Return known failures without redundant checks

[onechat] / turn.go

diff --git a/turn.go b/turn.go
index 76f524eefc5fff41012cb391fc144850d9cede2a..c08e94a0eed7d94a22e787d7bbaf97a69ce89289 100644 (file)
--- a/turn.go
+++ b/turn.go
@@ -2,100 +2,236 @@ package main
 
 import (
        "context"
 
 import (
        "context"

-       "encoding/json"

        "errors"
        "fmt"
        "errors"
        "fmt"

-       "strings"
+       "log/slog"
+       "path/filepath"

        "sync"
        "time"
 )
 
        "sync"
        "time"
 )
 

-// modelClient sends model calls.
-type modelClient interface {
-       Create(ctx context.Context, request Request) (Response, error)
-}
-

 // Event tells the clients what the agent does as it happens. Kind is the kind
 // Event tells the clients what the agent does as it happens. Kind is the kind

-// of a logged message (user, reply, tool, echo), or one of busy, idle, and
-// error. ID is the ID of a logged message.
+// of a logged message (user, reply, tool, echo, note, work), or one of ready,
+// busy, idle, error, and subagent, which tells that the subagent Text starts a
+// run. ID is the ID of a logged message. Source names the client that sent a
+// user message, such as terminal or telegram, so that each client can show the
+// messages of the others, or the subagent of a report. Status tells how the run
+// of a report ended: finished, stopped, or failed.

 type Event struct {
 type Event struct {

-       Kind string `json:"kind"`
-       ID   int    `json:"id,omitempty"`
-       Text string `json:"text,omitempty"`
+       Kind   string `json:"kind"`
+       ID     int    `json:"id,omitempty"`
+       Text   string `json:"text,omitempty"`
+       Source string `json:"source,omitempty"`
+       Status string `json:"status,omitempty"`

 }
 
 }
 

-// loop runs the turns of one agent. A user message goes to the log when it
+// TurnLoop runs the turns of one agent. A user message goes to the log when it

 // comes, and it wakes the loop. A turn waits until every message before it is
 // comes, and it wakes the loop. A turn waits until every message before it is

-// summarized. Then its first request holds the prompt, the tools, and the
-// input of turnInput. The loop runs the tool calls of each response and sends
-// their results back, until a response has no tool call: the final answer.
+// summarized. Then the turn runs a Chain whose first request holds the prompt,
+// the tools, and the input of turnInput.

 //
 //

-// Each later request of a turn begins with the whole input of the request
-// before it, then its output items, then the tool results. The output items
-// hold the reasoning with its encrypted content, because the calls do not
-// store it. So each request begins with all of the previous one and reads it
-// from the prompt cache. User messages that come during a turn join the next
-// request after the tool results. The replies, the tool calls, and the results
-// go to the log as they happen. The thoughts never do.
-type loop struct {
-       agent  *Agent
-       memory *Memory
-       client modelClient
-       tools  *toolbox
-       clock  func() time.Time
-       notify func(Event)
-       wake   chan struct{}
+// The view goes out as the messages of a ViewLayout, so the first request of a
+// turn reads the view of the turn before from the prompt cache. User messages
+// that come during a turn join the next request of its chain after the tool
+// results, and so do the reports of the subagents. The subagents run until the
+// loop ends. Parent and child messages stay queued until the turn finishes.
+// active belongs to the running turn. The mutex protects pending, cancel, and
+// failure.
+type TurnLoop struct {
+       agent     *Agent
+       memory    *Memory
+       client    ChainClient
+       tools     *Toolbox
+       runs      *SubagentRuns
+       layout    *ViewLayout
+       clock     func() time.Time
+       logger    *slog.Logger
+       notify    func(Event)
+       wake      chan struct{}
+       messages  *DaemonMessages
+       openChild func(context.Context, *Agent) (*DaemonAgent, error)
+       active    []Message

