Replace locale dirs with a data-folder namespace model: New([]paths), folder dot-paths, last-wins merge

README.md

@@ -7,10 +7,11 @@ lacked the locale coverage and format control we needed.
 - **Standards first** — formats, names and structures follow international and
   local standards first and foremost.
 - **Locale-aware** — names, addresses, postal codes and phone numbers follow
-  per-locale data and formats. Locales are always full language + territory
+  per-locale data and formats. The shipped data is organised by full locale tag
-  tags (`sv_SE`, never `sv`).
+  (`sv_SE`), but the engine treats folders as plain namespaces — name yours
+  anything.
 - **Data lives in JSON** — all source data is recursive JSON on disk, read when
-  you create a faker and then served from memory. Add or change locales without
+  you create a faker and then served from memory. Add or change data without
   touching the library. Behavior belongs in data too: the engine grows a
   built-in function only for what data can't express (a checksum, a time-based
   id), never for what character classes and choices already do.
@@ -23,25 +24,27 @@ lacked the locale coverage and format control we needed.
 
 ## CLI
 
-Install the `fakes` command, then point it at a locale directory and a category
+Install the `fakes` command, then point it at one or more data directories and a
-path — it prints one value to stdout. The first path segment names a category (a
+path — it prints one value to stdout. Each dot segment descends one level:
-JSON file); deeper dotted segments descend into it.
+folders, then the category (a JSON file), then fields inside it.
 
 ```sh
 go install github.com/Timewave-AB/fakes/cmd/fakes@latest
 
-fakes ./locales/sv_SE person        # Sara Eriksson
+fakes ./data/sv_SE person        # Sara Eriksson
-fakes ./locales/sv_SE person.last   # Eriksson  (dotted path into a category)
+fakes ./data/sv_SE person.last   # Eriksson  (dotted path into a category)
-fakes -seed 42 ./locales/sv_SE address
+fakes ./data sv_SE.person        # point at the tree; the folder is a segment
+fakes ./data/sv_SE ./mydata word # layer dirs; the last wins a name clash
+fakes -seed 42 ./data/sv_SE address
 ```
 
 Without installing, run it from a checkout with `go run ./cmd/fakes …`. Exit
-codes: `0` success, `1` runtime error (bad locale, unknown path), `2` misuse.
+codes: `0` success, `1` runtime error (missing dir, unknown path), `2` misuse.
 
 ### Generating a file from a custom template
 
-A category is just a JSON file in the locale directory, so you can drop in your
+A category is just a JSON file in a data directory, so you can drop in your
-own and render it — no code change. Save this as `locales/sv_SE/sql.json`:
+own and render it — no code change. Save this as `data/sv_SE/sql.json`:
 
 ```json
 {
@@ -58,10 +61,10 @@ own and render it — no code change. Save this as `locales/sv_SE/sql.json`:
 `sql-username` renders `'{username}'` `repeat` times and joins the results with
 the `),(` separator; the outer `V#ALUES(…)` wraps that into one valid row list.
 (`#A` escapes the literal `A`, which a format string would otherwise read as a
-letter token — see [Locale data format](#locale-data-format).)
+letter token — see [Data format](#data-format).)
 
 ```sh
-fakes -seed 1 ./locales/sv_SE sql
+fakes -seed 1 ./data/sv_SE sql
 # INSERT INTO users VALUES('zoom'),('wahoo'),('blip');
 ```
 
@@ -69,7 +72,7 @@ Raise `repeat` for more rows per statement, or loop in the shell to build a
 whole seed file:
 
 ```sh
-for _ in $(seq 100); do fakes ./locales/sv_SE sql; done > seed.sql
+for _ in $(seq 100); do fakes ./data/sv_SE sql; done > seed.sql
 ```
 
 ## Library
@@ -78,9 +81,9 @@ for _ in $(seq 100); do fakes ./locales/sv_SE sql; done > seed.sql
 go get github.com/Timewave-AB/fakes   # requires Go 1.22+ (for math/rand/v2)
 ```
 
-Point `New` at a locale directory, then generate values by category path with
+Point `New` at one or more data directories, then generate values by path with
-`Fake`. The first path segment names a category (a JSON file); deeper dotted
+`Fake`. Each dot segment descends one level: folders, then the category (a JSON
-segments descend into it.
+file), then fields inside it.
 
 ```go
 package main
@@ -93,7 +96,7 @@ import (
 )
 
 func main() {
-	f, err := fakes.New("./locales/sv_SE")
+	f, err := fakes.New([]string{"./data/sv_SE"})
 	if err != nil {
 		log.Fatal(err)
 	}
@@ -120,8 +123,8 @@ Seed a faker for reproducible output — same seed + locale yields an identical
 sequence, handy for stable tests:
 
 ```go
-a, _ := fakes.New("./locales/sv_SE", fakes.WithSeed(42))
+a, _ := fakes.New([]string{"./data/sv_SE"}, fakes.WithSeed(42))
-b, _ := fakes.New("./locales/sv_SE", fakes.WithSeed(42))
+b, _ := fakes.New([]string{"./data/sv_SE"}, fakes.WithSeed(42))
 av, _ := a.Fake("person")
 bv, _ := b.Fake("person")
 av == bv // true
@@ -129,25 +132,34 @@ av == bv // true
 
 A `*Fakes` is **not** safe for concurrent use — create one per goroutine.
 
