Files
petal/internal/suggestions/chat.go
T
prosolis 336cae93e0 Phase 19: the copy stops being hardcoded Mandarin
Every `中文 · English` string moves out of ~29 components into
web/src/i18n: one Pack type, a verbatim zh pack, and two ways to read
it — usePack() for components, pack() for the modules that build a line
when something happens rather than when something renders.

Anything with a value in it is a function on the pack rather than a
template at the call site, English pluralisation included: word order
isn't universal, and a pack author has to be able to move the number.
The roster constants (tones, rewrite styles, export formats, companions)
keep only value + emoji, so a label can't drift from its key.

On the server, internal/llm/lang.go replaces "Simplified Chinese" in the
three prompts that actually name her language. pt-PT is spelled
"European Portuguese (pt-PT, never Brazilian Portuguese)" in the prompt
itself, and each Lang carries her word for "why" so the tutor prompt
still recognises the question when she asks it her way.

pair_lang reaches the model through the row-scoped query each handler
already ran — the one that proves she owns the document — rather than a
second lookup that could disagree with it.

Also records Phase 18's deploy: migration 0011 rehearsed against a copy
of the live VPS database, then applied for real.
2026-07-27 08:37:05 -07:00

131 lines
4.4 KiB
Go

package suggestions
import (
"database/sql"
"encoding/json"
"errors"
"net/http"
"strings"
"github.com/go-chi/chi/v5"
"gitea.parodia.dev/drwily/petal/internal/auth"
"gitea.parodia.dev/drwily/petal/internal/httputil"
"gitea.parodia.dev/drwily/petal/internal/llm"
)
// chatRequest is the body the AskPetal panel posts: the full conversation so
// far. The suggestion context is loaded server-side and never trusted from the
// client (spec Note #10).
type chatRequest struct {
Messages []llm.Message `json:"messages"`
}
// chat streams an Ask Petal conversational reply over SSE. It loads the
// suggestion and its parent document's surrounding paragraph, injects them as
// the tutor system prompt, then relays the model's tokens to the browser as
// `data:` events. History persistence lives entirely in the client.
func (h *Handler) chat(w http.ResponseWriter, r *http.Request) {
sugID := chi.URLParam(r, "id")
var body chatRequest
if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
httputil.ErrorJSON(w, http.StatusBadRequest, "invalid request body")
return
}
// One query for the suggestion fields and the parent document's plain text,
// scoped to the local user so a stray id can't read another user's doc.
var (
original, replacement, explanation, typ string
fromPos int
contentText string
pairLang string
)
err := h.DB.QueryRow(
`SELECT s.original, s.replacement, s.explanation, s.type, s.from_pos, d.content_text,
COALESCE(u.pair_lang, '')
FROM suggestions s
JOIN documents d ON d.id = s.doc_id
JOIN users u ON u.id = d.user_id
WHERE s.id = ? AND d.user_id = ?`,
sugID, auth.UserID(r.Context()),
).Scan(&original, &replacement, &explanation, &typ, &fromPos, &contentText, &pairLang)
if errors.Is(err, sql.ErrNoRows) {
httputil.ErrorJSON(w, http.StatusNotFound, "suggestion not found")
return
}
if err != nil {
httputil.ServerError(w, err)
return
}
paragraph := surroundingParagraph(contentText, fromPos)
systemPrompt := llm.AskPetalSystemPrompt(original, replacement, typ, explanation, paragraph, llm.LangFor(pairLang))
// SSE requires an unbuffered, flushable writer. chi's middleware writers pass
// Flush through; bail with a plain error if somehow they don't.
flusher, ok := w.(http.Flusher)
if !ok {
httputil.ServerError(w, errors.New("streaming unsupported"))
return
}
ch, err := llm.StreamAskPetal(r.Context(), h.Client, systemPrompt, body.Messages)
if err != nil {
// The stream never opened (e.g. LLM unreachable) — a normal JSON error is
// still appropriate since we haven't written SSE headers yet.
httputil.ErrorJSON(w, http.StatusBadGateway, "chat failed: "+err.Error())
return
}
w.Header().Set("Content-Type", "text/event-stream")
w.Header().Set("Cache-Control", "no-cache")
w.Header().Set("Connection", "keep-alive")
w.Header().Set("X-Accel-Buffering", "no") // disable proxy buffering (e.g. nginx)
w.WriteHeader(http.StatusOK)
flusher.Flush()
for chunk := range ch {
writeSSE(w, "token", map[string]string{"text": chunk})
flusher.Flush()
}
// Signal a clean end so the client can stop reading without waiting on EOF.
writeSSE(w, "done", map[string]bool{"done": true})
flusher.Flush()
}
// writeSSE emits one named SSE event with a JSON data payload. JSON-encoding the
// data keeps token text (which may contain newlines) from breaking SSE framing.
func writeSSE(w http.ResponseWriter, event string, data any) {
payload, err := json.Marshal(data)
if err != nil {
return
}
_, _ = w.Write([]byte("event: " + event + "\ndata: "))
_, _ = w.Write(payload)
_, _ = w.Write([]byte("\n\n"))
}
// surroundingParagraph returns the paragraph of contentText containing the
// plain-text offset from. Tiptap flattens blocks with blank-line separators, so
// paragraphs are bounded by "\n\n". When the offset is unknown (original wasn't
// located, from == -1) it falls back to the latency-capped document so the tutor
// still has context to work with.
func surroundingParagraph(contentText string, from int) string {
if from < 0 || from > len(contentText) {
return strings.TrimSpace(llm.TruncateDoc(contentText))
}
start := strings.LastIndex(contentText[:from], "\n\n")
if start < 0 {
start = 0
} else {
start += 2
}
end := len(contentText)
if rel := strings.Index(contentText[from:], "\n\n"); rel >= 0 {
end = from + rel
}
return strings.TrimSpace(contentText[start:end])
}