RFM: ground-scalar specimen for generic_formal_unjudged_against_a_sibling_bound_argument (conflicting actuals, advisory-mitigated)

[gunbai-bot]

Updates the existing row generic_formal_unjudged_against_a_sibling_bound_argument with a ground-scalar specimen. No code changes.

Correction. This PR first filed a new row (generic_parameter_keeps_the_first_of_conflicting_actuals) claiming that conflicting actuals are accepted silently. A diagnostic census at main ed68b5573b (gunbc compile --entry) shows that is wrong: take_s(s: pick(a: 'x', b: 2)) and pick(a: 1, b: 's') each draw DeclaredTypeInhabitanceUndecided (UndecidableGenericFormal) at a and b. That is exactly this existing class, which is mitigatable under gunbc.rung_drop.direct_call_unbound_carried_generic_compat.

What this head (bc4e34d1d6) does:

• deletes the duplicate row;
• adds the specimen to the existing row, with one new fact: for ground kernel scalars the conflict needs no nominal-identity carrier to decide.

The PR was dequeued for this change and needs a fresh review.

🤖 Generated with Claude Code

[briansrls]

LAND. This row isolates a real, silently accepted class at the exact producer: unify_generics inserts the first binding and its Present arm returns acc without judging the later actual. The two opposite-order specimens establish the first-write behavior rather than merely one bad return inference.

The ground-scalar boundary is useful and honest. String versus Int is decidable without the declaration-identity carrier needed by the bare-marker and unit-variant populations, so this row can carry a concrete structurally-guaranteed ceiling and a narrower trigger without claiming those siblings repaired. It also states #13330 as unmerged and explicitly records that its repair neither causes nor closes this class.

The next-rung evidence is properly discriminating: both mismatched orderings refuse at the second argument, while pick(a: 1, b: 2) remains accepted. No code repair or enrolled wall is claimed by this filing.

[gunbai-bot]

Reworked: this PR's first version claimed silent acceptance, and a diagnostic census at main ed68b55 shows each conflicting argument draws the UndecidableGenericFormal advisory. So it duplicated generic_formal_unjudged_against_a_sibling_bound_argument. The new row is deleted and the specimen is folded into that row; the body is updated.

— sent from jolly-pike-330

[briansrls]

LAND at this head. The prior approval’s separate-class verdict is superseded by the diagnostic census: the conflicting ground actual is not silent; it reaches the existing typed, located DeclaredTypeInhabitanceUndecided / UndecidableGenericFormal population. Deleting generic_parameter_keeps_the_first_of_conflicting_actuals and adding the ground-scalar occurrence to generic_formal_unjudged_against_a_sibling_bound_argument restores one authority for the class instead of keeping two rows distinguished only by an incorrect rung.

The specimen belongs in the existing class: the first actual fixes T, the later incompatible actual is nevertheless admitted at the same raw generic formal, and the advisory makes the path mitigatable under direct_call_unbound_carried_generic_compat. Ground String-vs-Int remains a useful narrower fact because its conflict is decidable without nominal declaration identity, but that supports an earlier repair subset, not a separate failure class.

The current PR diff contains only the addition to the established row; the duplicate row is absent. Queue only after this exact head's required witnesses complete successfully.