add tests

tests/windows_overflow_test.py

@@ -0,0 +1,97 @@
+"""
+Regression test for the Windows year-2038 OverflowError crash.
+
+Background
+----------
+On Windows ``time_t`` is 32-bit, so ``datetime.fromtimestamp(ts)`` raises
+``OverflowError: Python int too large to convert to C int`` for any timestamp
+past 2038 (e.g. ``NLOCKTIME_MAX = 2**32 - 1``, used as the default/sentinel
+locktime).  On 64-bit Linux the same call succeeds, which is why the bug only
+showed up on the user's Windows build: ``BalWindow.__init__`` ->
+``create_heirs_tab`` -> ``WillSettingsWidget`` -> ``on_locktime_change`` ->
+``BalTimestamp.to_date`` -> ``datetime.fromtimestamp(NLOCKTIME_MAX)`` crashed,
+which aborted ``init_menubar`` / ``load_wallet`` and left the Will/Heirs tabs
+and the menu entry half-built (the garbled/condensed element under the logo).
+
+This test forces ``datetime.fromtimestamp`` to behave like the Windows 32-bit
+implementation, then exercises ``BalTimestamp`` with NLOCKTIME_MAX to prove the
+overflow-safe conversion no longer raises and clamps to INT32_MAX.
+
+Run with:
+    QT_QPA_PLATFORM=offscreen PYTHONPATH=<electrum-src> \
+        python3 tests/windows_overflow_test.py <PLUGIN_IMPORT_NAME>
+"""
+import datetime as _datetime_mod
+import importlib
+import sys
+
+PKG = sys.argv[1] if len(sys.argv) > 1 else "electrum.plugins.bal"
+
+INT32_MAX = 2 ** 31 - 1
+NLOCKTIME_MAX = 2 ** 32 - 1  # 4294967295, the value seen in the crash log
+
+_real_datetime = _datetime_mod.datetime
+
+
+class _WindowsLikeDatetime(_real_datetime):
+    """A datetime subclass whose fromtimestamp emulates Windows' 32-bit limit."""
+
+    @classmethod
+    def fromtimestamp(cls, ts, tz=None):
+        if tz is None and (ts > INT32_MAX or ts < 0):
+            raise OverflowError("Python int too large to convert to C int")
+        return _real_datetime.fromtimestamp(ts, tz)
+
+
+def main():
+    plugin_base = importlib.import_module(f"{PKG}.core.plugin_base")
+    BalTimestamp = plugin_base.BalTimestamp
+
+    # 1) Sanity: with the real (Linux 64-bit) datetime, large ts works already.
+    bt = BalTimestamp(NLOCKTIME_MAX)
+    d = bt.to_date()
+    assert isinstance(d, _real_datetime), d
+    print("[OK] BalTimestamp(NLOCKTIME_MAX).to_date() works on this platform")
+
+    # 2) Now emulate Windows: patch datetime in the plugin_base module so that
+    #    fromtimestamp raises OverflowError past 2038, exactly like Windows.
+    original = plugin_base.datetime
+    plugin_base.datetime = _WindowsLikeDatetime
+    try:
+        # 2a) Absolute sentinel timestamp (the exact crash path from the log).
+        bt = BalTimestamp(NLOCKTIME_MAX)
+        d = bt.to_date()  # must NOT raise OverflowError anymore
+        assert d.year <= 2038, f"expected clamp to <=2038, got {d!r}"
+        print("[OK] to_date(NLOCKTIME_MAX) no longer raises (clamped to INT32_MAX)")
+
+        # 2b) to_timestamp must also be safe.
+        ts = bt.to_timestamp()
+        assert ts <= INT32_MAX, ts
+        print("[OK] to_timestamp(NLOCKTIME_MAX) clamped & safe")
+
+        # 2c) __str__ / __repr__ must not raise either.
+        _ = str(bt)
+        _ = repr(bt)
+        print("[OK] str()/repr() on out-of-range timestamp are safe")
+
+        # 2d) Relative durations that overflow when added (e.g. huge 'd').
+        bt_rel = BalTimestamp(f"{10 ** 9}d")  # ~2.7M years -> overflow
+        d2 = bt_rel.to_date()
+        assert d2 is not None
+        print("[OK] huge relative duration no longer raises")
+
+        # 2e) Normal values are unchanged (behaviour-preserving check).
+        bt_norm = BalTimestamp("90d")
+        d3 = bt_norm.to_date()
+        # 90 days from now, normalised to midnight
+        assert d3.hour == 0 and d3.minute == 0 and d3.second == 0
+        print("[OK] normal '90d' value still resolves to a midnight datetime")
+    finally:
+        plugin_base.datetime = original
+
+    print(f"\n[OK] Windows overflow regression passed for package {PKG!r}")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())