Files
petal/internal/auth/users.go
T
prosolis 3cc23b8ea4 The advice arrived in the language she was trying to read her way out of
Reported as "the Portuguese option isn't translating the advice in
English — it's just reprinting Portuguese." Nothing was wrong with
targetFor. It was reading a direction the account could not leave.

learnerPairs held only zh, so SetPair refused learning_pair for pt-PT
and every Portuguese account was learning_en by force. targetFor then
did exactly what it says: explanations follow the half of the pair she
is not learning, which for a forced learning_en account is Portuguese.
A Portuguese document, corrected in Portuguese, explained in
Portuguese, with no way to ask for English — correct behaviour derived
from a fact about the roster that was no longer true.

The note in learnerPairs was written one phase too early to see it. It
said turning a pair around needs a word list and a dictionary reading
into English, and that fr, es and pt-PT had neither. Portuguese has
both. Word boundaries are spaces — the megabyte jieba needs is a
property of a writing system that doesn't use them, not a debt every
pair owes. And the dictionary arrived with dict.db, which reads pt→en
as readily as en→pt; dreamProvider.reverse has been answering that
question since the pair shipped. What was actually blocking the pair a
native English speaker learning Portuguese needs was this list.

So pt-PT joins it, and the pt-PT pack gets the learner copy the control
renders from — each label in the language of whoever would pick it,
since someone on the wrong side of that switch cannot read the side
they are reaching for. fr and es clear the same two bars through the
same dict.db and stay out: their packs carry no learner block yet,
which is a translation question rather than a data one, and the server
should keep saying no until one is written.

Two things that assumed learning_pair meant Chinese, now that it
doesn't. The segmenter gate reads the pair as well as the direction, or
a Portuguese learner would load a megabyte of Chinese word list and
hover Portuguese words at /api/hanzi. And that endpoint's own comment
justified skipping providerFor with a guarantee it no longer has; the
real guarantee was always the caller's — it is only ever asked about
tokens the Chinese segmenter found — and a stray lookup was already
safe, answering a miss with an empty 200.

Tests in both packages. The auth test that pinned pt-PT's refusal now
pins its acceptance, with fr and es still refused; the suggestions test
pins the consequence where it actually lands, which is the language she
reads her advice in.

Claude-Session: https://claude.ai/code/session_01GJHNvirh7Hzhc9RL3HAvz7
2026-07-29 18:40:22 -07:00

