use hurray_core::{
    BufferHandle, DeviceTag, ElementType, LayoutDescriptor, Shape, SyncMode,
    TensorDescriptor, MIN_BUFFER_ALIGNMENT,
};

fn main() -> Result<(), Box<dyn std::error::Error>> {
    // Describe a float32 [2, 3] row-major tensor.
    let shape = Shape::new(vec![2u64, 3])?;
    let buffer = BufferHandle::new(
        24,                     // 6 elements × 4 bytes
        MIN_BUFFER_ALIGNMENT,   // 64-byte SIMD alignment
        DeviceTag::Cpu,
        SyncMode::ProducerSynced,
    )?;
    let desc = TensorDescriptor::new(
        1, 0,                   // descriptor format version 1.0
        ElementType::Float32,
        shape,
        0,                      // byte_offset to element [0, 0]
        LayoutDescriptor::RowMajor,
        vec![buffer],
        None, None, None, None, // no quantization / shard / statistics / extension-type
    )?;

    // Encode to the self-delimiting wire format …
    let bytes = desc.encode()?;
    println!("descriptor: {} bytes", bytes.len());

    // … and decode it straight back.
    let decoded = TensorDescriptor::decode(&bytes)?;
    assert_eq!(decoded, desc);
    println!(
        "shape = {:?}, dtype = {:?}",
        decoded.shape.dims(),
        decoded.element_type,
    );
    Ok(())
}

import os
import tempfile

import numpy as np
import hurray

# Build a float32 [2, 3] tensor, zero-copy from a NumPy array.
arr = np.arange(6, dtype=np.float32).reshape(2, 3)
t = hurray.from_numpy(arr)
print("shape =", t.shape, "dtype =", t.dtype, "device =", t.device)

# Hand it back to NumPy zero-copy via DLPack (dense Tier-1 tensors share the buffer).
view = np.from_dlpack(t)
assert np.array_equal(view, arr)

# Round-trip through the Hurray file format.
path = os.path.join(tempfile.gettempdir(), "quickstart.hrry")
hurray.save(path, {"x": t})
loaded = hurray.load(path)
print("loaded:", list(loaded.keys()), "→", loaded["x"].shape)
os.unlink(path)