-## Locales
+## Data
 
-The library ships a ready-to-use set under [`locales/`](locales) (`en_US`,
+The library ships a ready-to-use set under [`data/`](data), organised by locale
-`sv_SE`). Point either tool at one of them, a copy, or your own directory —
+(`en_US`, `sv_SE`). Point either tool at the whole tree, a single locale, a
-anywhere on disk.
+copy, or your own directory — anywhere on disk; no naming rules.
 
-Each ships these categories, formatted per locale (e.g. `date` is `MM/DD/YYYY`
+A directory is just a namespace. Each JSON file is a category named after the
-in `en_US`, `YYYY-MM-DD` in `sv_SE`; `ssn` is a US SSN vs a Swedish
+file; each subdirectory is a dot-path segment — folders nest exactly like JSON
-personnummer): `address`, `color`, `company`, `date`, `email`, `ip`, `person`,
+objects do. So `data/sv_SE/person.json` is `Fake("person")` when you point at
-`phone`, `price`, `sentence`, `ssn`, `time`, `url`, `username`, `uuid`,
+`data/sv_SE`, or `Fake("sv_SE.person")` when you point at `data`.
-`version`, `word`.
 
-The directory's name is the locale and must be a full tag (`sv_SE`, never
+Pass several directories and they merge, left to right: matching folders combine
-`sv`). Any casing or separator is accepted and canonicalised (`sv-se`,
+by their children, and any other clash is won by the last directory loaded. That
-`SV_SE` → `sv_SE`); a non-full name returns an error.
+lets you layer your own data over the built-ins without copying them:
 
-## Locale data format
+```go
+fakes.New([]string{"./data/sv_SE", "./mydata"}) // mydata overrides on a clash
+```
 
-Each JSON file in a locale directory is a **category** named after the file
+Each shipped locale carries these categories, formatted per locale (e.g. `date`
+is `MM/DD/YYYY` in `en_US`, `YYYY-MM-DD` in `sv_SE`; `ssn` is a US SSN vs a
+Swedish personnummer): `address`, `color`, `company`, `date`, `email`, `ip`,
+`person`, `phone`, `price`, `sentence`, `ssn`, `time`, `url`, `username`,
+`uuid`, `version`, `word`.
+
+## Data format
+
+Each JSON file in a data directory is a **category** named after the file
 (`address.json` → `address`), rendered by `Fake("address")`. Drop in a new file
 or folder — no code change, no recompile.
 
@@ -297,13 +309,13 @@ docker compose run --rm test                   # latest
 ```
 fakes.go        Fakes, New, options, seeding
 template.go     Fake, the recursive renderer (choices, format strings, paths)
-locale.go       locale loading + tag parsing
+data.go         data loading: folders/files -> namespace tree, multi-path merge
 cmd/fakes/      the `fakes` CLI (New + Fake over stdout)
-locales/        shipped locale data (JSON)
+data/           shipped data (JSON), organised by locale
 ```
 
-To add a category, drop a JSON file into a locale directory; to add a locale,
+To add a category, drop a JSON file into a data directory; to add a locale, add
-add a directory named with its full tag.
+a subdirectory of JSON files.
 
 ## License
 

cmd/fakes/main.go

@@ -1,10 +1,12 @@
-// Command fakes prints one fake value from a locale directory.
+// Command fakes prints one fake value from one or more data directories.
 //
-//	fakes ./locales/sv_SE person        # a full person
+//	fakes ./data/sv_SE person          # a full person
-//	fakes ./locales/sv_SE person.last   # just the surname (dotted path)
+//	fakes ./data/sv_SE person.last     # just the surname (dotted path)
-//	fakes -seed 42 ./locales/sv_SE address
+//	fakes ./data sv_SE.person          # point at the tree, address by folder
+//	fakes ./data/sv_SE ./mydata person # layer custom data; last dir wins
+//	fakes -seed 42 ./data/sv_SE address
 //
-// It is a thin CLI over the fakes library: New(dir) then Fake(path).
+// It is a thin CLI over the fakes library: New(dirs) then Fake(path).
 package main
 
 import (
@@ -16,11 +18,11 @@ import (
 	"github.com/Timewave-AB/fakes"
 )
 
-const usage = `Usage: fakes [-seed N] <locale-dir> <path>
+const usage = `Usage: fakes [-seed N] <data-dir>... <path>
 
-  <locale-dir>  a locale directory, e.g. ./locales/sv_SE
+  <data-dir>  one or more data directories, e.g. ./data/sv_SE (last wins on clash)
-  <path>        a category, or a dotted path into one (person, person.last)
+  <path>      a category, or a dotted path into one (person, person.last)
-  -seed N       seed for reproducible output`
+  -seed N     seed for reproducible output`
 
 func main() { os.Exit(run(os.Args[1:], os.Stdout, os.Stderr)) }
 
@@ -34,7 +36,7 @@ func run(args []string, stdout, stderr io.Writer) int {
 	if err := fs.Parse(args); err != nil {
 		return 2
 	}
-	if fs.NArg() != 2 {
+	if fs.NArg() < 2 {
 		fs.Usage()
 		return 2
 	}
@@ -46,12 +48,13 @@ func run(args []string, stdout, stderr io.Writer) int {
 		}
 	})
 
-	f, err := fakes.New(fs.Arg(0), opts...)
+	dirs, path := fs.Args()[:fs.NArg()-1], fs.Arg(fs.NArg()-1)
+	f, err := fakes.New(dirs, opts...)
 	if err != nil {
 		fmt.Fprintln(stderr, err)
 		return 1
 	}
-	v, err := f.Fake(fs.Arg(1))
+	v, err := f.Fake(path)
 	if err != nil {
 		fmt.Fprintln(stderr, err)
 		return 1

cmd/fakes/main_test.go

@@ -6,7 +6,10 @@ import (
 	"testing"
 )
 
