
-- a simple module.
-- (LuaDoc)
-- @class module
-- @name simple
------------
-- a simple module.
-- (LDoc)
-- @module simple
This is because type names (like 'function', 'module', 'table', etc) can function as tags. LDoc also provides a means to add new types (e.g. 'macro') using a configuration file which can be shipped with the source. If you become bored with typing 'param' repeatedly then you can define an alias for it, such as 'p'. This can also be specified in the configuration file.
LDoc will also work with C/C++ files, since extension writers clearly have the same documentation needs as Lua module writers.
LDoc gives the documenter the option to use Markdown to parse the contents of comments.
Adding new Tags
LDoc tries to be faithful to LuaDoc, but provides some extensions. Aliases for tags can be defined, and new types declared.
--- zero function. Two new ldoc features here; item types
-- can be used directly as tags, and aliases for tags
-- can be defined in config.ld.
-- @function zero_fun
-- @p k1 first
-- @p k2 second
Here an alias for 'param' has been defined. If a file config.ld
is found in the source, then it will be loaded as Lua data. For example, the configuration for the above module provides a title and defines an alias for 'param':
title = "testmod docs"
project = "testmod"
alias("p","param")
Extra tag types can be defined:
new_type("macro","Macros")
And then used as any other type:
-----
-- A useful macro. This is an example of a custom type.
-- @macro first_macro
-- @see second_function
This will also create a new module section called 'Macros'.
Inferring more from Code
The qualified name of a function will be inferred from any function
keyword following the doc comment. LDoc goes further with this kind of code analysis, however.
Instead of:
--- first table.
-- @table one
-- @field A alpha
-- @field B beta
M.one = {
A = 1,
B = 2;
}
you can write:
--- first table
-- @table one
M.one = {
A = 1, -- alpha
B = 2; -- beta
}
Simularly, function parameter comments can be directly used:
------------
-- third function. Can also provide parameter comments inline,
-- provided they follow this pattern.
function mod1.third_function(
alpha, -- correction A
beta, -- correction B
gamma -- factor C
)
...
end
As always, explicit tags can override this behaviour if it is inappropriate.
Extension modules written in C
LDoc can process C/C++ files:
@plain
/***
Create a table with given array and hash slots.
@function createtable
@param narr initial array slots, default 0
@param nrec initial hash slots, default 0
@return the new table
*/
static int l_createtable (lua_State *L) {
....
Both /**
and ///
are recognized as starting a comment block. Otherwise, the tags are processed in exactly the same way. It is necessary to specify that this is a function with a given name, since this cannot be reliably be inferred from code. Such a file will need a module comment, which is treated exactly as in Lua.
An unknown extension can be associated with a language using a call like add_language_extension('lc','c')
in config.ld
. (Currently the language can only be 'c' or 'lua'.)
See 'tests/examples/mylib.c' for the full example.
Basic Usage
For example, to process all files in the 'lua' directory:
$ ldoc lua
output written to docs/
Thereafter the docs
directory will contain index.html
which points to individual modules in the modules
subdirectory. The --dir
flag can specify where the output is generated, and will ensure that the directory exists. The output structure is like LuaDoc: there is an index.html
and the individual modules are in the modules
subdirectory. This applies to all project-level types, so that you can also get scripts
, examples
and topics
directories.
If your modules use module(...)
then the module name has to be deduced. If ldoc
is run from the root of the package, then this deduction does not need any help - e.g. if your package was foo
then ldoc foo
will work as expected. If we were actually in the foo
directory then ldoc -b .. .
will correctly deduce the module names. An example would be generating documentation for LuaDoc itself:
$ ldoc -b .. /path/to/luadoc
Without the -b
setting the base of the package to the parent of the directory, implicit modules like luadoc.config
will be incorrectly placed in the global namespace.
