data: one trie for maps and dictionaries, one loader

The hash map served a single exact-match lookup that the trie already answers, at the price of a second container, a second file loader with its own drift, and a Str-to-UTF-8 conversion on every chain probe. The files are small (kanji.dict is 7.5k lines), so the trie holds both. trieopen validates keys and each space-separated value word against Str and reports path:line; trielookup takes the Str every caller holds and treats a nil trie as an unloaded map; trienew/trieput exist for tests. The overflow guards on growth, Trie.root (always 0) and the per-language init loop written twice are gone.
2026-08-16 15:56:32 +09:00
parent eb0f88f764
commit c7718ece52
15 changed files with 206 additions and 609 deletions
--- a/2
+++ b/2
@@ -19,7 +19,7 @@ DOCKER_IMAGE = strans-build
 DOCKER_RUN = docker run --rm --user "$$(id -u):$$(id -g)" \
 	-v "$(CURDIR):/src" $(DOCKER_IMAGE)

-SRCS = dict.c font.c hash.c ibus.c ipc.c ko.c main.c popup_layout.c \
+SRCS = dict.c font.c ibus.c ipc.c ko.c main.c popup_layout.c \
 	srv.c str.c strans.c trie.c vi.c win.c
 XIMSRCS = xim/xim.c xim/keymap.c xim/ximtext.c
 XIMOBJS = $(XIMSRCS:.c=.o)
--- a/dat.h
+++ b/dat.h
@@ -57,31 +57,19 @@ struct Emit
 	Str	dict;
 };

-typedef struct Hnode Hnode;
-struct Hnode
-{
-	int	filled;
-	int	next;
-	char	*key;
-	int	klen;
-	char	*val;
-	int	vlen;
-};
-
 typedef struct Tnode Tnode;
 struct Tnode
 {
+	char	*val;
+	int	vlen;
 	int	child;
 	int	sibling;
 	char	c;
-	char	*val;
-	int	vlen;
 };

 typedef struct Trie Trie;
 struct Trie
 {
-	int	root;
 	Tnode	*nodes;
 	int	n;
 	int	cap;
@@ -94,16 +82,6 @@ enum
 	TrieExact,
 };

-typedef struct Hmap Hmap;
-struct Hmap
-{
-	int	nbs;
-	int	nsz;
-	int	len;
-	int	cap;
-	uchar	*nodes;
-};
-
 typedef struct Lang Lang;
 typedef struct Im Im;
 struct Lang
@@ -115,7 +93,7 @@ struct Lang
 	void	(*back)(Im*);
 	void	(*dictq)(Im*);
 	Trie	*map;
-	Hmap	*dict;
+	Trie	*dict;
 };

 struct Im
--- a/dict.c
+++ b/dict.c
@@ -1,59 +1,34 @@
-#include <errno.h>
 #include "dat.h"
 #include "fn.h"

-static char*
-readline(Biobuf *b, char *path)
-{
-	char *line;
-
-	errno = 0;
-	line = Brdstr(b, '\n', 1);
-	if(errno != 0)
-		die("can't read %s: %s", path, strerror(errno));
-	return line;
-}
-
+/* Candidates are the space-separated words of the value; the reading itself
+ * is a candidate only for the emoji dictionary. */
 void
 dictlookup(Dictreq *req, Dictres *res)
 {
 	Lang *l;
-	Hmap *dict;
-	Hnode *n;
 	char *p, *e, *sp;
 	Str tmp;
+	int vlen;

 	res->key = req->pre;
 	res->nkouho = 0;
 	res->lang = req->lang;
 	res->seq = req->seq;
-	if(req->key.n == 0)
-		return;
 	l = getlang(req->lang);
-	dict = l ? l->dict : nil;
-	if(dict == nil)
+	if(req->key.n == 0 || l == nil ||
+	   trielookup(l->dict, &req->key, &p, &vlen) != TrieExact)
 		return;
-	n = hmapget(dict, &req->key);
-	if(n == nil || n->vlen == 0)
-		return;
-	p = n->val;
-	e = p + n->vlen;
+	e = p + vlen;
 	while(res->nkouho < Maxkouho && p < e){
 		while(p < e && *p == ' ')
 			p++;
-		if(p == e)
-			break;
 		sp = p;
 		while(p < e && *p != ' ')
 			p++;
-		if(!sinit(&tmp, sp, p - sp)){
-			res->nkouho = 0;
-			return;
-		}
-		if(req->lang == LangEMOJI || scmp(&tmp, &req->key) != 0)
+		if(sinit(&tmp, sp, p - sp) && tmp.n > 0 &&
+		   (req->lang == LangEMOJI || scmp(&tmp, &req->key) != 0))
 			res->kouho[res->nkouho++] = tmp;
-		if(p < e)
-			p++;
 	}
 }

@@ -74,70 +49,29 @@ dictthread(void*)
 	}
 }

-static Hmap*
-dictopen(char *path)
+static Trie*
+langopen(char *dir, char *name, char *ext)
 {
-	Hmap *h;
-	Biobuf *b;
-	Str key, tmp;
-	char *line, *tab, *p, *e;
-	int len, lineno;
+	char *path;
+	Trie *t;