-const svSE = "../../locales/sv_SE"
+const (
+	svSE = "../../data/sv_SE"
+	enUS = "../../data/en_US"
+)
 
 // runOut runs the CLI and returns exit code, stdout, stderr.
 func runOut(args ...string) (int, string, string) {
@@ -54,8 +57,9 @@ func TestRunSeedDeterministic(t *testing.T) {
 	}
 }
 
-func TestRunUsageOnWrongArgCount(t *testing.T) {
+func TestRunUsageOnTooFewArgs(t *testing.T) {
-	for _, args := range [][]string{{}, {svSE}, {svSE, "person", "extra"}} {
+	// Need at least one data dir plus a path; fewer is misuse.
+	for _, args := range [][]string{{}, {svSE}} {
 		code, _, errb := runOut(args...)
 		if code != 2 {
 			t.Errorf("run(%v) = %d, want 2", args, code)
@@ -66,6 +70,17 @@ func TestRunUsageOnWrongArgCount(t *testing.T) {
 	}
 }
 
+func TestRunMultipleDirs(t *testing.T) {
+	// Several data dirs then a final path: all but the last arg are dirs.
+	code, out, errb := runOut(enUS, svSE, "person")
+	if code != 0 {
+		t.Fatalf("run(multi-dir) = %d, stderr=%q", code, errb)
+	}
+	if strings.TrimSpace(out) == "" {
+		t.Fatal("empty output for multi-dir run")
+	}
+}
+
 func TestRunUnknownCategoryFails(t *testing.T) {
 	code, _, errb := runOut(svSE, "nope")
 	if code != 1 {
@@ -76,8 +91,8 @@ func TestRunUnknownCategoryFails(t *testing.T) {
 	}
 }
 
-func TestRunBadLocaleFails(t *testing.T) {
+func TestRunMissingDirFails(t *testing.T) {
-	code, _, errb := runOut("../../locales/sv", "person")
+	code, _, errb := runOut("../../data/nope", "person")
 	if code != 1 {
 		t.Fatalf("run = %d, want 1", code)
 	}

data.go

@@ -0,0 +1,93 @@
+package fakes
+
+import (
+	"encoding/json"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+)
+
+// loadData loads every given directory into one namespace tree and returns its
+// root children. A directory becomes a group; each *.json file in it compiles to
+// a node keyed by its base name (address.json -> "address"); each subdirectory
+// becomes a nested group, so folders turn into dot-path segments. Paths are
+// merged left to right: matching groups merge by their children, and any other
+// clash (a leaf, or a leaf-vs-group) is won by the last directory loaded.
+func loadData(paths []string) (map[string]node, error) {
+	root := map[string]node{}
+	for _, p := range paths {
+		g, err := loadDir(p)
+		if err != nil {
+			return nil, err
+		}
+		mergeChildren(root, g.children)
+	}
+	if len(root) == 0 {
+		return nil, fmt.Errorf("no .json data found in %v", paths)
+	}
+	return root, nil
+}
+
+// loadDir compiles one directory into a group. os.ReadDir yields entries sorted
+// by name, so the tree is built deterministically. Empty subdirectories (no JSON
+// anywhere under them) are skipped rather than added as empty namespaces.
+func loadDir(dir string) (*group, error) {
+	info, err := os.Stat(dir)
+	if err != nil {
+		return nil, err
+	}
+	if !info.IsDir() {
+		return nil, fmt.Errorf("%s is not a directory", dir)
+	}
+	entries, err := os.ReadDir(dir)
+	if err != nil {
+		return nil, err
+	}
+	g := &group{children: map[string]node{}}
+	for _, e := range entries {
+		full := filepath.Join(dir, e.Name())
+		if e.IsDir() {
+			child, err := loadDir(full)
+			if err != nil {
+				return nil, err
+			}
+			if len(child.children) > 0 {
+				g.children[e.Name()] = child
+			}
+			continue
+		}
+		if !strings.HasSuffix(e.Name(), ".json") {
+			continue
+		}
+		b, err := os.ReadFile(full)
+		if err != nil {
+			return nil, err
+		}
+		var raw any
+		if err := json.Unmarshal(b, &raw); err != nil {
+			return nil, fmt.Errorf("%s: %w", full, err)
+		}
+		n, err := compile(raw)
+		if err != nil {
+			return nil, fmt.Errorf("%s: %w", full, err)
+		}
+		g.children[strings.TrimSuffix(e.Name(), ".json")] = n
+	}
+	return g, nil
+}
+
+// mergeChildren overlays src onto dst. Two groups under the same key merge
+// recursively (so locales/categories from several paths combine); every other
+// key is replaced, making the last-loaded directory win on a conflict.
+func mergeChildren(dst, src map[string]node) {
+	for k, v := range src {
+		if dg, ok := dst[k].(*group); ok {
+			if sg, ok := v.(*group); ok {
+				mergeChildren(dg.children, sg.children)
+				continue
+			}
+		}
+		dst[k] = v
+	}
+}

data_test.go

@@ -52,8 +52,8 @@ func TestShippedDataCategories(t *testing.T) {
 			regexp.MustCompile(`^\d{1,3}( \d{3})?(,\d{2})? kr$`)},
 	}
 
