Support snippet completions

In general this is terrible, compared to working normal completion, but
some servers (json, yaml) will only ever return snippet completions, and
they will not be convinved to change.

The approach is to put the snippet text in the extra_data and have the
client expand the snippet in-place on selecting the completion item.

As many servers will agressively over-use snippets when the client
claims they support it, we _dont'_ claim we support it, but provide a
mechanism for extra_conf files to override the capabilities that we
declare. This allows users to enable snippets where they make sense,
i.e. for the json LS.

ycmd/completers/language_server/language_server_completer.py

@@ -1441,7 +1441,7 @@ def _CandidatesFromCompletionItems( self,
         this_tem_is_resolved = True
 
       try:
-        insertion_text, extra_data, start_codepoint = (
+        insertion_text, snippet, extra_data, start_codepoint = (
           _InsertionTextForItem( request_data, item ) )
       except IncompatibleCompletionException:
         LOGGER.exception( 'Ignoring incompatible completion suggestion %s',
@@ -1459,6 +1459,13 @@ def _CandidatesFromCompletionItems( self,
         # We'll use this later to do the resolve.
         extra_data[ 'item' ] = item
 
+      if snippet:
+        extra_data = {} if extra_data is None else extra_data
+        extra_data[ 'snippet' ] = {
+          'snippet': snippet,
+          'trigger_string': insertion_text
+        }
+
       min_start_codepoint = min( min_start_codepoint, start_codepoint )
 
       # Build a ycmd-compatible completion for the text as we received it. Later
@@ -2182,9 +2189,11 @@ def _SendInitialize( self, request_data ):
       # the settings on the Initialize request are somehow subtly different from
       # the settings supplied in didChangeConfiguration, though it's not exactly
       # clear how/where that is specified.
+      extra_capabilities = self._settings.get( 'capabilities' , {} )
+      extra_capabilities.update( self.ExtraCapabilities() )
       msg = lsp.Initialize( request_id,
                             self._project_directory,
-                            self.ExtraCapabilities(),
+                            extra_capabilities,
                             self._settings.get( 'ls', {} ) )
 
       def response_handler( response, message ):
@@ -2783,13 +2792,21 @@ def _InsertionTextForItem( request_data, item ):
 
   Returns a tuple (
      - insertion_text   = the text to insert
+     - snippet          = optional snippet text
      - fixits           = ycmd fixit which needs to be applied additionally when
                           selecting this completion
      - start_codepoint  = the start column at which the text should be inserted
   )"""
   start_codepoint = request_data[ 'start_codepoint' ]
+  label = item[ 'label' ]
+  insertion_text_is_snippet = False
   # We will always have one of insertText or label
   if 'insertText' in item and item[ 'insertText' ]:
+    # 1 = PlainText
+    # 2 = Snippet
+    if lsp.INSERT_TEXT_FORMAT[ item.get( 'insertTextFormat', 1 ) ] == 'Snippet':
+      insertion_text_is_snippet = True
+
     insertion_text = item[ 'insertText' ]
   else:
     insertion_text = item[ 'label' ]
@@ -2808,7 +2825,7 @@ def _InsertionTextForItem( request_data, item ):
 
     insertion_text = text_edit[ 'newText' ]
 
-    if '\n' in insertion_text:
+    if '\n' in insertion_text and not insertion_text_is_snippet:
       # FIXME: If this logic actually worked, then we should do it for
       # everything and not just hte multi-line completions ? Would that allow us
       # to avoid the _GetCompletionItemStartCodepointOrReject logic ? Possibly,
@@ -2830,9 +2847,17 @@ def _InsertionTextForItem( request_data, item ):
       #    insertion
       #  - insert another textEdit in additionalTextEdits which applies this
       #    textedit
+      #
+      # On the other hand, if the insertion text is a snippet, then the snippet
+      # system will handle the expansion.
       insertion_text = item[ 'label' ]
       start_codepoint = request_data[ 'start_codepoint' ]
 
+      # FIXME:
+      # So forced-completion breaks here, as the textEdit is formulated to
+      # remove the existing "prefix". E.g. typing getT<ctrl-space>, the edit
+      # attempts to replace getT with the new code.
+
       # Add a fixit which removes the inserted label
       #
       # TODO: Perhaps we should actually supply a completion with an empty
@@ -2864,7 +2889,7 @@ def _InsertionTextForItem( request_data, item ):
       contents = GetFileLines( request_data, filepath )
       completion_fixit_chunks = [
         responses.FixItChunk(
-          text_edit[ 'newText' ],
+          text_edit[ 'newText' ], # FIXME: This could also be a Snippet
           _BuildRange( contents, filepath, text_edit[ 'range' ] )
         )
       ]
@@ -2898,7 +2923,11 @@ def _InsertionTextForItem( request_data, item ):
     fixits.append( responses.FixIt( chunks[ 0 ].range.start_, chunks ) )
 
   extra_data = responses.BuildFixItResponse( fixits ) if fixits else None
-  return insertion_text, extra_data, start_codepoint
+
+  if insertion_text_is_snippet:
+    return label, insertion_text, extra_data, start_codepoint
+
+  return insertion_text, None, extra_data, start_codepoint
 
 
 def FindOverlapLength( line_value, insertion_text ):

ycmd/responses.py

@@ -312,7 +312,6 @@ def BuildFixItData( fixit ):
         'text': fixit.text,
         'kind': fixit.kind,
         'resolve': fixit.resolve,
-        'is_completion': fixit.is_completion
       }
     else:
       result = {

ycmd/utils.py

@@ -25,6 +25,7 @@
 import tempfile
 import time
 import threading
+import collections.abc as collections_abc
 
 LOGGER = logging.getLogger( 'ycmd' )
 ROOT_DIR = os.path.normpath( os.path.join( os.path.dirname( __file__ ), '..' ) )