Fix NOQA wrap mode accumulating spaces before the comment on each run

isort/comments.py

@@ -30,6 +30,14 @@ def add_to_line(
         if comment not in unique_comments:
             unique_comments.append(comment)
     comment_text = "; ".join(unique_comments)
+    base, existing_comment = parse(original_string)
+    if existing_comment is not None:
+        # An existing comment was stripped off, leaving the whitespace that used to
+        # precede it (the previous ``comment_prefix`` spacing). Drop it so re-adding a
+        # comment does not accumulate leading spaces on every run. Bases without a
+        # comment (e.g. the indentation-only strings used by the multi-line wrap modes)
+        # are left untouched so their indentation is preserved.
+        base = base.rstrip()
     if comment_text:
-        return f"{parse(original_string)[0]}{comment_prefix} {comment_text}"
-    return f"{parse(original_string)[0]}{comment_prefix}"
+        return f"{base}{comment_prefix} {comment_text}"
+    return f"{base}{comment_prefix}"

tests/unit/test_regressions.py

@@ -2094,3 +2094,26 @@ def test_noqa_wrap_mode_idempotent_with_existing_comment():
     second_pass = isort.code(first_pass, multi_line_output=7)
     assert second_pass == first_pass
     assert second_pass.count("NOQA") == 1
+
+
+def test_noqa_wrap_mode_does_not_accumulate_spaces_with_as_import():
+    """A long ``as`` import in NOQA mode must not grow extra spaces before ``# NOQA``.
+
+    With ``force_single_line`` an aliased import that overflows the line length is emitted
+    on its own line and gets a ``# NOQA`` appended.  The auto-added comment is re-parsed on
+    the next run and put back through ``add_to_line``, which used to keep the whitespace that
+    had preceded the stripped ``#`` and then add the comment prefix on top of it - so every
+    pass inserted two more spaces (``import x  # NOQA`` -> ``import x    # NOQA`` -> ...),
+    never reaching a fixpoint. See issue #2394.
+    """
+    to_sort = "from my_package.my_module import super_long_file_name as super_long_alias\n"
+
+    first_pass = isort.code(to_sort, multi_line_output=7, force_single_line=True, line_length=40)
+    assert first_pass == (
+        "from my_package.my_module import super_long_file_name as super_long_alias  # NOQA\n"
+    )
+
+    second_pass = isort.code(
+        first_pass, multi_line_output=7, force_single_line=True, line_length=40
+    )
+    assert second_pass == first_pass