Merge pull request #46 from int19h/2079

Fix microsoft/ptvsd#2079: sys.argv sends in unicode instead of string

Author: Pavel Minaev

src/debugpy/server/cli.py

@@ -171,9 +171,16 @@ def do(arg, it):
 # fmt: on
 
 
-def parse(args):
+def consume_argv():
+    while len(sys.argv) >= 2:
+        value = sys.argv[1]
+        del sys.argv[1]
+        yield value
+
+
+def parse_argv():
     seen = set()
-    it = (compat.filename(arg) for arg in args)
+    it = consume_argv()
 
     while True:
         try:
@@ -215,14 +222,12 @@ def parse(args):
     assert options.target_kind is not None
     assert options.address is not None
 
-    return it
-
 
 def start_debugging(argv_0):
     # We need to set up sys.argv[0] before invoking either listen() or connect(),
     # because they use it to report the "process" event. Thus, we can't rely on
     # run_path() and run_module() doing that, even though they will eventually.
-    sys.argv[0] = compat.filename(argv_0)
+    sys.argv[0] = compat.filename_str(argv_0)
     log.debug("sys.argv after patching: {0!r}", sys.argv)
 
     debugpy.configure(options.config)
@@ -392,7 +397,7 @@ def attach_to_pid():
 def main():
     original_argv = list(sys.argv)
     try:
-        sys.argv[1:] = parse(sys.argv[1:])
+        parse_argv()
     except Exception as ex:
         print(HELP + "\nError: " + str(ex), file=sys.stderr)
         sys.exit(2)

tests/debugpy/server/test_cli.py

@@ -23,26 +23,30 @@ def cli_parser():
         from debugpy.server import cli
 
         try:
-            sys.argv[1:] = cli.parse(sys.argv[1:])
+            cli.parse_argv()
         except Exception as exc:
             os.write(1, pickle.dumps(exc))
             sys.exit(1)
-        else:
-            # We only care about options that correspond to public switches.
-            options = {
-                name: getattr(cli.options, name)
-                for name in [
-                    "address",
-                    "config",
-                    "log_to",
-                    "log_to_stderr",
-                    "mode",
-                    "target",
-                    "target_kind",
-                    "wait_for_client",
-                ]
-            }
-            os.write(1, pickle.dumps([sys.argv[1:], options]))
+
+        # Check that sys.argv has the correct type after parsing - there should be
+        # no bytes on Python 3, nor unicode on Python 2.
+        assert all(isinstance(s, str) for s in sys.argv)
+
+        # We only care about options that correspond to public switches.
+        options = {
+            name: getattr(cli.options, name)
+            for name in [
+                "address",
+                "config",
+                "log_to",
+                "log_to_stderr",
+                "mode",
+                "target",
+                "target_kind",
+                "wait_for_client",
+            ]
+        }
+        os.write(1, pickle.dumps([sys.argv[1:], options]))
 
     def parse(args):
         log.debug("Parsing argv: {0!r}", args)