Case 219: Public Virtual Method's Override Set Changes¶

Field	Value
Verdict	🟡 COMPATIBLE_WITH_RISK
Category	Risk
Classification	Rule
Platforms	Linux
Flags	—
Detected `ChangeKind`s	`public_virtual_dispatch_set_changed`
Source files	`catalog/cases/case219_public_virtual_dispatch_set_changed/`
Rule family	`public-virtual-dispatch-set-changed`
Subject	Vtable and virtual dispatch changes

Category: Build/Source Evidence (L5) | Verdict: 🟡 COMPATIBLE_WITH_RISK

This case ships a hand-built pair of evidence-model fixtures (old.json + new.json — SourceGraphSummary dumps) instead of compilable v1/v2 sources, so the corpus is also validated compiler-free by tests/test_l3l4l5_examples.py. See scripts/gen_l3l4l5_examples.py for how they were generated.

Verdict and consumer impact¶

The exported virtual method demo::Shape::area() keeps its symbol, signature and call site. In v2 a second override, demo::Square::area(), joins demo::Circle::area() among the implementations a virtual call through Shape::area() may dispatch to. Nothing about the shipped ABI changes, so this is not a compatibility break. It is a risk: the set of possible runtime targets behind a stable public virtual changed, and the new target can alter behavior for code that calls through the base without any symbol or signature changing. The set is an approximate, over-approximated join of the source graph, so the finding never upgrades to a break on its own.

Old/new diff¶

v1 source graph	v2 source graph
`demo::Shape::area()` declared by `include/demo/shape.h`, a virtual call from `demo::render()` dispatches to `{Circle::area}`	same call and declaration; dispatch set now `{Circle::area, Square::area}`
`demo::Circle::area()` overrides `Shape::area()`	unchanged
—	new: `demo::Square::area()` overrides `Shape::area()`, and a `VIRTUAL_CALL_MAY_DISPATCH_TO` edge joins it to the call

The detector compares the candidate set per public virtual base as a set, and only when the call, override and virtual-dispatch passes ran to completion on both sides. A reordered set, or a side whose passes were narrowed or degraded, produces no finding (see case 220 for the reordered twin).

Why this is a risk, not a break¶

The override set is an over-approximation of a runtime decision: the source graph lists every override a virtual call through Shape::area() may reach, not the one a given call takes. A changed set therefore says the possible behavior behind a stable symbol moved. It does not prove that any consumer's call now lands in different code, so the finding is COMPATIBLE_WITH_RISK and never gates the exit code on its own.

What the detector does not claim¶

It makes no clean-compatibility claim when the evidence is incomplete. A side whose call, override or virtual-dispatch pass was narrowed or degraded produces no finding, and that silence means the set was not established, not that it is unchanged.

Source files¶

new.json
old.json