textmachine/platform/internal/httpapi/problem.go

package httpapi

import (
	"encoding/json"
	"log/slog"
	"net/http"
	"strconv"
	"time"

	"textmachine/platform/internal/reqid"
)

// problem.go: the contract's error model (canon 0.3.0 §Problem, §ErrorCode; companion §2.17).
//
// What a client dispatches on is the machine `code`, never the words: `title` and `detail` are
// written for a developer and a log and a client MUST NOT show either. The phrase the user reads is
// drawn by the CLIENT from the code, in the language of its interface — which is why no wording
// here is a product decision and why none of it is translated.
//
// The vocabulary is TWO-LEVEL and that is the whole extensibility mechanism: the root `code` is
// closed for this version, `cause.code` is not. A new special case is added as a cause and breaks
// no client; inventing a root code is a contract change and belongs to a ratification, not to this
// package.

// Code is a root error code. Closed for contract version 0.3.0.
type Code string

const (
	CodeInvalidRequest      Code = "invalid_request"
	CodeUnauthenticated     Code = "unauthenticated"
	CodeForbidden           Code = "forbidden"
	CodeNotFound            Code = "not_found"
	CodeGone                Code = "gone"
	CodeRequestTimeout      Code = "request_timeout"
	CodePayloadTooLarge     Code = "payload_too_large"
	CodeRunInFlight         Code = "run_in_flight"
	CodeBookNotReady        Code = "book_not_ready"
	CodeRunNotStoppable     Code = "run_not_stoppable"
	CodeRunNotResumable     Code = "run_not_resumable"
	CodeCeilingUnavailable  Code = "ceiling_unavailable"
	CodeIdempotencyConflict Code = "idempotency_conflict"
	CodeContentRefused      Code = "content_refused"
	CodeServiceUnavailable  Code = "service_unavailable"
	CodeInternalError       Code = "internal_error"
	// The two bank codes of 0.5.0. They exist because their remedies have different ADDRESSEES: a
	// refused set is re-decided by the person, an incomplete write is re-sent verbatim by the
	// machine (canon §applyBankCorrections).
	CodeBankCorrectionsRefused    Code = "bank_corrections_refused"
	CodeBankCorrectionsIncomplete Code = "bank_corrections_incomplete"
)

// Second-level causes this deployment gives. NOT a closed vocabulary — a client that does not know
// one falls back to the root code and loses only precision — so a value added here is not a
// contract change.
const (
	// CauseCeilingReached — THIS RUN is finished with: it stopped at the limit it bought, or spent
	// all of it. The remedy is a NEW run (canon §resumeRun).
	CauseCeilingReached = "ceiling_reached"
	// CauseCreditUnavailable — the run has room left and the ACCOUNT cannot cover the rest of it. The
	// remedy is to top up, after which the same run continues; the two never stand in for one another.
	CauseCreditUnavailable = "credit_unavailable"
	// CauseBoundsMoved — the scale moved between the read and the call.
	CauseBoundsMoved = "bounds_moved"
	// CauseRePassUnavailable — a re-pass was asked for and there is nothing to re-pass: no
	// correction was recorded since the last run, or a resnapshot run has already walked it in
	// (canon §RunRequest.re_pass, 0.7.0).
	CauseRePassUnavailable = "re_pass_unavailable"
	// CauseNotPriced — this deployment has no price projection for the book yet, so it cannot quote an
	// order over it. It is a state that CLEARS by itself: the projection lands with the book's reading
	// surface, which the materializer owes and retries. Its own cause rather than `bounds_moved`,
	// because the remedy is opposite — wait, do not re-read and retry in a loop.
	CauseNotPriced = "not_priced"
	// CauseChapterOrdersUnavailable — an order phrased in CHAPTERS over a book whose chapter cut is
	// not trustworthy enough to sell against (see ingest.ChapterOrdersOffered). The remedy is in the
	// options read: `order.chapter_orders` is false and `order.structure` says why.
	CauseChapterOrdersUnavailable = "chapter_orders_unavailable"
	// CauseCreditHeld — another book of this account holds the credit; `blocked` names it.
	CauseCreditHeld = "credit_held"
	// CauseKeyReused / CauseKeyInFlight — the two halves of `Idempotency-Key`.
	CauseKeyReused   = "key_reused"
	CauseKeyInFlight = "key_in_flight"
	// CauseCursorInvalid — a cursor from a structure that no longer exists (canon §NextCursor).
	CauseCursorInvalid = "cursor_invalid"
	// CauseVersionTooOld — a delta watermark that predates a wholesale replacement.
	CauseVersionTooOld = "version_too_old"
	// The two ways a request is refused before authorization. They are DIFFERENT facts and the
	// canon gives them one root code (`forbidden`), so this is the only place they are told apart —
	// which is what the second level is for.
	CauseClientHeaderMissing = "client_header_missing"
	CauseOriginRejected      = "origin_rejected"
)