295 lines
11 KiB
Go

package auth
import (
"database/sql"
"encoding/json"
"errors"
"net/http"
"strings"
"gitea.parodia.dev/drwily/petal/internal/db"
"gitea.parodia.dev/drwily/petal/internal/httputil"
)
// UserStore provisions and reads accounts. Petal has no signup flow: a row
// appears the first time someone Authentik vouches for signs in, and that is
// the only way one is ever created.
type UserStore struct {
db *sql.DB
}
// NewUserStore returns a store backed by the given database.
func NewUserStore(sqlDB *sql.DB) *UserStore { return &UserStore{db: sqlDB} }
// Upsert records the account behind an OIDC login, keyed by the issuer's
// subject id.
//
// The subject is the id — not the email, which people change and which
// Authentik does not promise is stable. Email and display name are refreshed on
// every login so a rename upstream shows up here; pair_lang is deliberately not
// touched, because it is Petal's own setting rather than the IdP's.
func (u *UserStore) Upsert(sub, email, displayName string) error {
if sub == "" {
return errors.New("oidc: empty subject")
}
if displayName == "" {
displayName = email
}
_, err := u.db.Exec(
`INSERT INTO users (id, email, display_name) VALUES (?, ?, ?)
ON CONFLICT(id) DO UPDATE SET
email = excluded.email,
display_name = excluded.display_name`,
sub, email, displayName,
)
return err
}
// Get loads one account.
func (u *UserStore) Get(id string) (db.User, error) {
var user db.User
err := u.db.QueryRow(
`SELECT id, email, COALESCE(display_name, ''), created_at, pair_lang, direction
FROM users WHERE id = ?`, id,
).Scan(&user.ID, &user.Email, &user.DisplayName, &user.CreatedAt, &user.PairLang, &user.Direction)
return user, err
}
// MeHandler reports who the caller is. The frontend uses it to namespace
// per-account browser state and to show the signed-in writer; it sits behind
// the auth middleware, so reaching it at all already proves a valid session.
func (u *UserStore) MeHandler() http.HandlerFunc {
return func(w http.ResponseWriter, r *http.Request) {
user, err := u.Get(UserID(r.Context()))
if err != nil {
httputil.ErrorJSON(w, http.StatusUnauthorized, "not signed in")
return
}
httputil.WriteJSON(w, http.StatusOK, user)
}
}
// The pairs a writer may actually choose, in the order the picker offers them.
//
// This is deliberately *not* internal/llm's list of languages. That one names
// every pair the prompts know how to talk about, which is a cheap thing to add;
// this one names the pairs Petal can render itself in, which requires a langpack
// on the frontend. Accepting a code with no pack would leave her looking at
// Chinese with no way back except another guess, so the server refuses it. es
// joined on the day its pack landed, not before.
//
// These four are now every pair PairLang names on the frontend, which makes the
// two lists look redundant. They are not: the next pair will exist in the type
// and in the prompts long before it has copy, and this list is the one that
// says a writer may actually be sent there.
var shippedPairs = []string{"zh", "pt-PT", "fr", "es"}
func pairIsShipped(lang string) bool {
for _, p := range shippedPairs {
if p == lang {
return true
}
}
return false
}
// The two directions a pair can be travelled in. `DirectionLearningEn` is the
// original assumption made explicit: the writer is native in X and practising
// English. `DirectionLearningPair` is the other way round.
const (
DirectionLearningEn = "learning_en"
DirectionLearningPair = "learning_pair"
)
// The pairs whose *learner* direction Petal can actually serve, which is a
// narrower thing than a shipped pair and narrower again than a langpack.
//
// Turning a pair around needs data no langpack carries: a way to find word
// boundaries, and a dictionary that reads from the pair language into English. A
// pair missing either would leave a writer looking at an editor that silently
// does nothing when she hovers — worse than a missing pack, which at least reads
// as a bug rather than as an absence. So the server refuses, for the same reason
// and by the same mechanism as `shippedPairs`.
//
// Chinese has both as of Phase 26 (CC-CEDICT + jieba). Portuguese turns out to
// have both as well, and the original note here — "French, Spanish and
// Portuguese have neither" — was written one phase too early to see it:
//
// - Word boundaries are spaces. The megabyte word list jieba needs is a
// property of a writing system that doesn't use them, not a debt every pair
// owes; a Latin-script pair needs nothing loaded to be segmented.
// - The dictionary arrived with dict.db, which reads pt→en as readily as
// en→pt (see lexicon.dreamProvider.reverse). The reverse lookup the hover
// and the word card need is already there and already answering.
//
// So the pair a native English speaker learning Portuguese needs is real, and
// what was actually blocking it was this list. French and Spanish clear the same
// two bars through the same dict.db; they are held back only by their packs
// carrying no `learner` copy yet (see Pack.learner), which is a translation
// question rather than a data one.
//
// This list is still expected to grow one pair at a time and never to be
// inferred: segmentation is a property of a writing system, and there is no rule
// that derives "has a word list" from a language code.
var learnerPairs = []string{"zh", "pt-PT"}
// SupportsLearnerDirection reports whether a pair can be turned around.
func SupportsLearnerDirection(lang string) bool {
for _, p := range learnerPairs {
if p == lang {
return true
}
}
return false
}
func directionIsKnown(d string) bool {
