mirror of
https://github.com/inspec/inspec
synced 2024-11-10 15:14:23 +00:00
02e435b6d0
Signed-off-by: Dominik Richter <dominik.richter@gmail.com>
192 lines
4.9 KiB
Ruby
192 lines
4.9 KiB
Ruby
# encoding: utf-8
|
|
# author: Dominik Richter
|
|
# author: Christoph Hartmann
|
|
|
|
require 'parslet'
|
|
|
|
class ErlangParser < Parslet::Parser
|
|
root :outermost
|
|
# only designed for rabbitmq config files for now:
|
|
rule(:outermost) { filler? >> array.maybe >> dot.maybe }
|
|
|
|
rule(:exp) {
|
|
(tuple | array | binary | string | bool | identifier | float | integer) >> filler?
|
|
}
|
|
|
|
rule(:array) {
|
|
str('[') >> filler? >> (
|
|
exp.repeat(1) >>
|
|
(comma >> exp).repeat
|
|
).maybe.as(:array) >> str(']') >> filler?
|
|
}
|
|
|
|
rule(:tuple) {
|
|
str('{') >> filler? >> (
|
|
exp.repeat(1) >> filler? >>
|
|
(comma >> exp).repeat
|
|
).maybe.as(:tuple) >> str('}') >> filler?
|
|
}
|
|
|
|
rule(:filler?) { space.repeat }
|
|
rule(:space) { match('\s+') | match["\n"] | comment }
|
|
|
|
rule(:comment) { str('%') >> (match["\n\r"].absent? >> any).repeat }
|
|
rule(:comma) { str(',') >> filler? }
|
|
rule(:dot) { str('.') >> filler? }
|
|
rule(:bool) { str('true').as(:bool) | str('false').as(:bool) }
|
|
|
|
rule(:identifier) {
|
|
(match('[a-zA-Z]') >> match('[a-zA-Z0-9_]').repeat).as(:identifier) >> filler?
|
|
}
|
|
|
|
rule(:float) {
|
|
(
|
|
integer >> (
|
|
str('.') >> match('[0-9]').repeat(1) |
|
|
str('e') >> match('[0-9]').repeat(1)
|
|
).as(:e)
|
|
).as(:float) >> filler?
|
|
}
|
|
|
|
rule(:integer) {
|
|
((str('+') | str('-')).maybe >> match('[0-9]').repeat(1)).as(:integer) >> filler?
|
|
}
|
|
|
|
rule(:string) { stringS | stringD }
|
|
|
|
rule(:stringS) {
|
|
str("'") >> (
|
|
str('\\') >> any | str("'").absent? >> any
|
|
).repeat.as(:string) >> str("'") >> filler?
|
|
}
|
|
|
|
rule(:stringD) {
|
|
str('"') >> (
|
|
str('\\') >> any | str('"').absent? >> any
|
|
).repeat.as(:string) >> str('"') >> filler?
|
|
}
|
|
|
|
rule(:binary_item) {
|
|
(string | integer) >>
|
|
(str(':') >> integer).maybe.as(:size) >>
|
|
(str('/') >> identifier).maybe.as(:type) >>
|
|
filler?
|
|
}
|
|
|
|
rule(:binary) {
|
|
str('<<') >> filler? >> (
|
|
binary_item.repeat(1) >>
|
|
(comma >> binary_item).repeat
|
|
).maybe.as(:binary) >> str('>>') >> filler?
|
|
}
|
|
end
|
|
|
|
class ErlangBitstream
|
|
def initialize
|
|
@data = [] # a stream of 8-bit numbers
|
|
@cur_bits = '' # a string of binary bits 10010010...
|
|
end
|
|
|
|
TYPES = {
|
|
'integer' => 8,
|
|
'float' => 8*8,
|
|
'utf8' => 8,
|
|
'utf16' => 8*2,
|
|
'utf32' => 8*4,
|
|
}.freeze
|
|
|
|
def bit_size(size, type)
|
|
raise 'Cannot specify size and type at the same time.' if !type.nil? && !size.nil?
