Yichun Zhang (章亦春)


Makefile::DOM - Simple DOM parser for Makefiles


This document describes Makefile::DOM 0.006 released on 28 August 2011.


This libary can serve as an advanced lexer for (GNU) makefiles. It parses makefiles as "documents" and the parsing is lossless. The results are data structures similar to DOM trees. The DOM trees hold every single bit of the information in the original input files, including white spaces, blank lines and makefile comments. That means it's possible to reproduce the original makefiles from the DOM trees. In addition, each node of the DOM trees is modifiable and so is the whole tree, just like the PPI module used for Perl source parsing and the HTML::TreeBuilder module used for parsing HTML source.

If you're looking for a true GNU make parser that generates an AST, please see Makefile::Parser::GmakeDB instead.

The interface of Makefile::DOM mimics the API design of PPI. In fact, I've directly stolen the source code and POD documentation of PPI::Node, PPI::Element, and PPI::Dumper, with the full permission from the author of PPI, Adam Kennedy.

Makefile::DOM tries to be independent of specific makefile's syntax. The same set of DOM node types is supposed to get shared by different makefile DOM generators. For example, MDOM::Document::Gmake parses GNU makefiles and returns an instance of MDOM::Document, i.e., the root of the DOM tree while the NMAKE makefile lexer in the future, MDOM::Document::Nmake, also returns instances of the MDOM::Document class. Later, I'll also consider adding support for dmake and bsdmake.

Structure of the DOM

Makefile DOM (MDOM) is a structured set of a series of data types. They provide a flexible document model conformed to the makefile syntax. Below is a complete list of the 19 MDOM classes in the current implementation where the indentation indicates the class inheritance relationships.


It's not hard to see that all of the MDOM classes inherit from the MDOM::Element class. MDOM::Token and MDOM::Node are its direct children. The former represents a string token which is atomic from the perspective of the lexer while the latter represents a structured node, which usually has one or more children, and serves as the container for other DOM::Element objects.

Next we'll show a few examples to demonstrate how to map DOM trees to particular makefiles.

Case 1

Consider the following simple "hello, world" makefile:

    all : ; echo "hello, world"

We can use the MDOM::Dumper class provided by Makefile::DOM to dump out the internal structure of its corresponding MDOM tree:

        MDOM::Token::Bare         'all'
        MDOM::Token::Whitespace   ' '
        MDOM::Token::Separator    ':'
        MDOM::Token::Whitespace   ' '
          MDOM::Token::Separator    ';'
          MDOM::Token::Whitespace   ' '
          MDOM::Token::Bare         'echo "hello, world"'
          MDOM::Token::Whitespace   '\n'

In this example, speparators : and ; are all instances of the MDOM::Token::Separator class while spaces and new line characters are all represented as MDOM::Token::Whitespace. The other two leaf nodes, all and echo "hello, world" both belong to MDOM::Token::Bare.

It's worth mentioning that, the space characters in the rule command echo "hello, world" were not represented as MDOM::Token::Whitespace. That's because in makefiles, the spaces in commands do not make any sense to make in syntax; those spaces are usually sent to shell programs verbatim. Therefore, the DOM parser does not try to recognize those spaces specifially so as to reduce memory use and the number of nodes. However, leading spaces and trailing new lines will still be recognized as MDOM::Token::Whitespace.

On a higher level, it's a MDOM::Rule::Simple instance holding several Token and one MDOM::Command. On the highest level, it's the root node of the whole DOM tree, i.e., an instance of MDOM::Document::Gmake.

Case 2

Below is a relatively complex example:

    a: foo.c  bar.h $(baz) # hello!
        @echo ...

