term: move the term core to subprojects/term

The core now builds, tests, and benchmarks alone, as a Meson subproject
with no dependency on the main tree.

- include/term/ holds term.h and the header-only unicode/. src/ holds
  the core and its private headers. The main tree gets only
  include/term through term_dep, so it cannot reach a private header.
- The term tests, the term benchmarks, the Kitty parser suite, and
  tools/amalgamate.py and tools/generate_unicode.py move with it. The
  tests run as term:<name>.
- The main build passes its -march (arch_flag) and b_ndebug to the
  subproject. A directory configured before this commit needs
  meson configure -Dterm:arch_flag=-march=... once.
- The term tests compile with -UNDEBUG. They do work inside assert.
- The subproject exports term_test_dep with term_api.h, for the main
  tree's render, thumbnail, color, and selection tests.
- The Kitty bridge used struct vector2i and has not compiled since the
  public API redesign. It builds again. 2 of its 14 tests fail: REP
  after a tab, and a cursor report after a resize.
- dirgraph.py drops the term and unicode layers. Its self-test uses pty.
- The symbols test ignores the __odr_asan symbols that ASan adds.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

1 files changed, 227 insertions, 0 deletions

diff --git a/tests/gstore_test.c b/tests/gstore_test.c
new file mode 100644
index 0000000..637b136
--- /dev/null
+++ b/tests/gstore_test.c
@@ -0,0 +1,227 @@
+/*
+ * Tests for the per-terminal grapheme store (gstore).
+ *
+ * Verifies:
+ * - Two terminals do not share clusters: freeing one leaves the other intact.
+ * - Clusters that remain visible on screen survive GC sweeps.
+ * - More than 65,536 distinct clusters can be interned as old ones scroll
+ *   out of history, without reaching any cap or exhausting memory.
+ * - Direct unit tests of intern, expand, deduplication, and sweep.
+ */
+#include "gstore.h"
+#include "internal.h"
+#include "term/term.h"
+#include "term/unicode/utf8.h"
+
+#include <assert.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+static void test_direct_unit(void)
+{
+    struct gstore *g = term_gstore_new(NULL);
+    assert(g != NULL);
+    assert(term_gstore_count(g) == 0);
+
+    /* Single code point returns the code point itself. */
+    uint32_t single = 'A';
+    assert(term_gstore_intern(g, &single, 1) == 'A');
+    assert(term_gstore_count(g) == 0);
+
+    /* Cluster returns a handle. */
+    uint32_t c1[2] = { 'e', 0x0301 };
+    uint32_t h1 = term_gstore_intern(g, c1, 2);
+    assert(grapheme_is_handle(h1));
+    assert(term_gstore_count(g) == 1);
+
+    /* Identical cluster deduplicates to the same handle. */
+    uint32_t h1_dup = term_gstore_intern(g, c1, 2);
+    assert(h1_dup == h1);
+    assert(term_gstore_count(g) == 1);
+
+    /* Different cluster gets a distinct handle. */
+    uint32_t c2[2] = { 'a', 0x0300 };
+    uint32_t h2 = term_gstore_intern(g, c2, 2);
+    assert(grapheme_is_handle(h2));
+    assert(h2 != h1);
+    assert(term_gstore_count(g) == 2);
+
+    /* Expand both handles. */
+    uint32_t out[GRAPHEME_MAX_LEN] = {0};
+    int n = term_gstore_expand(g, h1, out, GRAPHEME_MAX_LEN);
+    assert(n == 2);
+    assert(out[0] == 'e' && out[1] == 0x0301);
+
+    n = term_gstore_expand(g, h2, out, GRAPHEME_MAX_LEN);
+    assert(n == 2);
+    assert(out[0] == 'a' && out[1] == 0x0300);
+
+    /* Sweep with h1 marked and h2 unmarked. */
+    term_gstore_mark(g, h1);
+    term_gstore_sweep(g, NULL);
+    assert(term_gstore_count(g) == 1);
+
+    /* h1 is still valid. */
+    n = term_gstore_expand(g, h1, out, GRAPHEME_MAX_LEN);
+    assert(n == 2);
+    assert(out[0] == 'e' && out[1] == 0x0301);
+
+    /* h2 was reclaimed. */
+    n = term_gstore_expand(g, h2, out, GRAPHEME_MAX_LEN);
+    assert(n == 1 && out[0] == 0xFFFD);
+
+    /* New cluster can reuse the reclaimed slot. */
+    uint32_t c3[3] = { 'o', 0x0308, 0x0301 };
+    uint32_t h3 = term_gstore_intern(g, c3, 3);
+    assert(grapheme_is_handle(h3));
+    assert(term_gstore_count(g) == 2);
+    n = term_gstore_expand(g, h3, out, GRAPHEME_MAX_LEN);
+    assert(n == 3);
+    assert(out[0] == 'o' && out[1] == 0x0308 && out[2] == 0x0301);
+
+    term_gstore_free(g);
+}
+
+static void test_two_terminals_isolation(void)
+{
+    struct term_options opt = { .size = { .x = 20, .y = 4 } };
+    struct term *t1 = term_new(&opt);
+    struct term *t2 = term_new(&opt);
