Function index_thread_fn

src/ui/http_server.c:974–1145

Index via subprocess — isolates crashes from the main process. */

/* Index via subprocess — isolates crashes from the main process. */
static void *index_thread_fn(void *arg) {
    index_job_t *job = arg;
    cbm_log_info("ui.index.start", "path", job->root_path);

    /* Use stored binary path, or try to find it */
    const char *bin = g_binary_path;
    char self_path[1024] = {0};
    if (!bin[0]) {
        cbm_http_server_resolve_binary_path(NULL, self_path, sizeof(self_path));
        bin = self_path[0] ? self_path : "codebase-memory-mcp";
    }

    char log_file[256];

    /* JSON-escape root_path and optional project name. */
    char escaped_path[2048];
    cbm_json_escape(escaped_path, (int)sizeof(escaped_path), job->root_path);
    char escaped_name[512];
    cbm_json_escape(escaped_name, (int)sizeof(escaped_name), job->project_name);
    char json_arg[4096];
    if (job->project_name[0]) {
        snprintf(json_arg, sizeof(json_arg), "{\"repo_path\":\"%s\",\"name\":\"%s\"}", escaped_path,
                 escaped_name);
    } else {
        snprintf(json_arg, sizeof(json_arg), "{\"repo_path\":\"%s\"}", escaped_path);
    }

#ifdef _WIN32
    snprintf(log_file, sizeof(log_file), "%s\\cbm_index_%d.log",
             getenv("TEMP") ? getenv("TEMP") : ".", (int)_getpid());

    /* Build command line for CreateProcess through the shared MS-CRT quoter so the
     * JSON arg's embedded quotes survive the child's argv re-parse — a naive
     * `"%s"` wrap dropped them, corrupting {"repo_path":"…"} into {repo_path:…}.
     * --index-worker: this http_server spawn is already the crash-isolation layer,
     * so the child runs indexing in-process rather than spawning its own supervisor
     * (avoids redundant process nesting). */
    char cmdline[2048];
    const char *const idx_argv[] = {bin,      "cli", "--index-worker", "index_repository",
                                    json_arg, NULL};
    if (!cbm_build_win_cmdline(cmdline, sizeof(cmdline), idx_argv)) {
        snprintf(job->error_msg, sizeof(job->error_msg), "index command line too long");
        atomic_store(&job->status, 3);
        return NULL;
    }
    /* Wide command line: CreateProcessA would re-mangle the UTF-8 repo path through the
     * ANSI code page at the spawn boundary, so a non-ASCII repo path never reaches the
     * worker intact (#423/#20). Convert and spawn via CreateProcessW. */
    wchar_t *wcmd = cbm_utf8_to_wide(cmdline);
    if (!wcmd) {
        snprintf(job->error_msg, sizeof(job->error_msg), "index command line conversion failed");
        atomic_store(&job->status, 3);
        return NULL;
    }

    cbm_log_info("ui.index.spawn", "bin", bin, "log", log_file);

    HANDLE hlog = CreateFileA(log_file, GENERIC_WRITE, FILE_SHARE_READ, NULL, CREATE_ALWAYS,