For new-style modules, that don't use module()
, it is recommended that the module comment has an explicit @module PACKAGE.NAME
. If it does not, then ldoc
will still attempt to deduce the module name, but may need help with --package/-b
as above.
format = 'markdown'
can be used in your config.ld
and will be used to process summaries and descriptions. This requires markdown.lua by Niklas Frykholm to be installed (this can be most easily done with luarocks install markdown
.) A much faster alternative is lua-discount which you can use by setting format
to 'discount' after installing using luarocks install lua-discount
) The discount Markdown processor additionally has more features than the pure Lua version, such as PHP-Extra style tables. As a special case, LDoc will fall back to using markdown.lua
if it cannot find discount
.
A special case is if you simply say 'ldoc .'. Then there must be a config.ld
file available in the directory, and it can specify the file:
file = "mymod.lua"
title = "mymod documentation"
description = "mymod does some simple but useful things"
file
can of course point to a directory, just as with the --file
option. This mode makes it particularly easy for the user to build the documentation, by allowing you to specify everything explicitly in the configuration.
In config.ld
, file
may be a Lua table, containing file names or directories; if it has an exclude
field then that will be used to exclude files from the list, for example {'examples', exclude = {'examples/slow.lua'}}
.
Processing Single Modules
--output
can be used to give the output file a different name. This is useful for the special case when a single module file is specified. Here an index would be redundant, so the single HTML file generated contains the module documentation.
$ ldoc mylib.lua --> results in docs/index.html
$ ldoc --output mylib mylib.lua --> results in docs/mylib.html
$ ldoc --output mylib --dir html mylib.lua --> results in html/mylib.html
The default sections used by LDoc are 'Functions', 'Tables' and 'Fields', corresponding to the built-in types 'function', 'table' and 'field'. If config.ld
contains something like new_type("macro","Macros")
then this adds a new section 'Macros' which contains items of 'macro' type - 'macro' is registered as a new valid tag name. The default template then presents items under their corresponding section titles, in order of definition.
Getting Help about a Module
There is an option to simply dump the results of parsing modules. Consider the C example `tests/example/mylib.c':
@plain
$ ldoc --dump mylib.c
----
module: mylib A sample C extension.
Demonstrates using ldoc's C/C++ support. Can either use /// or /*** */ etc.
function createtable(narr, nrec)
Create a table with given array and hash slots.
narr initial array slots, default 0
nrec initial hash slots, default 0
function solve(a, b, c)
Solve a quadratic equation.
a coefficient of x^2
b coefficient of x
c constant
return {"first root","second root"}
This is useful to quickly check for problems; here we see that createable
did not have a return tag.
LDoc takes this idea of data dumping one step further. If used with the -m
flag it will look up an installed Lua module and parse it. If it has been marked up in LuaDoc-style then you will get a handy summary of the contents:
@plain
$ ldoc -m pl.pretty
----
module: pl.pretty Pretty-printing Lua tables.
* read(s) - read a string representation of a Lua table.
* write(tbl, space, not_clever) - Create a string representation of a Lua table.
* dump(t, ...) - Dump a Lua table out to a file or stdout.
You can specify a fully qualified function to get more information:
@plain
$ ldoc -m pl.pretty.write
function write(tbl, space, not_clever)
create a string representation of a Lua table.
tbl {table} Table to serialize to a string.
space {string} (optional) The indent to use.
Defaults to two spaces.
not_clever {bool} (optional) Use for plain output, e.g {['key']=1}.
Defaults to false.
LDoc knows about the basic Lua libraries, so that it can be used as a handy console reference:
@plain
$> ldoc -m assert
function assert(v, message)
Issues an error when the value of its argument `v` is false (i.e.,
nil or false); otherwise, returns all its arguments.
`message` is an error
message; when absent, it defaults to "assertion failed!"
v
message
Thanks to mitchell's TextAdept project, LDoc has a set of .luadoc
files for all the standard tables, plus LuaFileSystem and LPeg.
@plain
$> ldoc -m lfs.lock
function lock(filehandle, mode, start, length)
Locks a file or a part of it.
This function works on open files; the file
handle should be specified as the first argument. The string mode could be
either r (for a read/shared lock) or w (for a write/exclusive lock). The
optional arguments start and length can be used to specify a starting point
and its length; both should be numbers.
