`tf.raw_ops.BlockLSTMGradV2` crashes when `seq_len_max` exceeds time dimension of inputs (missing shape validation)

Issue type

Bug

Have you reproduced the bug with TensorFlow Nightly?

Yes

Source

source

TensorFlow version

tf 2.19

Custom code

Yes

OS platform and distribution

No response

Mobile device

No response

Python version

No response

Bazel version

No response

GCC/compiler version

No response

CUDA/cuDNN version

No response

GPU model and memory

No response

Current behavior?

When calling tf.raw_ops.BlockLSTMGradV2 with a seq_len_max value larger than the actual time dimension of inputs, the process crashes (segmentation fault) instead of raising a Python exception.

The following code would crash on tf 2.19. To reproduce the issue, I provided that a colab notebook to reproduce the error.

Standalone code to reproduce the issue

import tensorflow as tf
import numpy as np

seq_len_max = 3
batch_size = 2
input_size = 4
num_units = 5

x = tf.constant(np.random.randn(seq_len_max - 1, batch_size, input_size).astype(np.float32)) # x: (2,2,4), seq_len_max:3
cs_prev = tf.constant(np.random.randn(batch_size, num_units).astype(np.float32))
h_prev = tf.constant(np.random.randn(batch_size, num_units).astype(np.float32))
w = tf.constant(np.random.randn(input_size + num_units, 4 * num_units).astype(np.float32))
wci = tf.constant(np.random.randn(num_units).astype(np.float32))
wcf = tf.constant(np.random.randn(num_units).astype(np.float32))
wco = tf.constant(np.random.randn(num_units).astype(np.float32))
b = tf.constant(np.random.randn(4 * num_units).astype(np.float32))
i = tf.constant(np.random.randn(seq_len_max - 1, batch_size, num_units).astype(np.float32))
cs = tf.constant(np.random.randn(seq_len_max - 1, batch_size, num_units).astype(np.float32))
f = tf.constant(np.random.randn(seq_len_max - 1, batch_size, num_units).astype(np.float32))
o = tf.constant(np.random.randn(seq_len_max - 1, batch_size, num_units).astype(np.float32))
ci = tf.constant(np.random.randn(seq_len_max - 1, batch_size, num_units).astype(np.float32))
co = tf.constant(np.random.randn(seq_len_max - 1, batch_size, num_units).astype(np.float32))
h = tf.constant(np.random.randn(seq_len_max - 1, batch_size, num_units).astype(np.float32))
cs_grad = tf.constant(np.random.randn(seq_len_max - 1, batch_size, num_units).astype(np.float32))
h_grad = tf.constant(np.random.randn(seq_len_max - 1, batch_size, num_units).astype(np.float32))
use_peephole = True

result = tf.raw_ops.BlockLSTMGradV2(
    seq_len_max=seq_len_max, # seq_len_max:3
    x=x, #  x: (2,2,4)
    cs_prev=cs_prev,
    h_prev=h_prev,
    w=w,
    wci=wci,
    wcf=wcf,
    wco=wco,
    b=b,
    i=i,
    cs=cs,
    f=f,
    o=o,
    ci=ci,
    co=co,
    h=h,
    cs_grad=cs_grad,
    h_grad=h_grad,
    use_peephole=use_peephole
)

Relevant log output

2025-12-17 11:28:27.238936: F tensorflow/core/framework/tensor.cc:1078] Check failed: limit <= dim0_size (3 vs. 2)
[1]    2650852 abort (core dumped)