|
|
return (size || 8).to_i if type.nil?
|
|
TYPES[type] || raise("Cannot handle binary-stream type #{type}")
|
|
end
|
|
|
|
def add(i)
|
|
if i[:integer].nil? && i[:string].nil?
|
|
raise 'No data provided, internal error for binary-stream processing!'
|
|
end
|
|
s = bit_size(i[:size], i[:type])
|
|
unless i[:string].nil?
|
|
str2int(i[:string].to_s, i[:type]).map { |e| add_bits(int2bits(e, 8)) }
|
|
else
|
|
add_int(i[:integer], s)
|
|
end
|
|
rescue RuntimeError => e
|
|
raise 'Error processing Erlang bit string '\
|
|
"'#{i[:string] || i[:integer]}:#{i[:size]}/#{i[:type]}'. #{e.message}"
|
|
end
|
|
|
|
def str2int(s, type)
|
|
case type
|
|
when 'utf8' then s.encode('utf-8').unpack('C*')
|
|
when 'utf16' then s.encode('utf-16').unpack('C*').drop(2)
|
|
when 'utf32' then s.encode('utf-32').unpack('C*').drop(4)
|
|
when 'integer', 'float' then raise "Cannot handle bit string as type #{type}"
|
|
else s.split('').map { |x| x.ord & 0xff }
|
|
end
|
|
end
|
|
|
|
def int2bits(i, len)
|
|
format("%0#{len}b", i)
|
|
end
|
|
|
|
def add_int(v, size)
|
|
x = v.to_i & (2**size - 1) # only get the bits specified in size
|
|
add_bits(int2bits(x, size))
|
|
end
|
|
|
|
def add_bits(s)
|
|
b = (@cur_bits + s).scan(/.{1,8}/)
|
|
@data += b[0..-2].map { |x| x.to_i(2) }
|
|
@cur_bits = b.last
|
|
end
|
|
|
|
def value(encoding = 'utf-8')
|
|
# fill in the rest
|
|
rest = '0' * (8 - @cur_bits.length) + @cur_bits
|
|
arr = @data + [rest.to_i(2)]
|
|
s = arr.pack('C*')
|
|
s.force_encoding(encoding) unless encoding.nil?
|
|
s
|
|
end
|
|
end
|
|
|
|
class ErlangTransform < Parslet::Transform
|
|
class Tuple < Array; end
|
|
class Identifier < String; end
|
|
|
|
def self.assemble_binary(seq)
|
|
b = ErlangBitstream.new
|
|
seq.each { |i| b.add(i) }
|
|
b.value
|
|
end
|
|
|
|
rule(string: simple(:x)) { x.to_s }
|
|
rule(string: []) { '' }
|
|
rule(integer: simple(:x)) { x.to_i }
|
|
rule(float: { integer: simple(:a), e: simple(:b) }) { (a+b).to_f }
|
|
rule(bool: 'true') { true }
|
|
rule(bool: 'false') { false }
|
|
rule(binary: subtree(:x)) { x.nil? ? '' : ErlangTransform.assemble_binary(x) }
|
|
rule(identifier: simple(:x)) { Identifier.new(x.to_s) }
|
|
rule(array: subtree(:x)) { Array(x) }
|
|
rule(tuple: subtree(:x)) {
|
|
x.nil? ? Tuple.new : Tuple.new(x)
|
|
}
|
|
end
|
|
|
|
class ErlangConfigFile
|
|
def self.parse(content)
|
|
lex = ErlangParser.new.parse(content)
|
|
tree = ErlangTransform.new.apply(lex)
|
|
turn_to_hash(tree)
|
|
end
|
|
|
|
def self.turn_to_hash(t)
|
|
if t.is_a?(Array) && t.all? { |x| x.class == ErlangTransform::Tuple && x.length == 2 }
|
|
Hash[t.map { |i| [i[0], turn_to_hash(i[1])] }]
|
|
else
|
|
t
|
|
end
|
|
end
|
|
end
|