-	b = Bopen(path, OREAD);
-	if(b == nil)
-		die("can't open %s: %s", path, strerror(errno));
-	h = hmapalloc(4096);
-	lineno = 0;
-	while((line = readline(b, path)) != nil){
-		lineno++;
-		len = Blinelen(b);
-		if(memchr(line, '\0', len) != nil)
-			die("NUL in dictionary: %s:%d", path, lineno);
-		if(len > 0 && line[len-1] == '\r')
-			line[--len] = '\0';
-		if(len == 0 || line[0] == ';'){
-			free(line);
-			continue;
-		}
-		tab = memchr(line, '\t', len);
-		if(tab == nil || tab == line || tab >= line + len - 1 ||
-		   memchr(tab+1, '\t', line+len-(tab+1)) != nil)
-			die("malformed dictionary: %s:%d", path, lineno);
-		*tab = '\0';
-		if(!sinit(&key, line, tab-line))
-			die("invalid or oversized dictionary key: %s:%d",
-				path, lineno);
-		for(p = tab+1; p < line+len; p = e+1){
-			e = memchr(p, ' ', line+len-p);
-			if(e == nil)
-				e = line+len;
-			if(!sinit(&tmp, p, e-p))
-				die("invalid or oversized dictionary candidate: %s:%d",
-					path, lineno);
-			if(e == line+len)
-				break;
-		}
-		hmapset(&h, &key, tab+1, len - (tab - line) - 1);
-		free(line);
-	}
-	if(Bterm(b) < 0)
-		die("can't close %s: %s", path, strerror(errno));
-	return h;
+	path = smprint("%s/%s.%s", dir, name, ext);
+	if(path == nil)
+		die("out of memory");
+	t = trieopen(path);
+	free(path);
+	return t;
 }

 void
-dictinit(char *dir)
+langinit(char *dir)
 {
-	char *path;
-	int i;
+	Lang *l;

-	for(i = 0; i < nlang; i++){
-		if(langs[i].dictname == nil)
-			continue;
-		path = smprint("%s/%s.dict", dir, langs[i].dictname);
-		if(path == nil)
-			die("out of memory");
-		langs[i].dict = dictopen(path);
-		free(path);
+	for(l = langs; l < langs + nlang; l++){
+		if(l->mapname != nil)
+			l->map = langopen(dir, l->mapname, "map");
+		if(l->dictname != nil)
+			l->dict = langopen(dir, l->dictname, "dict");
 	}
 }
--- a/fn.h
+++ b/fn.h
@@ -9,19 +9,16 @@ int	scmp(Str*, Str*);
 int	stoutf(Str*, char*, int);
 Rune	slastr(Str*);

-Hmap*	hmapalloc(int);
-void	hmapfree(Hmap*);
-void	hmapset(Hmap**, Str*, const char*, int);
-Hnode*	hmapget(Hmap*, Str*);
 int	mapget(Trie*, Str*, Str*);

+Trie*	trienew(void);
+void	trieput(Trie*, char*, int, char*, int);
 Trie*	trieopen(char*);
 void	trieclose(Trie*);
-int	trielookup(Trie*, char*, int, char**, int*);
+int	trielookup(Trie*, Str*, char**, int*);

 Lang*	getlang(int);
-void	mapinit(char*);
-void	dictinit(char*);
+void	langinit(char*);

