source: subversion/sites/rails_port/lib/diff_reader.rb @ 15052

Last change on this file since 15052 was 15042, checked in by zere, 10 years ago

Added a consistency check that both ids are valid and match. Fixed diff upload code where this wasn't being set.

File size: 8.3 KB
Line 
1##
2# DiffReader reads OSM diffs and applies them to the database.
3#
4# Uses the streaming LibXML "Reader" interface to cut down on memory
5# usage, so hopefully we can process fairly large diffs.
6class DiffReader
7  include ConsistencyValidations
8
9  # maps each element type to the model class which handles it
10  MODELS = { 
11    "node"     => Node, 
12    "way"      => Way, 
13    "relation" => Relation
14  }
15
16  ##
17  # Construct a diff reader by giving it a bunch of XML +data+ to parse
18  # in OsmChange format. All diffs must be limited to a single changeset
19  # given in +changeset+.
20  def initialize(data, changeset)
21    @reader = XML::Reader.string(data)
22    @changeset = changeset
23  end
24
25  ##
26  # Reads the next element from the XML document. Checks the return value
27  # and throws an exception if an error occurred.
28  def read_or_die
29    # NOTE: XML::Reader#read returns false for EOF and raises an
30    # exception if an error occurs.
31    begin
32      @reader.read
33    rescue LibXML::XML::Error => ex
34      raise OSM::APIBadXMLError.new("changeset", xml, ex.message)
35    end
36  end
37
38  ##
39  # An element-block mapping for using the LibXML reader interface.
40  #
41  # Since a lot of LibXML reader usage is boilerplate iteration through
42  # elements, it would be better to DRY and do this in a block. This
43  # could also help with error handling...?
44  def with_element
45    # if the start element is empty then don't do any processing, as
46    # there won't be any child elements to process!
47    unless @reader.empty_element?
48      # read the first element
49      read_or_die
50
51      while @reader.node_type != 15 do # end element
52        # because we read elements in DOM-style to reuse their DOM
53        # parsing code, we don't always read an element on each pass
54        # as the call to @reader.next in the innermost loop will take
55        # care of that for us.
56        if @reader.node_type == 1 # element
57          yield @reader.name
58        else
59          read_or_die
60        end
61      end 
62    end
63    read_or_die
64  end
65
66  ##
67  # An element-block mapping for using the LibXML reader interface.
68  #
69  # Since a lot of LibXML reader usage is boilerplate iteration through
70  # elements, it would be better to DRY and do this in a block. This
71  # could also help with error handling...?
72  def with_model
73    with_element do |model_name|
74      model = MODELS[model_name]
75      raise OSM::APIBadUserInput.new("Unexpected element type #{model_name}, " +
76                                     "expected node, way or relation.") if model.nil?
77      yield model, @reader.expand
78      @reader.next
79    end
80  end
81
82  ##
83  # Checks a few invariants. Others are checked in the model methods
84  # such as save_ and delete_with_history.
85  def check(model, xml, new)
86    raise OSM::APIBadXMLError.new(model, xml) if new.nil?
87    unless new.changeset_id == @changeset.id
88      raise OSM::APIChangesetMismatchError.new(new.changeset_id, @changeset.id)
89    end
90  end
91
92  ##
93  # Consume the XML diff and try to commit it to the database. This code
94  # is *not* transactional, so code which calls it should ensure that the
95  # appropriate transaction block is in place.
96  #
97  # On a failure to meet preconditions (e.g: optimistic locking fails)
98  # an exception subclassing OSM::APIError will be thrown.
99  def commit
100
101    # data structure used for mapping placeholder IDs to real IDs
102    node_ids, way_ids, rel_ids = {}, {}, {}
103    ids = { :node => node_ids, :way => way_ids, :relation => rel_ids}
104
105    # take the first element and check that it is an osmChange element
106    @reader.read
107    raise APIBadUserInput.new("Document element should be 'osmChange'.") if @reader.name != 'osmChange'
108
109    result = OSM::API.new.get_xml_doc
110    result.root.name = "diffResult"
111
112    # loop at the top level, within the <osmChange> element
113    with_element do |action_name|
114      if action_name == 'create'
115        # create a new element. this code is agnostic of the element type
116        # because all the elements support the methods that we're using.
117        with_model do |model, xml|
118          new = model.from_xml_node(xml, true)
119          check(model, xml, new)
120
121          # when this element is saved it will get a new ID, so we save it
