rose-ash/tests/playwright/hs-behavioral.spec.js

// @ts-check
/**
 * Hyperscript behavioral tests — SX tests in Playwright sandbox.
 *
 * Tests are registered during file load (deferred), then each is run
 * individually via page.evaluate with a 3s Promise.race timeout.
 * Hanging tests fail with TIMEOUT. After a timeout, the page is
 * closed and a fresh one is created to avoid cascading hangs.
 */
const { test, expect } = require('playwright/test');
const fs = require('fs');
const path = require('path');

const PROJECT_ROOT = path.resolve(__dirname, '../..');
const WASM_DIR = path.join(PROJECT_ROOT, 'shared/static/wasm');
const SX_DIR = path.join(WASM_DIR, 'sx');

const WEB_MODULES = [
  'render', 'core-signals', 'signals', 'deps', 'router',
  'page-helpers', 'freeze', 'dom', 'browser',
  'adapter-html', 'adapter-sx', 'adapter-dom',
  'boot-helpers', 'hypersx', 'engine', 'orchestration', 'boot',
];
const HS_MODULES = [
  'hs-tokenizer', 'hs-parser', 'hs-compiler', 'hs-runtime', 'hs-integration',
];

// Cache module sources — avoid re-reading files on reboot
const MODULE_CACHE = {};
function getModuleSrc(mod) {
  if (MODULE_CACHE[mod]) return MODULE_CACHE[mod];
  const sxPath = path.join(SX_DIR, mod + '.sx');
  const libPath = path.join(PROJECT_ROOT, 'lib/hyperscript', mod.replace(/^hs-/, '') + '.sx');
  try {
    MODULE_CACHE[mod] = fs.existsSync(sxPath) ? fs.readFileSync(sxPath, 'utf8') : fs.readFileSync(libPath, 'utf8');
  } catch(e) { MODULE_CACHE[mod] = null; }
  return MODULE_CACHE[mod];
}

// Cache test file sources
const TEST_FILES = ['spec/harness.sx', 'spec/tests/test-framework.sx', 'spec/tests/test-hyperscript-behavioral.sx'];
const TEST_FILE_CACHE = {};
for (const f of TEST_FILES) {
  TEST_FILE_CACHE[f] = fs.readFileSync(path.join(PROJECT_ROOT, f), 'utf8');
}

async function bootSandbox(page) {
  await page.goto('about:blank');
  const kernelSrc = fs.readFileSync(path.join(WASM_DIR, 'sx_browser.bc.js'), 'utf8');
  await page.addScriptTag({ content: kernelSrc });
  await page.waitForFunction('!!window.SxKernel', { timeout: 10000 });

  await page.evaluate(() => {
    const K = window.SxKernel;
    K.registerNative('host-global', a => { const n=a[0]; return (n in globalThis)?globalThis[n]:null; });
    K.registerNative('host-get', a => { if(a[0]==null)return null; const v=a[0][a[1]]; return v===undefined?null:v; });
    K.registerNative('host-set!', a => { if(a[0]!=null)a[0][a[1]]=a[2]; return a[2]; });
    K.registerNative('host-call', a => {
      const[o,m,...r]=a;
      if(o==null){const f=globalThis[m];return typeof f==='function'?f.apply(null,r):null;}
      if(typeof o[m]!=='function')return null;
      try{const v=o[m].apply(o,r);return v===undefined?null:v;}catch(e){return null;}
    });
    K.registerNative('host-new', a => {
      const C=typeof a[0]==='string'?globalThis[a[0]]:a[0];
      return typeof C==='function'?new C(...a.slice(1)):null;
    });
    K.registerNative('host-callback', a => {
      const fn=a[0];
      if(typeof fn==='function'&&fn.__sx_handle===undefined)return fn;
      if(fn&&fn.__sx_handle!==undefined){
        return function(){const r=K.callFn(fn,Array.from(arguments));if(window._driveAsync)window._driveAsync(r);return r;};
      }
      return function(){};
    });
    K.registerNative('host-typeof', a => {
      const o=a[0]; if(o==null)return'nil';
      if(o instanceof Element)return'element'; if(o instanceof Text)return'text';
      if(o instanceof DocumentFragment)return'fragment'; if(o instanceof Document)return'document';
      if(o instanceof Event)return'event'; if(o instanceof Promise)return'promise';
      return typeof o;
    });
    K.registerNative('host-await', a => {
      const[p,cb]=a;if(p&&typeof p.then==='function'){const f=(cb&&cb.__sx_handle!==undefined)?v=>K.callFn(cb,[v]):()=>{};p.then(f);}
    });
    K.registerNative('load-library!', () => false);
    window._driveAsync = function driveAsync(result) {
      if(!result||!result.suspended)return;
      const req=result.request;const items=req&&(req.items||req);
      const op=items&&items[0];const opName=typeof op==='string'?op:(op&&op.name)||String(op);
      const arg=items&&items[1];
      function doResume(val,delay){setTimeout(()=>{try{const r=result.resume(val);driveAsync(r);}catch(e){}},delay);}
      if(opName==='io-sleep'||opName==='wait')doResume(null,Math.min(typeof arg==='number'?arg:0,10));
      else if(opName==='io-fetch')doResume({ok:true,text:''},1);
    };
    K.eval('(define SX_VERSION "hs-test-1.0")');
    K.eval('(define SX_ENGINE "ocaml-vm-sandbox")');
    K.eval('(define parse sx-parse)');
    K.eval('(define serialize sx-serialize)');
  });