It's corresponding DOM structure is

      MDOM::Token::Bare         'a'
      MDOM::Token::Separator    ':'
      MDOM::Token::Whitespace   ' '
      MDOM::Token::Bare         'foo.c'
      MDOM::Token::Whitespace   '  '
      MDOM::Token::Bare         'bar.h'
      MDOM::Token::Whitespace   '\t'
      MDOM::Token::Interpolation   '$(baz)'
      MDOM::Token::Whitespace      ' '
      MDOM::Token::Comment         '# hello!'
      MDOM::Token::Whitespace      '\n'
      MDOM::Token::Separator    '\t'
      MDOM::Token::Modifier     '@'
      MDOM::Token::Bare         'echo ...'
      MDOM::Token::Whitespace   '\n'

Compared to the previous example, here appears several new node types.

The variable interpolation $(baz) on the first line of the original makefile corresponds to a MDOM::Token::Interpolation node in its MDOM tree. Similarly, the comment # hello corresponds to a MDOM::Token::Comment node.

On the second line, the rule command indented by a tab character is still represented by a MDOM::Command object. Its first child node (or its first element) is also an MDOM::Token::Seperator instance corresponding to that tab. The command modifier @ follows the Separator immediately, which is of type MDOM::Token::Modifier.

Case 3

Now let's study a sample makefile with various global structures:

  a: b
  foo = bar
      # hello!

Here on the top level, there are three language structures: one rule "a: b", one assignment statement "foo = bar", and one comment # hello!.

Its MDOM tree is shown below:

      MDOM::Token::Bare                  'a'
      MDOM::Token::Separator            ':'
      MDOM::Token::Whitespace           ' '
      MDOM::Token::Bare                   'b'
      MDOM::Token::Whitespace           '\n'
      MDOM::Token::Bare                  'foo'
      MDOM::Token::Whitespace           ' '
      MDOM::Token::Separator            '='
      MDOM::Token::Whitespace           ' '
      MDOM::Token::Bare                  'bar'
      MDOM::Token::Whitespace           '\n'
    MDOM::Token::Whitespace            '\t'
    MDOM::Token::Comment               '# hello!'
    MDOM::Token::Whitespace            '\n'

We can see that below the root node MDOM::Document::Gmake, there are MDOM::Rule::Simple, MDOM::Assignment, and MDOM::Comment three elements, as well as two MDOM::Token::Whitespace objects.

It can be observed from the examples above that the MDOM representation for the makefile's lexical elements is rather loose. It only provides very limited structural representation instead of making a bad guess.


Generating an MDOM tree from a GNU makefile only requires two lines of Perl code:

    use MDOM::Document::Gmake;
    my $dom = MDOM::Document::Gmake->new('Makefile');

If the makefile source code being parsed is already stored in a Perl variable, say, $var, then we can construct an MDOM via the following code:

    my $dom = MDOM::Document::Gmake->new(\$var);

Now $dom becomes the reference to the root of the MDOM tree and its type is now MDOM::Document::Gmake, which is also an instance of the MDOM::Node class.

Just as mentioned above, MDOM::Node is the container for other MDOM::Element instances. So we can retrieve some element node's value via its child method:

    $node = $dom->child(3);
    # or $node = $dom->elements(0);

And we may also use the elements method to obtain the values of all the nodes:

    @elems = $dom->elements;

For every MDOM node, its corresponding makefile source can be generated by invoking its content method.


The current implemenation of the MDOM::Document::Gmake lexer is based on a hand-written state machie. Although the efficiency of the engine is not bad, the code is rather complicated and messy, which hurts both extensibility and maintanabilty. So it's expected to rewrite the parser using some grammatical tools like the Perl 6 regex engine Pugs::Compiler::Rule or a yacc-style one like Parse::Yapp.


You can always get the latest source code of this module from its GitHub repository:


If you want a commit bit, please let me know.


Zhang "agentzh" Yichun (章亦春) <agentzh@gmail.com>


Copyright 2006-2011 by Zhang "agentzh" Yichun (章亦春).

This library is free software; you can redistribute it and/or modify it under the same terms as Perl itself.


MDOM::Document, MDOM::Document::Gmake, PPI, Makefile::Parser::GmakeDB, makesimple.

Hosting generously
sponsored by Bytemark