Fix/corrige build docker by daniloflorenzano · Pull Request #16 · daniloflorenzano/Melissa

daniloflorenzano · 2025-11-27T02:19:44Z

Esse PR corrige problemas com o deploy usando docker
Tambem habilita o container do ollama a usar GPU Nvidia (nao testado para AMD)

…issões de bibliotecas

…anscrição

…cess

Copilot

Pull request overview

This PR addresses Docker deployment issues and adds mobile audio support with GPU acceleration for the Ollama container. The changes include critical fixes for Docker builds, a new audio decoding pipeline for mobile clients, and CORS configuration for cross-origin requests.

Key Changes:

Enables Nvidia GPU support for the Ollama container in Docker Compose
Adds AAC-to-PCM audio decoding for mobile audio input using FFmpeg
Configures CORS to allow mobile app requests and adds automatic database migrations on startup

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 22 comments.

Show a summary per file

File	Description
compose.yaml	Removed HolidaysCsvPath environment variable, added timezone configuration, and configured Nvidia GPU support for Ollama
Dockerfile	Added native dependencies for Whisper, fixed permission issues, and added holidays CSV file copying with path configuration
Program.cs	Added DbContext DI registration, CORS configuration with unrestricted origin policy, automatic database migrations, and startup model download
MelissaHub.cs	Extracted model configuration to constants, added new `AskMelissaAudioFromMobile` endpoint with AAC decoding, significant code duplication between audio methods
AudioDecoder.cs	New static class for AAC-to-PCM conversion using FFmpeg (not installed in Docker image)
AppEndpoints.cs	Added `GetNationalHolidaysAsTxt` method to return holidays as string instead of file export
HolidayService.cs	Implemented `GetNationalHolidaysAsTxt` to query and format national holidays
SettingsWindow.axaml	Removed trailing whitespace
Melissa.cs	Updated system message to exclude formatting/emojis (contains spelling error)

Comments suppressed due to low confidence (2)

src/Melissa/Melissa.WebServer/MelissaHub.cs:178

Using the null-forgiving operator (!) on cnVoice without null checking is unsafe. If no voice matching "pt-BR-FranciscaNeural" is found, cnVoice will be null and accessing cnVoice!.Name will throw a NullReferenceException.

Add proper null checking:

var voices = await edgeTts.GetVoices();
var cnVoice = voices.FirstOrDefault(v => v.ShortName == "pt-BR-FranciscaNeural");

if (cnVoice == null)
{
    Log.Warning("Voice pt-BR-FranciscaNeural not found, using default voice");
    cnVoice = voices.FirstOrDefault() ?? throw new InvalidOperationException("No voices available");
}

var options = new TTSOption
(
    voice: cnVoice.Name,
    // ...
);

        var voices = await edgeTts.GetVoices();
        var cnVoice = voices.FirstOrDefault(v => v.ShortName == "pt-BR-FranciscaNeural");
        var options = new TTSOption
        (
            voice: cnVoice!.Name,

src/Melissa/Melissa.WebServer/MelissaHub.cs:206

There's significant code duplication between AskMelissaAudio and AskMelissaAudioFromMobile. The only difference is the AAC decoding step (lines 132-134). The rest of the transcription, assistant interaction, and TTS logic is identical (90+ lines duplicated).

Consider refactoring to extract the common logic into a shared private method:

private async IAsyncEnumerable<byte[]> ProcessAudioQuery(
    byte[] pcmBytes, 
    MelissaAssistant melissa,
    [EnumeratorCancellation] CancellationToken cancellationToken)
{
    // Common logic for transcription, assistant query, and TTS
}

Then call it from both methods after obtaining the PCM bytes.

    public async IAsyncEnumerable<byte[]> AskMelissaAudioFromMobile(IAsyncEnumerable<byte[]> audioStream,
        [FromServices] MelissaAssistant melissa,
        [EnumeratorCancellation] CancellationToken cancellationToken = default)
    {
        using var ms = new MemoryStream();

        await foreach (var chunk in audioStream.WithCancellation(cancellationToken))
        {
            await ms.WriteAsync(chunk, cancellationToken);
        }