 void	dictthread(void*);
 void	dictlookup(Dictreq*, Dictres*);
--- a/hash.c
+++ b/hash.c
@@ -1,175 +0,0 @@
-#include <limits.h>
-#include "dat.h"
-#include "fn.h"
-
-enum {
-	Tagsize = sizeof(Hnode),
-};
-
-static uvlong
-hash(Str *s)
-{
-	uvlong h;
-	int i;
-
-	h = 7;
-	for(i = 0; i < s->n; i++)
-		h = h*31 + s->r[i];
-	return h;
-}
-
-Hmap*
-hmapalloc(int nbuckets)
-{
-	void *store;
-	Hmap *h;
-	int nsz;
-
-	if(nbuckets < 1)
-		return nil;
-	nsz = Tagsize;
-	if((ulong)nbuckets > (ULONG_MAX-sizeof(*h))/(ulong)nsz)
-		return nil;
-	store = emalloc(sizeof(*h) + (ulong)nbuckets * nsz);
-	h = store;
-	h->nbs = nbuckets;
-	h->nsz = nsz;
-	h->len = h->cap = nbuckets;
-	h->nodes = (uchar*)store + sizeof(*h);
-	return h;
-}
-
-static int
-keycmp(Hnode *n, Str *key)
-{
-	char buf[Maxutf];
-	int len;
-
-	len = stoutf(key, buf, sizeof(buf));
-	if(n->klen != len)
-		return 1;
-	return memcmp(n->key, buf, len);
-}
-
-Hnode*
-hmapget(Hmap *h, Str *key)
-{
-	Hnode *n;
-	uchar *v;
-
-	if(h == nil || key == nil || key->n < 0 || key->n > Maxrunes)
-		return nil;
-	v = h->nodes + (hash(key) % h->nbs) * (ulong)h->nsz;
-	for(;;){
-		n = (Hnode*)v;
-		if(n->filled && keycmp(n, key) == 0)
-			return n;
-		if(n->next == 0)
-			break;
-		v = h->nodes + (ulong)n->next * h->nsz;
-	}
-	return nil;
-}
-
-static char*
-sdup(Str *s, int *len)
-{
-	char buf[Maxutf];
-	char *p;
-	int n;
-
-	n = stoutf(s, buf, sizeof(buf));
-	p = emalloc((ulong)n + 1);
-	memmove(p, buf, n);
-	p[n] = '\0';
-	*len = n;
-	return p;
-}
-
-static char*
-memdup(const char *src, int n)
-{
-	char *p;
-
-	if(n == 0)
-		return nil;
-	p = emalloc((ulong)n + 1);
-	memmove(p, src, n);
-	p[n] = '\0';
-	return p;
-}
-
-void
-hmapfree(Hmap *h)
-{
-	Hnode *n;
-	int i;
-
-	if(h == nil)
-		return;
-	for(i = 0; i < h->len; i++){
-		n = (Hnode*)(h->nodes + (ulong)i * h->nsz);
-		if(!n->filled)
-			continue;
-		free(n->key);
-		free(n->val);
-	}
-	free(h);
-}
-
-void
-hmapset(Hmap **store, Str *key, const char *val, int vlen)
-{
-	char *newval;
-	Hnode *n;
-	uchar *v;
-	Hmap *h;
-	int next;
-	ulong diff;
-
-	if(store == nil || *store == nil || key == nil ||
-	   key->n < 0 || key->n > Maxrunes || vlen < 0 ||
-	   (vlen > 0 && val == nil))
-		return;
-	newval = memdup(val, vlen);
-	h = *store;
-	v = h->nodes + (hash(key) % h->nbs) * (ulong)h->nsz;
-	for(;;){
-		n = (Hnode*)v;
-		next = n->next;
-		if(n->filled == 0)
-			goto replace;
-		if(keycmp(n, key) == 0)
-			goto replace;
-		if(next == 0)
-			break;
-		v = h->nodes + (ulong)next * h->nsz;
-	}
-	if(h->cap == h->len){
-		diff = v - h->nodes;
-		if(h->cap > INT_MAX/2 ||
-		   (ulong)h->cap > (ULONG_MAX-sizeof(*h))/(2*(ulong)h->nsz))
-			die("hash table is too large");
-		h->cap *= 2;
-		*store = erealloc(*store,
-			sizeof(*h) + (ulong)h->cap * h->nsz);
-		h = *store;
-		h->nodes = (uchar*)*store + sizeof(*h);
-		v = h->nodes + diff;
-		n = (Hnode*)v;
-	}
-	n->next = h->len;
-	memset(h->nodes + (ulong)h->len * h->nsz, 0, h->nsz);
-	h->len++;
-	v = h->nodes + (ulong)n->next * h->nsz;
-	n = (Hnode*)v;
-replace:
-	if(n->filled == 0){
-		n->key = sdup(key, &n->klen);
-		n->filled = 1;
-	}
-	n->next = next;
-	free(n->val);
-	n->val = newval;
-	n->vlen = vlen;
-}
--- a/main.c
+++ b/main.c
@@ -58,8 +58,7 @@ threadmain(int argc, char **argv)
 	keyc = chancreate(sizeof(Keyreq), 0);
 	dictreqc = chancreate(sizeof(Dictreq), 4);
 	dictresc = chancreate(sizeof(Dictres), 0);
-	mapinit(argv[1]);
-	dictinit(argv[1]);
+	langinit(argv[1]);
 	srvinit();
 	proccreate(drawthread, nil, 16384);
 	proccreate(srvthread, nil, 16384);
--- a/strans.c
+++ b/strans.c
@@ -405,18 +405,16 @@ getlang(int lang)
 static int
 maplookup(Trie *t, Str *key, Str *out)
 {
-	char buf[Maxutf], *v;
-	int klen, match, vlen;
+	char *v;
+	int match, vlen;

-	if(out != nil)
-		sclear(out);
-	if(t == nil || key == nil || key->n == 0)
+	sclear(out);
+	if(key->n == 0)
 		return 0;
-	klen = stoutf(key, buf, sizeof(buf));
-	match = trielookup(t, buf, klen, &v, &vlen);
+	match = trielookup(t, key, &v, &vlen);
 	if(match == TrieMiss)
 		return 0;
-	if(match == TrieExact && out != nil && !sinit(out, v, vlen))
+	if(match == TrieExact && !sinit(out, v, vlen))
 		return 0;
 	return 1;
 }
@@ -1052,33 +1050,11 @@ imthread(void*)
 int
 mapget(Trie *t, Str *key, Str *out)
 {
-	char buf[Maxutf], *v;
-	int klen, match, vlen;
+	char *v;
+	int vlen;

-	if(out != nil)
-		sclear(out);
-	if(key == nil || out == nil || key->n == 0)
-		return 0;
-	klen = stoutf(key, buf, sizeof(buf));
-	match = trielookup(t, buf, klen, &v, &vlen);
-	if(match != TrieExact)
+	sclear(out);
+	if(key->n == 0 || trielookup(t, key, &v, &vlen) != TrieExact)
 		return 0;
 	return sinit(out, v, vlen);
 }
-
-void
-mapinit(char *dir)
-{
-	char *path;
-	int i;
-
-	for(i = 0; i < nelem(langs); i++){
-		if(langs[i].mapname == nil)
-			continue;
-		path = smprint("%s/%s.map", dir, langs[i].mapname);
-		if(path == nil)
-			die("out of memory");
-		langs[i].map = trieopen(path);
-		free(path);
-	}
-}
--- a/tests/Makefile
+++ b/tests/Makefile
@@ -28,11 +28,11 @@ SMOKE = ibus_live_test ibus_client_smoke gtk_live_test xim_live_test \
 	ipc_live_test
 FAULT = daemon_collision_test daemon_failure_test daemon_restart_test
 LIVE = $(SMOKE) $(FAULT)
-TESTSRC = test_util.c str_test.c hash_test.c trie_test.c \
+TESTSRC = test_util.c str_test.c trie_test.c \
 	ko_test.c vi_test.c engine_test.c dict_test.c ipc_test.c \
 	popup_test.c font_test.c ibus_test.c server_test.c xim_adapter_test.c
 TESTOBJ = $(TESTSRC:.c=.o)
-PARENTSRC = str.c hash.c trie.c dict.c ko.c vi.c ipc.c popup_layout.c \
+PARENTSRC = str.c trie.c dict.c ko.c vi.c ipc.c popup_layout.c \
 	font.c
 PARENTOBJ = $(PARENTSRC:%.c=unit_%.o)
 XIMOBJ = unit_xim_keymap.o unit_ximtext.o
--- a/tests/dict_test.c
+++ b/tests/dict_test.c
@@ -3,21 +3,20 @@
 void
 dictionary_candidates(struct ct *t)
 {
-	static char malformed[] = { 'o', 'k', ' ', (char)0x80 };
 	char many[512], item[8];
 	char *p;
 	Dictreq req;
 	Dictres res;
-	Hmap *saved;
+	Trie *saved;
 	Lang *lang;
 	Str key;
 	int i;