-	en := newFakes(t, "locales/en_US", WithSeed(1))
+	en := newFakes(t, "data/en_US", WithSeed(1))
-	sv := newFakes(t, "locales/sv_SE", WithSeed(1))
+	sv := newFakes(t, "data/sv_SE", WithSeed(1))
 	for _, c := range cases {
 		for i := 0; i < 200; i++ {
 			if v := fake(t, en, c.path); !c.en.MatchString(v) {
@@ -70,7 +70,7 @@ func TestShippedDataCategories(t *testing.T) {
 // is a real calendar date (so month-length variants never emit e.g. Apr 31 or
 // Feb 30) and the trailing digit is a valid Luhn checksum over the other nine.
 func TestSwedishPersonnummer(t *testing.T) {
-	sv := newFakes(t, "locales/sv_SE", WithSeed(1))
+	sv := newFakes(t, "data/sv_SE", WithSeed(1))
 	sawLongMonthEnd := false
 	for i := 0; i < 2000; i++ {
 		v := fake(t, sv, "ssn")
@@ -93,6 +93,69 @@ func TestSwedishPersonnummer(t *testing.T) {
 	}
 }
 
+func TestShippedSwedishPhone(t *testing.T) {
+	f := newFakes(t, "data/sv_SE", WithSeed(11))
+	re := regexp.MustCompile(`^0\d{1,2}-\d{3} \d{2} \d{2}$`)
+	for i := 0; i < 50; i++ {
+		if n := fake(t, f, "phone"); !re.MatchString(n) {
+			t.Fatalf("phone %q does not match %s", n, re)
+		}
+	}
+}
+
+func TestShippedSwedishAddress(t *testing.T) {
+	f := newFakes(t, "data/sv_SE", WithSeed(3))
+	digit := regexp.MustCompile(`\d`)
+	for i := 0; i < 30; i++ {
+		a := fake(t, f, "address")
+		if !regexp.MustCompile(`\n`).MatchString(a) || !digit.MatchString(a) {
+			t.Fatalf("address %q is not a multi-line address with a number", a)
+		}
+	}
+
+	locality := regexp.MustCompile(`^\p{L}+( \p{L}+)*$`)
+	for i := 0; i < 30; i++ {
+		if c := fake(t, f, "address.locality"); !locality.MatchString(c) {
+			t.Fatalf("locality %q is not a Swedish place name", c)
+		}
+	}
+}
+
+func TestShippedPersonHasParts(t *testing.T) {
+	for _, dir := range []string{"data/sv_SE", "data/en_US"} {
+		f := newFakes(t, dir, WithSeed(7))
+		for i := 0; i < 30; i++ {
+			if name := fake(t, f, "person"); len(name) < 3 || !regexp.MustCompile(`\S \S`).MatchString(name) {
+				t.Fatalf("%s person %q lacks first and last name", dir, name)
+			}
+		}
+	}
+}
+
+func TestShippedUSPhone(t *testing.T) {
+	f := newFakes(t, "data/en_US", WithSeed(11))
+	re := regexp.MustCompile(`^(\(\d{3}\) \d{3}-\d{4}|\d{3}-\d{3}-\d{4})$`)
+	for i := 0; i < 50; i++ {
+		if n := fake(t, f, "phone"); !re.MatchString(n) {
+			t.Fatalf("phone %q does not match %s", n, re)
+		}
+	}
+}
+
+// TestShippedNamespacedTree loads the whole data/ tree (not a single locale) and
+// reaches each locale through its folder segment: data/sv_SE/person -> sv_SE.person.
+func TestShippedNamespacedTree(t *testing.T) {
+	f := newFakes(t, "data", WithSeed(1))
+	for _, path := range []string{"sv_SE.person", "en_US.person", "sv_SE.address.locality"} {
+		if got := fake(t, f, path); got == "" {
+			t.Fatalf("Fake(%q) returned empty", path)
+		}
+	}
+	if _, err := f.Fake("person"); err == nil {
+		t.Fatal("Fake(person) = nil error; categories should be namespaced under the locale folder")
+	}
+}
+
 func digitsOnly(s string) string {
 	var b strings.Builder
 	for _, r := range s {

edge_test.go

@@ -54,11 +54,11 @@ func TestNewErrors(t *testing.T) {
 	if err := os.WriteFile(file, []byte("{}"), 0o644); err != nil {
 		t.Fatal(err)
 	}
-	if _, err := New(file); err == nil {
+	if _, err := New([]string{file}); err == nil {
 		t.Error("New(file) = nil error, want not-a-directory error")
 	}
 	// Invalid JSON in a category file fails.
-	if _, err := New(writeLocale(t, "xx_XX", map[string]string{"broken": `{ not json`})); err == nil {
+	if _, err := New([]string{writeData(t, map[string]string{"broken": `{ not json`})}); err == nil {
 		t.Error("New(invalid JSON) = nil error")
 	}
 }
@@ -92,7 +92,7 @@ func TestDescendIntoLiteralErrors(t *testing.T) {
 // --- category root shapes ---
 
 func TestCategoryRootShapes(t *testing.T) {
-	dir := writeLocale(t, "xx_XX", map[string]string{
+	dir := writeData(t, map[string]string{
 		"obj": `{"format":"00"}`, // object root
 		"lit": `"hello"`,         // bare-string root
 	})
@@ -117,7 +117,7 @@ func TestLongStringList(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	f := newFakes(t, writeLocale(t, "xx_XX", map[string]string{"name": string(list)}), WithSeed(1))
+	f := newFakes(t, writeData(t, map[string]string{"name": string(list)}), WithSeed(1))
 
 	valid := map[string]bool{}
 	for _, v := range names {
@@ -145,7 +145,7 @@ var swedishName = regexp.MustCompile(`^\p{L}+([ -]\p{L}+)*$`)
 