        // ✅ NOVO: Decodificar AAC para PCM
        var aacBytes = ms.ToArray();
        var pcmBytes = await AudioDecoder.DecodeAACToPCM(aacBytes);
        
        if (!File.Exists(ModelFileName))
        {
            await DownloadModel(ModelFileName, GgmlType);
        }

        using var whisperFactory = WhisperFactory.FromPath("ggml-medium.bin");
        await using var processor = whisperFactory.CreateBuilder()
            .WithLanguage("pt")
            .Build();

        // ✅ Usar PCM decodificado em vez de tentar converter
        var wavBytes = GenerateWav(pcmBytes);

        var wavStream = new MemoryStream(wavBytes);
        wavStream.Seek(0, SeekOrigin.Begin);
        
        Log.Information("Iniciando transcrição de áudio...");

        var msgBuilder = new StringBuilder();
        await foreach (var result in processor.ProcessAsync(wavStream, cancellationToken))
        {
            msgBuilder.Append(result.Text);
        }

        var message = msgBuilder.ToString();
        Log.Information("Usuário: {0}", message);

        var question = new Question(message, "AudioHub", DateTimeOffset.Now);
        string melissaReply;
        var (isAvailable, statusMessage) = await melissa.CanUse();

        if (isAvailable)
            melissaReply =
                await MelissaHub.SafeAskMelissaWithErrorHandlingAndRetry(melissa, question, cancellationToken);
        else
            melissaReply = statusMessage;

        var edgeTts = new EdgeTTSNet();
        var voices = await edgeTts.GetVoices();
        var cnVoice = voices.FirstOrDefault(v => v.ShortName == "pt-BR-FranciscaNeural");
        var options = new TTSOption
        (
            voice: cnVoice!.Name,
            pitch: "+0Hz",
            rate: "+25%",
            volume: "+0%"
        );

        Log.Information("Assistente: {0}", melissaReply);

        edgeTts = new EdgeTTSNet(options);
        var channel = Channel.CreateUnbounded<byte[]>();

        _ = Task.Run(async () =>
        {
            await edgeTts.TTS(melissaReply, (metaObj) =>
            {
                if (metaObj.Type == TTSMetadataType.Audio)
                {
                    channel.Writer.TryWrite(metaObj.Data);
                }
            }, cancellationToken);

            channel.Writer.Complete();
        }, cancellationToken);

        await foreach (var audioChunk in channel.Reader.ReadAllAsync(cancellationToken))
        {
            yield return audioChunk;
        }
    }

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2025-11-27T02:27:41Z

 var assistantFactory = new AssistantFactory();
 var melissa = await assistantFactory.TryCreateMelissa(TimeSpan.FromSeconds(10));

+await MelissaHub.DownloadModel(MelissaHub.ModelFileName, MelissaHub.GgmlType);


The model download is called synchronously during application startup before the services are built. If the model download fails or takes a long time, it will block the entire application startup process and could cause deployment failures.

Consider moving this to a background service or lazy loading it on first use, with proper error handling:

// Remove this line from here // await MelissaHub.DownloadModel(MelissaHub.ModelFileName, MelissaHub.GgmlType);

The model is already downloaded on-demand in AskMelissaAudio and AskMelissaAudioFromMobile (lines 46-48, 136-138 in MelissaHub.cs), so this pre-download is redundant.

Suggested change

await MelissaHub.DownloadModel(MelissaHub.ModelFileName, MelissaHub.GgmlType);

Copilot · 2025-11-27T02:27:41Z

+RUN mkdir -p /app/Data
+COPY ./Melissa.WebServer/Data/holidays_2025.csv /app/Data/holidays_2025.csv


The removal of the HolidaysCsvPath environment variable from compose.yaml combined with the Dockerfile copying the file to /app/Data/holidays_2025.csv (capital D) creates a path mismatch.

The application defaults to "data/holidays_2025.csv" (lowercase 'd') when HolidaysCsvPath is not set (line 73 in Program.cs), which will be resolved to /app/data/holidays_2025.csv at runtime. However, the Dockerfile copies the file to /app/Data/holidays_2025.csv (capital 'D').

