Overview

This PR fixes two bugs in tf.ragged.constant:

1. Empty lists produce incorrect static shape ((3,0) instead of (3, None)) – discovered while fixing bug 2.
2. Uniform outer dimensions are not reflected in static shape (issue tf.ragged.constant does not detect dense dimensions #40159).

The Problem

• Bug 1 (empty list): tf.ragged.constant([[], [], []]) returned shape (3,0) – a dense dimension of size 0, which is wrong because the inner dimension should be ragged (None). This bug was found during testing of the fix for bug 2.
• Bug 2 (uniform outer dimensions): tf.ragged.constant([[[1],[2,3],[4]],[[5,6],[],[7]]]) returned shape (2, None, None) instead of the expected (2, 3, None). Uniform outer dimensions (2 and 3) were lost.

Key Changes

• File ragged_factory_ops.py:
  • In _default_inner_shape_for_pylist, added if not flat_values: return () to prevent a zero‑length dense inner shape for empty lists (fixes bug 1).
  • In _constant_value, added static shape inference that:
    • Detects whether the last ragged dimension is uniform.
    • Sets concrete lengths for outer ragged dimensions only when the tensor is truly ragged (last dimension non‑uniform).
    • Leaves the last ragged dimension and all inner_shape dimensions as None, preserving existing behavior for dense tensors (fixes bug 2).
• File ragged_const_op_test.py:
  • Added new test method testUniformOuterDimensionDetection with five cases covering 3D, 4D, empty inner lists, all‑ragged, and explicit ragged_rank/inner_shape.

Verification

Below is a screenshot of the command line after running the test file:

python ragged_const_op_test.py

Results Summary

• Bug 1 fixed: tf.ragged.constant([[], [], []]) → (3, None) ✅
• Bug 2 fixed: tf.ragged.constant([[[1],[2,3],[4]],[[5,6],[],[7]]]) → (2, 3, None) ✅
• All 81 existing tests pass, including dense tensors (e.g., [[1,2,3],[4,5,6],[7,8,9]] → (3, None) unchanged).
• New test passes in graph and eager modes.