// Item codes of `errors[]` (canon §ErrorItem). Not closed either.
const (
	ItemMissing         = "missing"
	ItemMissingOrLate   = "missing_or_late"
	ItemMalformed       = "malformed"
	ItemTooLong         = "too_long"
	ItemUnsupportedPair = "unsupported_pair"
	ItemOutOfRange      = "out_of_range"
	// ItemUnknown has no producer in this build: the one branch that raised it went with the bank's
	// per-term write path (D39.144). Kept because this block mirrors the canon's vocabulary rather
	// than this build's use of it.
	ItemUnknown = "unknown"
	// The two answers the intake's own cut gives about the file. Told apart because the remedies are
	// opposite; `errors[]` is an open vocabulary by the canon's own words.
	//
	// ItemNoBook — read and cut, and there is no book in it. The remedy is a different file.
	ItemNoBook = "no_book"
	// ItemNoChapterStructure — the book is there and nothing cut it into chapters, so a reader would
	// get one document of any length. Temporary and about our reach (backlog rows 283/325).
	ItemNoChapterStructure = "no_chapter_structure"
)

// codes is the closed root vocabulary of this contract version: every code, the status it names and
// its developer-facing title, in ONE table.
//
// One table because the two are a PAIR: the canon says each code names its status, and a code whose
// status and title were written in two switches is the class of divergence a reviewer cannot see —
// both halves look right on their own. The title is never shown to a user and never translated; the
// sentence a reader gets is the client's, drawn from the code.
var codes = map[Code]struct {
	status int
	title  string
}{
	CodeInvalidRequest:      {http.StatusBadRequest, "Invalid request"},
	CodeContentRefused:      {http.StatusBadRequest, "Content refused"},
	CodeUnauthenticated:     {http.StatusUnauthorized, "Unauthenticated"},
	CodeForbidden:           {http.StatusForbidden, "Forbidden"},
	CodeNotFound:            {http.StatusNotFound, "Not found"},
	CodeGone:                {http.StatusGone, "Gone"},
	CodeRequestTimeout:      {http.StatusRequestTimeout, "Request timeout"},
	CodePayloadTooLarge:     {http.StatusRequestEntityTooLarge, "Payload too large"},
	CodeRunInFlight:         {http.StatusConflict, "Run in flight"},
	CodeBookNotReady:        {http.StatusConflict, "Book not ready"},
	CodeRunNotStoppable:     {http.StatusConflict, "Run not stoppable"},
	CodeRunNotResumable:     {http.StatusConflict, "Run not resumable"},
	CodeCeilingUnavailable:  {http.StatusConflict, "Ceiling unavailable"},
	CodeIdempotencyConflict: {http.StatusConflict, "Idempotency conflict"},
	CodeServiceUnavailable:  {http.StatusServiceUnavailable, "Service unavailable"},
	CodeInternalError:       {http.StatusInternalServerError, "Internal error"},

