Acquire lock when using Jedi.

This should ensure thread safety, but might not give the best
performance. It might be possible to skip processing of some
requests if too many arrive in a short interval instead of processing
all of them with a delay once the lock becomes available for each
request.

Addresses #975.

CHANGES.rst

@@ -22,6 +22,7 @@ Release History
 0.4.3 (unreleased)
 ==================
 
+- Bugfix: thread-safety for jedi autocompletion
 - Bugfix: Handle authentication when connecting to multiple servers
 - Bugfix: Fail gracefully when binding server
 

nengo_gui/completion.py

@@ -1,3 +1,4 @@
+import threading
 import warnings
 
 try:
@@ -14,6 +15,9 @@ def completions(self):
             return []
 
 
+_jedi_lock = threading.Lock()
+
 def get_completions(code, line, column, path=None):
-    script = Script(code, line, column, path=path)
-    return script.completions()
+    with _jedi_lock:
+        script = Script(code, line, column, path=path)
+        return script.completions()