  const loadErrors = [];
  await page.evaluate(() => { if (window.SxKernel.beginModuleLoad) window.SxKernel.beginModuleLoad(); });
  for (const mod of [...WEB_MODULES, ...HS_MODULES]) {
    const src = getModuleSrc(mod);
    if (!src) { loadErrors.push(mod); continue; }
    const err = await page.evaluate(s => {
      try { window.SxKernel.load(s); return null; } catch(e) { return e.message; }
    }, src);
    if (err) loadErrors.push(mod + ': ' + err);
  }
  await page.evaluate(() => { if (window.SxKernel.endModuleLoad) window.SxKernel.endModuleLoad(); });

  // Deferred test registration + helpers
  await page.evaluate(() => {
    const K = window.SxKernel;
    K.eval('(define _test-registry (list))');
    K.eval('(define _test-suite "")');
    K.eval('(define push-suite (fn (name) (set! _test-suite name)))');
    K.eval('(define pop-suite (fn () (set! _test-suite "")))');
    K.eval(`(define try-call (fn (thunk)
      (set! _test-registry (append _test-registry (list {:suite _test-suite :thunk thunk})))
      {:ok true}))`);
    K.eval(`(define report-pass (fn (name)
      (let ((i (- (len _test-registry) 1)))
        (when (>= i 0) (dict-set! (nth _test-registry i) "name" name)))))`);
    K.eval(`(define report-fail (fn (name error)
      (let ((i (- (len _test-registry) 1)))
        (when (>= i 0) (dict-set! (nth _test-registry i) "name" name)))))`);
    // eval-hs: compile and evaluate a hyperscript expression/command, return its value.
    // If src contains 'return', use as-is. If it starts with a command keyword (set/put/get),
    // use as-is (the last expression is the result). Otherwise wrap in 'return'.
    K.eval(`(define eval-hs (fn (src)
      (let ((has-cmd (or (string-contains? src "return ")
                         (string-contains? src "then ")
                         (= "set " (slice src 0 4))
                         (= "put " (slice src 0 4))
                         (= "get " (slice src 0 4)))))
        (let ((wrapped (if has-cmd src (str "return " src))))
          (let ((sx (hs-to-sx-from-source wrapped)))
            (eval-expr sx))))))`);
  });

  for (const f of TEST_FILES) {
    const err = await page.evaluate(s => {
      try { window.SxKernel.load(s); return null; } catch(e) { return e.message; }
    }, TEST_FILE_CACHE[f]);
    if (err) loadErrors.push(f + ': ' + err);
  }
  return loadErrors;
}

// ===========================================================================
test.describe('Hyperscript behavioral tests', () => {
  test.describe.configure({ timeout: 600000 });

  test('upstream conformance', async ({ browser }) => {
    let page = await browser.newPage();
    let loadErrors = await bootSandbox(page);
    expect(loadErrors).toEqual([]);

