tai/taicpp/tests/mplex_flow_test.cpp

// Differential test: validate_flow_spec and PlexerFlow.run against tai.py.
// usage: mplex_flow_test <corpus.jsonl produced by mplex_flow_gen.py>
#include <unistd.h>

#include <cstdio>
#include <cstdlib>
#include <cstring>
#include <fstream>
#include <iostream>
#include <string>
#include <vector>

#include "mplex_plexer_flow.h"

using detz::mplex::Json;
using namespace detz::mplex_flow;

namespace {

class Script : public PlexerController {
 public:
  explicit Script(std::vector<Json> replies_in) : replies(std::move(replies_in)) { kind = "script"; }
  std::vector<Json> replies;
  size_t next = 0;
  Json send_keys(const std::string&, bool) override { return Json(); }
  Json read_pane(int) override { return Json(); }
  Json status() override { return Json(); }
  Json ask(const Json&, double) override {
    if (next >= replies.size()) {
      Json out = Json::object();
      out.set("status", Json::string("cancelled"));
      return out;
    }
    return replies[next++];
  }
};

}  // namespace

// Real tmux 3.6 round trip (tmux is the oracle): prints one line per observation so the same
// steps run through tai.py's TmuxPlexerController can be diffed (see mplex_tmux_oracle.py).
int tmux_mode() {
  char dir[] = "/tmp/mplex-tmux-XXXXXX";
  if (!mkdtemp(dir)) return 2;
  setenv("TMUX_TMPDIR", dir, 1);
  unsetenv("TMUX");
  if (std::system("tmux new-session -d -s t -x 80 -y 24 'cat' >/dev/null 2>&1") != 0) return 2;
  FILE* pipe = popen("tmux list-panes -t t -F '#{pane_id}'", "r");
  char buf[64] = {0};
  if (!pipe || !std::fgets(buf, sizeof buf, pipe)) return 2;
  pclose(pipe);
  std::string pane(buf);
  while (!pane.empty() && (pane.back() == '\n' || pane.back() == '\r')) pane.pop_back();
  TmuxPlexerController c(pane);
  std::printf("send %s\n", c.send_keys("hello world", true).dumps().c_str());
  usleep(300000);
  std::string content = c.read_pane(5).as_string();
  std::printf("read %s\n", Json::string(content).dumps().c_str());
  std::printf("status %s\n", c.status().dumps().c_str());
  try {
    TmuxPlexerController bad("%99");
    bad.read_pane(3);
  } catch (const PlexerError& e) {
    std::printf("error %s\n", e.what());
  }
  Json screen = Json::object();
  screen.set("kind", Json::string("weird"));
  try {
    c.ask(screen, 5);
  } catch (const PlexerError& e) {
    std::printf("ask %s\n", e.what());
  }
  screen.set("kind", Json::string("choice"));
  screen.set("title", Json::string("T"));
  screen.set("options", Json::loads("[\"a\", \"b\"]"));
  try {
    std::printf("choice %s\n", c.ask(screen, 5).dumps().c_str());
  } catch (const PlexerError& e) {
    std::printf("choice error %s\n", e.what());
  }
  screen.set("kind", Json::string("text"));
  screen.set("text", Json::string("question"));
  try {
    std::printf("text %s\n", c.ask(screen, 5).dumps().c_str());
  } catch (const PlexerError& e) {
    std::printf("text error %s\n", e.what());
  }
  if (std::system("tmux kill-server >/dev/null 2>&1") != 0) return 3;
  return 0;
}

// Drives the native controller against a running mplex server (env TAI_PLEXER_SOCKET/PANE):
// status keys, send, read, then a whole PlexerFlow app answered by the user at the terminal.
int native_mode() {
  auto controller = PlexerControllerFactory::detect();
  if (!controller) {
    std::puts("no controller");
    return 2;
  }
  Json status = controller->status();
  std::string keys;
  for (const auto& item : status.as_object()) keys += item.first + ",";
  std::printf("status-keys %s kind %s\n", keys.c_str(), controller->kind.c_str());
  std::printf("send %s\n", controller->send_keys("echo native-$((20+22))", true).dumps().c_str());
  usleep(600000);
  std::string tail = controller->read_pane(30).as_string();
  std::printf("read-has-marker %d\n", tail.find("native-42") != std::string::npos);
  Json spec = Json::loads(
      "{\"title\": \"Deploy\", \"timeout\": 30, \"screens\": [{\"id\": \"main\", \"title\": \"Where\", \"body\": \"Pick.\","
      " \"fields\": [{\"id\": \"env\", \"type\": \"choice\", \"label\": \"Environment\", \"options\": [\"staging\", \"prod\"]},"
      " {\"id\": \"tag\", \"type\": \"text\", \"label\": \"Tag\", \"initial\": \"v1\", \"show_if\": {\"env\": \"prod\"}}],"
      " \"actions\": [{\"id\": \"go\", \"label\": \"Deploy\"}, {\"id\": \"back\", \"label\": \"Back\"}]}]}");
  auto [app, err] = validate_flow_spec(spec);
  if (!err.empty()) {
    std::printf("spec error %s\n", err.c_str());
    return 2;
  }
  std::fflush(stdout);
  PlexerFlow flow(app);
  std::printf("outcome %s\n", flow.run(*controller).dumps().c_str());
  return 0;
}

int main(int argc, char** argv) {
  if (argc >= 2 && std::strcmp(argv[1], "--tmux") == 0) return tmux_mode();
  if (argc >= 2 && std::strcmp(argv[1], "--native") == 0) return native_mode();
  if (argc < 2) {
    std::fputs("usage: mplex_flow_test corpus.jsonl\n", stderr);
    return 2;
  }
  std::ifstream in(argv[1]);
  std::string line;
  int specs = 0, flows = 0, accepted = 0, failures = 0;
  while (std::getline(in, line)) {
    if (line.empty()) continue;
    Json item = Json::loads(line);
    const std::string kind = item.get("kind")->as_string();
    if (kind == "spec") {
      ++specs;
      auto [app, err] = validate_flow_spec(*item.get("spec"));
      const Json& want_app = *item.get("app");
      const std::string& want_err = item.get("err")->as_string();
      std::string got_app = app.is_null() ? "null" : app.dumps();
      if (!app.is_null()) ++accepted;
      if (got_app != want_app.dumps() || err != want_err) {
        ++failures;
        if (failures < 15)
          std::cerr << "SPEC MISMATCH\n spec: " << item.get("spec")->dumps() << "\n want: " << want_err << " | "
                    << want_app.dumps() << "\n got:  " << err << " | " << got_app << "\n";
      }
    } else if (kind == "flow") {
      ++flows;
      std::vector<Json> replies = item.get("replies")->as_array();
      Script script(replies);
      PlexerFlow flow(*item.get("app"));
      Json outcome = flow.run(script);
      if (outcome.dumps() != item.get("outcome")->dumps()) {
        ++failures;
        if (failures < 15)
          std::cerr << "FLOW MISMATCH\n app: " << item.get("app")->dumps() << "\n want: " << item.get("outcome")->dumps()
                    << "\n got:  " << outcome.dumps() << "\n";
      }
    }
  }
  std::printf("spec cases %d (accepted %d), flow cases %d, failures %d\n", specs, accepted, flows, failures);
  return failures ? 1 : 0;
}