// Licensed to the .NET Foundation under one or more agreements.
// The .NET Foundation licenses this file to you under the MIT license.
 
using System;
using System.Collections.Generic;
using System.Diagnostics;
using System.Diagnostics.CodeAnalysis;
using System.Linq;
 
#if NET
using System.Numerics.Tensors;
#endif
using System.Threading.Tasks;
using Microsoft.Extensions.Caching.Distributed;
using Microsoft.Extensions.Caching.Memory;
using Microsoft.TestUtilities;
using OpenTelemetry.Trace;
using Xunit;
 
#pragma warning disable CA2214 // Do not call overridable methods in constructors
#pragma warning disable S3967  // Multidimensional arrays should not be used
 
namespace Microsoft.Extensions.AI;
 
public abstract class EmbeddingGeneratorIntegrationTests : IDisposable
{
    private readonly IEmbeddingGenerator<string, Embedding<float>>? _embeddingGenerator;
 
    protected EmbeddingGeneratorIntegrationTests()
    {
        _embeddingGenerator = CreateEmbeddingGenerator();
    }
 
    public void Dispose()
    {
        _embeddingGenerator?.Dispose();
        GC.SuppressFinalize(this);
    }
 
    protected abstract IEmbeddingGenerator<string, Embedding<float>>? CreateEmbeddingGenerator();
 
    [ConditionalFact]
    public virtual async Task GenerateEmbedding_CreatesEmbeddingSuccessfully()
    {
        SkipIfNotEnabled();
 
        var embeddings = await _embeddingGenerator.GenerateAsync("Using AI with .NET");
 
        Assert.NotNull(embeddings.Usage);
        Assert.NotNull(embeddings.Usage.InputTokenCount);
        Assert.NotNull(embeddings.Usage.TotalTokenCount);
        Assert.Single(embeddings);
        Assert.Equal(_embeddingGenerator.Metadata.ModelId, embeddings[0].ModelId);
        Assert.NotEmpty(embeddings[0].Vector.ToArray());
    }
 
    [ConditionalFact]
    public virtual async Task GenerateEmbeddings_CreatesEmbeddingsSuccessfully()
    {
        SkipIfNotEnabled();
 
        var embeddings = await _embeddingGenerator.GenerateAsync([
            "Red",
            "White",
            "Blue",
        ]);
 
        Assert.Equal(3, embeddings.Count);
        Assert.NotNull(embeddings.Usage);
        Assert.NotNull(embeddings.Usage.InputTokenCount);
        Assert.NotNull(embeddings.Usage.TotalTokenCount);
        Assert.All(embeddings, embedding =>
        {
            Assert.Equal(_embeddingGenerator.Metadata.ModelId, embedding.ModelId);
            Assert.NotEmpty(embedding.Vector.ToArray());
        });
    }
 
    [ConditionalFact]
    public virtual async Task Caching_SameOutputsForSameInput()
    {
        SkipIfNotEnabled();
 
        using var generator = new EmbeddingGeneratorBuilder<string, Embedding<float>>()
            .UseDistributedCache(new MemoryDistributedCache(Options.Options.Create(new MemoryDistributedCacheOptions())))
            .UseCallCounting()
            .Use(CreateEmbeddingGenerator()!);
 
        string input = "Red, White, and Blue";
        var embedding1 = await generator.GenerateAsync(input);
        var embedding2 = await generator.GenerateAsync(input);
        var embedding3 = await generator.GenerateAsync(input + "... and Green");
        var embedding4 = await generator.GenerateAsync(input);
 
        var callCounter = generator.GetService<CallCountingEmbeddingGenerator>();
        Assert.NotNull(callCounter);
 
        Assert.Equal(2, callCounter.CallCount);
    }
 
    [ConditionalFact]
    public virtual async Task OpenTelemetry_CanEmitTracesAndMetrics()
    {
        SkipIfNotEnabled();
 
        string sourceName = Guid.NewGuid().ToString();
        var activities = new List<Activity>();
        using var tracerProvider = OpenTelemetry.Sdk.CreateTracerProviderBuilder()
            .AddSource(sourceName)
            .AddInMemoryExporter(activities)
            .Build();
 
        var embeddingGenerator = new EmbeddingGeneratorBuilder<string, Embedding<float>>()
            .UseOpenTelemetry(sourceName: sourceName)
            .Use(CreateEmbeddingGenerator()!);
 