 func TestShippedStreetComposition(t *testing.T) {
 	// street is a choice of composed {first}{last} templates and literal names.
-	f := newFakes(t, "locales/sv_SE", WithSeed(5))
+	f := newFakes(t, "data/sv_SE", WithSeed(5))
 	for i := 0; i < 300; i++ {
 		if s := fake(t, f, "address.street"); !swedishName.MatchString(s) {
 			t.Fatalf("street %q is not a Swedish street name", s)
@@ -155,7 +155,7 @@ func TestShippedStreetComposition(t *testing.T) {
 
 func TestShippedLastNameComposition(t *testing.T) {
 	// last is a choice of patronymic {first}sson templates and literal surnames.
-	f := newFakes(t, "locales/sv_SE", WithSeed(6))
+	f := newFakes(t, "data/sv_SE", WithSeed(6))
 	for i := 0; i < 300; i++ {
 		if s := fake(t, f, "person.last"); !swedishName.MatchString(s) {
 			t.Fatalf("last name %q is not a Swedish surname", s)
@@ -165,7 +165,7 @@ func TestShippedLastNameComposition(t *testing.T) {
 
 func TestShippedStreetNumberFormats(t *testing.T) {
 	// Reachable via a hyphenated path; covers all five weighted number variants.
-	f := newFakes(t, "locales/sv_SE", WithSeed(8))
+	f := newFakes(t, "data/sv_SE", WithSeed(8))
 	re := regexp.MustCompile(`^[1-9]\d{0,2}[A-Z]?$`)
 	for i := 0; i < 300; i++ {
 		if n := fake(t, f, "address.street-number"); !re.MatchString(n) {

fakes.go

@@ -1,15 +1,17 @@
-// Package fakes generates fake data from recursive, locale-specific JSON
+// Package fakes generates fake data from recursive JSON templates.
-// templates.
 //
-// Locale data lives in JSON on disk, not in Go. Point [New] at a locale
+// Data lives in JSON on disk, not in Go. Point [New] at one or more data
-// directory, then generate values by category path:
+// directories; folders and files become a dot-path namespace, then generate
+// values by path:
 //
-//	f, _ := fakes.New("./locales/sv_SE", fakes.WithSeed(42))
+//	f, _ := fakes.New([]string{"./data/sv_SE"}, fakes.WithSeed(42))
 //	f.Fake("address")          // "Storgatan 12\n234 56 Göteborg"
 //	f.Fake("address.locality") // "Göteborg"
 //
-// The directory's name is the locale and must be a full tag ("sv_SE", never
+// A subdirectory is a namespace segment, so pointing at "./data" instead reaches
-// "sv"). The JSON template format is documented in the README.
+// a category as "sv_SE.address". Several directories are merged left to right;
+// the last one wins on a name clash, so you can layer custom data over the
+// built-ins. The JSON template format is documented in the README.
 //
 // A [Fakes] is not safe for concurrent use; create one per goroutine.
 package fakes
@@ -19,14 +21,12 @@ import (
 	"encoding/binary"
 	"fmt"
 	"math/rand/v2"
-	"path/filepath"
 )
 
-// Fakes generates fake data for a single locale. Create one with [New].
+// Fakes generates fake data from a loaded namespace tree. Create one with [New].
 type Fakes struct {
 	rand       *rand.Rand
-	locale     string
+	categories map[string]node // root namespace: name -> compiled node tree
-	categories map[string]node // category name -> compiled node tree
 }
 
 type config struct {
@@ -43,29 +43,23 @@ func WithSeed(seed uint64) Option {
 	return func(c *config) { c.seed, c.seeded = seed, true }
 }
 
-// New builds a faker from a locale directory (e.g. "./locales/sv_SE"). The
+// New builds a faker from one or more data directories (e.g. "./data/sv_SE").
-// directory's name is the locale tag and must be a full tag like "sv_SE"; each
+// Each JSON file becomes a category named after the file (address.json ->
-// JSON file in it becomes a category named after the file (address.json ->
+// "address") and each subdirectory a namespace segment; directories are merged
-// "address"). It errors on a non-full tag, a missing directory, or invalid JSON.
+// in order, the last winning a name clash. It errors on a missing directory,
-func New(dir string, opts ...Option) (*Fakes, error) {
+// invalid JSON, or no data found.
-	name, ok := canonicalLocale(filepath.Base(dir))
+func New(paths []string, opts ...Option) (*Fakes, error) {
-	if !ok {
+	cats, err := loadData(paths)
-		return nil, fmt.Errorf("fakes: %q is not a full locale tag like \"sv_SE\"", filepath.Base(dir))
-	}
-	cats, err := loadLocale(dir)
 	if err != nil {
-		return nil, fmt.Errorf("fakes: %s: %w", name, err)
+		return nil, fmt.Errorf("fakes: %w", err)
 	}
 	var c config
 	for _, opt := range opts {
 		opt(&c)
 	}
-	return &Fakes{rand: newRand(c.seed, c.seeded), locale: name, categories: cats}, nil
+	return &Fakes{rand: newRand(c.seed, c.seeded), categories: cats}, nil
 }
 
-// Locale returns the faker's canonical locale tag.
-func (f *Fakes) Locale() string { return f.locale }
-
 func newRand(seed uint64, seeded bool) *rand.Rand {
 	if !seeded {
 		var b [16]byte

fakes_test.go

@@ -5,12 +5,18 @@ import (
 	"testing"
 )
 