This mismatch will cause the file not to be found. Either:

Re-add the environment variable in compose.yaml:

- HolidaysCsvPath=/app/Data/holidays_2025.csv

OR

Change the Dockerfile to use lowercase:

RUN mkdir -p /app/data COPY ./Melissa.WebServer/Data/holidays_2025.csv /app/data/holidays_2025.csv

Suggested change

RUN mkdir -p /app/Data

COPY ./Melissa.WebServer/Data/holidays_2025.csv /app/Data/holidays_2025.csv

RUN mkdir -p /app/data

COPY ./Melissa.WebServer/Data/holidays_2025.csv /app/data/holidays_2025.csv

Copilot · 2025-11-27T02:27:42Z

+            using (var process = System.Diagnostics.Process.Start(processInfo))
+            {
+                await process.WaitForExitAsync();
+                if (process.ExitCode != 0)
+                {
+                    var error = process.StandardError.ReadToEnd();
+                    throw new InvalidOperationException($"FFmpeg failed: {error}");
+                }
+            }


The Process object is not properly disposed. While the using statement is used, the process should be explicitly disposed to ensure resources are released properly.

Change:

using var process = System.Diagnostics.Process.Start(processInfo); await process.WaitForExitAsync(); if (process.ExitCode != 0) { var error = await process.StandardError.ReadToEndAsync(); throw new InvalidOperationException($"FFmpeg failed: {error}"); }

This ensures the process resources are cleaned up via the using statement's Dispose call.

Suggested change

using (var process = System.Diagnostics.Process.Start(processInfo))

{

await process.WaitForExitAsync();

if (process.ExitCode != 0)

{

var error = process.StandardError.ReadToEnd();

throw new InvalidOperationException($"FFmpeg failed: {error}");

}

}

var process = System.Diagnostics.Process.Start(processInfo);

if (process == null)

throw new InvalidOperationException("Failed to start FFmpeg process.");

try

{

await process.WaitForExitAsync();

if (process.ExitCode != 0)

{

var error = await process.StandardError.ReadToEndAsync();

throw new InvalidOperationException($"FFmpeg failed: {error}");

}

}

finally

{

process.Dispose();

}

Copilot · 2025-11-27T02:27:42Z

+            using (var process = System.Diagnostics.Process.Start(processInfo))
+            {
+                await process.WaitForExitAsync();
+                if (process.ExitCode != 0)
+                {
+                    var error = process.StandardError.ReadToEnd();
+                    throw new InvalidOperationException($"FFmpeg failed: {error}");
+                }


The AudioDecoder.DecodeUsingFFmpeg method executes an external FFmpeg process without proper input validation or sanitization. While the current implementation uses GUIDs for temp files, there's a potential command injection risk if the file paths are manipulated.

Additionally:

There's no check if FFmpeg is installed, which will cause runtime failures

The error message from FFmpeg is read using ReadToEnd() after the process has exited, but StandardError should be read asynchronously to avoid deadlocks

Temp files may not be cleaned up if an exception occurs before the finally block

Consider:

if (process.ExitCode != 0) { var error = await process.StandardError.ReadToEndAsync(); throw new InvalidOperationException($"FFmpeg failed: {error}"); }

Also consider checking for FFmpeg availability at startup or providing a clear error message.

Copilot · 2025-11-27T02:27:42Z

+    /// Decodifica AAC/M4A para PCM 16-bit
+    /// </summary>


Missing documentation for the public method DecodeAACToPCM. Public APIs should have XML documentation comments explaining:

What the method does

The expected format/encoding of the input aacData

What PCM format is returned (sample rate, bit depth, channels)

What exceptions can be thrown (e.g., if FFmpeg is not installed)

Example:

/// <summary> /// Decodes AAC/M4A audio data to PCM 16-bit format. /// </summary> /// <param name="aacData">The AAC/M4A encoded audio data.</param> /// <returns>PCM audio data in 16-bit, 16kHz, mono format.</returns> /// <exception cref="InvalidOperationException">Thrown when FFmpeg fails to decode the audio or is not installed.</exception> public static async Task<byte[]> DecodeAACToPCM(byte[] aacData)

Suggested change

/// Decodifica AAC/M4A para PCM 16-bit

/// </summary>

/// Decodes AAC or M4A encoded audio data to raw PCM 16-bit audio.

/// </summary>

/// <param name="aacData">The AAC or M4A encoded audio data as a byte array.</param>

/// <returns>

/// A byte array containing the decoded PCM audio data in 16-bit, 16kHz, mono format (little-endian).

/// </returns>

/// <exception cref="InvalidOperationException">

/// Thrown when FFmpeg fails to decode the audio or is not installed on the system.

