LDoc2tl/docs/doc.md

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 module is 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 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 naturally less can be inferred from the code itself.

LDoc can provide integrated documentation, with traditional function comments, any readme in Markdown format, and specified source examples. Lua source in examples and the readme 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 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.) By default it will process any file ending in .lua or .luadoc.

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.

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') you can define a new section ('Macros') 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()
...

(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, 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:

@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) {
....

LDoc does not pretend to understand C/C++, so in this case it is necessary to specify the name and type explicitly.

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 tag:

-----
-- A useful macro. This is an example of a custom 'kind'.
-- @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 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:

/***
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.

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

The command-line options are:

@plain
ldoc, a documentation generator for Lua, vs 0.5
  -d,--dir (default docs) output directory
  -o,--output  (default 'index') output name
  -v,--verbose          verbose
  -a,--all              show local functions, etc, in docs
  -q,--quiet            suppress output
  -m,--module           module docs as text
  -s,--style (default !) directory for style sheet (ldoc.css)
  -l,--template (default !) directory for template (ldoc.ltp)
  -1,--one              use one-column output layout
  -p,--project (default ldoc) project name
  -t,--title (default Reference) page title
  -f,--format (default plain) formatting - can be markdown, discount or plain
  -b,--package  (default .) top-level package basename (needed for module(...))
  -x,--ext (default html) output file extension
  -c,--config (default config.ld) configuration name
  --dump                debug output dump
  --filter (default none) filter output as Lua data (e.g pl.pretty.dump)
  <file> (string) source file or directory containing source

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.

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. Another 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, then 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)

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.

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.

There is a more customizable way to process the data, 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.

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. ?ref to config.ld for LM

Including examples and a readme file

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.

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. 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 well, unless the first indented line is '@plain'.

Fields allowed in config.ld

These mostly have the same meaning as the corresponding parameters:

• file a file or directory containing sources. In config.ld this can also be a table of files and directories.
• project name of project, used as title in top left
• title page title, default 'Reference'
• package
• all show local functions, etc as well in the docs
• format 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 layout
• style, template together these specify the directories for the style and and the template. In config.ld they may also be true, meaning use the same directory as the configuration file.

These only appear in config.ld:

• description a project description used under the project title
• examples a directory or file: can be a table
• readme name of readme file (to be processed with Markdown)

Available functions are:

• alias(a,tag) provide an alias a for the tag tag, for instance p as short for param
• add_language_extension(ext,lang) here lang may be either 'c' or 'lua', and ext is an extension to be recognized as this language
• add_section
• new_type(tag,header,project_level) used to add new tags, which are put in their own section header. They may be 'project level'.

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 ought 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.