Summary

[AI-assisted WIP experiment — no review yet]

Broadens isdict() from isinstance(x, dict) to isinstance(x, Mapping) so non-dict Mapping types get structured assertion diff output (#14461).

• One-line fix in src/_pytest/assertion/util.py
• _compare_eq_dict already accepts Mapping in its type signature, so only the guard needed changing
• Backward-compatible: dict is a subclass of Mapping

Mapping types surveyed

Now get structured diff (previously fell through to generic iterable diff):

• types.MappingProxyType (stdlib)
• collections.ChainMap (stdlib)
• collections.UserDict (stdlib)
• immutables.Map
• bidict / frozenbidict
• httpx.Headers (normalizes keys on iteration, so set() works)

Already handled (dict subclasses):

• collections.OrderedDict
• python-box.Box / dynaconf.DynaBox
• sortedcontainers.SortedDict

Known-problematic — xfail tests added as documentation:

• requests.CaseInsensitiveDict — iterates original-cased keys, so set(left) & set(right) misses common keys when casing differs (misleading, not crash)
• multidict.MultiDict — set() collapses duplicate keys, __getitem__ returns only first value (silently lossy, not crash)
• multidict.CIMultiDict — both problems combined

Key insight

The failure modes for non-dict Mappings are misleading output, not crashes. The _compare_eq_dict algorithm (set(keys) intersection/difference + mapping[k] lookups) works correctly for the vast majority of real-world Mapping implementations. The problematic types (case-insensitive, multi-valued) produce suboptimal but still safe output — which is better than the current state of no structured diff at all.

Test plan

• 6 new passing tests for stdlib mapping types (MappingProxyType, ChainMap, UserDict, custom Mapping)
• 3 xfail tests documenting known-problematic external types
• All 61 existing TestAssert_reprcompare* tests pass (no regression)
• CI validation

Made with Cursor