 	lang = getlang(LangJP);
 	saved = lang->dict;
-	lang->dict = hmapalloc(1);
+	lang->dict = trienew();
 	key = mkstr("かな");
-	hmapset(&lang->dict, &key, "  候補1  かな 候補2  ",
+	trieput(lang->dict, "かな", strlen("かな"), "  候補1  かな 候補2  ",
 		strlen("  候補1  かな 候補2  "));
 	memset(&req, 0, sizeof req);
 	req.key = key;
@@ -42,7 +41,7 @@ dictionary_candidates(struct ct *t)
 		p += strlen(item);
 	}
 	*p = '\0';
-	hmapset(&lang->dict, &key, many, strlen(many));
+	trieput(lang->dict, "key", 3, many, strlen(many));
 	req.key = key;
 	req.pre = mkstr("preedit-two");
 	req.seq = 29;
@@ -54,13 +53,8 @@ dictionary_candidates(struct ct *t)
 	CT_EQ_INT(t, 0, scmp(&req.pre, &res.key));
 	checkstr(t, "first capped candidate", "c00", &res.kouho[0]);
 	checkstr(t, "last capped candidate", "c31", &res.kouho[31]);
-	key = mkstr("malformed");
-	hmapset(&lang->dict, &key, malformed, sizeof malformed);
-	req.key = key;
-	dictlookup(&req, &res);
-	CT_EQ_INT(t, 0, res.nkouho);
 cleanup:
-	hmapfree(lang->dict);
+	trieclose(lang->dict);
 	lang->dict = saved;
 }

@@ -77,13 +71,13 @@ dictionary_misses_clear_result(struct ct *t)
 	};
 	Dictreq req;
 	Dictres res;
-	Hmap *saved;
+	Trie *saved;
 	Lang *lang;
 	int i;

 	lang = getlang(LangJP);
 	saved = lang->dict;
-	lang->dict = hmapalloc(1);
+	lang->dict = trienew();
 	for(i = 0; i < nelem(cases); i++){
 		memset(&res, 0xa5, sizeof res);
 		req.key = mkstr(cases[i].key);
@@ -98,7 +92,7 @@ dictionary_misses_clear_result(struct ct *t)
 		CT_EQ_UINT(t, 0xf00d0000U + i, res.seq);
 		checkstr(t, cases[i].name, cases[i].pre, &res.key);
 	}
-	hmapfree(lang->dict);
+	trieclose(lang->dict);
 	lang->dict = saved;
 }

@@ -107,15 +101,15 @@ dictionary_emoji_identity(struct ct *t)
 {
 	Dictreq req;
 	Dictres res;
-	Hmap *saved;
+	Trie *saved;
 	Lang *lang;
 	Str key;

 	lang = getlang(LangEMOJI);
 	saved = lang->dict;
-	lang->dict = hmapalloc(2);
+	lang->dict = trienew();
 	key = mkstr("é");
-	hmapset(&lang->dict, &key, "é", strlen("é"));
+	trieput(lang->dict, "é", strlen("é"), "é", strlen("é"));
 	memset(&req, 0, sizeof req);
 	req.key = key;
 	req.pre = key;
@@ -124,6 +118,6 @@ dictionary_emoji_identity(struct ct *t)
 	dictlookup(&req, &res);
 	if(CT_EQ_INT(t, 1, res.nkouho))
 		checkstr(t, "emoji identity candidate", "é", &res.kouho[0]);
-	hmapfree(lang->dict);
+	trieclose(lang->dict);
 	lang->dict = saved;
 }
--- a/tests/engine_test.c
+++ b/tests/engine_test.c
@@ -311,14 +311,14 @@ engine_rejects_stale_dictionary_results(struct ct *t)
 {
 	Dictreq a, b;
 	Dictres res;
-	Hmap *saved;
+	Trie *saved;
 	Lang *jp;
 	char one, two;
 	int n;

 	jp = getlang(LangJP);
 	saved = jp->dict;
-	jp->dict = hmapalloc(1);
+	jp->dict = trienew();
 	init();
 	im.l = jp;
 	while(channbrecv(dictreqc, &a) > 0)
@@ -362,7 +362,7 @@ engine_rejects_stale_dictionary_results(struct ct *t)
 cleanup:
 	while(channbrecv(dictreqc, &a) > 0)
 		;
-	hmapfree(jp->dict);
+	trieclose(jp->dict);
 	jp->dict = saved;
 }