-// newFakes creates a faker for a locale directory, failing the test on error.
+// newFakes creates a faker over a single data directory, failing on error. Most
+// tests load one dir; newFakesN loads several (last-loaded wins on conflicts).
 func newFakes(t *testing.T, dir string, opts ...Option) *Fakes {
 	t.Helper()
-	f, err := New(dir, opts...)
+	return newFakesN(t, []string{dir}, opts...)
+}
+
+func newFakesN(t *testing.T, dirs []string, opts ...Option) *Fakes {
+	t.Helper()
+	f, err := New(dirs, opts...)
 	if err != nil {
-		t.Fatalf("New(%q): %v", dir, err)
+		t.Fatalf("New(%q): %v", dirs, err)
 	}
 	return f
 }
@@ -25,23 +31,15 @@ func fake(t *testing.T, f *Fakes, path string) string {
 	return s
 }
 
-func TestNewRejectsNonFullLocale(t *testing.T) {
-	for _, bad := range []string{"locales/sv", "locales/se", "locales/swedish", "locales/sv_S"} {
-		if _, err := New(bad); err == nil {
-			t.Errorf("New(%q) = nil error, want full-locale error", bad)
-		}
-	}
-}
-
 func TestNewMissingDirectory(t *testing.T) {
-	_, err := New("locales/de_DE")
+	_, err := New([]string{"data/de_DE"})
 	if err == nil || !strings.Contains(err.Error(), "de_DE") {
-		t.Fatalf("New(missing) error = %v, want it to name the locale", err)
+		t.Fatalf("New(missing) error = %v, want it to name the path", err)
 	}
 }
 
 func TestWithSeedIsDeterministic(t *testing.T) {
-	a, b := newFakes(t, "locales/sv_SE", WithSeed(42)), newFakes(t, "locales/sv_SE", WithSeed(42))
+	a, b := newFakes(t, "data/sv_SE", WithSeed(42)), newFakes(t, "data/sv_SE", WithSeed(42))
 	for i := 0; i < 50; i++ {
 		if x, y := fake(t, a, "person"), fake(t, b, "person"); x != y {
 			t.Fatalf("same seed diverged at %d: %q != %q", i, x, y)
@@ -50,7 +48,7 @@ func TestWithSeedIsDeterministic(t *testing.T) {
 }
 
 func TestDifferentSeedsDiffer(t *testing.T) {
-	a, b := newFakes(t, "locales/en_US", WithSeed(1)), newFakes(t, "locales/en_US", WithSeed(2))
+	a, b := newFakes(t, "data/en_US", WithSeed(1)), newFakes(t, "data/en_US", WithSeed(2))
 	for i := 0; i < 50; i++ {
 		if fake(t, a, "person") != fake(t, b, "person") {
 			return

loading_test.go

@@ -0,0 +1,117 @@
+package fakes
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+)
+
+// writeData builds a temp data directory from a map of relative path (without
+// ".json") -> file content, creating parent folders as needed. The directory's
+// name carries no meaning anymore, so callers pick any layout they like —
+// including nested folders, which become dot-path segments.
+func writeData(t *testing.T, files map[string]string) string {
+	t.Helper()
+	dir := t.TempDir()
+	for name, content := range files {
+		p := filepath.Join(dir, name+".json")
+		if err := os.MkdirAll(filepath.Dir(p), 0o755); err != nil {
+			t.Fatal(err)
+		}
+		if err := os.WriteFile(p, []byte(content), 0o644); err != nil {
+			t.Fatal(err)
+		}
+	}
+	return dir
+}
+
+func TestNewLoadsAnyDirName(t *testing.T) {
+	// No locale tag required: a directory named anything loads fine.
+	dir := writeData(t, map[string]string{"greeting": `["hej", "hallå"]`})
+	f := newFakes(t, dir, WithSeed(1))
+	if got := fake(t, f, "greeting"); got != "hej" && got != "hallå" {
+		t.Fatalf("greeting = %q, want hej or hallå", got)
+	}
+}
+
+func TestNewEmptyDirErrors(t *testing.T) {
+	if _, err := New([]string{writeData(t, nil)}); err == nil {
+		t.Fatal("New(empty dir) = nil error")
+	}
+}
+
+// TestFoldersBecomeDotPaths checks the core of the new model: a subfolder is a
+// namespace, so data/<loc>/person.json is reachable as "<loc>.person".
+func TestFoldersBecomeDotPaths(t *testing.T) {
+	dir := writeData(t, map[string]string{
+		"sv_SE/greeting": `["hej"]`,
+		"en_US/greeting": `["hi"]`,
+	})
+	f := newFakes(t, dir, WithSeed(1))
+	if got := fake(t, f, "sv_SE.greeting"); got != "hej" {
+		t.Fatalf("sv_SE.greeting = %q, want hej", got)
+	}
+	if got := fake(t, f, "en_US.greeting"); got != "hi" {
+		t.Fatalf("en_US.greeting = %q, want hi", got)
+	}
+}
+
+// TestNestedFoldersAndJSON crosses both nesting kinds in one dot path: folders
+// a/b/c, then deep JSON fields inside the file at the end of that path.
+func TestNestedFoldersAndJSON(t *testing.T) {
+	dir := writeData(t, map[string]string{
+		"a/b/c/thing": `{"format":"{x}","x":{"format":"{y}","y":{"format":"{z}","z":["leaf"]}}}`,
+	})
+	f := newFakes(t, dir, WithSeed(1))
+	// Folders a.b.c, file thing, then JSON fields x.y.z — one continuous path.
+	if got := fake(t, f, "a.b.c.thing.x.y.z"); got != "leaf" {
+		t.Fatalf("a.b.c.thing.x.y.z = %q, want leaf", got)
+	}
+	// Rendering the file resolves the same chain top-down.
+	if got := fake(t, f, "a.b.c.thing"); got != "leaf" {
+		t.Fatalf("a.b.c.thing = %q, want leaf", got)
+	}
+}
+
+func TestRenderingAFolderErrors(t *testing.T) {
+	dir := writeData(t, map[string]string{"sv_SE/greeting": `["hej"]`})
+	f := newFakes(t, dir, WithSeed(1))
+	if _, err := f.Fake("sv_SE"); err == nil {
+		t.Fatal("Fake(folder) = nil error, want a not-a-value error")
+	}
+}
+
+// TestMultiPathLastWins loads two dirs; on a name clash the later dir wins, and
+// non-clashing entries from both are reachable (data combines).
+func TestMultiPathLastWins(t *testing.T) {
+	a := writeData(t, map[string]string{"greeting": `["from-a"]`, "only-a": `["a"]`})
+	b := writeData(t, map[string]string{"greeting": `["from-b"]`, "only-b": `["b"]`})
+	f := newFakesN(t, []string{a, b}, WithSeed(1))
+	if got := fake(t, f, "greeting"); got != "from-b" {
+		t.Fatalf("greeting = %q, want from-b (last loaded wins)", got)
+	}
+	if got := fake(t, f, "only-a"); got != "a" {
+		t.Fatalf("only-a = %q, want a", got)
+	}
+	if got := fake(t, f, "only-b"); got != "b" {
+		t.Fatalf("only-b = %q, want b", got)
+	}
+}
+
+// TestMultiPathMergesFolders checks that clashing folders merge by their
+// children rather than replacing wholesale: each dir adds a file to sv_SE, and
+// a per-file clash inside still resolves last-wins.
+func TestMultiPathMergesFolders(t *testing.T) {
+	a := writeData(t, map[string]string{"sv_SE/person": `["from-a"]`, "sv_SE/shared": `["a"]`})
+	b := writeData(t, map[string]string{"sv_SE/company": `["from-b"]`, "sv_SE/shared": `["b"]`})
+	f := newFakesN(t, []string{a, b}, WithSeed(1))
+	if got := fake(t, f, "sv_SE.person"); got != "from-a" {
+		t.Fatalf("sv_SE.person = %q, want from-a (folder merged, not replaced)", got)
+	}
+	if got := fake(t, f, "sv_SE.company"); got != "from-b" {
+		t.Fatalf("sv_SE.company = %q, want from-b", got)
+	}
+	if got := fake(t, f, "sv_SE.shared"); got != "b" {
+		t.Fatalf("sv_SE.shared = %q, want b (last loaded wins)", got)
+	}
+}