Returns true if the operation was successful; in case of error, it returns
nil plus an error string.
filehandle
mode
start
length
Anatomy of a LDoc-generated Page
winapi can be used as a good example of a module that uses extended LDoc features.
The navigation section down the left has several parts:
- The project name ('project' in the config)
- A project description ('description')
- ''Contents'' of the current page
- ''Modules'' listing all the modules in this project
Note that description
will be passed through Markdown, if it has been specified for the project. This gives you an opportunity to make lists of links, etc; any '##' headers will be formatted like the other top-level items on the navigation bar.
'Contents' is automatically generated. It will contain any explicit sections, if they have been used. Otherwise you will get the usual categories: 'Functions', 'Tables' and 'Fields'.
'Modules' will appear for any project providing Lua libraries; there may also be a 'Scripts' section if the project contains Lua scripts. For example, LuaMacro has a driver script luam
in this section. The builtin module only defines macros, which are defined as a custom tag type.
The content section on the right shows:
- The module summary and description
- The contents summary, per section as above
- The detailed documentation for each item
As before, the description can use Markdown. The summary contains the contents of each section as a table, with links to the details. This is where the difference between an item's summary and an item's description is important; the first will appear in the contents summary. The item details show the item name and its summary again, followed by the description. There are then sections for the following tags: 'param', 'usage', 'return' and 'see' in that order. (For tables, 'Fields' is used instead of 'Parameters' but internally fields of a table are stored as the 'param' tag.)
You can of course customize the default template, but there are some parameters that can control what the template will generate. Setting one
to true
in your configuration file will give a one-column layout, which can be easier to use as a programming reference. You can suppress the contents summary with no_summary
.
Customizing the Page
Setting no_return_or_parms
to true
will suppress the display of 'param' and 'return' tags. This may appeal to programmers who dislike the traditional @tag soup xDoc style and prefer to comment functions just with a description. This is particularly useful when using Markdown in a stylized way to specify arguments:
---------
-- This extracts the shortest common substring from the strings _s1_ and _s2_
function M.common_substring(s1,s2)
Here I've chosen to italicise parameter names; the main thing is to be consistent.
This style is close to the Python documentation standard, especially when used with no_summary
.
It is also very much how the Lua documentation is ordered. For instance, this configuration file formats the built-in documentation for the Lua global functions in a way which is close to the original:
project = 'Lua'
description = 'Lua Standard Libraries'
file = {'ldoc/builtin',exclude = {'ldoc/builtin/globals.lua'}}
no_summary = true
no_return_or_parms = true
format = 'discount'
Generally, using Markdown gives you the opportunity to structure your documentation in any way you want; particularly if using lua-discount and its table syntax; the desired result can often be achieved then by using a custom style sheet.
Examples
It has been long known that documentation generated just from the source is not really adequate to explain how to use a library. People like reading narrative documentation, and they like looking at examples. Previously I found myself dealing with source-generated and writer-generated documentation using different tools, and having to match these up.
LDoc allows for source examples to be included in the documentation. For example, see the online documentation for winapi. The function utf8_expand
has a @see
reference to 'testu.lua' and following that link gives you a pretty-printed version of the code.
The line in the config.ld
that enables this is:
examples = {'examples', exclude = {'examples/slow.lua'}}
That is, all files in the examples
folder are to be pretty-printed, except for slow.lua
which is meant to be called from one of the examples. The see-reference to testu.lua
resolves to 'examples/testu.lua.html'.
Examples may link back to the API documentation, for instance the example input.lua
has a @{spawn_process} inline reference.
Readme files
Like all good Github projects, Winapi has a readme.md
:
readme = "readme.md"
This goes under the 'Topics' global section; the 'Contents' of this document is generated from the second-level (##) headings of the readme.
Readme files are always processed with Markdown, but may also contain @{} references back to the documentation and to example files. Any symbols within backticks will be expanded as references, if possible. As with doc comments, a link to a standard Lua function like @{os.execute} will work as well. Any code sections will be pretty-printed as Lua, unless the first indented line is '@plain'. (See the source for this readme to see how it's used.)