	CodeBankCorrectionsRefused:    {http.StatusConflict, "Bank corrections refused"},
	CodeBankCorrectionsIncomplete: {http.StatusServiceUnavailable, "Bank corrections incomplete"},
}

func statusOf(c Code) int {
	if d, ok := codes[c]; ok {
		return d.status
	}
	return http.StatusInternalServerError
}

func title(c Code) string {
	if d, ok := codes[c]; ok {
		return d.title
	}
	return "Internal error"
}

// Problem is an RFC 9457 error body with the contract's extension members.
//
// `type` stays `about:blank`: a URI that resolves nowhere is a promise we do not keep and a URN
// that repeats `code` is a second copy of one fact (RFC 9457 §3.1.1 wants `type` to be the
// identifier, §3.2 allows extension members to be it instead — companion §2.17).
//
// Detail NEVER carries engine or database text. The engine's own detail strings read like "CJK leak
// in the ru output: 第一节" — pipeline vocabulary that must not cross this seam (canon §Boundaries).
type Problem struct {
	Type      string `json:"type"`
	Title     string `json:"title"`
	Status    int    `json:"status"`
	Detail    string `json:"detail,omitempty"`
	Code      Code   `json:"code,omitempty"`
	RequestID string `json:"request_id"`
	// Cause is the narrower reason within Code, when there is one to give.
	Cause *Cause `json:"cause,omitempty"`
	// Errors names the parts of the request that were wrong. Carried by `invalid_request`; absent
	// where there is no field to point at — a form with too many parts is reported by the root code
	// alone (canon §ErrorItem).
	Errors []Item `json:"errors,omitempty"`
	// Blocked is carried by `ceiling_unavailable` when another book of the account holds the credit.
	Blocked *Blocked `json:"blocked,omitempty"`
	// Refusals is carried by `bank_corrections_refused`: one entry per refused correction, or one
	// about the would-be result as a whole (canon §CorrectionRefusal).
	Refusals []CorrectionRefusal `json:"refusals,omitempty"`
	// RetryAfter is not a member: it is the header of the same name, and it is here so one writer
	// sets it (canon §Conflict).
	RetryAfter time.Duration `json:"-"`
}

// Cause is the second level of the code.
type Cause struct {
	Code string `json:"code"`
}

// Item is one thing wrong with the request: a JSON Pointer at the member, or `/<name>` naming a
// form part, plus a machine code for it.
type Item struct {
	Pointer string `json:"pointer"`
	Code    string `json:"code"`
}

// Blocked says what is holding the run scale down and which book is doing it. The same shape
// `RunOptions` answers with, because it is the same fact.
type Blocked struct {
	Code   string `json:"code"`
	BookID string `json:"book_id"`
}

// CorrectionRefusal is one refused correction: a JSON Pointer at it (`/corrections/3`), or the
// EMPTY string for a refusal about the result as a whole. `detail` is developer-facing free text,
// like Problem's own — a client marks the row and draws a neutral phrase (canon §CorrectionRefusal).
type CorrectionRefusal struct {
	Pointer string `json:"pointer"`
	Detail  string `json:"detail"`
}

// Fail writes an error identified by its root code.
func Fail(w http.ResponseWriter, r *http.Request, c Code) {
	WriteProblem(w, r, Problem{Code: c})
}

// FailCause writes an error with its narrower cause.
func FailCause(w http.ResponseWriter, r *http.Request, c Code, cause string) {
	WriteProblem(w, r, Problem{Code: c, Cause: &Cause{Code: cause}})
}

// Invalid writes a 400 naming the parts of the request that were wrong.
func Invalid(w http.ResponseWriter, r *http.Request, items ...Item) {
	WriteProblem(w, r, Problem{Code: CodeInvalidRequest, Errors: items})
}

