Name the municipality when the CEP does not existv0.2.0

Answering only "não encontrei" wastes the trip. A CEP that fails is
almost never a random number: it is someone who knows the city and
guessed the rest, most often at the "-000" that 957 municipalities never
received. The API now names that municipality on the 404, so the reply
names it too.

It stays an answer to a failure. No 📮, no address block, and no CEP
suggested from inside the range — the range is sparse (Areal's 118 CEPs
span nearly five thousand numbers), so the reply gives the count and says
so instead of letting the two ends read as a list of valid CEPs.

cep.is/go v0.1.0 discards the 404 body, so internal/lookup reads it with
one extra GET, on that path alone and never inside a retry. Everything
that can go wrong there — a slow answer, broken JSON, a faixa naming no
municipality — falls back to the reply the bot has always given. The
wrapper goes away the day the library exposes the field.

1 files changed, 94 insertions, 2 deletions

diff --git a/internal/bot/format.go b/internal/bot/format.go
index 9a9a22d..b94dee4 100644
--- a/internal/bot/format.go
+++ b/internal/bot/format.go
@@ -9,6 +9,8 @@ import (
 	"time"
 
 	cepis "cep.is/go"
+
+	"cep.is/telegrambot/internal/lookup"
 )
 
 // ParseMode is the formatting Telegram applies to every message this bot sends.
@@ -152,8 +154,7 @@ func formatRadius(metres int) string {
 func ErrorMessage(err error) string {
 	switch {
 	case errors.Is(err, cepis.ErrCEPNotFound):
-		return "🔍 CEP não encontrado na base dos Correios (DNE).\n" +
-			"Confira os oito dígitos — a base é atualizada a cada quinze dias."
+		return notFoundMessage(err)
 	case errors.Is(err, cepis.ErrInvalidCEP):
 		return "✋ CEP inválido. Preciso de oito dígitos: <code>01001000</code> " +
 			"ou <code>01001-000</code>."
@@ -173,6 +174,97 @@ func ErrorMessage(err error) string {
 	return "⚠️ Não consegui consultar este CEP agora. Tente de novo em alguns minutos."
 }
 
+// The two halves of the not-found answer, unchanged from the day the bot
+// shipped. Whatever is said about the municipality goes between them: the reply
+// opens by saying the CEP does not exist and closes by saying what to do, so
+// the middle cannot be mistaken for an address.
+const (
+	notFoundHead   = "🔍 CEP não encontrado na base dos Correios (DNE)."
+	notFoundAdvice = "Confira os oito dígitos — a base é atualizada a cada quinze dias."
+)
+
+// notFoundMessage answers a CEP that does not exist, naming the municipality it
+// would belong to when the API told us which one.
+//
+// It is worth the extra lines. The CEP that fails is almost never a random
+// number — it is someone who knows the city and guessed the rest, most often at
+// the "-000" that 957 municipalities never received. Telling that person only
+// "não encontrei" wastes the trip; telling them the number would have been in
+// Areal/RJ says which half of their guess was right.
+//
+// What it must never do is read like a hit. There is no 📮, no address block
+// and no CEP suggested from inside the range: the range is sparse — Areal's 118
+// CEPs are spread over nearly five thousand numbers — so its ends are the
+// extremes of what exists, not a list of what is valid, and the reply says so.
+func notFoundMessage(err error) string {
+	faixa := lookup.FaixaOf(err)
+	if faixa == nil || faixa.Localidade == "" {
+		return notFoundHead + "\n" + notFoundAdvice
+	}
+
+	var b strings.Builder
+	b.WriteString(notFoundHead + "\n")
+	b.WriteString("Ele cairia em <b>" + esc(faixa.Place()) + "</b>")
+
+	count := faixaCount(faixa)
+	span := faixaSpan(faixa)
+	switch {
+	case count != "" && span != "":
+		b.WriteString(", que tem " + count + ", " + span + ".\n")
+		b.WriteString("<i>A faixa é esparsa: nem todo número entre os dois é um CEP válido.</i>\n")
+	case count != "":
+		b.WriteString(", que tem " + count + ".\n")
+	case span != "":
+		b.WriteString(", com CEPs " + span + ".\n")
+		b.WriteString("<i>A faixa é esparsa: nem todo número entre os dois é um CEP válido.</i>\n")
+	default:
+		b.WriteString(".\n")
+	}
+
+	b.WriteString(notFoundAdvice)
+	return b.String()
+}
+
+// faixaCount renders how many CEPs the municipality has, or "" when the API did
+// not say. Without it the two ends of the range read as a count of one big
+// block, which is the misreading the number exists to prevent.
+func faixaCount(faixa *lookup.Faixa) string {
+	switch {
+	case faixa.CEPs <= 0:
+		return ""
+	case faixa.CEPs == 1:
+		return "1 CEP"
+	}
+	return strconv.Itoa(faixa.CEPs) + " CEPs"
+}
+
+// faixaSpan renders the ends of the range, or "" when they are missing or are
+// the same number — "de X a X" is not a range, and printing a single CEP there
+// would hand the user a guess to try.
+func faixaSpan(faixa *lookup.Faixa) string {
+	first, last := maskCEP(faixa.CEPIni), maskCEP(faixa.CEPFim)
+	if first == "" || last == "" || first == last {
+		return ""
+	}
+	return "de <code>" + esc(first) + "</code> a <code>" + esc(last) + "</code>"
+}
+
+// maskCEP puts the mask on eight digits, and returns "" for anything else: a
+// CEP the API sent in a shape this bot does not recognise is left out of the
+// sentence rather than printed raw.
+func maskCEP(cep string) string {
+	cep = strings.TrimSpace(cep)
+	if len(cep) != 8 {
+		return ""
+	}
+	for _, r := range cep {
+		if r < '0' || r > '9' {
+			return ""
+		}
+	}
+	return cep[:5] + "-" + cep[5:]
+}
+
 // RateLimitMessage is what a user held by the local token bucket is told, once
 // per warning window.
 func RateLimitMessage(retryAfter time.Duration) string {