locale.go

@@ -1,66 +0,0 @@
-package fakes
-
-import (
-	"encoding/json"
-	"fmt"
-	"os"
-	"path/filepath"
-	"regexp"
-	"sort"
-	"strings"
-)
-
-// loadLocale parses and compiles every JSON file in dir into a category node
-// tree, keyed by the file's base name without extension (address.json ->
-// "address"). Compiling here means structural errors surface at New, not at
-// Fake time.
-func loadLocale(dir string) (map[string]node, error) {
-	info, err := os.Stat(dir)
-	if err != nil {
-		return nil, err
-	}
-	if !info.IsDir() {
-		return nil, fmt.Errorf("%s is not a directory", dir)
-	}
-	files, err := filepath.Glob(filepath.Join(dir, "*.json"))
-	if err != nil {
-		return nil, err
-	}
-	sort.Strings(files)
-	if len(files) == 0 {
-		return nil, fmt.Errorf("no .json files in %s", dir)
-	}
-	categories := make(map[string]node, len(files))
-	for _, file := range files {
-		b, err := os.ReadFile(file)
-		if err != nil {
-			return nil, err
-		}
-		var raw any
-		if err := json.Unmarshal(b, &raw); err != nil {
-			return nil, fmt.Errorf("%s: %w", filepath.Base(file), err)
-		}
-		n, err := compile(raw)
-		if err != nil {
-			return nil, fmt.Errorf("%s: %w", filepath.Base(file), err)
-		}
-		categories[strings.TrimSuffix(filepath.Base(file), ".json")] = n
-	}
-	return categories, nil
-}
-
-var localeTag = regexp.MustCompile(`^[a-z]{2,3}_[A-Z]{2}$`)
-
-// canonicalLocale normalises a tag to "lang_REGION" form, accepting '-' or '_'
-// separators and any casing. It reports whether the result is a full tag.
-func canonicalLocale(s string) (string, bool) {
-	parts := strings.Split(strings.ReplaceAll(strings.TrimSpace(s), "-", "_"), "_")
-	if len(parts) != 2 {
-		return "", false
-	}
-	tag := strings.ToLower(parts[0]) + "_" + strings.ToUpper(parts[1])
-	if !localeTag.MatchString(tag) {
-		return "", false
-	}
-	return tag, true
-}

