1

I'm trying to break open a Packages.gz using Ruby's treetop and I've having trouble making keywords and values unambiguous. Here's my treetop grammar:

grammar Debian
  rule collection
    entry+
  end
  rule entry
    (tag space value)
  end

  rule package_details
    tag value &[^$]
  end
  rule tag 
    [A-Za-z0-9\-]+ ":"
  end
  rule value
    (!tag value_line+ "\n")+
  end
  rule value_line
    ([A-Za-z0-9 <>@()=\.\-|/,_"':])+
  end
  rule space
    [ \t]+
  end
end

And here's my sample input:

Package: acct
Priority: optional
Section: admin
Installed-Size: 352
Maintainer: Ubuntu Developers <ubuntu-devel-discuss@lists.ubuntu.com>
Original-Maintainer: Mathieu Trudel <mathieu.tl@gmail.com>
Architecture: i386
Version: 6.5.4-2ubuntu1
Depends: dpkg (>= 1.15.4) | install-info, libc6 (>= 2.4)
Filename: pool/main/a/acct/acct_6.5.4-2ubuntu1_i386.deb
Size: 111226
MD5sum: 10cba1458ace8c31169c0e9e915c9a0f
SHA1: 6c2dcdc480144a9922329cd4fa22c7d1cb83fcbb
SHA256: bf8d8bb8eef3939786a1cefc39f94079f43464b71099f4a59b61b24cafdbc010
Description: The GNU Accounting utilities for process and login accounting
 GNU Accounting Utilities is a set of utilities which reports and summarizes
 data about user connect times and process execution statistics.
 .
 "Login accounting" provides summaries of system resource usage based on connect
 time, and "process accounting" provides summaries based on the commands
 executed on the system.
 .
 The 'last' command is provided by the sysvinit package and not included here.
Homepage: http://www.gnu.org/software/acct/
Bugs: https://bugs.launchpad.net/ubuntu/+filebug
Origin: Ubuntu
Supported: 18m

This works almost 100%, but then it fails when examining a url. The problem is that a URL contains a ":" which I don't seem to be able to snag. When I edit the sample's Homepage entry and substitue a "_" for the ":" it goes right through.

This is my first PEG grammar but I can tell that I need to make it less ambiguous/more succinct. Looking over the advanced documentation I would like to define a tag as

rule tag
  !(!'\n' .) [A-Za-z0-9\-]+ ":"
end

But I don't understand completely what it is doing. The tag must not(not have a new line OR anything) which I take to mean, (either a newline or nothing). The subtleties escape me...

Would switching to that format help me out? Does someone know why that doesn't match?

xrl
  • 2,155
  • 5
  • 26
  • 40

1 Answers1

0

I appear to have gotten one working grammar at this point:

grammar Debian
  # The file is too big for us to emit a package_list. Look at parser.rb to see how I just split the string.
  #rule package_list
  #  (package "\n"?)+ <DebianSyntaxNode::PackageList>
  #end
  rule package
    (tag / value)+ <DebianSyntaxNode::Package>
  end

  rule tag
    tag_value tag_stop <DebianSyntaxNode::Tag>
  end
  rule tag_value
    [\w\-]+ <DebianSyntaxNode::TagValue>
  end
  rule tag_stop
    ": " <DebianSyntaxNode::TagStop>
  end

  rule value
    value_line value_stop <DebianSyntaxNode::Value>
    # value_line value_stop <DebianSyntaxNode::Value>
  end
  rule value_line
    (!"\n" .)+ <DebianSyntaxNode::ValueLine>
    # ([\w \. " , \- ' : / < > @ ( ) = | \[ \] + ; ~ í á * % `])+ <DebianSyntaxNode::ValueLine>
  end
  rule value_stop
    "\n"? <DebianSyntaxNode::ValueStop>
  end
end

Problem is that now the value_line doesn't include the "\n" when it's a multiline entry. Plus I have to combine multi-line entries in the parser.

If you want to see where this code is going, check out the little github project I started: https://github.com/derdewey/Debian-Packages-Parser

xrl
  • 2,155
  • 5
  • 26
  • 40