 
        mutex   sync.Mutex
        pending []Message
        cancel  context.CancelFunc
 
        mutex   sync.Mutex
        pending []Message
        cancel  context.CancelFunc

+       failure error

 }
 
 }
 

-// functionCall holds the fields of a function call output item.
-type functionCall struct {
-       Type      string `json:"type"`
-       Name      string `json:"name"`
-       Arguments string `json:"arguments"`
-       CallID    string `json:"call_id"`
-}
-
-// functionCallOutput is the input item with the result of a function call.
-type functionCallOutput struct {
-       Type   string `json:"type"`
-       CallID string `json:"call_id"`
-       Output string `json:"output"`
-}
+// newTurnLoop returns the loop of agent with its memory, its subagents, and the
+// WhatsApp link and the schedules of its tools. Its calls go through client,
+// and notify gets each event. The clock gives the date of a turn. logger gets
+// the failed turns and the ends of the subagents. A child has no WhatsApp
+// tools, even when the caller supplies a link.
+func newTurnLoop(agent *Agent, memory *Memory, subagents *Subagents, whatsapp *WhatsApp, schedules *Schedules, client ChainClient, clock func() time.Time, logger *slog.Logger, notify func(Event)) (*TurnLoop, error) {
+       layout, err := loadViewLayout(filepath.Join(agent.dir, viewLayoutFileName))
+       if err != nil {
+               return nil, err
+       }

 
 

-// newLoop returns the loop of agent with its memory. Its calls go through
-// client, and notify gets each event. The clock gives the date of a turn.
-func newLoop(agent *Agent, memory *Memory, client modelClient, clock func() time.Time, notify func(Event)) *loop {
-       l := &loop{
+       l := &TurnLoop{

                agent:  agent,
                memory: memory,
                client: client,
                agent:  agent,
                memory: memory,
                client: client,

+               layout: layout,

                clock:  clock,
                clock:  clock,

+               logger: logger,

                notify: notify,
                wake:   make(chan struct{}, 1),
        }
 
                notify: notify,
                wake:   make(chan struct{}, 1),
        }
 

-       l.tools = &toolbox{
-               log:     l,
-               chat:    memory,
-               workdir: agent.config.Workdir,
+       l.runs = &SubagentRuns{
+               agent:     agent,
+               memory:    memory,
+               subagents: subagents,
+               layout:    layout,
+               client:    client,
+               state:     l.state,
+               report:    l.Report,
+               note:      l.note,
+               fail:      l.fail,
+               logger:    logger,
+               notify:    notify,
+               running:   make(map[string]*SubagentRun),
+       }
+
+       if agent.config.Parent != "" {
+               whatsapp = nil
+       }
+
+       l.tools = &Toolbox{
+               log:       l,
+               chat:      memory,
+               subagents: subagents,
+               runs:      l.runs,
+               whatsapp:  whatsapp,
+               schedules: schedules,
+               workdir:   agent.config.Workdir,
+               agentName: agent.config.Name,
+               images:    agent.Images(),
+       }
+
+       if agent.config.Parent == "" {
+               l.tools.imageShow = &ImageShow{store: agent.Images()}
+       }
+
+       return l, nil
+}
+
+// state returns the per-turn state: the end of the view, and the date and time
+// of the turn, then the names and responsibilities of its children.
+func (l *TurnLoop) state() string {
+       state := "</chat>\n\nNow: " + l.clock().Format("Monday 2006-01-02 15:04 -07:00")
+       if l.messages != nil {
+               state += l.messages.ChildState(l.agent.config.Name)
+       }
+
+       return state
+}
+
+// Report logs the report text of the subagent name as a work message and tells
+// the clients, with the status of the run. Unless the subagent was stopped, the
+// report joins the waiting messages and wakes the loop, as a user message of
+// Submit does. A failed write stops the loop, because the log refuses all
+// writes after it.
+func (l *TurnLoop) Report(name, text, status string) {
+       message, err := l.Work(name, text, status)
+       switch {
+       case err != nil:
+               l.fail(fmt.Errorf("%w: %w", errChainLogWrite, err))
+       case status != subagentStopped:
+               l.Wake(message)

        }
        }

+}
+
+// note logs a note and tells the clients. A note starts no turn.
+func (l *TurnLoop) note(text string) error {
+       _, err := l.record("note", "", text)

 
 

