Refactor llm providers to use internal http client (#227)

* refactor: Move http client into provider

* refactor: Rework ollama provider for work with internal http client

* refactor: Rework LM Studio provider to work with internal http client

* refactor: Rework Mistral AI to work with internal http client

* fix: Replace url and header to QNetworkRequest

* refactor: Rework Google provider to use internal http client

* refactor: OpenAI compatible providers switch to use internal http client

* fix: Remove m_requestHandler from tests

* refactor: Remove old handleData method

* fix: Remove LLMClientInterfaceTest
This commit is contained in:
Petr Mironychev
2025-09-03 10:56:05 +02:00
committed by GitHub
parent 5969d530bd
commit 76309be0a6
34 changed files with 1144 additions and 909 deletions

View File

@ -1,6 +1,6 @@
add_library(LLMCore STATIC
RequestType.hpp
Provider.hpp
Provider.hpp Provider.cpp
ProvidersManager.hpp ProvidersManager.cpp
ContextData.hpp
IPromptProvider.hpp
@ -10,8 +10,6 @@ add_library(LLMCore STATIC
PromptTemplate.hpp
PromptTemplateManager.hpp PromptTemplateManager.cpp
RequestConfig.hpp
RequestHandlerBase.hpp RequestHandlerBase.cpp
RequestHandler.hpp RequestHandler.cpp
OllamaMessage.hpp OllamaMessage.cpp
OpenAIMessage.hpp OpenAIMessage.cpp
ValidationUtils.hpp ValidationUtils.cpp

View File

@ -46,24 +46,11 @@ HttpClient::~HttpClient()
void HttpClient::onSendRequest(const HttpRequest &request)
{
QNetworkRequest networkRequest(request.url);
networkRequest.setTransferTimeout(300000);
networkRequest.setHeader(QNetworkRequest::ContentTypeHeader, "application/json");
networkRequest.setRawHeader("Accept", "text/event-stream");
networkRequest.setRawHeader("Cache-Control", "no-cache");
networkRequest.setRawHeader("Connection", "keep-alive");
if (request.headers.has_value()) {
for (const auto &[headername, value] : request.headers->asKeyValueRange()) {
networkRequest.setRawHeader(headername.toUtf8(), value.toUtf8());
}
}
QJsonDocument doc(request.payload);
LOG_MESSAGE(QString("HttpClient: Sending POST to %1").arg(request.url.toString()));
LOG_MESSAGE(QString("HttpClient: data: %1").arg(doc.toJson(QJsonDocument::Indented)));
QNetworkReply *reply = m_manager->post(networkRequest, doc.toJson(QJsonDocument::Compact));
QNetworkReply *reply
= m_manager->post(request.networkRequest, doc.toJson(QJsonDocument::Compact));
addActiveRequest(reply, request.requestId);
connect(reply, &QNetworkReply::readyRead, this, &HttpClient::onReadyRead);

View File

@ -32,10 +32,9 @@ namespace QodeAssist::LLMCore {
struct HttpRequest
{
QUrl url;
QNetworkRequest networkRequest;
QString requestId;
QJsonObject payload;
std::optional<QMap<QString, QString>> headers;
};
class HttpClient : public QObject

18
llmcore/Provider.cpp Normal file
View File

@ -0,0 +1,18 @@
#include "Provider.hpp"
namespace QodeAssist::LLMCore {
Provider::Provider(QObject *parent)
: QObject(parent)
, m_httpClient(std::make_unique<HttpClient>())
{
connect(m_httpClient.get(), &HttpClient::dataReceived, this, &Provider::onDataReceived);
connect(m_httpClient.get(), &HttpClient::requestFinished, this, &Provider::onRequestFinished);
}
HttpClient *Provider::httpClient() const
{
return m_httpClient.get();
}
} // namespace QodeAssist::LLMCore