locale_test.go

@@ -1,95 +0,0 @@
-package fakes
-
-import (
-	"os"
-	"path/filepath"
-	"regexp"
-	"testing"
-)
-
-// writeLocale creates a locale directory named tag under a temp root, with one
-// JSON file per category, and returns its path.
-func writeLocale(t *testing.T, tag string, categories map[string]string) string {
-	t.Helper()
-	dir := filepath.Join(t.TempDir(), tag)
-	if err := os.MkdirAll(dir, 0o755); err != nil {
-		t.Fatal(err)
-	}
-	for name, content := range categories {
-		if err := os.WriteFile(filepath.Join(dir, name+".json"), []byte(content), 0o644); err != nil {
-			t.Fatal(err)
-		}
-	}
-	return dir
-}
-
-func TestNewLoadsCustomLocale(t *testing.T) {
-	dir := writeLocale(t, "is_IS", map[string]string{"greeting": `["hej", "hallå"]`})
-	f := newFakes(t, dir, WithSeed(1))
-	if got := fake(t, f, "greeting"); got != "hej" && got != "hallå" {
-		t.Fatalf("greeting = %q, want hej or hallå", got)
-	}
-}
-
-func TestNewCanonicalizesDirName(t *testing.T) {
-	// A directory named with a lowercase region resolves to the canonical tag.
-	dir := writeLocale(t, "sv_se", map[string]string{"x": `["a"]`})
-	if got := newFakes(t, dir).Locale(); got != "sv_SE" {
-		t.Fatalf("Locale() = %q, want sv_SE", got)
-	}
-}
-
-func TestNewEmptyLocaleErrors(t *testing.T) {
-	if _, err := New(writeLocale(t, "xx_XX", nil)); err == nil {
-		t.Fatal("New(empty locale) = nil error")
-	}
-}
-
-func TestShippedSwedishPhone(t *testing.T) {
-	f := newFakes(t, "locales/sv_SE", WithSeed(11))
-	re := regexp.MustCompile(`^0\d{1,2}-\d{3} \d{2} \d{2}$`)
-	for i := 0; i < 50; i++ {
-		if n := fake(t, f, "phone"); !re.MatchString(n) {
-			t.Fatalf("phone %q does not match %s", n, re)
-		}
-	}
-}
-
-func TestShippedSwedishAddress(t *testing.T) {
-	f := newFakes(t, "locales/sv_SE", WithSeed(3))
-	digit := regexp.MustCompile(`\d`)
-	for i := 0; i < 30; i++ {
-		a := fake(t, f, "address")
-		if !regexp.MustCompile(`\n`).MatchString(a) || !digit.MatchString(a) {
-			t.Fatalf("address %q is not a multi-line address with a number", a)
-		}
-	}
-
-	locality := regexp.MustCompile(`^\p{L}+( \p{L}+)*$`)
-	for i := 0; i < 30; i++ {
-		if c := fake(t, f, "address.locality"); !locality.MatchString(c) {
-			t.Fatalf("locality %q is not a Swedish place name", c)
-		}
-	}
-}
-
-func TestShippedPersonHasParts(t *testing.T) {
-	for _, locale := range []string{"locales/sv_SE", "locales/en_US"} {
-		f := newFakes(t, locale, WithSeed(7))
-		for i := 0; i < 30; i++ {
-			if name := fake(t, f, "person"); len(name) < 3 || !regexp.MustCompile(`\S \S`).MatchString(name) {
-				t.Fatalf("%s person %q lacks first and last name", locale, name)
-			}
-		}
-	}
-}
-
-func TestShippedUSPhone(t *testing.T) {
-	f := newFakes(t, "locales/en_US", WithSeed(11))
-	re := regexp.MustCompile(`^(\(\d{3}\) \d{3}-\d{4}|\d{3}-\d{3}-\d{4})$`)
-	for i := 0; i < 50; i++ {
-		if n := fake(t, f, "phone"); !re.MatchString(n) {
-			t.Fatalf("phone %q does not match %s", n, re)
-		}
-	}
-}

template.go

@@ -24,6 +24,13 @@ type literal string
 
 func (literal) isNode() {}
 
+// group is a namespace of named children, built from a directory of JSON files
+// and subdirectories. It has no value of its own: descend into a named child by
+// dot path; rendering one is an error (see Fake).
+type group struct{ children map[string]node }
+
+func (*group) isNode() {}
+
 // choice picks one of its items. cum holds cumulative weights for a weighted
 // pick; when nil the choice is uniform and selection is O(1).
 type choice struct {
@@ -45,20 +52,18 @@ type template struct {
 
 func (*template) isNode() {}
 
-// Fake generates a value for a category path. The first segment names a
+// Fake generates a value for a dot path. Each segment descends one level: folder
-// category (a JSON file); further dot-separated segments descend into named
+// names and the category (JSON file) come first, then named fields within it,
-// fields, e.g. "address" or "address.street". Choices along the way are
+// e.g. "sv_SE.address" or "sv_SE.address.street". Choices along the way are
-// resolved at random.
+// resolved at random. A path naming a folder (no value of its own) is an error.
 func (f *Fakes) Fake(path string) (string, error) {
-	segments := strings.Split(path, ".")
+	n, err := descend(f.rand, &group{children: f.categories}, strings.Split(path, "."))
-	n, ok := f.categories[segments[0]]
-	if !ok {
-		return "", fmt.Errorf("fakes: unknown category %q", segments[0])
-	}
-	n, err := descend(f.rand, n, segments[1:])
 	if err != nil {
 		return "", fmt.Errorf("fakes: %s: %w", path, err)
 	}
+	if _, ok := n.(*group); ok {
+		return "", fmt.Errorf("fakes: %s names a folder, not a value", path)
+	}
 	return render(f.rand, n), nil
 }
 
@@ -70,6 +75,12 @@ func descend(r rng, n node, segments []string) (node, error) {
 		return n, nil
 	}
 	switch n := n.(type) {
+	case *group:
+		child, ok := n.children[segments[0]]
+		if !ok {
+			return nil, fmt.Errorf("no entry %q", segments[0])
+		}
+		return descend(r, child, segments[1:])
 	case *template:
 		child, ok := n.fields[segments[0]]
 		if !ok {