-       return l
+       return err

 }
 
 }
 

-// Submit logs a user message and wakes the loop. The message joins the next
-// request: the first one of a new turn, or the next one of a running turn.
-func (l *loop) Submit(text string) (Message, error) {
-       message, err := l.Append("user", text)
+// fail stops the loop with err.
+func (l *TurnLoop) fail(err error) {
+       l.mutex.Lock()
+
+       l.failure = errors.Join(l.failure, err)
+       l.mutex.Unlock()
+
+       select {
+       case l.wake <- struct{}{}:
+       default:
+       }
+}
+
+// Work logs text as a work message of source, such as the name of a subagent,
+// and tells the clients, with status. The message starts no turn until it goes
+// to Wake.
+func (l *TurnLoop) Work(source, text, status string) (Message, error) {
+       message, err := l.memory.Append("work", text)

        if err != nil {
                return Message{}, err
        }
 
        if err != nil {
                return Message{}, err
        }
 

+       l.notify(Event{
+               Kind:   "work",
+               ID:     message.ID,
+               Text:   text,
+               Source: source,
+               Status: status,
+       })
+
+       return message, nil
+}
+
+// Wake adds the logged messages to the waiting messages and wakes the loop
+// once, as a user message of Submit does. They join the first request of a new
+// turn together, or the next request of a running turn after the tool results.
+func (l *TurnLoop) Wake(messages ...Message) {
+       if len(messages) == 0 {
+               return
+       }
+

        l.mutex.Lock()
        l.mutex.Lock()

+
+       l.pending = append(l.pending, messages...)
+       l.mutex.Unlock()
+
+       select {
+       case l.wake <- struct{}{}:
+       default:
+       }
+}
+
+// record logs a message and tells the clients, with the client source of a user
+// message.
+func (l *TurnLoop) record(kind, source, text string) (Message, error) {
+       message, err := l.memory.Append(kind, text)
+       if err != nil {
+               return Message{}, err
+       }
+
+       l.notify(Event{
+               Kind:   kind,
+               ID:     message.ID,
+               Text:   text,
+               Source: source,
+       })
+
+       return message, nil
+}
+
+// Submit logs a user message from the client source and wakes the loop. The
+// message joins the next request: the first one of a new turn, or the next one
+// of a running turn.
+func (l *TurnLoop) Submit(source, text string) (Message, error) {
+       message, err := l.record("user", source, text)
+       if err != nil {
+               return Message{}, err
+       }
+
+       l.mutex.Lock()
+

        l.pending = append(l.pending, message)
        l.mutex.Unlock()
 
        l.pending = append(l.pending, message)
        l.mutex.Unlock()
 


@@ -107,9 +243,15 @@ func (l *loop) Submit(text string) (Message, error) {
        return message, nil
 }
 
        return message, nil
 }
 

+// StopSubagent ends the run of the subagent name and tells if it ran. Its
+// report is logged, but it starts no turn.
+func (l *TurnLoop) StopSubagent(name string) (bool, error) {
+       return l.runs.Stop(name)
+}
+

 // Interrupt ends the running turn. A running tool is killed, and the turn sends
 // no further request.
 // Interrupt ends the running turn. A running tool is killed, and the turn sends
 // no further request.

-func (l *loop) Interrupt() bool {
+func (l *TurnLoop) Interrupt() bool {

        l.mutex.Lock()
        defer l.mutex.Unlock()
 
        l.mutex.Lock()
        defer l.mutex.Unlock()
 


@@ -122,36 +264,47 @@ func (l *loop) Interrupt() bool {
        return true
 }
 
        return true
 }
 

-// Append logs a message and tells the clients. The toolbox logs its calls and
-// results through it.
-func (l *loop) Append(kind, text string) (Message, error) {
-       message, err := l.memory.Append(kind, text)
-       if err != nil {
-               return Message{}, err
-       }
+// Busy tells if a turn runs.
+func (l *TurnLoop) Busy() bool {
+       l.mutex.Lock()
+       defer l.mutex.Unlock()

 
 

-       l.notify(Event{
-               Kind: kind,
-               ID:   message.ID,
-               Text: text,
-       })
+       return l.cancel != nil
+}

 
 

-       return message, nil
+// Append logs a message and tells the clients. The toolbox logs its calls and
+// results through it.
+func (l *TurnLoop) Append(kind, text string) (Message, error) {
+       return l.record(kind, "", text)

 }
 
 // Flush makes the log durable.
 }
 
 // Flush makes the log durable.