Another name for readme
is topics
, which is more descriptive. From LDoc 1.2, readme/topics
can be a list of documents. These act as a top-level table-of-contents for your documentation. Currently, if you want them in a particular order, then use names like 01-introduction.md
etc which sort appropriately.
The first line of a document may be a Markdown #
title. If so, then LDoc will regard the next level as the subheadings, normally second-level ##
. But if the title is already second-level, then third-level headings will be used ###
, and so forth. The implication is that the first heading must be top-level relative to the headings that follow, and must start at the first line.
A reference like @{string.upper} is unambiguous, and will refer to the online Lua manual. In a project like Penlight, it can get tedious to have to write out fully qualified names like @{pl.utils.printf}. The first simplification is to use the package
field to resolve unknown references, which in this case is 'pl'. (Previously we discussed how package
is used to tell LDoc where the base package is in cases where the module author wishes to remain vague, but it does double-duty here.) A further level of simplification comes from the @lookup directive in documents, which must start at the first column on its own line. For instance, if I am talking about pl.utils
, then I can say "@lookup utils" and thereafter references like @{printf} will resolve correctly.
Remember that the default is for references in backticks to be resolved; unlike @ references, it is not an error if the reference cannot be found.
The sections of a document (the second-level headings) are also references. This particular section can be refered to as @{doc.md.Resolving_References_in_Documents} - the rule is that any non-alphabetic character is replaced by an underscore.
Tag Modifiers
New with this release is the idea of tag modifiers. For instance, you may say @\param[type=number] and this associates the modifier type
with value number
with this particular param tag. A shorthand can be introduced for this common case, which is "@tparam "; in the same way @\treturn is defined.
This is useful for larger projects where you want to provide the argument and return value types for your API, in a structured way that can be easily extracted later. There is a useful function for creating new tags that can be used in config.ld
:
tparam_alias('string','string')
That is, "@string" will now have the same meaning as "@tparam string".
The exact form of <type>
is not defined, but here is a suggested scheme:
number -- a plain type
Bonzo -- a known type; a reference link will be generated
{string,number} -- a 'list' tuple, built from type expressions
{A=string,N=number} -- a 'struct' tuple, ditto
{Bonzo,...} -- an array of Bonzo objects
{[string]=Bonzo,...} -- a map of Bonzo objects with string keys
Array(Bonzo) -- (assuming that Array is a container)
Currently the type
modifier is the only one known and used by LDoc when generating HTML output. However, any other modifiers are allowed and are available for use with your own templates or for extraction by your own tools.
The alias
function within configuration files has been extended so that alias tags can be defined as a tag plus a set of modifiers. So tparam
is defined as:
alias('tparam',{'param',modifiers={type="$1"}})
Fields allowed in config.ld
These mostly have the same meaning as the corresponding parameters:
file
a file or directory containing sources. Inconfig.ld
this can also be a table of files and directories.project
name of project, used as title in top lefttitle
page title, default 'Reference'package
explicit base package name; also used for resolving references in documentsall
show local functions, etc as well in the docsformat
markup processor, can be 'plain' (default), 'markdown' or 'discount'output
output name (default 'index')dir
directory for output files (default 'docs')ext
extension for output (default 'html')one
use a one-column layoutstyle
,template
together these specify the directories for the style and and the template. Inconfig.ld
they may also betrue
, meaning use the same directory as the configuration file.
These only appear in config.ld
:
description
a project description used under the project titleexamples
a directory or file: can be a tablereadme
name of readme file (to be processed with Markdown)no_return_or_parms
don't show parameters or return values in outputbacktick_references
whether references in backticks will be resolvedmanual_url
point to an alternative or local location for the Lua manual, e.g. 'file:///D:/dev/lua/projects/lua-5.1.4/doc/manual.html'
Available functions are:
alias(a,tag)
provide an aliasa
for the tagtag
, for instancep
as short forparam
add_language_extension(ext,lang)
herelang
may be either 'c' or 'lua', andext
is an extension to be recognized as this languageadd_section
new_type(tag,header,project_level)
used to add new tags, which are put in their own sectionheader
. They may be 'project level'.tparam_alias(name,type)
for instance, you may wish thatstring
means@\tparam string
.