@@ -720,7 +720,7 @@ struct Searchfix
 	Im	im;
 	Search	search;
 	Lang	*dictlang;
-	Hmap	*dict;
+	Trie	*dict;
 	int	activecap;
 	int	visible;
 	int	candidatechosen;
@@ -939,12 +939,9 @@ engine_candidate_completion(struct ct *t)
 }

 static void
-setdict(Hmap **dict, char *key, char *val)
+setdict(Trie *dict, char *key, char *val)
 {
-	Str s;
-
-	s = mkstr(key);
-	hmapset(dict, &s, val, strlen(val));
+	trieput(dict, key, strlen(key), val, strlen(val));
 }

 static void
@@ -964,22 +961,22 @@ static void
 emojibegin(Searchfix *f, int showpre)
 {
 	searchsave(f, LangEMOJI);
-	f->dictlang->dict = hmapalloc(32);
-	setdict(&f->dictlang->dict, "a", "A B");
-	setdict(&f->dictlang->dict, "ㅁ", "B C");
-	setdict(&f->dictlang->dict, "alpha", "α");
-	setdict(&f->dictlang->dict, "smil", "😀 😄");
-	setdict(&f->dictlang->dict, "smile", "😀 😄");
-	setdict(&f->dictlang->dict, "웃음", "😀 😄 😂");
-	setdict(&f->dictlang->dict, "えがお", "😀 😄 😊");
-	setdict(&f->dictlang->dict, "エガオ", "😀 😄 😊");
-	setdict(&f->dictlang->dict, "heart", "❤️");
-	setdict(&f->dictlang->dict, "coffee", "☕️");
-	setdict(&f->dictlang->dict, "^", "¹ ² ³ ⁴ ⁵ ⁶ ⁷ ⁸ ⁹ ⁽");
-	setdict(&f->dictlang->dict, "^0", "⁰");
-	setdict(&f->dictlang->dict, "_", "₁ ₂ ₃ ₄ ₅ ₆ ₇ ₈ ₉ ₍");
-	setdict(&f->dictlang->dict, "<", "← ≤ ♥ ≠");
-	setdict(&f->dictlang->dict, "many",
+	f->dictlang->dict = trienew();
+	setdict(f->dictlang->dict, "a", "A B");
+	setdict(f->dictlang->dict, "ㅁ", "B C");
+	setdict(f->dictlang->dict, "alpha", "α");
+	setdict(f->dictlang->dict, "smil", "😀 😄");
+	setdict(f->dictlang->dict, "smile", "😀 😄");
+	setdict(f->dictlang->dict, "웃음", "😀 😄 😂");
+	setdict(f->dictlang->dict, "えがお", "😀 😄 😊");
+	setdict(f->dictlang->dict, "エガオ", "😀 😄 😊");
+	setdict(f->dictlang->dict, "heart", "❤️");
+	setdict(f->dictlang->dict, "coffee", "☕️");
+	setdict(f->dictlang->dict, "^", "¹ ² ³ ⁴ ⁵ ⁶ ⁷ ⁸ ⁹ ⁽");
+	setdict(f->dictlang->dict, "^0", "⁰");
+	setdict(f->dictlang->dict, "_", "₁ ₂ ₃ ₄ ₅ ₆ ₇ ₈ ₉ ₍");
+	setdict(f->dictlang->dict, "<", "← ≤ ♥ ≠");
+	setdict(f->dictlang->dict, "many",
 		"c1 c2 c3 c4 c5 c6 c7 c8 c9 c10 c11 c12");
 	init();
 	im.l = getlang(LangKO);
@@ -989,11 +986,11 @@ emojibegin(Searchfix *f, int showpre)
 static void
 searchend(Searchfix *f)
 {
-	Hmap *dict;
+	Trie *dict;

 	dict = f->dictlang->dict;
 	f->dictlang->dict = f->dict;
-	hmapfree(dict);
+	trieclose(dict);
 	draindraw(nil);
 	im = f->im;
 	search = f->search;
@@ -1079,12 +1076,12 @@ static void
 hanjabegin(Searchfix *f, int lang)
 {
 	searchsave(f, LangHANJA);
-	f->dictlang->dict = hmapalloc(8);
-	setdict(&f->dictlang->dict, "한", "漢 韓");
-	setdict(&f->dictlang->dict, "가", "家");
-	setdict(&f->dictlang->dict, "ㄱ", "假");
-	setdict(&f->dictlang->dict, "한글", "文");
-	setdict(&f->dictlang->dict, "か", "仮");
+	f->dictlang->dict = trienew();
+	setdict(f->dictlang->dict, "한", "漢 韓");
+	setdict(f->dictlang->dict, "가", "家");
+	setdict(f->dictlang->dict, "ㄱ", "假");
+	setdict(f->dictlang->dict, "한글", "文");
+	setdict(f->dictlang->dict, "か", "仮");
 	init();
 	im.l = getlang(lang);
 }
@@ -1156,15 +1153,14 @@ engine_japanese_candidates(struct ct *t)
 {
 	Dictreq req;
 	Dictres res;
-	Hmap *saved;
+	Trie *saved;
 	Lang *jp;
-	Str com, key, shown;
+	Str com, shown;

 	jp = getlang(LangJP);
 	saved = jp->dict;
-	jp->dict = hmapalloc(8);
-	key = mkstr("かんじ");
-	hmapset(&jp->dict, &key, "漢字 幹事", strlen("漢字 幹事"));
+	jp->dict = trienew();
+	setdict(jp->dict, "かんじ", "漢字 幹事");
 	init();
 	im.l = jp;
 	sclear(&com);
@@ -1189,7 +1185,7 @@ engine_japanese_candidates(struct ct *t)
 	checkstr(t, "selected Kanji", "幹事", &com);
 cleanup:
 	newestrequest(&req);
-	hmapfree(jp->dict);
+	trieclose(jp->dict);
 	jp->dict = saved;
 }