-func (l *loop) Flush(ctx context.Context) error {
+func (l *TurnLoop) Flush(ctx context.Context) error {

        return l.memory.Flush(ctx)
 }
 
        return l.memory.Flush(ctx)
 }
 

-// Run runs a turn each time user messages wait, until ctx ends. It returns the
-// error of a failed log write, which stops the agent.
-func (l *loop) Run(ctx context.Context) error {
+// Run runs a turn each time user messages or reports wait, until ctx ends. It
+// returns the error of a failed log write, which stops the agent. At its start
+// it resumes the subagents that a stop of the daemon cut, and at its end it
+// ends the runs of the subagents and waits for them.
+func (l *TurnLoop) Run(ctx context.Context) error {
+       defer l.runs.Close()
+
+       err := l.runs.Resume(ctx)
+       if err != nil {
+               return fmt.Errorf("resume subagents: %w", err)
+       }
+

        for {
        for {

-               for l.waiting() {
-                       err := l.turn(ctx)
+               for ctx.Err() == nil && l.waiting() {
+                       err := l.failed()

                        if err != nil {
                        if err != nil {

-                               return err
+                               return fmt.Errorf("check turn loop failure: %w", err)
+                       }
+
+                       err = l.turn(ctx)
+                       if err != nil {
+                               return fmt.Errorf("run turn: %w", err)

                        }
                }
 
                        }
                }
 


@@ -160,42 +313,51 @@ func (l *loop) Run(ctx context.Context) error {
                case <-ctx.Done():
                        return ctx.Err()
                }
                case <-ctx.Done():
                        return ctx.Err()
                }

+
+               err := l.failed()
+               if err != nil {
+                       return fmt.Errorf("check turn loop failure: %w", err)
+               }

        }
 }
 
        }
 }
 

-// waiting tells if user messages wait for a request.
-func (l *loop) waiting() bool {
+// waiting tells if user messages or reports wait for a request.
+func (l *TurnLoop) waiting() bool {

        l.mutex.Lock()
        defer l.mutex.Unlock()
 
        return len(l.pending) > 0
 }
 
        l.mutex.Lock()
        defer l.mutex.Unlock()
 
        return len(l.pending) > 0
 }
 

-// take returns the user messages that wait for a request, and removes them.
-func (l *loop) take() []Message {
+// failed returns the error of a failed write of a report.
+func (l *TurnLoop) failed() error {

        l.mutex.Lock()
        defer l.mutex.Unlock()
 
        l.mutex.Lock()
        defer l.mutex.Unlock()
 

-       messages := l.pending
-       l.pending = nil
-
-       return messages
+       return l.failure

 }
 
 // turn runs one turn. A failed or interrupted model call ends the turn and
 }
 
 // turn runs one turn. A failed or interrupted model call ends the turn and

-// tells the clients. The messages of the turn stay in the log and the view, and
-// the next turn starts only with a new message, so a lasting failure such as a
-// needed login does not repeat. Only a failed log write is an error of turn.
-func (l *loop) turn(ctx context.Context) error {
+// tells the clients. A failure also goes to the logger, so that it is not lost
+// when no client is connected, but an interrupt does not. The messages of the
+// turn stay in the log and the view, and the next turn starts only with a new
+// message, so a lasting failure such as a needed login does not repeat. Only a
+// failed log or queue write is an error of turn. A daemon stop leaves taken
+// parent and child messages queued for restart. A user interrupt completes them
+// without a report.
+func (l *TurnLoop) turn(ctx context.Context) error {
+       runCtx := ctx

        ctx, cancel := context.WithCancel(ctx)
        defer cancel()
 
        l.mutex.Lock()
        ctx, cancel := context.WithCancel(ctx)
        defer cancel()
 
        l.mutex.Lock()

+

        l.cancel = cancel
        l.mutex.Unlock()
 
        defer func() {
                l.mutex.Lock()
        l.cancel = cancel
        l.mutex.Unlock()
 
        defer func() {
                l.mutex.Lock()

+

                l.cancel = nil
                l.mutex.Unlock()
        }()
                l.cancel = nil
                l.mutex.Unlock()
        }()