122          # to produce the mapping which is sent to other elements.
123          placeholder_id = xml['id'].to_i
124          raise OSM::APIBadXMLError.new(model, xml) if placeholder_id.nil?
125
126          # check if the placeholder ID has been given before and throw
127          # an exception if it has - we can't create the same element twice.
128          model_sym = model.to_s.downcase.to_sym
129          raise OSM::APIBadUserInput.new("Placeholder IDs must be unique for created elements.") if ids[model_sym].include? placeholder_id
130
131          # some elements may have placeholders for other elements in the
132          # diff, so we must fix these before saving the element.
133          new.fix_placeholders!(ids, placeholder_id)
134
135          # create element given user
136          new.create_with_history(@changeset.user)
137         
138          # save placeholder => allocated ID map
139          ids[model_sym][placeholder_id] = new.id
140
141          # add the result to the document we're building for return.
142          xml_result = XML::Node.new model.to_s.downcase
143          xml_result["old_id"] = placeholder_id.to_s
144          xml_result["new_id"] = new.id.to_s
145          xml_result["new_version"] = new.version.to_s
146          result.root << xml_result
147        end
148       
149      elsif action_name == 'modify'
150        # modify an existing element. again, this code doesn't directly deal
151        # with types, but uses duck typing to handle them transparently.
152        with_model do |model, xml|
153          # get the new element from the XML payload
154          new = model.from_xml_node(xml, false)
155          check(model, xml, new)
156
157          # if the ID is a placeholder then map it to the real ID
158          model_sym = model.to_s.downcase.to_sym
159          client_id = new.id
160          is_placeholder = ids[model_sym].include? client_id
161          id = is_placeholder ? ids[model_sym][client_id] : client_id
162
163          # and the old one from the database
164          old = model.find(id)
165
166          # translate any placeholder IDs to their true IDs.
167          new.fix_placeholders!(ids)
168          new.id = id
169
170          old.update_from(new, @changeset.user)
171
172          xml_result = XML::Node.new model.to_s.downcase
173          xml_result["old_id"] = client_id.to_s
174          xml_result["new_id"] = id.to_s
175          # version is updated in "old" through the update, so we must not
176          # return new.version here but old.version!
177          xml_result["new_version"] = old.version.to_s
178          result.root << xml_result
179        end
180
181      elsif action_name == 'delete'
182        # delete action. this takes a payload in API 0.6, so we need to do
183        # most of the same checks that are done for the modify.
184        with_model do |model, xml|
185          # delete doesn't have to contain a full payload, according to
186          # the wiki docs, so we just extract the things we need.
187          new_id = xml['id'].to_i
188          raise API::APIBadXMLError.new(model, xml, "ID attribute is required") if new_id.nil?
189
190          # if the ID is a placeholder then map it to the real ID
191          model_sym = model.to_s.downcase.to_sym
192          is_placeholder = ids[model_sym].include? new_id
193          id = is_placeholder ? ids[model_sym][new_id] : new_id
194
195          # build the "new" element by modifying the existing one
196          new = model.find(id)
197          new.changeset_id = xml['changeset'].to_i
198          new.version = xml['version'].to_i
199          check(model, xml, new)
200
201          # fetch the matching old element from the DB
202          old = model.find(id)
203
204          # can a delete have placeholders under any circumstances?
205          # if a way is modified, then deleted is that a valid diff?
206          new.fix_placeholders!(ids)
207          old.delete_with_history!(new, @changeset.user)
208
209          xml_result = XML::Node.new model.to_s.downcase
210          # oh, the irony... the "new" element actually contains the "old" ID
211          # a better name would have been client/server, but anyway...
212          xml_result["old_id"] = new_id.to_s
213          result.root << xml_result
214        end
215
216      else
217        # no other actions to choose from, so it must be the users fault!
218        raise OSM::APIChangesetActionInvalid.new(action_name)
219      end
220    end
221
222    # return the XML document to be rendered back to the client
223    return result
224  end
225
226end
Note: See TracBrowser for help on using the repository browser.