mirror of
https://github.com/Palm1r/QodeAssist.git
synced 2026-06-14 02:09:22 -04:00
doc: update architecture
This commit is contained in:
Petr Mironychev
@@ -1,35 +1,42 @@
|
|||||||
# QodeAssist Architecture
|
# QodeAssist Architecture
|
||||||
|
|
||||||
This document describes the runtime architecture of QodeAssist after the
|
This document describes the **current** runtime architecture, after the §10
|
||||||
migration of all LLM runtime paths onto the agent / `Session` stack
|
rework in `target-architecture.md` was completed. Every runtime LLM path —
|
||||||
("Stack B"). Every runtime LLM path — code completion, chat (send/stream +
|
code completion, chat (send/stream + compression + token counting), and quick
|
||||||
compression + token counting), and quick refactor — now goes through agents,
|
refactor — flows through one stack: agents, `Session`, and the
|
||||||
`Session`, and the `Providers::GenericProvider` layer.
|
`Providers::GenericProvider` layer. There is no legacy parallel path; the old
|
||||||
|
"Stack A" (root `providers/*`, `pluginllmcore/*`, `ConfigurationManager`, the
|
||||||
|
provider/model/template settings pages) has been removed.
|
||||||
|
|
||||||
> Legend: ✅ = on Stack B (active runtime), 🔴 = legacy Stack A (isolated, no
|
For the design rationale, layering contract, and cross-cutting policies, see
|
||||||
> runtime consumers left).
|
[`target-architecture.md`](target-architecture.md). This file documents how the
|
||||||
|
code is wired today.
|
||||||
|
|
||||||
---
|
---
|
||||||
|
|
||||||
## 1. Top level: ownership and dependency injection
|
## 1. Top level: ownership and dependency injection
|
||||||
|
|
||||||
The plugin (`qodeassist.cpp`) owns everything via `new` + parent (no plugin-wide
|
The plugin (`qodeassist.cpp`) owns everything via `new` + parent — no
|
||||||
singletons; each feature receives its dependencies explicitly).
|
plugin-wide singletons; each feature receives its dependencies explicitly.
|
||||||
|
|
||||||
```
|
```
|
||||||
QodeAssistPlugin
|
QodeAssistPlugin
|
||||||
Stack B infrastructure:
|
• Providers::registerBuiltinProviders() — client_api → provider table
|
||||||
• Providers::registerBuiltinProviders() — registers 13 client_api types
|
• ProviderInstanceFactory — provider instances from TOML
|
||||||
• ProviderInstanceFactory — 14 instances from TOML
|
• ProviderSecretsStore — secrets behind a port
|
||||||
• ProviderSecretsStore
|
• AgentFactory — agents from TOML + agent_models.json
|
||||||
• AgentFactory — agents from TOML
|
• SessionManager(agentFactory) — owns the ToolContributorRegistry
|
||||||
• SessionManager(agentFactory)
|
toolContributors().add(registerQodeAssistTools)
|
||||||
|
toolContributors().add(registerSkillTool)
|
||||||
|
toolContributors().add(McpClientsManager::registerToolsOn)
|
||||||
• m_engine (QQmlEngine)
|
• m_engine (QQmlEngine)
|
||||||
rootContext: "agentFactory", "sessionManager" — DI for chat (QML)
|
rootContext: "agentFactory", "sessionManager" — DI for chat (QML)
|
||||||
|
|
||||||
Wired into consumers:
|
Wired into consumers:
|
||||||
• QodeAssistClient ← LLMClientInterface(*sessionManager, *agentFactory)
|
• QodeAssistClient ← LLMClientInterface(generalSettings, completeSettings,
|
||||||