@@ -204,8 +366,24 @@ func (l *loop) turn(ctx context.Context) error {
                Kind: "busy",
        })
 
                Kind: "busy",
        })
 

-       err := l.chain(ctx)
-       if err != nil && !errors.Is(err, errLogWrite) {
+       if l.messages != nil {
+               l.logger.Info("turn started")
+       }
+
+       reply, err := l.chain(ctx)
+       if l.messages != nil && !errors.Is(err, errChainLogWrite) && (err == nil || runCtx.Err() == nil) {
+               finishErr := l.messages.Finish(l, l.active, reply)
+               if finishErr != nil {
+                       err = errors.Join(err, fmt.Errorf("%w: finish agent messages: %w", errChainLogWrite, finishErr))
+               }
+       }
+
+       l.active = nil
+       if err != nil && !errors.Is(err, errChainLogWrite) {
+               if ctx.Err() == nil {
+                       l.logger.Error("turn failed", "err", err)
+               }
+

                l.notify(Event{
                        Kind: "error",
                        Text: turnError(ctx, err),
                l.notify(Event{
                        Kind: "error",
                        Text: turnError(ctx, err),


@@ -214,6 +392,10 @@ func (l *loop) turn(ctx context.Context) error {
                err = nil
        }
 
                err = nil
        }
 

+       if l.messages != nil {
+               l.logger.Info("turn done")
+       }
+

        l.notify(Event{
                Kind: "idle",
        })
        l.notify(Event{
                Kind: "idle",
        })


@@ -221,24 +403,21 @@ func (l *loop) turn(ctx context.Context) error {
        return err
 }
 
        return err
 }
 

-// errLogWrite marks an error of a log write, which ends the loop.
-var errLogWrite = errors.New("log write failed")
-

 // chain runs the request chain of a turn until the final answer.
 // chain runs the request chain of a turn until the final answer.

-func (l *loop) chain(ctx context.Context) error {
+func (l *TurnLoop) chain(ctx context.Context) (string, error) {

        messages := l.take()
        if len(messages) == 0 {
        messages := l.take()
        if len(messages) == 0 {

-               return nil
+               return "", nil

        }
 
        err := l.memory.Wait(ctx)
        if err != nil {
        }
 
        err := l.memory.Wait(ctx)
        if err != nil {

-               return err
+               return "", fmt.Errorf("wait for turn memory: %w", err)

        }
 
        instructions, err := l.agent.Prompt()
        if err != nil {
        }
 
        instructions, err := l.agent.Prompt()
        if err != nil {

-               return err
+               return "", fmt.Errorf("read turn prompt: %w", err)

        }
 
        skip := make(map[int]bool, len(messages))
        }
 
        skip := make(map[int]bool, len(messages))


@@ -246,121 +425,52 @@ func (l *loop) chain(ctx context.Context) error {
                skip[message.ID] = true
        }
 
                skip[message.ID] = true
        }
 

-       input := turnInput(l.memory.ViewWithout(skip), l.state(), messages)
-       for {
-               response, err := l.client.Create(ctx, Request{
+       view, err := l.layout.Split(l.memory.ViewLines(skip))
+       if err != nil {
+               return "", fmt.Errorf("split turn view: %w", err)
+       }
+
+       chain := &Chain{
+               client: l.client,
+               tools:  l.tools,
+               request: ModelRequest{

                        Model:        l.agent.config.Model,
                        Effort:       l.agent.config.Effort,
                        Instructions: instructions,
                        Model:        l.agent.config.Model,
                        Effort:       l.agent.config.Effort,
                        Instructions: instructions,

-                       Tools:        agentTools(),
-                       Input:        input,
-               })
-               if err != nil {
-                       return err
-               }
-
-               input, err = l.handle(ctx, input, response)
-               if err != nil {
-                       return err
-               }
-
-               if len(input) == 0 {
-                       return nil
-               }
+                       Tools:        l.agent.Tools(),
+               },
+               take:   l.take,
+               images: l.agent.Images(),

        }
        }

-}
-
-// handle logs the replies of response and runs its tool calls. It returns the
-// input of the next request: input, the output items, the tool results, and
-// the user messages that came in the meantime. After the final answer it
-// returns no input.
-func (l *loop) handle(ctx context.Context, input []any, response Response) ([]any, error) {
-       var outputs []any
-       for _, raw := range response.Output {
-               item, err := inputItem(raw)
-               if err != nil {
-                       return nil, err
-               }
-
-               input = append(input, item)

