[tcoratger]

ssz.get_generalized_index now reads a path as the consensus specification does.
A field of a container is named, never given by its position, and the words __active_fields__ and __selector__ select the mixed-in nodes of a progressive container and of a union.

Part of #2480

The problem

The consensus specification computes generalized indices with get_generalized_index(type, *path).
In the release that tests/sources.toml pins, a part of the path that steps into a container must be a field name.
An integer there is an error: for a container C, the path (2,) raises "C has no field named 2".

The export added in #2508 took an integer as the position of the field.
So a caller that relied on the error, such as a negative test, got an index instead.

The same reader also names the field layout of a progressive container __active_fields__, and the selector of a union __selector__.
Both are the right child of their node.
The export knew only __len__, so it refused both words, although the SSZ library has a step for each.

The idea

The step reader of the export now refuses an integer when the type it has reached is a container or a progressive container.
Every other shape keeps its integer steps: elements of lists and vectors, bits of bitfields, and the selector of a union.
EthLib's own paths in Lean still select fields by position; only the Python path changes.
The two words map to the existing steps for the field layout and the selector, and behave like __len__ past the tree.

Example

For NESTED, a container whose second field pair is itself a container:

• get_generalized_index(NESTED, "pair") is 5, as before.
• get_generalized_index(NESTED, 1) was 5, and is now refused with path: no field 1.
• get_generalized_index(NESTED, "pair", 0) is refused with path: no field 0.
• For SQUARE, a progressive container, get_generalized_index(SQUARE, "__active_fields__") is 3, the right child of the root, as in the reference.

For reviewers

• The error is the one an unknown field name gets, so a caller sees one kind of fault for a field that a path does not name.
• A path that has already left the tree keeps its old error: the step reader sees no type there.
• The same release gives no branch for the root, generalized index 1: its build_proof raises "the root has no proof branch of its own".
  EthLib's export refuses index 1 too, so that point of feat(ssz): export generalized indices, Merkle proofs and chunk roots #2508 needs no change.

Checks

• lake build --wfail Exports, lake exe gen-bindings (contract regenerated for the changed doc): passed.
• scripts/check_lean_layout.py --base upstream/master, scripts/check_layers.py: passed.
• ruff check, ruff format --check, ty check, scripts/check_python_comments.py --all: passed.
• pytest bindings/python/tests/ssz: 4266 passed, 3 skipped.
• 57 type-and-path cases compared with the SSZ package of the pinned consensus-specs release: the same outcome and index on each.
  The two words were the one divergence that comparison found; they now give 3, as the reference does.
• Statement audit (FULL) and spec-fidelity review: no CRITICAL or MAJOR finding left.

🤖 Generated with Claude Code