← setSessionManager / setAgentFactory (for quick refactor)
|
agentFactory, sessionManager, documentReader,
|
||||||
|
performanceLogger)
|
||||||
|
← setSessionManager / setAgentFactory (quick refactor)
|
||||||
```
|
```
|
||||||
|
|
||||||
Chat lives in QML (`ChatRootView` is a `QML_ELEMENT`), so `AgentFactory` and
|
Chat lives in QML (`ChatRootView` is a `QML_ELEMENT`), so `AgentFactory` and
|
||||||
@@ -39,220 +46,262 @@ context** and resolved in `ChatRootView` via
|
|||||||
|
|
||||||
---
|
---
|
||||||
|
|
||||||
## 2. Stack B core (agent / Session)
|
## 2. Core (agent / Session)
|
||||||
|
|
||||||
```
|
```
|
||||||
AgentFactory.create(name)
|
AgentFactory.create(name)
|
||||||
configByName(name) → AgentConfig (TOML)
|
configByName(name) → AgentConfig (TOML, [body] table; model override from
|
||||||
providerInstance, model, endpoint, role, messageFormat,
|
agent_models.json applied here)
|
||||||
sampling, enableTools, enableThinking, match{filePatterns,...}
|
|
||||||
buildProviderForAgent:
|
buildProviderForAgent:
|
||||||
instance = ProviderInstanceFactory.instanceByName(cfg.providerInstance)
|
instance = ProviderInstanceFactory.instanceByName(cfg.providerInstance)
|
||||||
provider = ProviderFactory::create(instance.clientApi) ◄── keystone
|
provider = ProviderFactory::create(instance.clientApi)
|
||||||
provider.setUrl(instance.url)
|
provider.setUrl(instance.url)
|
||||||
provider.setApiKey(secrets.read(instance.apiKeyRef))
|
provider.setApiKey(secrets.read(instance.apiKeyRef))
|
||||||
▼
|
▼
|
||||||
Agent(config, provider)
|
Agent(config, provider)
|
||||||
promptTemplate = JsonPromptTemplate::fromConfig(cfg.messageFormat) (inja)
|
promptTemplate = JsonPromptTemplate::fromConfig(cfg) — compiles [body] (inja),
|
||||||
|
validated at load against a synthetic context
|
||||||
|
provider.setPromptCaching(cfg.cachePrompt, cfg.cacheTtl == "1h")
|
||||||
▼
|
▼
|
||||||
SessionManager.createSession(agentName) → Session(agent)
|
SessionManager — two ways to obtain a Session:
|
||||||
├─ ConversationHistory — messages as ContentBlocks
|
• createSession(agentName, externalHistory?) — chat: attaches a persistent,
|
||||||
├─ SystemPromptBuilder — layers: agent.role + caller layers
|
externally-owned history
|
||||||
└─ ResponseRouter(client) — emits ResponseEvent
|
• acquire(agentName) / release(session) — one-shot pipelines: a small
|
||||||
|
per-agent pool of internal-history
|
||||||
|
sessions; acquire hands out a
|
||||||
|
session with cleared history,
|
||||||
|
cleared system-prompt layers and
|
||||||
|
cleared client tools
|
||||||
|
▼
|
||||||
|
Session(agent[, externalHistory])
|
||||||
|
├─ ConversationHistory — messages as polymorphic ContentBlocks
|
||||||
|
├─ SystemPromptBuilder — ordered named layers (priority-sorted)
|
||||||
|
└─ ResponseRouter(client) — adapts client signals → typed ResponseEvent
|
||||||
|
|
||||||
Session API:
|
Session API:
|
||||||
• send(blocks, toolsOverride) — chat/refactor: append user msg + dispatch
|
• send(blocks, toolsOverride) — the ONLY dispatch entry point: append a user
|
||||||
• sendCompletion(ContextData) — completion: FIM prefix/suffix
|
message and dispatch. Completion/chat/refactor
|
||||||
• client() — agent's LLMQore::BaseClient (direct streaming)
|
differ only in block content + template.
|
||||||
• systemPrompt()->setLayer(...) — dynamic context layers
|
• cancel() — tears down in-flight; emits cancelled(id)
|
||||||
• supportsImages() — provider Image capability
|
• history() / systemPrompt() / client() / supportsImages()
|
||||||
• history() — for seeding from ChatModel
|
• setContentLoader(loader) — resolves Stored* attachment/image blocks
|
||||||
|
• lastError() → ErrorInfo — typed synchronous start-failure detail
|
||||||
|
|
||||||
|
Session signals (three-state, mutually exclusive per request):
|
||||||
|
• finished(id, stopReason)
|
||||||
|
• failed(id, ErrorInfo{category, message, providerDetail})
|
||||||
|
• cancelled(id)
|
||||||
|
+ event(ResponseEvent) — live delta stream for the chat UI
|
||||||
```
|
```
|
||||||
|
|
||||||
`Session::sendCompletion` and `dispatch` compose `SystemPromptBuilder` layers
|
`Session::dispatch` renders the agent's `system_prompt` into the `agent.system`
|
||||||
(`agent.role` + caller-provided) into the request system prompt.
|
layer, composes all `SystemPromptBuilder` layers into the request system prompt,
|
||||||
|
and substitutes `${MODEL}` in the endpoint before sending.
|
||||||
|
|
||||||
---
|
---
|
||||||
|
|
||||||
## 3. Provider layer — the keystone (implemented during migration)
|
## 3. Provider layer
|
||||||
|
|
||||||
The Stack B provider layer previously existed only as an abstract base +
|
One configuration-driven `GenericProvider` covers every API; it varies only by
|
||||||
empty factory (`registerType` was never called, no concrete providers). This
|
the LLMQore client factory and metadata. Request *shape* belongs to the agent's
|
||||||
blocked every agent from obtaining a working provider. It is now implemented
|
`JsonPromptTemplate` (the `[body]` table), never to the provider.
|
||||||
via a single configuration-driven `GenericProvider`.
|
|
||||||
|
|
||||||
```
|
```
|
||||||
ProviderFactory (sources/providers, namespace functions)
|
ProviderFactory (sources/providers, namespace functions)
|
||||||
registerType(name, fn) / create(name, parent) / knownNames()
|
registerType(name, fn) / create(name, parent) / knownNames()
|
||||||
▲
|
▲ registerBuiltinProviders() — client_api → provider table
|
||||||
│ registerBuiltinProviders() — client_api → provider table
|
|
||||||
│
|
|
||||||
GenericProvider : Providers::Provider
|
GenericProvider : Providers::Provider
|
||||||
• owns an LLMQore::BaseClient (created by a ClientFactory)
|
• owns an LLMQore::BaseClient (created by a ClientFactory)
|
||||||
• prepareRequest — inherited from Provider base:
|
• prepareRequest → PromptTemplate::buildFullRequest; injects tools when
|
||||||
delegates to PromptTemplate::buildFullRequest
|
enable_tools; applies ClaudeCacheControl when prompt caching is on
|
||||||
• client() / providerID() / capabilities() / getInstalledModels()
|
• client() / providerID() / capabilities() / getInstalledModels()
|
||||||
```
|
```
|
||||||
|
|
||||||
### client_api → provider table
|
### client_api → provider table
|
||||||
|
|
||||||
| client_api | LLMQore client | ProviderID | capabilities |
|
| client_api | LLMQore client | ProviderID | capabilities |
|
||||||
|--------------------------------|-------------------------|------------------|-------------------------|
|
|------------------------------|-----------------------|------------------|-----------------------------------|
|
||||||
| Claude | ClaudeClient | Claude | Tools·Thinking·Image·ModelListing |
|
| Claude | ClaudeClient | Claude | Tools·Thinking·Image·ModelListing |
|
||||||
| Google AI | GoogleAIClient | GoogleAI | Tools·Thinking·Image·ModelListing |
|
| Google AI | GoogleAIClient | GoogleAI | Tools·Thinking·Image·ModelListing |
|
||||||
| llama.cpp | LlamaCppClient | LlamaCpp | Tools·Thinking·Image·ModelListing |
|
| llama.cpp | LlamaCppClient | LlamaCpp | Tools·Thinking·Image·ModelListing |
|
||||||
| Mistral AI | MistralClient | MistralAI | Tools·Thinking·Image·ModelListing |
|
| Mistral AI | MistralClient | MistralAI | Tools·Thinking·Image·ModelListing |
|
||||||
| Codestral | MistralClient | MistralAI | Tools·Image |
|
| Codestral | MistralClient | MistralAI | Tools·Image |
|
||||||
| Ollama (Native) | OllamaClient | Ollama | Tools·Thinking·Image·ModelListing |
|
| Ollama (Native) | OllamaClient | Ollama | Tools·Thinking·Image·ModelListing |
|
||||||
| Ollama (OpenAI-compatible) | OpenAIClient | OpenAICompatible | Tools·Thinking·Image·ModelListing |
|
| Ollama (OpenAI-compatible) | OpenAIClient | OpenAICompatible | Tools·Thinking·Image·ModelListing |
|
||||||
| OpenAI (Chat Completions) | OpenAIClient | OpenAI | Tools·Thinking·Image·ModelListing |
|
| OpenAI (Chat Completions) | OpenAIClient | OpenAI | Tools·Thinking·Image·ModelListing |
|
||||||
| OpenAI (Responses API) | OpenAIResponsesClient | OpenAIResponses | Tools·Thinking·Image·ModelListing |
|
| OpenAI (Responses API) | OpenAIResponsesClient | OpenAIResponses | Tools·Thinking·Image·ModelListing |
|
||||||
| OpenAI Compatible | OpenAIClient | OpenAICompatible | Tools·Image·Thinking |
|
| OpenAI Compatible | OpenAIClient | OpenAICompatible | Tools·Image·Thinking |
|
||||||
| OpenRouter | OpenAIClient | OpenRouter | Tools·Image·Thinking·ModelListing |
|
| OpenRouter | OpenAIClient | OpenRouter | Tools·Image·Thinking·ModelListing |
|
||||||
| LM Studio (Chat Completions) | OpenAIClient | LMStudio | Tools·Thinking·Image·ModelListing |
|
| LM Studio (Chat Completions) | OpenAIClient | LMStudio | Tools·Thinking·Image·ModelListing |
|
||||||
| LM Studio (Responses API) | OpenAIResponsesClient | OpenAIResponses | Tools·Thinking·Image·ModelListing |
|
| LM Studio (Responses API) | OpenAIResponsesClient | OpenAIResponses | Tools·Thinking·Image·ModelListing |
|
||||||
|
|
||||||
Request *shape* comes from the agent's prompt template (jinja `messageFormat`),
|
|
||||||
so a single provider class covers every API by varying only the client factory
|
|
||||||
and metadata.
|
|
||||||
|
|
||||||
---
|
---
|
||||||
|
|
||||||
## 4. Runtime paths (all on Stack B)
|
## 4. Configuration model
|
||||||
|
|
||||||
### 4a. Code completion ✅
|
|
||||||
|
|
||||||
```
|
|
||||||
Qt Creator LSP (getCompletionsCycling)
|
|
||||||
▼
|
|
||||||
LLMClientInterface
|
|
||||||
pickCompletionAgent: AgentRouter.pickAgent(roster.codeCompletion, {file, project})
|
|
||||||
session = sessionManager.createSession(agent)
|
|
||||||
ctx = Templates::ContextData{ prefix, suffix,
|
|
||||||
systemPrompt = fileContext + openFiles }
|
|
||||||
session.sendCompletion(ctx)
|
|
||||||
▼ stream from session.client():
|
|
||||||
requestCompleted → sendCompletionToClient → CodeHandler → LSP
|
|
||||||
system prompt = agent.role; FIM template renders prefix/suffix
|
|
||||||
```
|
|
||||||
|
|
||||||
### 4b. Chat ✅
|
|
||||||
|
|
||||||
```
|
|
||||||
ChatRootView (QML)
|
|
||||||
resolve agentFactory()/sessionManager() = qmlEngine(this)->rootContext()
|
|
||||||
ChatAgentController: agent list (configNames), active agent (persisted),
|
|
||||||
supportsThinking/Tools
|
|
||||||
QML agent picker (TopBar.agentSelector) — replaced provider/model/template combos
|
|
||||||
▼ dispatchSend
|
|
||||||
ClientInterface
|
|
||||||
session = sessionManager.createSession(currentChatAgent)
|
|
||||||
registerQodeAssistTools(session.client().tools()) + registerSkillTool
|
|
||||||
systemPrompt layer "chat.context" = project info + skills + linked files
|
|
||||||
seedHistory(session.history() ← ChatModel: user/assistant/tool-call+result)
|
|
||||||
session.send(userBlocks{text + images}, useTools)
|
|
||||||
▼ stream from session.client() → existing handlers → ChatModel:
|
|
||||||
chunk→addMessage thinking→addThinkingBlock
|
|
||||||
tool→addToolExecutionStatus / updateToolResult
|
|
||||||
finalized→usage completed→messageReceivedCompletely → removeSession
|
|
||||||
|
|
||||||
ChatCompressor → createSession(agent) → seed history → layer "compression" → send(prompt)
|
|
||||||
InputTokenCounter → estimate without provider (calibrated by server usage)
|
|
||||||
```
|
|
||||||
|
|
||||||
### 4c. Quick refactor ✅
|
|
||||||
|
|
||||||
```
|
|
||||||
QodeAssistClient.requestQuickRefactor → QuickRefactorHandler (setSessionManager/setAgentFactory)
|
|
||||||
pickRefactorAgent: AgentRouter.pickAgent(roster.quickRefactor, {file, project})
|
|
||||||
session = createSession(agent)
|
|
||||||
if useTools: registerQodeAssistTools(session.client().tools())
|
|
||||||
systemPrompt layer "refactor" = buildSystemPrompt(tagged content +
|
|
||||||
output requirements + indentation rules)
|
|
||||||
session.send(blocks{instructions}, useTools)
|
|
||||||
▼ stream from session.client():
|
|
||||||
requestCompleted → ResponseCleaner → RefactorResult → insert into editor
|
|
||||||
```
|
|
||||||
|
|
||||||
---
|
|
||||||
|
|
||||||
## 5. Configuration sources
|
|
||||||
|
|
||||||
```
|
```
|
||||||
~/.config/.../qodeassist/config/
|
~/.config/.../qodeassist/config/
|
||||||
providers/*.toml → ProviderInstance { name, client_api, url, api_key_ref }
|
providers/*.toml → ProviderInstance { name, client_api, url, api_key_ref }
|
||||||
agents/*.toml → AgentConfig { providerInstance, model, endpoint, role,
|
agents/*.toml → AgentConfig { schema_version, providerInstance, model,
|
||||||
messageFormat, sampling, match, enable* }
|
endpoint, system_prompt, [body], match,
|
||||||
|
enable_tools, enable_thinking, cache_prompt,
|
||||||
|
extends, abstract, hidden, tags }
|
||||||
|
agent_models.json → per-agent model override (applied by AgentFactory)
|
||||||
|
agent_roles/*.json → role text, pulled into system_prompt via {{ agent_role(id) }}
|
||||||
pipelines rosters → codeCompletion / chatAssistant / chatCompression / quickRefactor
|
pipelines rosters → codeCompletion / chatAssistant / chatCompression / quickRefactor
|
||||||
consumed by AgentRouter.pickAgent(roster, {filePath, projectName})
|
consumed by AgentRouter.pickAgent(roster, {filePath, projectName})
|
||||||
|
|
||||||
Editor policy (NOT agent config):
|
Editor policy (NOT agent config):
|
||||||
CodeCompletionSettings — triggers, modelOutputHandler, context extraction,
|
CodeCompletionSettings — triggers, modelOutputHandler, context extraction,
|
||||||
useOpenFilesContext
|
useOpenFilesContext
|
||||||
(sampling / prompt-generation fields removed)
|
```
|
||||||
|
|
||||||
|
`[body]` **is** the request body (deep-mergeable through `extends`; Jinja-bearing
|
||||||
|
string values render and splice as raw JSON, literals pass through, empty renders
|
||||||
|
drop the key). `include` resolves only sandboxed partial roots. Profiles validate
|
||||||
|
at load: a referenced partial must resolve and the assembled body must parse as
|
||||||
|
JSON against a synthetic context — config errors surface in the agents settings
|
||||||
|
page, never as a silent runtime drop. Full spec:
|
||||||
|
[`agent-templates-design.md`](agent-templates-design.md).
|
||||||
|
|
||||||
|
---
|
||||||
|
|
||||||
|
## 5. Runtime paths
|
||||||
|
|
||||||
|
`AgentRouter.pickAgent(roster, {file, project})` is the only agent picker; every
|
||||||
|
pipeline resolves its agent through a roster.
|
||||||
|
|
||||||
|
### 5a. Code completion
|
||||||
|
|
||||||
|
```
|
||||||
|
Qt Creator LSP (getCompletionsCycling)
|
||||||
|
▼
|
||||||
|
LLMClientInterface
|
||||||
|
agent = AgentRouter.pickAgent(roster.codeCompletion, {file, project})
|
||||||
|
session = sessionManager.acquire(agent) — pooled
|
||||||
|
systemPrompt layer "completion.context" = fileContext + open-files context
|
||||||
|
session.send( blocks{ CompletionContent(prefix, suffix) }, tools=off )
|
||||||
|
▼ on Session::finished:
|
||||||
|
history().lastAssistantText() → CodeHandler (output-mode) → LSP items
|
||||||
|
→ sessionManager.release(session)
|
||||||
|
```
|
||||||
|
|
||||||
|
The completion context travels as a `CompletionContent` block; the template
|
||||||
|
exposes it as `ctx.prefix` / `ctx.suffix`. FIM vs instruct is purely agent
|
||||||
|
config (the body), not feature code. Completion never touches the delta stream —
|
||||||
|
it waits for `finished` and reads the last message.
|
||||||
|
|
||||||
|
### 5b. Chat
|
||||||
|
|
||||||
|
`ChatRootView` owns one persistent `ConversationHistory` for the whole chat view
|
||||||
|
and injects it into every collaborator. **History is the single source of truth.**
|
||||||
|
|
||||||
|
```
|
||||||
|
ChatRootView (QML) — owns ConversationHistory m_history
|
||||||
|
ChatModel.setHistory(m_history) — ChatModel is a PROJECTION:
|
||||||
|
subscribes to messageAdded/Updated/cleared/reset, flattens blocks→rows,
|
||||||
|
overlays file-edit status from ChangesManager, holds a per-message usage map
|
||||||
|
ChatAgentController — agent list filtered to the
|
||||||
|
chatAssistant roster; active agent persisted
|
||||||
|
▼ dispatchSend
|
||||||
|
ClientInterface
|
||||||
|
session = sessionManager.createSession(activeAgent, m_history)
|
||||||
|
sessionManager.toolContributors().contribute(client.tools()) — builtin+skills+MCP
|
||||||
|
session.setContentLoader(ChatSerializer::loadContentFromStorage)
|
||||||
|
systemPrompt layer "chat.context" = project info + skills + linked files
|
||||||
|
session.send( blocks{ TextContent + StoredAttachmentContent + StoredImageContent } )
|
||||||
|
▼ consumes Session signals (NOT raw client signals):
|
||||||
|
event(Usage) → ChatModel.setMessageUsage + token-counter calibration
|
||||||
|
finished(id) → ChangesManager.applyPendingEditsForRequest + persist;
|
||||||
|
removeSession (the persistent history survives)
|
||||||
|
failed(id, ErrorInfo) → surface error; removeSession
|
||||||
|
|
||||||
|
ChatCompressor → acquire(chatCompression-roster agent) → seed history from the
|
||||||
|
chat's messages → "compression" layer → send → read summary from
|
||||||
|
the compression session's own history → release
|
||||||
|
InputTokenCounter → estimates over ConversationHistory (calibrated by Usage events)
|
||||||
|
ChatSerializer → persists ConversationHistory via MessageSerializer (v0.3);
|
||||||
|
imports legacy v0.1/v0.2 files
|
||||||
|
```
|
||||||
|
|
||||||
|
`ChatModel`'s QML role surface (roleType / content / attachments / images /
|
||||||
|
isRedacted / token roles) is unchanged, so the QML delegates were untouched. The
|
||||||
|
projection's incremental updates avoid model resets on the streaming hot path.
|
||||||
|
|
||||||
|
### 5c. Quick refactor
|
||||||
|
|
||||||
|
```
|
||||||
|
QodeAssistClient.requestQuickRefactor → QuickRefactorHandler
|
||||||
|
agent = AgentRouter.pickAgent(roster.quickRefactor, {file, project})
|
||||||
|
session = sessionManager.acquire(agent)
|
||||||
|
if useTools: sessionManager.toolContributors().contribute(client.tools())
|
||||||
|
systemPrompt layer "refactor" = tagged selection + output + indentation rules
|
||||||
|
session.send(blocks{instructions}, useTools)
|
||||||
|
▼ on Session::finished:
|
||||||
|
history().lastAssistantText() → ResponseCleaner → RefactorResult → editor insert
|
||||||
|
→ sessionManager.release(session)
|
||||||
|
on Session::failed(ErrorInfo) → RefactorResult{error}
|
||||||
```
|
```
|
||||||
|
|
||||||
---
|
---
|
||||||
|
|
||||||
## 6. Remaining Stack A (runtime does NOT depend on it)
|
## 6. Context layer
|
||||||
|
|
||||||
|
The context services sit behind IDE-agnostic ports; Qt Creator API use lives in
|
||||||
|
the adapters.
|
||||||
|
|
||||||
```
|
```
|
||||||
🔴 Settings UI: provider/model/template selection pages
|
EditorContext — IDocumentReader (port) ← DocumentReaderQtCreator (TextEditor API)
|
||||||
(ccProvider / caProvider / qrProvider) + ConfigurationManager
|
ProjectContext — IProjectScanner (port) ← ProjectScannerQtCreator (ProjectExplorer
|
||||||
→ use ProvidersManager
|
+ Core::DocumentModel + the IgnoreManager for .qodeassistignore)
|
||||||
🔴 root providers/* (PluginLLMCore::Provider, 14 classes)
|
TokenEstimator — TokenUtils (pure) ← InputTokenCounter (thin UI consumer)
|
||||||
→ read only chat/quick-refactor sampling settings
|
|
||||||
🔴 pluginllmcore/* (ProvidersManager, PromptTemplateManager, ResponseCleaner,
|
|
||||||
PromptProviderChat/Fim, ContextData)
|
|
||||||
🔴 qodeassist.cpp:144-146 registerProviders() / registerTemplates() (Stack A registration)
|
|
||||||
🔴 qodeassist.cpp:185 MCP skill-tool loop on Stack A providers (effectively dead)
|
|
||||||
🔴 ChatAssistantSettings / QuickRefactorSettings — sampling fields (read only by root providers)
|
|
||||||
|
|
||||||
ResponseCleaner (pluginllmcore) is still used by QuickRefactorHandler as a text
|
|
||||||
utility — orthogonal to the provider stack.
|
|
||||||
```
|
```
|
||||||
|
|
||||||
### Removed during the migration
|
`ContextManager` is now Qt-Creator-free: it delegates open-file enumeration and
|
||||||
|
ignore filtering to an injected `IProjectScanner` (defaulting to the QtC adapter),
|
||||||
- Rules subsystem (`RulesLoader` + chat "active rules" UI + QuickRefactor rules block)
|
and keeps only filesystem reads + formatting. `ContextManager::shouldIgnore(path)`
|
||||||
- `ChatConfigurationController`, `AgentRoleController` (chat config/role presets)
|
replaced the previously exposed `ignoreManager()`.
|
||||||
- `m_promptProvider` (`PromptProviderFim`) in the plugin
|
|
||||||
- `RequestType::CodeCompletion` branch in all 14 root providers
|
|
||||||
- Sampling / prompt-generation fields in `CodeCompletionSettings`
|
|
||||||
- ChatView no longer links `PluginLLMCore`
|
|
||||||
|
|
||||||
---
|
---
|
||||||
|
|
||||||
## 7. Dependency summary
|
## 7. Cross-cutting
|
||||||
|
|
||||||
```
|
- **Request lifecycle** — a session has at most one in-flight request; `send()`
|
||||||
┌──────────────── Stack B (active runtime) ────────────────┐
|
while in flight cancels the previous. Every request ends in exactly one of
|
||||||
LLMClientInterface ─┐ │
|
`finished` / `failed` / `cancelled`. Cancellation is not an error; no consumer
|
||||||
ClientInterface ────┼─► SessionManager ─► Session ─► Agent ─► GenericProvider ─► LLMQore::*Client
|
string-matches a message to tell them apart.
|
||||||
QuickRefactorHandler─┘ │ │ │ │
|
- **Typed errors** — `ErrorInfo { category ∈ {Config, Auth, Network, Provider,
|
||||||
ChatCompressor ──────────────┘ │ AgentFactory ProviderFactory
|
Validation, Tool}, message, providerDetail }`. `ResponseRouter` categorizes wire
|
||||||
AgentRouter (rosters) │ │
|
errors (best-effort) at the boundary; `Session::failed` carries the typed value.
|
||||||
ProviderInstanceFactory (TOML)
|
- **Tools** — `SessionManager` owns a `ToolContributorRegistry`; built-in ToolKit,
|
||||||
└──────────────────────────────────────────────────────────┘
|
the skill tool, and MCP client tools register once and are contributed to chat
|
||||||
|
and quick-refactor session clients uniformly.
|
||||||
Stack A (settings UI + ConfigurationManager + MCP loop) — isolated,
|
- **Threading** — the core runs on the GUI thread; concurrency is the Qt event
|
||||||
no runtime consumers remain.
|
loop plus async network I/O. Blocking work hides behind L3 ports.
|
||||||
```
|
|
||||||
|
|
||||||
---
|
---
|
||||||
|
|
||||||
## 8. Open follow-ups (optional)
|
## 8. Tests
|
||||||
|
|
||||||
1. **Chat picker filtering** — show only `chatAssistant`-roster agents (currently
|
`test/` (GTest + Qt::Test) covers the two engines most affected by the rework:
|
||||||
lists all non-hidden agents; the auto-default may land on a FIM agent).
|
|
||||||
Requires wiring ChatView to `PipelinesConfig` (watch for OBJECT-library
|
- `JsonPromptTemplateTest` — the `[body]` engine: jinja render + JSON splice,
|
||||||
symbol duplication).
|
literal passthrough, empty-render key drop, nested literals, and load-time
|
||||||
2. **MCP tools on agent clients** — MCP skill tools are registered only on Stack A
|
rejection of bodies that render invalid JSON.
|
||||||
providers; to expose MCP tools to chat agents, register them on the session
|
- `ResponseRouterTest` — a fake `BaseClient` replays a recorded provider stream;
|
||||||
client alongside `registerQodeAssistTools`.
|
asserts the assistant message is stamped with the request id, history is built
|
||||||
3. **Physical Stack A teardown** — remove the provider/model/template settings UI,
|
correctly (thinking + text + tool use/result), the typed event stream is emitted,
|
||||||
`ConfigurationManager`, root `providers/*`, `pluginllmcore/*`, and the
|
and wire errors are categorized.
|
||||||
registration + MCP loop in `qodeassist.cpp`. Runtime no longer depends on them.
|
|
||||||
4. **Per-message session cost** — chat/refactor create a fresh agent/provider/client
|
---
|
||||||
(and read secrets) per request; a session pool could reduce latency.
|
|
||||||
|
## 9. Remaining follow-ups (optional)
|
||||||
|
|
||||||
|
1. **Qt-Creator-free core build + CI** — `AgentFactory` / `ContextRenderer` still
|
||||||
|
call `Core::ICore::userResourcePath`, so the core targets link `QtCreator::Core`.
|
||||||
|
A `ResourcePaths` port + adapter would let the core build without Qt Creator and
|
||||||
|
enable a CI job that fails on a layering-violating include, plus golden
|
||||||
|
rendered-body snapshots over the bundled agents loaded through the real loader.
|
||||||
|
2. **§9 target module layout** — the `core/ ide/ features/ hosts/` physical target
|
||||||
|
split in `target-architecture.md` is not yet reflected in the directory layout.
|
||||||
```
|
```
|
||||||
|
|||||||
Reference in New Issue
Block a user