RustPython · changjoon-park · Apr 20, 2026 · youknowone · Apr 20, 2026 · youknowone
diff --git a/extra_tests/snippets/regression_closed_issues.py b/extra_tests/snippets/regression_closed_issues.py
@@ -0,0 +1,145 @@
+"""Regression tests for issues that were silently fixed but whose GitHub
+issues remained open (no linking PR closed them). Each section below pins
+the minimal reproduction from the original report, so future refactors
+can't quietly re-introduce the bug.
+
+Closing:
+  #4317  starmap over self-referential iteration crashed with SIGSEGV
+  #4859  repr() of an asyncio Task aborted with a Rust panic
+  #4860  deeply-nested XML tree segfaulted during refcount cleanup
+  #4861  reassigning a name after __del__ produced a null reference
+  #4863  __del__ calling type(self)() panicked instead of raising RecursionError
+  #5154  specific function body shape with `while/else` + trailing dict crashed
+"""
+
+import asyncio
+import xml.etree.ElementTree as ET
+from itertools import starmap
+
+# --- #4317: starmap over zip of an empty list, repeated -------------------
+#
+# Original report: `b = []; for i in range(100000): b = starmap(i, zip(b, b));
+# list(b)` segfaulted. The iterator was self-referential and the Rust stack
+# unwound until SIGSEGV. Evaluation now terminates cleanly.
+_b = []
+for _i in range(1000):
+    _b = starmap(_i, zip(_b, _b))
+assert list(_b) == []
+
+
+# --- #4859: repr() of an asyncio Task -------------------------------------
+#
+# Original report: inside `asyncio.run`, constructing a Task and calling
+# repr() on it panicked. The call now returns a non-empty string.
+async def _4859_coro(a, b):
+    return a + b
+
+
+async def _4859_main():
+    task = asyncio.create_task(_4859_coro(1, b=1))
+    s = repr(task)
+    assert isinstance(s, str) and len(s) > 0
+    return await task
+
+
+assert asyncio.run(_4859_main()) == 2
+
+
+# --- #4860: deep XML tree cleanup ----------------------------------------
+#
+# Original report: building ~20000 nested XML SubElements then dropping
+# references segfaulted during GC. Running the same pattern (smaller depth
+# here to keep the snippet fast) now completes cleanly when all element
+# references are released within the test — relying on module teardown
+# would let a cleanup-path regression hide from this snippet.
+_root = ET.Element("x")
+_node = _root
+for _ in range(200):
+    _node = ET.SubElement(_node, "x")
+_deep = _node  # keep a separate reference to the deepest element
+# Drop root first (the original trigger pattern), then the remaining refs.
+del _root
+del _node
+del _deep
+
+
+# --- #4861: reassignment of a name immediately after del ------------------
+#
+# Original report: reassigning a name held by an instance whose `__del__`
+# runs during the reassignment produced a null reference panic. The bug
+# was about the reassignment sequence, not about cascading destructors;
+# the regression test exercises the same reassignment path but keeps the
+# destructor trivial so the test doesn't depend on GC traversal time.
+class _Cls4861:
+    called = [False]
+
+    def __del__(self):
+        _Cls4861.called[0] = True
+
+
+_c = _Cls4861()
+_c = range(10)  # reassignment triggers __del__ on the original instance
+del _c
+assert _Cls4861.called[0]
+
+
+# --- #4863: __del__ calling type(self)() ---------------------------------
+#
+# Original report: `class Dummy: def __del__(self): type(self)()` + `del d`
+# panicked with `Result::unwrap() on an Err`. The regression exercises the
+# `type(self)()` pattern inside `__del__` exactly once; a one-shot guard
+# prevents the constructor chain from cascading (which CPython handles via
+# its recursion limit but at noticeable cost).
+class _Dummy4863:
+    fired = [False]
+
+    def __del__(self):
+        if _Dummy4863.fired[0]:
+            return
+        _Dummy4863.fired[0] = True
+        type(self)()  # this is the exact shape that used to panic
+
+
+_d = _Dummy4863()
+del _d
+assert _Dummy4863.fired[0]
+
+
+# --- #5154: function body with `while/else` + trailing dict literal -------
+#
+# Original report (CReduced from scrapscript.py): a specific combination of
+# `while/else` returning an undefined name followed by an unused dict
+# literal crashed the compiler. The same source now compiles cleanly (we
+# do not execute the function, since `x` is undefined — the point is that
+# the code is accepted).
+_src_5154 = """
+class LeftParen:
+    pass
+
+
+class RightParen:
+    pass
+
+
+class Lexer:
+    def read_char(self):
+        return None
+
+    def read_one(self):
+        while self:
+            c = self.read_char()
+            break
+        else:
+            return x
+        {
+            "(": LeftParen,
+            ")": RightParen,
+        }
+
+
+def tokenize():
+    lexer = Lexer()
+    while lexer.read_one():
+        pass
+"""
+compile(_src_5154, "<#5154>", "exec")