--- a/tests/hash_test.c
+++ b/tests/hash_test.c
@@ -1,112 +0,0 @@
-#include "test.h"
-
-void
-hmap_set_replace_and_grow(struct ct *t)
-{
-	char keybuf[16], valbuf[16], source[] = "copied";
-	Hmap *h;
-	Hnode *n;
-	Str key;
-	int i;
-
-	h = hmapalloc(1);
-	for(i = 0; i < 4; i++){
-		snprint(keybuf, sizeof keybuf, "key%d", i);
-		snprint(valbuf, sizeof valbuf, "value%d", i);
-		key = mkstr(keybuf);
-		hmapset(&h, &key, valbuf, strlen(valbuf));
-	}
-	for(i = 0; i < 4; i++){
-		snprint(keybuf, sizeof keybuf, "key%d", i);
-		snprint(valbuf, sizeof valbuf, "value%d", i);
-		key = mkstr(keybuf);
-		n = hmapget(h, &key);
-		if(n == nil || strcmp(n->val, valbuf) != 0)
-			CT_ERRORF(t, "%s: collision chain lost value", keybuf);
-	}
-	key = mkstr("key1");
-	hmapset(&h, &key, source, strlen(source));
-	source[0] = 'X';
-	n = hmapget(h, &key);
-	if(!CT_CHECK(t, n != nil))
-		goto cleanup;
-	CT_EQ_STR(t, "copied", n->val);
-	key = mkstr("key2");
-	n = hmapget(h, &key);
-	if(!CT_CHECK(t, n != nil))
-		goto cleanup;
-	CT_EQ_STR(t, "value2", n->val);
-	key = mkstr("key1");
-	hmapset(&h, &key, nil, 0);
-	n = hmapget(h, &key);
-	if(!CT_CHECK(t, n != nil))
-		goto cleanup;
-	CT_EQ_INT(t, 0, n->vlen);
-	CT_EQ_PTR(t, nil, n->val);
-	key = mkstr("missing");
-	CT_EQ_PTR(t, nil, hmapget(h, &key));
-cleanup:
-	hmapfree(h);
-}
-
-void
-hmap_long_utf8_keys(struct ct *t)
-{
-	Hmap *h;
-	Hnode *n;
-	Str a, b;
-	int i;
-
-	for(i = 0; i < Maxrunes-1; i++)
-		a.r[i] = b.r[i] = 0x1f600;
-	a.r[Maxrunes-1] = 0x1f601;
-	b.r[Maxrunes-1] = 0x1f602;
-	a.n = b.n = Maxrunes;
-	h = hmapalloc(1);
-	hmapset(&h, &a, "first", 5);
-	hmapset(&h, &b, "second", 6);
-	n = hmapget(h, &a);
-	if(!CT_CHECK(t, n != nil))
-		goto cleanup;
-	CT_EQ_STR(t, "first", n != nil ? n->val : nil);
-	n = hmapget(h, &b);
-	if(!CT_CHECK(t, n != nil))
-		goto cleanup;
-	CT_EQ_STR(t, "second", n != nil ? n->val : nil);
-cleanup:
-	hmapfree(h);
-}
-
-void
-hmap_binary_keys_and_invalid_lengths(struct ct *t)
-{
-	Hmap *h;
-	Hnode *n;
-	Str key, other;
-
-	h = hmapalloc(1);
-	CT_CHECK(t, h != nil);
-	key.n = 3;
-	key.r[0] = 'a';
-	key.r[1] = 0;
-	key.r[2] = 'b';
-	other = key;
-	other.r[2] = 'c';
-	hmapset(&h, &key, "one", 3);
-	hmapset(&h, &other, "two", 3);
-	n = hmapget(h, &key);
-	if(CT_CHECK(t, n != nil)){
-		CT_EQ_INT(t, 3, n->klen);
-		CT_EQ_MEM(t, "one", n->val, n->vlen);
-	}
-	n = hmapget(h, &other);
-	if(CT_CHECK(t, n != nil))
-		CT_EQ_MEM(t, "two", n->val, n->vlen);
-	hmapset(&h, &key, "bad", -1);
-	hmapset(&h, &key, nil, 1);
-	n = hmapget(h, &key);
-	if(CT_CHECK(t, n != nil))
-		CT_EQ_MEM(t, "one", n->val, n->vlen);
-	CT_EQ_PTR(t, nil, hmapalloc(0));
-	hmapfree(h);
-}
--- a/tests/test.h
+++ b/tests/test.h
@@ -12,11 +12,8 @@ void	str_init_utf8(struct ct*);
 void	str_edit_and_alias(struct ct*);
 void	str_utf8_capacity(struct ct*);
 void	str_invalid_and_full_appends(struct ct*);
-void	hmap_set_replace_and_grow(struct ct*);
-void	hmap_long_utf8_keys(struct ct*);
-void	hmap_binary_keys_and_invalid_lengths(struct ct*);
 void	trie_exact_prefix_and_duplicate(struct ct*);
-void	trie_optional_outputs_and_invalid_lengths(struct ct*);
+void	trie_put_and_unloaded(struct ct*);
 void	popup_layout(struct ct*);
 void	font_render(struct ct*);
 void	production_maps_load(struct ct*);
--- a/tests/trie_test.c
+++ b/tests/trie_test.c
@@ -5,26 +5,26 @@ trie_exact_prefix_and_duplicate(struct ct *t)
 {
 	static const struct {
 		char *key;
-		int klen;
 		int match;
 		char *want;
 	} cases[] = {
-		{ "a", 1, TrieExact, "alpha" },
-		{ "dupli", 5, TriePrefix, nil },
-		{ "duplicate", 9, TrieExact, "second" },
-		{ "missing", 7, TrieMiss, nil },
-		{ nil, 1, TrieMiss, nil },
-		{ "a", -1, TrieMiss, nil },
+		{ "a", TrieExact, "alpha" },
+		{ "ab", TrieExact, "beta" },
+		{ "한", TrieExact, "값" },
+		{ "dupli", TriePrefix, nil },
+		{ "duplicate", TrieExact, "second" },
+		{ "missing", TrieMiss, nil },
+		{ "", TriePrefix, nil },
 	};
 	char *v;
 	Trie *trie;
+	Str key;
 	int i, match, n;