+    assert(t1 && t2);
+
+    const char *cluster_text = "e\xcc\x81"; /* 'e' with acute */
+    term_feed(t1, cluster_text, strlen(cluster_text), 0);
+    term_feed(t2, cluster_text, strlen(cluster_text), 0);
+
+    /* Free terminal 1. */
+    term_free(t1);
+
+    /* Terminal 2 must still expand its cluster correctly. */
+    struct term_snapshot *snap = term_snapshot_new();
+    assert(snap);
+    assert(term_snapshot_update(snap, t2));
+
+    const struct term_cell *row = term_snapshot_row(snap, 0);
+    assert(row[0].text & TERM_TEXT_CLUSTER);
+
+    uint32_t cps[GRAPHEME_MAX_LEN] = {0};
+    int n = term_snapshot_cluster(snap, row[0].text, cps, GRAPHEME_MAX_LEN);
+    assert(n == 2);
+    assert(cps[0] == 'e' && cps[1] == 0x0301);
+
+    term_snapshot_free(snap);
+    term_free(t2);
+}
+
+static void test_screen_clusters_survive_sweep(void)
+{
+    struct term_options opt = { .size = { .x = 40, .y = 10 }, .scrollback = 10 };
+    struct term *t = term_new(&opt);
+    assert(t);
+
+    /* Write 5 clusters on the top row. */
+    term_feed(t, "a\xcc\x80" "b\xcc\x81" "c\xcc\x82" "d\xcc\x83" "e\xcc\x84", 15, 0);
+
+    /* Write ephemeral clusters on row 1, then overwrite them. */
+    for (int i = 0; i < 50; i++) {
+        term_feed(t, "\033[2;1H\033[2K", 8, 0);
+        char buf[32];
+        int len = snprintf(buf, sizeof(buf), "z\xcc\x80\xcc\x81%d", i);
+        term_feed(t, buf, (size_t)len, 0);
+    }
+    term_feed(t, "\033[2;1H\033[2K", 8, 0);
+
+    /* Explicit sweep. The clusters on row 0 are on screen and must survive. */
+    term_gstore_sweep(t->gstore, t);
+
+    struct term_snapshot *snap = term_snapshot_new();
+    assert(snap);
+    assert(term_snapshot_update(snap, t));
+
+    /* Row 0 of screen has "a\xcc\x80...". */
+    const struct term_cell *row = term_snapshot_row(snap, 0);
+    assert(row[0].text & TERM_TEXT_CLUSTER);
+
+    uint32_t cps[GRAPHEME_MAX_LEN] = {0};
+    int n = term_snapshot_cluster(snap, row[0].text, cps, GRAPHEME_MAX_LEN);
+    assert(n == 2);
+    assert(cps[0] == 'a' && cps[1] == 0x0300);
+
+    n = term_snapshot_cluster(snap, row[1].text, cps, GRAPHEME_MAX_LEN);
+    assert(n == 2);
+    assert(cps[0] == 'b' && cps[1] == 0x0301);
+
+    n = term_snapshot_cluster(snap, row[2].text, cps, GRAPHEME_MAX_LEN);
+    assert(n == 2);
+    assert(cps[0] == 'c' && cps[1] == 0x0302);
+
+    term_snapshot_free(snap);
+    term_free(t);
+}
+
+static void test_more_than_65536_clusters(void)
+{
+    /* Bounded history so old lines scroll out. */
+    struct term_options opt = { .size = { .x = 80, .y = 24 }, .scrollback = 10 };
+    struct term *t = term_new(&opt);
+    assert(t);
+
+    /*
+     * Generate 66,000 distinct clusters.
+     * Base 'a' (0x61) + mark1 (0x0300..0x034F, 80 options) + mark2 (0x0350..0x036F + 0xFE00..0xFE0F, 830 options)
+     * 256 * 260 = 66,560 distinct clusters.
+     */
+    char buf[64];
+    uint32_t total = 0;
+    const uint32_t target = 66000;
+
+    for (uint32_t i = 0; total < target; i++) {
+        char base = (char)('a' + (i % 26));
+        uint32_t cp1 = 0x0300 + ((i / 26) % 55);
+        uint32_t cp2 = 0x0340 + ((i / (26 * 55)) % 47);
+
+        size_t len = 0;
+        buf[len++] = base;
+        len += (size_t)utf8_encode(cp1, buf + len);
+        len += (size_t)utf8_encode(cp2, buf + len);
+        buf[len++] = '\r';
+        buf[len++] = '\n';
+
+        term_feed(t, buf, len, 0);
+        total++;
+    }
+
+    assert(total >= 65536);
+
+    /* Verify the latest cluster on screen reads back correctly. */
+    struct term_snapshot *snap = term_snapshot_new();
+    assert(snap);
+    assert(term_snapshot_update(snap, t));
+
+    struct term_cursor c = term_cursor(t);
+    int y = c.pos.y > 0 ? c.pos.y - 1 : 0;
+    const struct term_cell *row = term_snapshot_row(snap, y);
+    assert(row[0].text & TERM_TEXT_CLUSTER);
+
+    uint32_t cps[GRAPHEME_MAX_LEN] = {0};
+    int n = term_snapshot_cluster(snap, row[0].text, cps, GRAPHEME_MAX_LEN);
+    assert(n == 3);
+
+    /* Live cluster count must be small (only bounded screen and history). */
+    uint32_t count = term_gstore_count(t->gstore);
+    assert(count < 2000);
+
+    term_snapshot_free(snap);
+    term_free(t);
+}
+
+int main(void)
+{
+    test_direct_unit();
+    test_two_terminals_isolation();
+    test_screen_clusters_survive_sweep();
+    test_more_than_65536_clusters();
+    puts("PASS: gstore tests");
+    return 0;
+}