Annotations and Searching for Tags
Annotations are special tags that can be used to keep track of internal development status. The known annotations are 'todo', 'fixme' and 'warning'. They may occur in regular function/table doc comments, or on their own anywhere in the code.
--- Testing annotations
-- @module annot1
...
--- first function.
-- @todo check if this works!
function annot1.first ()
if boo then
end
--- @fixme what about else?
end
Although not currently rendered by the template as HTML, they can be extracted by the --tags
command, which is given a comma-separated list of tags to list.
@plain
D:\dev\lua\LDoc\tests> ldoc --tags todo,fixme annot1.lua
d:\dev\lua\ldoc\tests\annot1.lua:14: first: todo check if this works!
d:\dev\lua\ldoc\tests\annot1.lua:19: first-fixme1: fixme what about else?
Generating HTML
LDoc, like LuaDoc, generates output HTML using a template, in this case ldoc_ltp.lua
. This is expanded by the powerful but simple preprocessor devised originally by Rici Lake which is now part of Penlight. There are two rules - any line starting with '#' is Lua code, which can also be embedded with '$(...)'.
<h2>Contents</h2>
<ul>
# for kind,items in module.kinds() do
<li><a href="#$(no_spaces(kind))">$(kind)</a></li>
# end
</ul>
This is then styled with ldoc.css
. Currently the template and stylesheet is very much based on LuaDoc, so the results are mostly equivalent; the main change that the template has been more generalized. The default location (indicated by '!') is the directory of ldoc.lua
.
You may customize how you generate your documentation by specifying an alternative style sheet and/or template, which can be deployed with your project. The parameters are --style
and --template
, which give the directories where ldoc.css
and ldoc.ltp
are to be found. If config.ld
contains these variables, they are interpreted slightly differently; if they are true, then it means 'use the same directory as config.ld'; otherwise they must be a valid directory relative to the ldoc invocation.
An example of fully customized documentation is tests/example/style
: this is what you could call 'minimal Markdown style' where there is no attempt to tag things (except emphasizing parameter names). The narrative alone can to be sufficient, if it is written appropriately.
Of course, there's no reason why LDoc must always generate HTML. --ext
defines what output extension to use; this can also be set in the configuration file. So it's possible to write a template that converts LDoc output to LaTex, for instance. The separation of processing and presentation makes this kind of new application possible with LDoc.
Internal Data Representation
The --dump
flag gives a rough text output on the console. But there is a more customizeable way to process the output data generated by LDoc, using the --filter
parameter. This is understood to be a fully qualified function (module + name). For example, try
$ ldoc --filter pl.pretty.dump mylib.c
to see a raw dump of the data. (Simply using dump
as the value here would be a shorthand for pl.pretty.dump
.) This is potentially very powerful, since you may write arbitrary Lua code to extract the information you need from your project.
For instance, a file custom.lua
like this:
return {
filter = function (t)
for _, mod in ipairs(t) do
print(mod.type,mod.name,mod.summary)
end
end
}
Can be used like so:
~/LDoc/tests/example$ ldoc --filter custom.filter mylib.c
module mylib A sample C extension.
The basic data structure is straightforward: it is an array of 'modules' (project-level entities, including scripts) which each contain an item
array (functions, tables and so forth).
For instance, to find all functions which don't have a @return tag:
return {
filter = function (t)
for _, mod in ipairs(t) do
for _, item in ipairs(mod.items) do
if item.type == 'function' and not item.ret then
print(mod.name,item.name,mod.file,item.lineno)
end
end
end
end
}
The internal naming is not always so consistent; ret
corresponds to @return, and params
corresponds to @param. item.params
is an array of the function parameters, in order; it is also a map from these names to the individual descriptions of the parameters.
item.modifiers
is a table where the keys are the tags and the values are arrays of modifier tables. The standard tag aliases tparam
and treturn
attach a type
modifier to their tags.