ponder-lab / Hybridize-Functions-Refactoring / #3312

88%

push

github

See a guard through the user frames between it and the guarded call (#1022)

* See a guard through the user frames between it and the guarded call

The expected-failure walk stopped at the first user-code frame calling
the function, so a guard written further up the stack was not seen. A
call that its own frame neither guards nor handles raises out of that
frame, so the walk now continues to the frame's callers. A frame with
no callers keeps its call, so a root's unguarded call is still evidence.
The walk memoizes each frame's sites, so a frame reached twice in one
walk is not read as unguarded the second time.

This matters once the analysis dispatches a user `__call__` override,
which then sits between a test's guard and the layer's `call`.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NTnpjPjJj2aZy1ZkUD1eAn

* Pin the walk's memo with a frame reached along two paths

A frame calling the next one from two sites gives each site its own
node, so the `called_twice` case exercised what `forwarded` already
does. Contexts merge two frames up, though, so a frame above a helper
called twice is reached along two paths in one walk, and the memo is
what keeps the second path from reading as unguarded. The `diamond`
case replaces `called_twice` and fails when the memo is reverted to a
`seen` set.

The walk's Javadoc now says what the memo is for, and notes that a
hopped call counts as guarded whether or not its argument comes from
the guarded call's.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NTnpjPjJj2aZy1ZkUD1eAn

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

28 of 29 new or added lines in 1 file covered. (96.55%)

4787 of 5445 relevant lines covered (87.92%)

0.88 hits per line