 
 

-               var call functionCall
-               err = json.Unmarshal(raw, &call)
+       input := turnInput(view, l.state(), messages)
+       for n, message := range messages {
+               input[len(input)-len(messages)+n], err = imageUserMessage(l.agent.Images(), message.Text)

                if err != nil {
                if err != nil {

-                       return nil, fmt.Errorf("decode output item: %w", err)
+                       return "", fmt.Errorf("read turn images: %w", err)

                }
                }

-
-               switch call.Type {
-               case "message":
-                       text, err := outputText([]json.RawMessage{raw})
-                       if err != nil {
-                               return nil, err
-                       }
-
-                       text = strings.TrimSpace(text)
-                       if text == "" {
-                               continue
-                       }
-
-                       _, err = l.Append("reply", text)
-                       if err != nil {
-                               return nil, fmt.Errorf("%w: %w", errLogWrite, err)
-                       }
-               case "function_call":
-                       output, err := l.tools.Call(ctx, call.Name, call.Arguments)
-                       if err != nil {
-                               return nil, fmt.Errorf("%w: %w", errLogWrite, err)
-                       }
-
-                       outputs = append(outputs, functionCallOutput{
-                               Type:   "function_call_output",
-                               CallID: call.CallID,
-                               Output: output,
-                       })
-               }
-       }
-
-       if len(outputs) == 0 {
-               return nil, nil

        }
 
        }
 

-       input = append(input, outputs...)
-       for _, message := range l.take() {
-               input = append(input, userMessage(message.Text))
+       reply, err := chain.Run(ctx, input)
+       if err != nil {
+               return "", fmt.Errorf("run turn chain: %w", err)

        }
 
        }
 

-       return input, nil
+       return reply, nil

 }
 
 }
 

-// state returns the per-turn state: the date and time of the turn.
-func (l *loop) state() string {
-       return "Now: " + l.clock().Format("Monday 2006-01-02 15:04 -07:00")
-}
-
-// turnInput returns the input of the first request of a turn: one user message
-// with the view, the per-turn state, and each new message whole. The view
-// leaves out the leaf lines of the new messages, which come whole after it.
-func turnInput(view, state string, messages []Message) []any {
-       parts := []string{view + "\n", state + "\n"}
-       for _, message := range messages {
-               parts = append(parts, message.Text)
-       }
-
-       return []any{userMessage(parts...)}
-}
+// take returns the user messages that wait for a request, and removes them.
+func (l *TurnLoop) take() []Message {
+       l.mutex.Lock()
+       defer l.mutex.Unlock()

 
 

-// inputItem returns an output item as an input item for the next request. The
-// calls do not store their items, so the item goes back whole, without its id.
-func inputItem(raw json.RawMessage) (map[string]any, error) {
-       var item map[string]any
-       err := json.Unmarshal(raw, &item)
-       if err != nil {
-               return nil, fmt.Errorf("decode output item: %w", err)
+       messages := l.pending
+       l.pending = nil
+       if l.messages != nil {
+               l.active = append(l.active, messages...)

        }
 
        }
 

-       delete(item, "id")
-
-       return item, nil
+       return messages

 }
 
 // turnError returns the text of the error that ended a turn.
 }
 
 // turnError returns the text of the error that ended a turn.


@@ -371,3 +481,21 @@ func turnError(ctx context.Context, err error) string {
 
        return err.Error()
 }
 
        return err.Error()
 }

+
+// turnInput returns the input of the first request of a turn, stable parts
+// first: a user message for each view message, one with the per-turn state,
+// and one for each new message, whole. The view leaves out the leaf lines of
+// the new messages.
+func turnInput(view []string, state string, messages []Message) []any {
+       input := make([]any, 0, len(view)+1+len(messages))
+       for _, text := range view {
+               input = append(input, modelUserMessage(text))
+       }
+
+       input = append(input, modelUserMessage(state))
+       for _, message := range messages {
+               input = append(input, modelUserMessage(message.Text))
+       }
+
+       return input
+}