        _ = await embeddingGenerator.GenerateAsync("Hello, world!");
 
        Assert.Single(activities);
        var activity = activities.Single();
        Assert.StartsWith("embedding", activity.DisplayName);
        Assert.StartsWith("http", (string)activity.GetTagItem("server.address")!);
        Assert.Equal(embeddingGenerator.Metadata.ProviderUri?.Port, (int)activity.GetTagItem("server.port")!);
        Assert.NotNull(activity.Id);
        Assert.NotEmpty(activity.Id);
        Assert.NotEqual(0, (int)activity.GetTagItem("gen_ai.response.input_tokens")!);
 
        Assert.True(activity.Duration.TotalMilliseconds > 0);
    }
 
#if NET
    [ConditionalFact]
    public async Task Quantization_Binary_EmbeddingsCompareSuccessfully()
    {
        SkipIfNotEnabled();
 
        using IEmbeddingGenerator<string, BinaryEmbedding> generator =
            new QuantizationEmbeddingGenerator(
                CreateEmbeddingGenerator()!);
 
        var embeddings = await generator.GenerateAsync(["dog", "cat", "fork", "spoon"]);
        Assert.Equal(4, embeddings.Count);
 
        long[,] distances = new long[embeddings.Count, embeddings.Count];
        for (int i = 0; i < embeddings.Count; i++)
        {
            for (int j = 0; j < embeddings.Count; j++)
            {
                distances[i, j] = TensorPrimitives.HammingBitDistance(embeddings[i].Bits.Span, embeddings[j].Bits.Span);
            }
        }
 
        for (int i = 0; i < embeddings.Count; i++)
        {
            Assert.Equal(0, distances[i, i]);
        }
 
        Assert.True(distances[0, 1] < distances[0, 2]);
        Assert.True(distances[0, 1] < distances[0, 3]);
        Assert.True(distances[0, 1] < distances[1, 2]);
        Assert.True(distances[0, 1] < distances[1, 3]);
 
        Assert.True(distances[2, 3] < distances[0, 2]);
        Assert.True(distances[2, 3] < distances[0, 3]);
        Assert.True(distances[2, 3] < distances[1, 2]);
        Assert.True(distances[2, 3] < distances[1, 3]);
    }
 
    [ConditionalFact]
    public async Task Quantization_Half_EmbeddingsCompareSuccessfully()
    {
        SkipIfNotEnabled();
 
        using IEmbeddingGenerator<string, Embedding<Half>> generator =
            new QuantizationEmbeddingGenerator(
                CreateEmbeddingGenerator()!);
 
        var embeddings = await generator.GenerateAsync(["dog", "cat", "fork", "spoon"]);
        Assert.Equal(4, embeddings.Count);
 
        var distances = new Half[embeddings.Count, embeddings.Count];
        for (int i = 0; i < embeddings.Count; i++)
        {
            for (int j = 0; j < embeddings.Count; j++)
            {
                distances[i, j] = TensorPrimitives.CosineSimilarity(embeddings[i].Vector.Span, embeddings[j].Vector.Span);
            }
        }
 
        for (int i = 0; i < embeddings.Count; i++)
        {
            Assert.Equal(1.0, (double)distances[i, i], 0.001);
        }
 
        Assert.True(distances[0, 1] > distances[0, 2]);
        Assert.True(distances[0, 1] > distances[0, 3]);
        Assert.True(distances[0, 1] > distances[1, 2]);
        Assert.True(distances[0, 1] > distances[1, 3]);
 
        Assert.True(distances[2, 3] > distances[0, 2]);
        Assert.True(distances[2, 3] > distances[0, 3]);
        Assert.True(distances[2, 3] > distances[1, 2]);
        Assert.True(distances[2, 3] > distances[1, 3]);
    }
#endif
 
    [MemberNotNull(nameof(_embeddingGenerator))]
    protected void SkipIfNotEnabled()
    {
        if (_embeddingGenerator is null)
        {
            throw new SkipTestException("Generator is not enabled.");
        }
    }
}