// WriteProblem renders an error response.
//
// The status comes from the code and the title from the code, so neither can disagree with it. A
// caller that has something to add fills Detail (for the log), Cause, Errors or Blocked.
func WriteProblem(w http.ResponseWriter, r *http.Request, p Problem) {
	// The versioned surface answers a code on every response, `500` included. `Code` is omitempty
	// only for the ungoverned surface below, and without this the omission leaked back here as a
	// silent 500 with no code at all.
	if p.Code == "" {
		p.Code = CodeInternalError
	}
	p.Type = "about:blank"
	p.Status = statusOf(p.Code)
	p.Title = title(p.Code)
	if r != nil {
		p.RequestID = reqid.FromContext(r.Context())
	}
	if p.RequestID == "" {
		// The middleware stamps every request, so an empty id means this response was produced
		// outside it — a test double, or a handler mounted above reqid.Middleware. The field is
		// required by the schema and a client MAY show it, so it gets a value rather than an empty
		// string that reads as "this request had no id".
		p.RequestID = reqid.New()
	}
	h := w.Header()
	h.Set("Content-Type", "application/problem+json")
	if p.Status == http.StatusUnauthorized {
		// RFC 9110 §15.5.2 requires a challenge on every 401, and the canon declares the header
		// REQUIRED on that response. Bearer is the scheme this API actually accepts; the session
		// cookie is not an HTTP authentication scheme and cannot be named in a challenge. A client
		// holding no session sends the user to the sign-in flow rather than parsing this.
		h.Set("WWW-Authenticate", `Bearer realm="textmachine"`)
	}
	if p.RetryAfter > 0 {
		h.Set("Retry-After", strconv.Itoa(int(p.RetryAfter.Round(time.Second).Seconds())))
	}
	w.WriteHeader(p.Status)
	if err := json.NewEncoder(w).Encode(p); err != nil {
		slog.Debug("problem body not delivered", "err", err) // the client went away mid-write
	}
}

// WriteStatusProblem is the error writer for the surfaces the contract does NOT govern: the sign-in
// mechanics under /auth and the operational endpoints.
//
// ⚠ It emits NO `code`, and that is a decision rather than a gap. The sign-in surface has no machine
// vocabulary: a client shows ONE neutral phrase for every refusal there, with a single exception that
// needs no vocabulary at all — `429`, whose remedy travels in `Retry-After`, which is exactly what
// HTTP has that header for. Ratified in the companion's §2.14: «различать причины отказа клиент не
// может по замыслу».
//
// The transport defences that run BEFORE any handler — origin and marker-header checks — still
// answer the versioned `forbidden`: they guard every surface alike, and a defence that answered
// differently by path would be a second rule to keep in step.
//
// One writer per surface and no INVERSE function: the versioned surface derives its status from a
// code, this one takes the status it means at the call site. A status→code inverse would be a second
// source of truth — a code added to `statusOf` would never appear in it.
func WriteStatusProblem(w http.ResponseWriter, r *http.Request, status int, developerTitle, detail string) {
	p := Problem{Type: "about:blank", Status: status, Title: developerTitle, Detail: detail}
	if r != nil {
		p.RequestID = reqid.FromContext(r.Context())
	}
	if p.RequestID == "" {
		p.RequestID = reqid.New()
	}
	w.Header().Set("Content-Type", "application/problem+json")
	w.WriteHeader(status)
	if err := json.NewEncoder(w).Encode(p); err != nil {
		slog.Debug("problem body not delivered", "err", err) // the client went away mid-write
	}
}

// ProblemHandler is a static problem response, for the middleware that must be handed a denier.
func ProblemHandler(c Code) http.Handler {
	return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
		Fail(w, r, c)
	})
}

// CauseHandler is the same for a denier that has a narrower cause to give — the two ways a request
// is refused before authorization (canon §Forbidden).
func CauseHandler(c Code, cause string) http.Handler {
	return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
		FailCause(w, r, c, cause)
	})
}