// Copyright (C) 2024-2026 Petr Mironychev
// SPDX-License-Identifier: GPL-3.0-or-later
// Additional attribution terms under GPLv3 §7(b) apply — see LICENSE

#include "InputTokenCounter.hpp"

#include <algorithm>

#include <LLMQore/ToolsManager.hpp>
#include <QDateTime>
#include <QFileInfo>
#include <QJsonArray>
#include <QJsonDocument>

#include <utils/aspects.h>

#include "ChatAssistantSettings.hpp"
#include "GeneralSettings.hpp"
#include "Logger.hpp"
#include "providers/ProvidersManager.hpp"
#include "context/ContextManager.hpp"
#include "context/TokenUtils.hpp"
#include "session/Session.hpp"

namespace QodeAssist::Chat {

InputTokenCounter::InputTokenCounter(
    Session::Session *session, Context::ContextManager *contextManager, QObject *parent)
    : QObject(parent)
    , m_session(session)
    , m_contextManager(contextManager)
{
    auto &settings = Settings::chatAssistantSettings();
    connect(
        &settings.useSystemPrompt,
        &Utils::BaseAspect::changed,
        this,
        &InputTokenCounter::recompute);
    connect(
        &settings.systemPrompt, &Utils::BaseAspect::changed, this, &InputTokenCounter::recompute);
    connect(
        &settings.enableChatTools,
        &Utils::BaseAspect::changed,
        this,
        &InputTokenCounter::recompute);

    connect(&Settings::generalSettings().caProvider, &Utils::BaseAspect::changed, this, [this]() {
        rewireToolsChangedConnection();
        recompute();
    });

    m_recomputeTimer.setSingleShot(true);
    m_recomputeTimer.setInterval(150);
    connect(&m_recomputeTimer, &QTimer::timeout, this, &InputTokenCounter::recompute);

    rewireToolsChangedConnection();
    recompute();
}

void InputTokenCounter::recomputeSoon()
{
    m_recomputeTimer.start();
}

int InputTokenCounter::inputTokens() const
{
    return m_inputTokens;
}

void InputTokenCounter::setMessage(const QString &message)
{
    m_messageTokens = Context::TokenUtils::estimateTokens(message);
    recomputeSoon();
}

void InputTokenCounter::setAttachments(const QStringList &attachments)
{
    m_attachments = attachments;
    recompute();
}

void InputTokenCounter::rewireToolsChangedConnection()
{
    if (m_toolsChangedConn)
        QObject::disconnect(m_toolsChangedConn);
    m_toolsChangedConn = {};

    const auto providerName = Settings::generalSettings().caProvider();
    auto *provider = Providers::ProvidersManager::instance().getProviderByName(providerName);
    if (!provider)
        return;
    auto *tm = provider->toolsManager();
    if (!tm)
        return;

    m_toolsChangedConn = connect(
        tm, &::LLMQore::ToolRegistry::toolsChanged, this, &InputTokenCounter::recompute);
}

int InputTokenCounter::estimateFileTokens(const QStringList &paths)
{
    if (paths.isEmpty())
        return 0;

    int total = 0;
    QStringList uncached;

    for (const QString &path : paths) {
        const QDateTime modified = QFileInfo(path).lastModified();
        const auto cached = m_fileTokens.constFind(path);
        if (cached != m_fileTokens.constEnd() && cached->modified == modified) {
            total += cached->tokens;
            continue;
        }
        uncached.append(path);
    }

    if (m_fileTokens.size() > 256)
        m_fileTokens.clear();

    for (const QString &path : std::as_const(uncached)) {
        const int tokens = Context::TokenUtils::estimateFilesTokens(
            m_contextManager->getContentFiles({path}));
        total += tokens;
        m_fileTokens.insert(path, {QFileInfo(path).lastModified(), tokens});
    }

    return total;
}

void InputTokenCounter::recompute()
{
    int inputTokens = m_messageTokens;
    auto &settings = Settings::chatAssistantSettings();

    if (settings.useSystemPrompt()) {
        inputTokens += Context::TokenUtils::estimateTokens(settings.systemPrompt());
    }

    const auto splitImageEstimate = [](const QStringList &paths, QStringList &textPaths) {
        int imageTokens = 0;
        for (const QString &p : paths) {
            if (Context::TokenUtils::isImageFilePath(p))
                imageTokens += Context::TokenUtils::estimateImageAttachmentTokens(p);
            else
                textPaths.append(p);
        }
        return imageTokens;
    };

    if (!m_attachments.isEmpty()) {
        QStringList textPaths;
        inputTokens += splitImageEstimate(m_attachments, textPaths);
        inputTokens += estimateFileTokens(textPaths);
    }

    if (m_session) {
        for (const Session::MessageRow &row : m_session->rows()) {
            if (Session::rowTreatmentFor(Session::RowAudience::TokenCount, row.kind)
                == Session::RowTreatment::Omit)
                continue;
            inputTokens += Context::TokenUtils::estimateTokens(row.content);
            inputTokens += 4; // + role
        }
    }

    if (settings.enableChatTools()) {
        const auto providerName = Settings::generalSettings().caProvider();
        if (auto *provider = Providers::ProvidersManager::instance().getProviderByName(
                providerName)) {
            if (auto *tm = provider->toolsManager()) {
                const QJsonArray toolDefs = tm->getToolsDefinitions();
                if (!toolDefs.isEmpty()) {
                    const QByteArray serialized
                        = QJsonDocument(toolDefs).toJson(QJsonDocument::Compact);
                    inputTokens += static_cast<int>(serialized.size() / 4);
                }
            }
        }
    }

    m_inputTokens = static_cast<int>(inputTokens * m_calibrationFactor);
    emit inputTokensChanged();
}

void InputTokenCounter::recordSent()
{
    if (m_recomputeTimer.isActive()) {
        m_recomputeTimer.stop();
        recompute();
    }

    m_lastSentEstimate = m_calibrationFactor > 0.0
                             ? static_cast<int>(m_inputTokens / m_calibrationFactor)
                             : m_inputTokens;
}

void InputTokenCounter::recordServerUsage(int promptTokens)
{
    if (promptTokens <= 0 || m_lastSentEstimate <= 0)
        return;

    const double rawFactor
        = static_cast<double>(promptTokens) / static_cast<double>(m_lastSentEstimate);
    const double clamped = std::clamp(rawFactor, 0.5, 3.0);
    m_calibrationFactor = 0.5 * m_calibrationFactor + 0.5 * clamped;

    LOG_MESSAGE(QString("Token calibration: server=%1 estimated=%2 ratio=%3 ema=%4")
                    .arg(promptTokens)
                    .arg(m_lastSentEstimate)
                    .arg(rawFactor, 0, 'f', 3)
                    .arg(m_calibrationFactor, 0, 'f', 3));

    recompute();
}

} // namespace QodeAssist::Chat