View File

@ -21,9 +21,11 @@
#include <utils/environment.h>
#include <QNetworkRequest>
#include <QObject>
#include <QString>
#include "ContextData.hpp"
#include "HttpClient.hpp"
#include "PromptTemplate.hpp"
#include "RequestType.hpp"
@ -32,9 +34,12 @@ class QJsonObject;
namespace QodeAssist::LLMCore {
class Provider
class Provider : public QObject
{
Q_OBJECT
public:
explicit Provider(QObject *parent = nullptr);
virtual ~Provider() = default;
virtual QString name() const = 0;
@ -48,12 +53,28 @@ public:
LLMCore::ContextData context,
LLMCore::RequestType type)
= 0;
virtual bool handleResponse(QNetworkReply *reply, QString &accumulatedResponse) = 0;
virtual QList<QString> getInstalledModels(const QString &url) = 0;
virtual QList<QString> validateRequest(const QJsonObject &request, TemplateType type) = 0;
virtual QString apiKey() const = 0;
virtual void prepareNetworkRequest(QNetworkRequest &networkRequest) const = 0;
virtual ProviderID providerID() const = 0;
virtual void sendRequest(const QString &requestId, const QUrl &url, const QJsonObject &payload)
= 0;
HttpClient *httpClient() const;
public slots:
virtual void onDataReceived(const QString &requestId, const QByteArray &data) = 0;
virtual void onRequestFinished(const QString &requestId, bool success, const QString &error) = 0;
signals:
void partialResponseReceived(const QString &requestId, const QString &partialText);
void fullResponseReceived(const QString &requestId, const QString &fullText);
void requestFailed(const QString &requestId, const QString &error);
private:
std::unique_ptr<HttpClient> m_httpClient;
};
} // namespace QodeAssist::LLMCore

View File

@ -1,202 +0,0 @@
/*
* Copyright (C) 2024-2025 Petr Mironychev
*
* This file is part of QodeAssist.
*
* QodeAssist is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* QodeAssist is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with QodeAssist. If not, see <https://www.gnu.org/licenses/>.
*/
#include "RequestHandler.hpp"
#include "Logger.hpp"
#include <QJsonDocument>
#include <QNetworkReply>
#include <QThread>
namespace QodeAssist::LLMCore {
RequestHandler::RequestHandler(QObject *parent)
: RequestHandlerBase(parent)
, m_manager(new QNetworkAccessManager(this))
{
connect(
this,
&RequestHandler::doSendRequest,
this,
&RequestHandler::sendLLMRequestInternal,
Qt::QueuedConnection);
connect(
this,
&RequestHandler::doCancelRequest,
this,
&RequestHandler::cancelRequestInternal,
Qt::QueuedConnection);
}
RequestHandler::~RequestHandler()
{
for (auto reply : m_activeRequests) {
reply->abort();
reply->deleteLater();
}
m_activeRequests.clear();
m_accumulatedResponses.clear();
}
void RequestHandler::sendLLMRequest(const LLMConfig &config, const QJsonObject &request)
{
emit doSendRequest(config, request);
}
bool RequestHandler::cancelRequest(const QString &id)
{
emit doCancelRequest(id);
return true;
}
void RequestHandler::sendLLMRequestInternal(const LLMConfig &config, const QJsonObject &request)
{
LOG_MESSAGE(QString("Sending request to llm: \nurl: %1\nRequest body:\n%2")
.arg(
config.url.toString(),
QString::fromUtf8(
QJsonDocument(config.providerRequest).toJson(QJsonDocument::Indented))));
QNetworkRequest networkRequest(config.url);
networkRequest.setTransferTimeout(300000);
config.provider->prepareNetworkRequest(networkRequest);
QNetworkReply *reply
= m_manager->post(networkRequest, QJsonDocument(config.providerRequest).toJson());
if (!reply) {
LOG_MESSAGE("Error: Failed to create network reply");
return;
}
QString requestId = request["id"].toString();
m_activeRequests[requestId] = reply;
connect(reply, &QNetworkReply::readyRead, this, [this, reply, request, config]() {
handleLLMResponse(reply, request, config);
});
connect(
reply,
&QNetworkReply::finished,
this,
[this, reply, requestId]() {
m_activeRequests.remove(requestId);
if (reply->error() != QNetworkReply::NoError) {
QString errorMessage = reply->errorString();
int statusCode = reply->attribute(QNetworkRequest::HttpStatusCodeAttribute).toInt();
LOG_MESSAGE(
QString("Error details: %1\nStatus code: %2").arg(errorMessage).arg(statusCode));
emit requestFinished(requestId, false, errorMessage);
} else {
LOG_MESSAGE("Request finished successfully");
emit requestFinished(requestId, true, QString());
}
reply->deleteLater();
},
Qt::QueuedConnection);
}
void RequestHandler::handleLLMResponse(
QNetworkReply *reply, const QJsonObject &request, const LLMConfig &config)
{
QString &accumulatedResponse = m_accumulatedResponses[reply];
bool isComplete = config.provider->handleResponse(reply, accumulatedResponse);
if (config.requestType == RequestType::CodeCompletion) {
if (!config.multiLineCompletion
&& processSingleLineCompletion(reply, request, accumulatedResponse, config)) {
return;
}
if (isComplete) {
auto cleanedCompletion
= removeStopWords(accumulatedResponse, config.promptTemplate->stopWords());
emit completionReceived(cleanedCompletion, request, true);
}
} else if (config.requestType == RequestType::Chat) {
emit completionReceived(accumulatedResponse, request, isComplete);
}
if (isComplete)
m_accumulatedResponses.remove(reply);
}
void RequestHandler::cancelRequestInternal(const QString &id)
{
QMutexLocker locker(&m_mutex);
if (m_activeRequests.contains(id)) {
QNetworkReply *reply = m_activeRequests[id];
disconnect(reply, nullptr, this, nullptr);
reply->abort();
m_activeRequests.remove(id);
m_accumulatedResponses.remove(reply);
reply->deleteLater();
locker.unlock();
m_manager->clearConnectionCache();
m_manager->clearAccessCache();
emit requestCancelled(id);
}
}
bool RequestHandler::processSingleLineCompletion(
QNetworkReply *reply,
const QJsonObject &request,
const QString &accumulatedResponse,
const LLMConfig &config)
{
QString cleanedResponse = accumulatedResponse;
int newlinePos = cleanedResponse.indexOf('\n');
if (newlinePos != -1) {
QString singleLineCompletion = cleanedResponse.left(newlinePos).trimmed();
singleLineCompletion
= removeStopWords(singleLineCompletion, config.promptTemplate->stopWords());
emit completionReceived(singleLineCompletion, request, true);
m_accumulatedResponses.remove(reply);
reply->abort();
return true;
}
return false;
}
QString RequestHandler::removeStopWords(const QStringView &completion, const QStringList &stopWords)
{
QString filteredCompletion = completion.toString();
for (const QString &stopWord : stopWords) {
filteredCompletion = filteredCompletion.replace(stopWord, "");
}
return filteredCompletion;
}
} // namespace QodeAssist::LLMCore

View File

@ -1,71 +0,0 @@
/*
* Copyright (C) 2024-2025 Petr Mironychev
*
* This file is part of QodeAssist.
*
* QodeAssist is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* QodeAssist is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with QodeAssist. If not, see <https://www.gnu.org/licenses/>.
*/
#pragma once
#include <QJsonObject>
#include <QMutex>
#include <QNetworkAccessManager>
#include <QObject>
#include "RequestConfig.hpp"
#include "RequestHandlerBase.hpp"
class QNetworkReply;
namespace QodeAssist::LLMCore {
class RequestHandler : public RequestHandlerBase
{
Q_OBJECT
public:
explicit RequestHandler(QObject *parent = nullptr);
~RequestHandler() override;
void sendLLMRequest(const LLMConfig &config, const QJsonObject &request) override;
bool cancelRequest(const QString &id) override;
signals:
void doSendRequest(QodeAssist::LLMCore::LLMConfig config, QJsonObject request);
void doCancelRequest(QString id);
private slots:
void sendLLMRequestInternal(
const QodeAssist::LLMCore::LLMConfig &config, const QJsonObject &request);
void cancelRequestInternal(const QString &id);
void handleLLMResponse(
QNetworkReply *reply,
const QJsonObject &request,
const QodeAssist::LLMCore::LLMConfig &config);
private:
QMap<QString, QNetworkReply *> m_activeRequests;
QMap<QNetworkReply *, QString> m_accumulatedResponses;
QNetworkAccessManager *m_manager;
QMutex m_mutex;
bool processSingleLineCompletion(
QNetworkReply *reply,
const QJsonObject &request,
const QString &accumulatedResponse,
const LLMConfig &config);
QString removeStopWords(const QStringView &completion, const QStringList &stopWords);
};
} // namespace QodeAssist::LLMCore