Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.43%. Comparing base (299c641) to head (0d27095).
⚠️ Report is 5 commits behind head on main.

@@            Coverage Diff             @@
##             main     #714      +/-   ##
==========================================
+ Coverage   91.35%   91.43%   +0.07%     
==========================================
  Files           7        7              
  Lines        1979     1997      +18     
==========================================
+ Hits         1808     1826      +18     
  Misses        171      171              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

Nice one tracking that down.

Can we get some tests for it? Something like your reproducer that covers unicode, non-unicode as well as the default() function returning a structure containing more than one string? If you pass --leak-max-loops=10000 to pytest, it applies a loop + forced gc + tracemalloc auditing to each test so you don't need to do any of that yourself – just run the offending code paths.

Ofc, I'll be working on tests

@bwoodsend I added 2 relevant test cases for this PR. Here is the test result

Test Result

Setup

Run command:

python -m pytest --leak-max-loops=5000 tests/test_ujson.py

Before Fix

======================================= short test summary info =======================================
FAILED tests/test_ujson.py::test_no_memory_leak_default_non_ascii - Failed: 242633B leaked (96.97561950439648 per iteration)
=================================== 1 failed, 300 passed in 15.89s ====================================

After Fix

======================================== 301 passed in 11.28s =========================================

Thanks for figuring that one out.