/// </exception>

Copilot · 2025-11-27T02:27:46Z

+            if (File.Exists(tempAacFile)) File.Delete(tempAacFile);
+            if (File.Exists(tempPcmFile)) File.Delete(tempPcmFile);


Temporary files created for FFmpeg processing may not be cleaned up if an exception occurs during file writing or before the FFmpeg process starts. The finally block only runs if execution reaches the try block successfully.

Consider wrapping file operations in try-catch or using a more robust cleanup approach:

var tempAacFile = Path.Combine(Path.GetTempPath(), $"temp_{Guid.NewGuid()}.m4a"); var tempPcmFile = Path.Combine(Path.GetTempPath(), $"temp_{Guid.NewGuid()}.pcm"); try { await File.WriteAllBytesAsync(tempAacFile, aacData); // ... rest of the code } finally { try { if (File.Exists(tempAacFile)) File.Delete(tempAacFile); } catch { } try { if (File.Exists(tempPcmFile)) File.Delete(tempPcmFile); } catch { } }

Suggested change

if (File.Exists(tempAacFile)) File.Delete(tempAacFile);

if (File.Exists(tempPcmFile)) File.Delete(tempPcmFile);

try { if (File.Exists(tempAacFile)) File.Delete(tempAacFile); } catch { }

try { if (File.Exists(tempPcmFile)) File.Delete(tempPcmFile); } catch { }

Copilot · 2025-11-27T02:27:46Z

 builder.Services.AddSignalR(options => { options.DisableImplicitFromServicesParameters = true; });

+// Registrar DbContext no DI container
+builder.Services.AddDbContext<AppDbContext>();


DbContext is registered in DI but most of the codebase still creates instances directly using new AppDbContext(). This leads to:

Inconsistent usage patterns

Potential connection leaks

The DI registration having no effect except for the migration code

Either:

Inject AppDbContext wherever it's needed (recommended for proper DI and lifecycle management), OR

Remove the DI registration if you're not planning to use it consistently

Examples of direct instantiation can be found in:

HolidayService.cs line 8

TaskListService.cs line 13

DatabaseFeeder.cs lines 16, 95, 107

ConversationHistoryOllamaTool.cs line 20

Suggested change

builder.Services.AddDbContext<AppDbContext>();

// builder.Services.AddDbContext<AppDbContext>();

Copilot · 2025-11-27T02:27:46Z

+        var edgeTts = new EdgeTTSNet();
+
+        var voices = await edgeTts.GetVoices();
+        var cnVoice = voices.FirstOrDefault(v => v.ShortName == "pt-BR-FranciscaNeural");
+        var options = new TTSOption
+        (
+            voice: cnVoice!.Name,
+            pitch: "+0Hz",
+            rate: "+25%",
+            volume: "+0%"
+        );
+
+        Log.Information("Assistente: {0}", melissaReply);
+
+
+        edgeTts = new EdgeTTSNet(options);


EdgeTTSNet is instantiated twice - once to get voices (line 84) and again with options (line 99). The first instance is never disposed, potentially leaking resources.

Refactor to use a single instance:

var voices = await EdgeTTSNet.GetVoicesAsync(); // If static method exists // OR using var edgeTtsForVoices = new EdgeTTSNet(); var voices = await edgeTtsForVoices.GetVoices(); var cnVoice = voices.FirstOrDefault(v => v.ShortName == "pt-BR-FranciscaNeural"); // ... null check ... var options = new TTSOption( voice: cnVoice.Name, pitch: "+0Hz", rate: "+25%", volume: "+0%" ); using var edgeTts = new EdgeTTSNet(options);

This issue appears in both audio methods.

Copilot · 2025-11-27T02:27:46Z

+        var wavStream = new MemoryStream(wavBytes);
+        wavStream.Seek(0, SeekOrigin.Begin);
+
+        Log.Information("Iniciando transcrição de áudio...");
+
+        var msgBuilder = new StringBuilder();
+        await foreach (var result in processor.ProcessAsync(wavStream, cancellationToken))
+        {
+            msgBuilder.Append(result.Text);


Disposable 'MemoryStream' is created but not disposed.

Suggested change

var wavStream = new MemoryStream(wavBytes);

wavStream.Seek(0, SeekOrigin.Begin);

Log.Information("Iniciando transcrição de áudio...");

var msgBuilder = new StringBuilder();

await foreach (var result in processor.ProcessAsync(wavStream, cancellationToken))

{

msgBuilder.Append(result.Text);

using (var wavStream = new MemoryStream(wavBytes))

{

wavStream.Seek(0, SeekOrigin.Begin);

Log.Information("Iniciando transcrição de áudio...");

var msgBuilder = new StringBuilder();

await foreach (var result in processor.ProcessAsync(wavStream, cancellationToken))

{

msgBuilder.Append(result.Text);

}

Copilot · 2025-11-27T02:27:47Z

+        if (isAvailable)
+            melissaReply =
+                await MelissaHub.SafeAskMelissaWithErrorHandlingAndRetry(melissa, question, cancellationToken);
+        else
+            melissaReply = statusMessage;


Both branches of this 'if' statement write to the same variable - consider using '?' to express intent better.

Suggested change

if (isAvailable)

melissaReply =

await MelissaHub.SafeAskMelissaWithErrorHandlingAndRetry(melissa, question, cancellationToken);

else

melissaReply = statusMessage;

melissaReply = isAvailable

? await MelissaHub.SafeAskMelissaWithErrorHandlingAndRetry(melissa, question, cancellationToken)

: statusMessage;

…assistente Melissa

Victorrj22 and others added 6 commits November 24, 2025 19:25

Ajustes gerais visando melhorar o build do Docker

d5598f3

Adiciona suporte a GPU no Docker Compose

c9131a4

Adiciona dependências nativas para o runtime do Whisper e ajusta perm…

6b82784

…issões de bibliotecas

Adiciona configuração de fuso horário para o Docker Compose

300806a

Adiciona decodificador de áudio AAC para PCM e integra ao fluxo de tr…

d196d20

…anscrição

Remove temporary audio file saving for debugging in transcription pro…

73fb0c3

…cess

daniloflorenzano requested a review from Copilot November 27, 2025 02:19

daniloflorenzano self-assigned this Nov 27, 2025

Copilot started reviewing on behalf of daniloflorenzano November 27, 2025 02:20 View session

Copilot finished reviewing on behalf of daniloflorenzano November 27, 2025 02:24

Copilot AI reviewed Nov 27, 2025

View reviewed changes

Adiciona configuração de Kestrel e suporte a OllamaUrl na criação da …

84ff687

…assistente Melissa

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fix/corrige build docker#16

Fix/corrige build docker#16
daniloflorenzano wants to merge 7 commits into
mainfrom
fix/corrige-build-docker

daniloflorenzano commented Nov 27, 2025

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Nov 27, 2025

Uh oh!

Copilot AI Nov 27, 2025

Uh oh!

Copilot AI Nov 27, 2025

Uh oh!

Copilot AI Nov 27, 2025

Uh oh!

Copilot AI Nov 27, 2025

Uh oh!

Copilot AI Nov 27, 2025

Uh oh!

Copilot AI Nov 27, 2025

Uh oh!

Copilot AI Nov 27, 2025

Uh oh!

Copilot AI Nov 27, 2025

Uh oh!

Copilot AI Nov 27, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

		RUN mkdir -p /app/Data
		COPY ./Melissa.WebServer/Data/holidays_2025.csv /app/Data/holidays_2025.csv

-    /// Decodifica AAC/M4A para PCM 16-bit
-    /// </summary>
+    /// Decodes AAC or M4A encoded audio data to raw PCM 16-bit audio.
+    /// </summary>
+    /// <param name="aacData">The AAC or M4A encoded audio data as a byte array.</param>
+    /// <returns>
+    /// A byte array containing the decoded PCM audio data in 16-bit, 16kHz, mono format (little-endian).
+    /// </returns>
+    /// <exception cref="InvalidOperationException">
+    /// Thrown when FFmpeg fails to decode the audio or is not installed on the system.
+    /// </exception>

		if (File.Exists(tempAacFile)) File.Delete(tempAacFile);
		if (File.Exists(tempPcmFile)) File.Delete(tempPcmFile);

	builder.Services.AddDbContext<AppDbContext>();
	// builder.Services.AddDbContext<AppDbContext>();

Conversation

daniloflorenzano commented Nov 27, 2025

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Nov 27, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 27, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 27, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 27, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 27, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 27, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 27, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 27, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 27, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 27, 2025

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants