#513 deepcopy performance bug

[Harmen]

In comments.py please change

setattr(t, a, copy.deepcopy(getattr(self, a, memo)))

to

setattr(t, a, copy.deepcopy(getattr(self, a), memo))

otherwise it runs in quadratic complexity when deepcopying something as trivial as a list

Thanks!

[Harmen]

Any updates on this? It's a serious performance problem with a trivial fix.

[Anthon van der Neut]

I would not mind changing a performance issue, but the proposed change breaks a test that has been in existence since 2019:

_test/test_issues.py:879: AssertionError
================================================================= short test summary info =================================================================
FAILED _test/test_issues.py::TestIssues::test_issue_295 - assert "A:\n  b:\n  ... - l33: '5'\n" == "A:\n  b:\n  ... - l33: '5'\n"
================================================== 1 failed, 772 passed, 77 skipped, 22 xfailed in 4.31s ==================================================
roo ruamel/yaml »  hg diff
diff --git a/comments.py b/comments.py
--- a/comments.py
+++ b/comments.py
@@ -448,7 +448,7 @@
                   Tag.attrib, merge_attrib]:
             if hasattr(self, a):
                 if memo is not None:

-                    setattr(t, a, copy.deepcopy(getattr(self, a, memo)))
+                    setattr(t, a, copy.deepcopy(getattr(self, a), memo))
                 else:
                     setattr(t, a, getattr(self, a))
         return t

So it is not that trivial.

[Harmen]

Hi Anthon, happy new year! I've spent another couple minutes on it, and it seems that this is the full fix.

Would be happy to see this finally addressed :)

# HG changeset patch
# User Harmen Stoppels <me@harmenstoppels.nl>
# Date 1767266026 -3600
#      Thu Jan 01 12:13:46 2026 +0100
# Node ID 670ee0e7ad049a1d7d97f0cabb8b182eb029c57b
# Parent  7938d2acd452c3b5c8f4a1df2dcbc2c7661149ac
Fix quadratic complexity and broken referential integrity in deepcopy

This commit addresses two critical issues in the deepcopy implementation for
`CommentedSeq` and `CommentedBase`:


1. Referential integrity: in `CommentedBase.copy_attributes`, the `memo`
dictionary was incorrectly passed as the default value to `getattr()` instead
of being passed as the second argument to `copy.deepcopy()`. The call
`copy.deepcopy(getattr(self, a, memo))` should have been
`copy.deepcopy(getattr(self, a), memo)`. This caused `deepcopy` to start a
fresh memoization context for every attribute, breaking referential integrity
for objects shared between the sequence and its attributes (e.g., comments).


2. Quadratic complexity: in `CommentedSeq.__deepcopy__`, the call to
`self.copy_attributes` was likely accidentally indented inside the loop that
copies sequence items. Attributes (including comments) were deep-copied `N`
times for a sequence of length `N`. Combined with issue #1, this resulted in
severe `N^2` performance degradation. The line is unindented so it executes
once after the loop completes.

diff -r 7938d2acd452 -r 670ee0e7ad04 comments.py
--- a/comments.py   Wed Dec 31 17:46:54 2025 +0100
+++ b/comments.py   Thu Jan 01 12:13:46 2026 +0100
@@ -448,7 +448,7 @@
                   Tag.attrib, merge_attrib]:
             if hasattr(self, a):
                 if memo is not None:

-                    setattr(t, a, copy.deepcopy(getattr(self, a, memo)))
+                    setattr(t, a, copy.deepcopy(getattr(self, a), memo))
                 else:
                     setattr(t, a, getattr(self, a))
         return t
@@ -562,7 +562,7 @@
         memo[id(self)] = res
         for k in self:
             res.append(copy.deepcopy(k, memo))
-            self.copy_attributes(res, memo=memo)
+        self.copy_attributes(res, memo=memo)
         return res

     def __add__(self, other: Any) -> Any: