docs(runtime): internal/runtime/loop 44 个文件注释改中文（逐行 1:1，代码零字节改动）
主人 2026-10-03 拍板「注释一律中文」；本模块按长期分批协议（一个模块一次派工）执行。 注释行数 1329 → 1329 逐行 1:1（含中文 33 → 1304）；行尾注释 5 条一并译。 校验：机械校验（非注释改动行）= 空；独立 Go AST 程序（剥注释复渲）证明 44/44 文件代码全等； 全树 go test ./... 58 包 ok / 0 FAIL；交叉编译绿。 gofmt：capacity_loop_test.go / flow.go 改动前即未格式化，按铁律保留原样未动。
This commit is contained in:
AskaEth committed
1 parent
355f81d04a
commit
91e8d7a029
44 files changed
+1315
-1315
No files matched your search
@@ -7,37 +7,37 @@ import (
|
||||
"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/model"
|
||||
)
|
||||
|
||||
// This file owns the loop's per-turn role resolution (DevDocs/design/22 §4.2):
|
||||
// one process now hosts several roles at once, so everything that used to be read
|
||||
// off the Loop's construction-time binding — the system prompt (via the agent),
|
||||
// the memory namespace, the emotion machine and the reasoning knobs — is resolved
|
||||
// from the turn's SessionEnv instead.
|
||||
// 本文件拥有循环的按 turn 角色解析（DevDocs/design/22 §4.2）：
|
||||
// 一个进程现在同时承载多个角色，因此过去从
|
||||
// Loop 构造期绑定读取的一切 —— 系统提示词（经 agent）、
|
||||
// 记忆命名空间、情绪机与推理旋钮 —— 改为从该 turn 的
|
||||
// SessionEnv 解析。
|
||||
//
|
||||
// The resolution is deliberately three tiny functions over ctx rather than a
|
||||
// field on the Loop: the Loop is shared by every concurrent session, so storing a
|
||||
// resolved role on it would race, while ctx already carries the per-turn env.
|
||||
// Every role-scoped read must go through these helpers, so prompt, memory,
|
||||
// emotion and thinking can never disagree about "who is answering".
|
||||
// 解析刻意做成 ctx 上的三个小函数，而不是 Loop 上的
|
||||
// 字段：Loop 由所有并发会话共享，把解析出的角色存在
|
||||
// 它上面会产生数据竞争，而 ctx 本就携带按 turn 的 env。
|
||||
// 所有角色作用域读取都必须走这些辅助函数，使提示词、记忆、
|
||||
// 情绪与思考永远不会对「谁在回答」产生分歧。
|
||||
|
||||
// LoopKnobs are the per-role model/reasoning knobs injected into every request the
|
||||
// loop builds. It mirrors the three construction-time options
|
||||
// (Options.Thinking / Options.ReasoningEffort / Options.ThinkingShowToUser) so a
|
||||
// caller can resolve them per role while the options keep their meaning as the
|
||||
// default. A nil Thinking and an empty ReasoningEffort mean "do not intervene",
|
||||
// and a nil ThinkingShowToUser means "broadcast thinking frames as usual".
|
||||
// LoopKnobs 是注入循环构建的每个请求的按角色模型/推理
|
||||
// 旋钮。它镜像三个构造期选项
|
||||
// （Options.Thinking / Options.ReasoningEffort / Options.ThinkingShowToUser），使
|
||||
// 调用方能按角色解析它们，同时这些选项仍作为
|
||||
// 默认值生效。nil Thinking 与空 ReasoningEffort 表示「不干预」，
|
||||
// nil ThinkingShowToUser 表示「照常广播思考帧」。
|
||||
type LoopKnobs struct {
|
||||
// Thinking is the reasoning directive (budget / explicit switch) for the role.
|
||||
// Thinking 是该角色的推理指令（预算 / 显式开关）。
|
||||
Thinking *model.Thinking
|
||||
// ReasoningEffort is the OpenAI-compatible strength knob for the role.
|
||||
// ReasoningEffort 是该角色兼容 OpenAI 的强度旋钮。
|
||||
ReasoningEffort string
|
||||
// ThinkingShowToUser gates the thinking frames broadcast for the role's turns.
|
||||
// ThinkingShowToUser 控制该角色各 turn 的思考帧是否广播。
|
||||
ThinkingShowToUser *bool
|
||||
}
|
||||
|
||||
// personaFor returns the role the current turn speaks as (design/22 §2): the
|
||||
// session env's explicit role when the caller resolved one, otherwise the loop's
|
||||
// configured role (Options.PersonaID, the global default). Both the prompt/memory
|
||||
// and the emotion hook read it, so a turn can never mix two roles.
|
||||
// personaFor 返回当前 turn 发言所作为的角色（design/22 §2）：调用方
|
||||
// 解析出角色时取会话环境的显式角色，否则取循环
|
||||
// 配置的角色（Options.PersonaID，全局默认）。提示词/记忆
|
||||
// 与情绪钩子都读它，因此一个 turn 绝不会混用两个角色。
|
||||
func (l *Loop) personaFor(ctx context.Context) string {
|
||||
if pid := sessionEnvFrom(ctx).PersonaID; pid != "" {
|
||||
return pid
|
||||
@@ -45,12 +45,12 @@ func (l *Loop) personaFor(ctx context.Context) string {
|
||||
return l.personaID
|
||||
}
|
||||
|
||||
// agentFor returns the agent that carries the current turn's system prompt: the
|
||||
// per-role agent pool entry when one exists, otherwise the loop's default agent
|
||||
// (design/22 §4.2 方案 A). A session bound to a role that has no registered agent —
|
||||
// an unknown/deleted persona, or a caller that wired no pool — falls back to the
|
||||
// default agent rather than failing the turn, and its memory scope still follows
|
||||
// personaFor, so the fallback can never silently re-scope memory.
|
||||
// agentFor 返回承载当前 turn 系统提示词的 agent：存在
|
||||
// 按角色 agent 池条目时用它，否则用循环的默认 agent
|
||||
// （design/22 §4.2 方案 A）。会话绑定到一个没有注册 agent 的角色 ——
|
||||
// 未知/已删除的人格，或未接任何池的调用方 —— 时回退到
|
||||
// 默认 agent 而不是让该 turn 失败，且其记忆作用域仍跟随
|
||||
// personaFor，因此该回退绝不会悄悄改变记忆作用域。
|
||||
func (l *Loop) agentFor(ctx context.Context) *agent.Agent {
|
||||
if len(l.agents) > 0 {
|
||||
if ag := l.agents[l.personaFor(ctx)]; ag != nil {
|
||||
@@ -60,11 +60,11 @@ func (l *Loop) agentFor(ctx context.Context) *agent.Agent {
|
||||
return l.agent
|
||||
}
|
||||
|
||||
// knobsFor resolves the reasoning knobs for the current turn: the resolver's
|
||||
// per-role values when it knows the role, otherwise the loop's configured
|
||||
// defaults. A resolver that does not recognize the role (ok = false) deliberately
|
||||
// keeps the defaults rather than zeroing the knobs, so an unregistered id behaves
|
||||
// like the default role instead of silently disabling reasoning.
|
||||
// knobsFor 解析当前 turn 的推理旋钮：解析器
|
||||
// 认识该角色时用其按角色取值，否则用循环配置的
|
||||
// 默认值。不认识该角色的解析器（ok = false）刻意
|
||||
// 保留默认值而不是把旋钮清零，使未注册的 id 表现
|
||||
// 如同默认角色，而不是悄悄禁用推理。
|
||||
func (l *Loop) knobsFor(ctx context.Context) LoopKnobs {
|
||||
def := LoopKnobs{
|
||||
Thinking: l.thinking,
|
||||
|
||||
Reference in new issue
Block a user