 	trie = trieopen("data/trie.map");
 	for(i = 0; i < nelem(cases); i++){
-		v = "unchanged";
-		n = 77;
-		match = trielookup(trie, cases[i].key, cases[i].klen, &v, &n);
+		key = mkstr(cases[i].key);
+		match = trielookup(trie, &key, &v, &n);
 		if(match != cases[i].match){
 			CT_ERRORF(t, "case %d: want match %d, got %d",
 				i, cases[i].match, match);
@@ -42,22 +42,28 @@ trie_exact_prefix_and_duplicate(struct ct *t)
 }

 void
-trie_optional_outputs_and_invalid_lengths(struct ct *t)
+trie_put_and_unloaded(struct ct *t)
 {
 	char *v;
 	Trie *trie;
+	Str key;
 	int n;

-	trie = trieopen("data/trie.map");
-	CT_EQ_INT(t, TrieExact, trielookup(trie, "a", 1, &v, nil));
-	CT_CHECK(t, v != nil);
-	CT_EQ_INT(t, TrieExact, trielookup(trie, "a", 1, nil, &n));
-	CT_EQ_INT(t, 5, n);
-	CT_EQ_INT(t, TriePrefix,
-		trielookup(trie, "dupli", 5, nil, nil));
-	CT_EQ_INT(t, TriePrefix, trielookup(trie, nil, 0, &v, &n));
-	CT_EQ_PTR(t, nil, v);
-	CT_EQ_INT(t, 0, n);
+	key = mkstr("k");
+	CT_EQ_INT(t, TrieMiss, trielookup(nil, &key, &v, &n));
+	trie = trienew();
+	CT_EQ_INT(t, TrieMiss, trielookup(trie, &key, &v, &n));
+	trieput(trie, "k", 1, "one two", 7);
+	trieput(trie, "ka", 2, "", 0);
+	if(CT_EQ_INT(t, TrieExact, trielookup(trie, &key, &v, &n)))
+		CT_EQ_MEM(t, "one two", v, 7);
+	key = mkstr("ka");
+	if(CT_EQ_INT(t, TrieExact, trielookup(trie, &key, &v, &n)))
+		CT_EQ_INT(t, 0, n);
+	trieput(trie, "k", 1, "three", 5);
+	key = mkstr("k");
+	if(CT_EQ_INT(t, TrieExact, trielookup(trie, &key, &v, &n)))
+		CT_EQ_MEM(t, "three", v, 5);
 	trieclose(trie);
 }

--- a/tests/unit_test.c
+++ b/tests/unit_test.c
@@ -67,11 +67,8 @@ static const struct ct_test tests[] = {
 	{ "str/edit-and-alias", str_edit_and_alias },
 	{ "str/utf8-capacity", str_utf8_capacity },
 	{ "str/invalid-full-appends", str_invalid_and_full_appends },
-	{ "hmap/set-replace-grow", hmap_set_replace_and_grow },
-	{ "hmap/long-utf8-keys", hmap_long_utf8_keys },
-	{ "hmap/binary-invalid-lengths", hmap_binary_keys_and_invalid_lengths },
 	{ "trie/exact-prefix-duplicate", trie_exact_prefix_and_duplicate },
-	{ "trie/optional-invalid-lengths", trie_optional_outputs_and_invalid_lengths },
+	{ "trie/put-and-unloaded", trie_put_and_unloaded },
 	{ "popup/layout", popup_layout },
 	{ "font/render", font_render },
 	{ "map/production-lifecycle", production_maps_load },
--- a/trie.c
+++ b/trie.c
@@ -1,33 +1,21 @@
 #include <errno.h>
-#include <limits.h>
 #include "dat.h"
 #include "fn.h"

-static char*
-readline(Biobuf *b, char *path)
-{
-	char *line;
-
-	errno = 0;
-	line = Brdstr(b, '\n', 1);
-	if(errno != 0)
-		die("can't read %s: %s", path, strerror(errno));
-	return line;
-}
+/*
+ * A byte trie over UTF-8 keys.  Maps need prefix matches while composing;
+ * dictionaries need exact matches; both are small enough for one structure.
+ * Node 0 is the root.
+ */

 static int
 newnode(Trie *t)
 {
-	int cap;
 	int i;

-	if(t->n >= t->cap){
-		if(t->cap > INT_MAX/2 ||
-		   (ulong)t->cap > ULONG_MAX/(2*sizeof(Tnode)))
-			die("map is too large");
-		cap = t->cap * 2;
-		t->nodes = erealloc(t->nodes, (ulong)cap * sizeof(Tnode));
-		t->cap = cap;
+	if(t->n == t->cap){
+		t->cap *= 2;
+		t->nodes = erealloc(t->nodes, t->cap * sizeof(Tnode));
 	}
 	i = t->n++;
 	memset(&t->nodes[i], 0, sizeof(Tnode));
@@ -59,13 +47,25 @@ add(Trie *t, int ni, char c)
 	return pi;
 }

-static void
-insert(Trie *t, char *key, int klen, char *val, int vlen)
+Trie*
+trienew(void)
 {
-	int ni, ci;
-	int i;
+	Trie *t;