return d == DirectionLearningEn || d == DirectionLearningPair
}
// SetPair moves an account to another (English + X) pair, in a given direction.
//
// The two are written together because they constrain each other: a direction is
// only meaningful for a pair that can be travelled in it, and validating them a
// field at a time would let a two-step change pass through a state that neither
// step is allowed to leave behind.
func (u *UserStore) SetPair(id, lang, direction string) error {
if !pairIsShipped(lang) {
return errors.New("auth: unshipped pair language " + lang)
}
if !directionIsKnown(direction) {
return errors.New("auth: unknown direction " + direction)
}
if direction == DirectionLearningPair && !SupportsLearnerDirection(lang) {
return errors.New("auth: no learner direction for " + lang)
}
res, err := u.db.Exec(
`UPDATE users SET pair_lang = ?, direction = ? WHERE id = ?`, lang, direction, id)
if err != nil {
return err
}
if n, err := res.RowsAffected(); err == nil && n == 0 {
return sql.ErrNoRows
}
return nil
}
// UpdateMeHandler changes the caller's own settings: which language Petal
// speaks alongside her English, and which of the two she is learning.
//
// It answers with the whole updated user rather than an empty 204 so the client
// has one shape to trust: /api/me and this return the same thing, and the app
// re-reads the pair from the response instead of assuming its request took.
//
// The pair language reaches further than the UI copy — it picks her Hunspell
// dictionary, her read-aloud voice, which word-lookup provider answers, and the
// language the prompts ask the model to explain in. All of those read
// `users.pair_lang` at use time, so all of them follow from this one write.
//
// Both fields are optional and each defaults to what the account already has, so
// the picker can send one without knowing the other. That matters for the
// combination this endpoint exists to prevent: a client that sent only
// `pair_lang: "fr"` while the account sat on `learning_pair` would otherwise ask
// for French-with-segmentation, which does not exist. Here it is one decision
// with one validation, and the answer carries whatever actually landed.
func (u *UserStore) UpdateMeHandler() http.HandlerFunc {
return func(w http.ResponseWriter, r *http.Request) {
var body struct {
PairLang *string `json:"pair_lang"`
Direction *string `json:"direction"`
}
if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
httputil.BadRequest(w, "invalid request body")
return
}
id := UserID(r.Context())
current, err := u.Get(id)
if err != nil {
// Only a missing row means "not signed in". A dictionary-file or
// SQLite fault answered as 401 would trip the client's session
// interceptor and throw a writer out of an app she is still signed
// in to — the same distinction SetPair's error branch makes below.
if errors.Is(err, sql.ErrNoRows) {
httputil.ErrorJSON(w, http.StatusUnauthorized, "not signed in")
return
}
httputil.ServerError(w, err)
return
}
lang, direction := current.PairLang, current.Direction
if body.PairLang != nil {
lang = strings.TrimSpace(*body.PairLang)
}
if body.Direction != nil {
direction = strings.TrimSpace(*body.Direction)
}
if !pairIsShipped(lang) {
// Name the ones that work. A writer who lands here has picked from a
// stale client, and "not a language" tells her nothing.
httputil.BadRequest(w, "unsupported language pair — Petal speaks "+strings.Join(shippedPairs, ", "))
return
}
if !directionIsKnown(direction) {
httputil.BadRequest(w, "unknown direction — expected "+DirectionLearningEn+" or "+DirectionLearningPair)
return
}
if direction == DirectionLearningPair && !SupportsLearnerDirection(lang) {
// Refused rather than quietly downgraded to learning_en. A silent
// downgrade would leave the writer looking at an editor that behaves
// like the one she just tried to leave, with nothing to read as an
// explanation — and the caller cannot tell the two outcomes apart
// without diffing the response it was given.
httputil.BadRequest(w, "Petal can only be learned toward "+strings.Join(learnerPairs, ", ")+" so far")
return
}
if err := u.SetPair(id, lang, direction); err != nil {
if errors.Is(err, sql.ErrNoRows) {
httputil.ErrorJSON(w, http.StatusUnauthorized, "not signed in")
return
}
httputil.ServerError(w, err)
return
}
user, err := u.Get(id)
if err != nil {
httputil.ServerError(w, err)
return
}
httputil.WriteJSON(w, http.StatusOK, user)
}
}
// Allowlist decides which of Authentik's users may write in this Petal.
// Authentik fronts several applications; being a valid user there does not mean
// being a user here.
//
// An entry matches a subject id or an email address, case-insensitively. Both
// are accepted on purpose: a subject is an opaque uuid nobody can know before
// that person's first login, so a subject-only list means the operator must let
// someone in, read a log line, and edit config — whereas an email is knowable in
// advance. An empty list allows everyone the IdP authenticates, which is the
// right default for a single-household instance.
type Allowlist map[string]bool
// ParseAllowlist builds an Allowlist from a comma-separated env value.
func ParseAllowlist(raw string) Allowlist {
list := Allowlist{}
for _, part := range strings.Split(raw, ",") {
if p := strings.ToLower(strings.TrimSpace(part)); p != "" {
list[p] = true
}
}
return list
}
// Permits reports whether this login may proceed.
func (a Allowlist) Permits(sub, email string) bool {
if len(a) == 0 {
return true
}
return a[strings.ToLower(sub)] || (email != "" && a[strings.ToLower(email)])
}