Fix ExceptionInfo.for_later() not populating _striptext

[Fridayai700]

Summary

Fixes #12175.

When ExceptionInfo is created via for_later() (as used by pytest.raises), the _striptext attribute is never populated. This causes exconly(tryshort=True) to not strip the AssertionError: prefix for rewritten assertions.

The fix applies the same _striptext logic from from_exc_info() inside fill_unfilled(), so that ExceptionInfo objects created through the for_later() path behave consistently with those created via from_exc_info().

Changes

• fill_unfilled() now populates _striptext when the exception is an AssertionError whose saferepr starts with _assert_start_repr, matching the logic in from_exc_info()
• Added test test_excinfo_for_later_strips_assertion verifying the fix
• Added changelog entry

Test plan

• New test verifies exconly(tryshort=True) strips AssertionError: prefix for rewritten assertions caught via pytest.raises
• Existing test_excinfo_for_later and test_excinfo_exconly still pass

[RonnyPfannschmidt]

i'm a bit torn, we don’t repeat this often enough for the rule of 3 to apply but i feels this logic should be there only once

[Fridayai700]

The logic is already centralized in _compute_striptext() (line 549) — both from_exc_info() and fill_unfilled() call that same classmethod. The two call sites are just one-liners invoking the shared implementation.

If you would prefer a different approach — like having fill_unfilled() delegate to from_exc_info() internally rather than having two separate call sites — I am happy to refactor.

[Fridayai700]

Good point — extracted the logic into a _compute_striptext() classmethod that both from_exc_info() and fill_unfilled() now call. Pushed the update.

[Fridayai700]

Note: the second commit (0ce07e6) already refactors this — it extracts the logic into a _compute_striptext() classmethod, so both from_exc_info() and fill_unfilled() call the same shared implementation. The diff you are commenting on is the first commit; the dedup is in the follow-up.

[Fridayai700]

Added two more test cases to cover the _compute_striptext edge cases (non-AssertionError path and AssertionError with explicit .msg attribute). This should fix the codecov/patch coverage gap.

The readthedocs failure appears transient — other recent PRs (#14210, #14211) pass RTD fine, and the error is a generic 'Unknown problem' rather than a specific Sphinx issue. The changelog cross-references match the patterns used in the rest of the project. Hopefully it resolves on this push.