-	ni = t->root;
+	t = emalloc(sizeof(*t));
+	t->cap = 1024;
+	t->nodes = emalloc(t->cap * sizeof(Tnode));
+	t->n = 0;
+	newnode(t);
+	return t;
+}
+
+void
+trieput(Trie *t, char *key, int klen, char *val, int vlen)
+{
+	int ni, ci, i;
+
+	ni = 0;
 	for(i = 0; i < klen; i++){
 		ci = find(t, ni, key[i]);
 		if(ci < 0)
@@ -73,57 +73,70 @@ insert(Trie *t, char *key, int klen, char *val, int vlen)
 		ni = ci;
 	}
 	free(t->nodes[ni].val);
-	t->nodes[ni].val = emalloc((ulong)vlen + 1);
+	t->nodes[ni].val = emalloc(vlen + 1);
 	memmove(t->nodes[ni].val, val, vlen);
 	t->nodes[ni].val[vlen] = '\0';
 	t->nodes[ni].vlen = vlen;
 }

+static char*
+readline(Biobuf *b, char *path)
+{
+	char *line;
+
+	errno = 0;
+	line = Brdstr(b, '\n', 1);
+	if(errno != 0)
+		die("can't read %s: %r", path);
+	return line;
+}
+
+/*
+ * A file holds "key<tab>value" lines; ';' starts a comment.  Keys and each
+ * space-separated word of a value must fit a Str, which is what lookups
+ * hand back.
+ */
 Trie*
 trieopen(char *path)
 {
 	Trie *t;
 	Biobuf *b;
-	char *line, *tab, *key, *val;
 	Str s;
-	int klen, vlen;
+	char *e, *line, *p, *tab;
+	int len, lineno;

 	b = Bopen(path, OREAD);
 	if(b == nil)
-		die("can't open %s: %s", path, strerror(errno));
-	t = emalloc(sizeof(*t));
-	t->cap = 1024;
-	t->nodes = emalloc((ulong)t->cap * sizeof(Tnode));
-	t->n = 0;
-	t->root = newnode(t);
-	while((line = readline(b, path)) != nil){
-		vlen = Blinelen(b);
-		if(memchr(line, '\0', vlen) != nil)
-			die("NUL in map: %s", path);
-		if(vlen > 0 && line[vlen-1] == '\r')
-			line[--vlen] = '\0';
-		if(line[0] == '\0' || line[0] == ';'){
+		die("can't open %s: %r", path);
+	t = trienew();
+	for(lineno = 1; (line = readline(b, path)) != nil; lineno++){
+		len = Blinelen(b);
+		if(memchr(line, '\0', len) != nil)
+			die("%s:%d: NUL byte", path, lineno);
+		if(len > 0 && line[len-1] == '\r')
+			line[--len] = '\0';
+		if(len == 0 || line[0] == ';'){
 			free(line);
 			continue;
 		}
-		tab = memchr(line, '\t', vlen);
-		if(tab == nil || tab == line || tab == line+vlen-1 ||
-		   memchr(tab+1, '\t', line+vlen-(tab+1)) != nil)
-			die("malformed map: %s", path);
-		*tab = '\0';
-		key = line;
-		klen = tab - line;
-		if(!sinit(&s, key, klen))
-			die("invalid or oversized map key: %s", path);
-		val = tab + 1;
-		vlen = line + vlen - val;
-		if(!sinit(&s, val, vlen))
-			die("invalid or oversized map value: %s", path);
-		insert(t, key, klen, val, vlen);
+		tab = memchr(line, '\t', len);
+		if(tab == nil || tab == line || tab == line+len-1 ||
+		   memchr(tab+1, '\t', line+len-(tab+1)) != nil)
+			die("%s:%d: expected key, tab, value", path, lineno);
+		if(!sinit(&s, line, tab-line))
+			die("%s:%d: invalid or oversized key", path, lineno);
+		for(p = tab+1; p < line+len; p = e+1){
+			e = memchr(p, ' ', line+len-p);
+			if(e == nil)
+				e = line+len;
+			if(!sinit(&s, p, e-p))
+				die("%s:%d: invalid or oversized value", path, lineno);
+		}
+		trieput(t, line, tab-line, tab+1, line+len-(tab+1));
 		free(line);
 	}
 	if(Bterm(b) < 0)
-		die("can't close %s: %s", path, strerror(errno));
+		die("can't close %s: %r", path);
 	return t;
 }

@@ -140,30 +153,27 @@ trieclose(Trie *t)
 	free(t);
 }

+/* A nil trie is an unloaded map: every key misses. */
 int
-trielookup(Trie *t, char *key, int klen, char **val, int *vlen)
+trielookup(Trie *t, Str *key, char **val, int *vlen)
 {
-	int ni;
-	int i;
+	char buf[Maxutf];
+	int i, klen, ni;

-	if(val != nil)
-		*val = nil;
-	if(vlen != nil)
-		*vlen = 0;
-	if(t == nil || klen < 0 || (klen > 0 && key == nil))
+	*val = nil;
+	*vlen = 0;
+	if(t == nil)
 		return TrieMiss;
-	ni = t->root;
+	klen = stoutf(key, buf, sizeof buf);
+	ni = 0;
 	for(i = 0; i < klen; i++){
-		ni = find(t, ni, key[i]);
+		ni = find(t, ni, buf[i]);
 		if(ni < 0)
 			return TrieMiss;
 	}
-	if(t->nodes[ni].val != nil){
-		if(val != nil)
-			*val = t->nodes[ni].val;
-		if(vlen != nil)
-			*vlen = t->nodes[ni].vlen;
-		return TrieExact;
-	}
-	return TriePrefix;
+	if(t->nodes[ni].val == nil)
+		return TriePrefix;
+	*val = t->nodes[ni].val;
+	*vlen = t->nodes[ni].vlen;
+	return TrieExact;
 }