Repository navigation
call_historic() memorizes the caller's kwargs mapping, so later mutation changes what late plugins receive聽#808
Description
馃 Written by Claude Opus 5.5 via Claude Code for the pluggy maintainers; I prompted it, it did the work, I read it.
HookCaller.call_historic() appends kwargs to _call_history and replays it for plugins registered later. When the hookspec has no argument defaults, _apply_defaults() returns the caller's own mapping unchanged, so the history holds a reference to the caller's object:
d = {"arg": "at-call-time"}
pm.hook.myhook.call_historic(kwargs=d)
d["arg"] = "mutated"
pm.register(LatePlugin()) # LatePlugin.myhook receives arg="mutated"When the spec has defaults, _apply_defaults() builds a new dict, so those calls are already isolated. The behavior therefore depends on whether the spec happens to declare a default.
Fix: store dict(kwargs) after applying defaults. This is a shallow copy, so mutable argument values are still shared, and that's expected. Low priority: nobody has reported it in the wild, and pytest passes fresh dicts.
Found while reviewing #806.
Activity
rajnish-verma-22172 commentedon Oct 21, 2026
I'd like to work on this. I'll add a regression test showing that mutating the original kwargs after call_historic() must not change what a plugin registered later receives, then copy the mapping at the point history is stored. I'll also check the existing default-argument behavior and run the relevant test suite.
RonnyPfannschmidt commentedon Oct 21, 2026
please don't - the general misaligned ai pr backlog for pytest-dev is ridiculous already - currently we are forced to reject a lot of goodwill as most external ai assists dont just miss the target, they miss the target range - and it comes down to what and how prompts are done
we'd prefer to spend our tokens on actual changes instead of assessing ai assists -