CST-agent/archive/qt-app/tools/selftest.c

/* The app's C smoke test: does the engine embed the way app/ needs it to?
 *
 * It links libcstcore.a (or the cstcore target) and runs without Qt, so the
 * C side of the GUI stays verifiable on any machine -- including CI and the
 * Android cross-build. `cst-agent-app --selftest` runs the same checks inside
 * the real process. */

#define _POSIX_C_SOURCE 200809L

#include "cst_agent.h"
#include "cst_tools.h"
#include "os_compat.h"

#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#ifdef _WIN32
#include <process.h>
#include <direct.h>
#define getpid _getpid
#else
#include <sys/stat.h>
#include <unistd.h>
#endif

static int failures;

#define CHECK(cond, what)                                                      \
  do {                                                                         \
    if (!(cond)) {                                                             \
      fprintf(stderr, "FAIL %s:%d: %s\n", __FILE__, __LINE__, what);           \
      failures++;                                                              \
    }                                                                          \
  } while (0)

/* A tool the app would implement (the Linux build uses notify-send; here it
 * only has to prove the plumbing). */
static int ctx_runs;
static int app_tool_run(const cst_tool_ctx_t *ctx, const ctl_tool_call_t *call,
                        cst_tool_result_t *r, void *ud) {
  (void)ctx;
  (void)call;
  (void)ud;
  ctx_runs++;
  r->out = strdup("app tool ran");
  r->out_len = r->out ? strlen(r->out) : 0;
  r->exit_code = 0;
  return 0;
}

static int probe_missing(char *why, size_t why_sz) {
  snprintf(why, why_sz, "the desktop service is not running");
  return -1;
}

int main(void) {
  /* The scratch workspace goes where this host can actually write: /tmp on a
   * desktop, the app's own directory on Android (where /tmp is off limits,
   * so TMPDIR is pointed at it -- see app/src/main.cpp). */
  char tmp[1024] = "";
  if (os_temp_dir(tmp, sizeof(tmp)) != 0 || !tmp[0])
    snprintf(tmp, sizeof(tmp), "%s", "/tmp");
  char ws[1200];
  snprintf(ws, sizeof(ws), "%s/cst-app-selftest-%d", tmp, (int)getpid());
#ifdef _WIN32
  _mkdir(ws); /* the engine wants its workspace to exist */
#else
  mkdir(ws, 0700);
#endif

  cst_agent_config_t cfg;
  memset(&cfg, 0, sizeof(cfg));
  snprintf(cfg.workspace_root, sizeof(cfg.workspace_root), "%s", ws);
  snprintf(cfg.session_path, sizeof(cfg.session_path), "%s/s.jsonl", ws);
  snprintf(cfg.provider, sizeof(cfg.provider), "deepseek");
  snprintf(cfg.model, sizeof(cfg.model), "deepseek-flash");
  snprintf(cfg.api_base, sizeof(cfg.api_base), "https://api.deepseek.com/v1");
  snprintf(cfg.api_key, sizeof(cfg.api_key), "test-key");
  snprintf(cfg.mode, sizeof(cfg.mode), "general");
  cfg.max_steps = 1;

  cst_agent_t *a = cst_agent_create(&cfg);
  CHECK(a != NULL, "the app can start an engine in its own process");
  if (!a) {
    fprintf(stderr, "cst-app-selftest: cannot start the engine\n");
    return 1;
  }

  /* 1. a tool that exists here (no probe) is offered to the model */
  cst_tool_def_t plain;
  memset(&plain, 0, sizeof(plain));
  plain.name = "app_ping";
  plain.desc = "A tool the app provides.";
  plain.perm = CST_TP_READ;
  plain.run_ex = app_tool_run;
  plain.owner = "app:selftest";
  CHECK(cst_tool_register(&plain) == 0, "register an app tool");
  const cst_tool_def_t *t = cst_tool_find("app_ping");
  CHECK(t != NULL, "it is findable");
  CHECK(t && cst_tool_active(cst_agent_config(a), t), "and it is offered");
  ctl_tool_call_t call;
  memset(&call, 0, sizeof(call));
  snprintf(call.name, sizeof(call.name), "app_ping");
  cst_tool_result_t res;
  cst_tool_ctx_t ctx;
  memset(&ctx, 0, sizeof(ctx));
  ctx.workspace_root = ws;
  ctx.agent = a;
  cst_tool_run(&ctx, &call, &res);
  CHECK(ctx_runs == 1, "the app callback ran");
  CHECK(res.out && strcmp(res.out, "app tool ran") == 0, "its result came back");
  cst_tool_result_free(&res);

  /* 2. a tool whose platform service is missing is never offered ... */
  cst_tool_def_t missing;
  memset(&missing, 0, sizeof(missing));
  missing.name = "app_no_service";
  missing.desc = "Needs a service that is not here.";
  missing.perm = CST_TP_ASK;
  missing.perm_name = "notify";
  missing.available = probe_missing;
  missing.run_ex = app_tool_run;
  missing.owner = "app:selftest";
  CHECK(cst_tool_register(&missing) == 0, "register a tool with a probe");
  const cst_tool_def_t *mt = cst_tool_find("app_no_service");
  char why[160] = "";
  CHECK(mt && cst_tool_available(mt, why, sizeof(why)) != 0,
        "the probe says it cannot run");
  CHECK(strstr(why, "desktop service") != NULL, "and passes its reason on");
  CHECK(mt && !cst_tool_active(cst_agent_config(a), mt),
        "so it stays out of tools[] and the prompt");
  snprintf(call.name, sizeof(call.name), "app_no_service");
  cst_tool_run(&ctx, &call, &res);
  CHECK(res.out && strstr(res.out, "cannot run here") != NULL,
        "asking for it anyway is answered with the reason");
  CHECK(ctx_runs == 1, "and its callback never ran");
  cst_tool_result_free(&res);

  /* 3. the app can drop its tools again */
  CHECK(cst_tool_unregister_owner("app:selftest") == 2, "unregister both");
  CHECK(cst_tool_find("app_ping") == NULL, "app_ping is gone");
  CHECK(cst_tool_find("read_file") != NULL, "the built-ins are untouched");

  cst_agent_destroy(a);
  {
    char cmd[1200];
    snprintf(cmd, sizeof(cmd), "rm -rf %s", ws);
    if (system(cmd) != 0)
      fprintf(stderr, "note: could not clean %s\n", ws);
  }
  if (failures) {
    fprintf(stderr, "cst-app-selftest: %d failure(s)\n", failures);
    return 1;
  }
  printf("cst-app-selftest: OK\n");
  return 0;
}