    // Get test list
    const testList = await page.evaluate(() => {
      const K = window.SxKernel;
      const count = K.eval('(len _test-registry)');
      const tests = [];
      for (let i = 0; i < count; i++) {
        tests.push({
          s: K.eval(`(get (nth _test-registry ${i}) "suite")`) || '',
          n: K.eval(`(get (nth _test-registry ${i}) "name")`) || `test-${i}`,
        });
      }
      return tests;
    });

    // Run each test individually with timeout
    const results = [];
    let consecutiveTimeouts = 0;

    for (let i = 0; i < testList.length; i++) {
      const t = testList[i];

      // If page is dead (after timeout), reboot
      if (consecutiveTimeouts > 0) {
        // After a timeout, the page.evaluate from Promise.race is orphaned.
        // We must close + reopen to get a clean page.
        try { await page.close(); } catch(_) {}
        page = await browser.newPage();
        loadErrors = await bootSandbox(page);
        if (loadErrors.length > 0) {
          for (let j = i; j < testList.length; j++)
            results.push({ s: testList[j].s, n: testList[j].n, p: false, e: 'reboot failed' });
          break;
        }
        consecutiveTimeouts = 0;
      }

      let result;
      try {
        result = await Promise.race([
          page.evaluate(async (idx) => {
            const K = window.SxKernel;
            const newBody = document.createElement('body');
            document.documentElement.replaceChild(newBody, document.body);

            const thunk = K.eval(`(get (nth _test-registry ${idx}) "thunk")`);
            if (!thunk) return { p: false, e: 'no thunk' };

            let lastErr = null;
            const orig = console.error;
            console.error = function() {
              const m = Array.from(arguments).join(' ');
              if (m.startsWith('[sx]')) lastErr = m;
              orig.apply(console, arguments);
            };

            // Drive async suspension chains (wait, fetch, etc.)
            let pending = 0;
            const oldDrive = window._driveAsync;
            window._driveAsync = function driveAsync(result) {
              if (!result || !result.suspended) return;
              pending++;
              const req = result.request;
              const items = req && (req.items || req);
              const op = items && items[0];
              const opName = typeof op === 'string' ? op : (op && op.name) || String(op);
              const arg = items && items[1];
              function doResume(val, delay) {
                setTimeout(() => {
                  try { const r = result.resume(val); pending--; driveAsync(r); }
                  catch(e) { pending--; }
                }, delay);
              }
              if (opName === 'io-sleep' || opName === 'wait') doResume(null, Math.min(typeof arg === 'number' ? arg : 0, 10));
              else if (opName === 'io-fetch') doResume({ok: true, text: ''}, 1);
              else if (opName === 'io-settle') doResume(null, 5);
              else if (opName === 'io-wait-event') doResume(null, 5);
              else pending--;
            };

            try {
              const r = K.callFn(thunk, []);
              // If thunk itself suspended, drive it
              if (r && r.suspended) window._driveAsync(r);
              // Wait for all pending async chains to settle
              if (pending > 0) {
                await new Promise(resolve => {
                  let waited = 0;
                  const check = () => {
                    if (pending <= 0 || waited > 2000) resolve();
                    else { waited += 10; setTimeout(check, 10); }
                  };
                  setTimeout(check, 10);
                });
              }
              console.error = orig;
              window._driveAsync = oldDrive;
              return lastErr ? { p: false, e: lastErr.replace(/[\\"]/g, ' ').slice(0, 150) } : { p: true, e: null };
            } catch(e) {
              console.error = orig;
              window._driveAsync = oldDrive;
              return { p: false, e: (e.message || '').replace(/[\\"]/g, ' ').slice(0, 150) };
            }
          }, i),
          new Promise(resolve => setTimeout(() => resolve({ p: false, e: 'TIMEOUT' }), 3000))
        ]);
      } catch(e) {
        result = { p: false, e: 'CRASH: ' + (e.message || '').slice(0, 80) };
      }

      if (result.e === 'TIMEOUT' || (result.e && result.e.startsWith('CRASH'))) {
        consecutiveTimeouts++;
      }

      results.push({ s: t.s, n: t.n, p: result.p, e: result.e });
    }

    try { await page.close(); } catch(_) {}

    // Tally
    let passed = 0, failed = 0;
    const cats = {};
    const errTypes = {};
    for (const r of results) {
      if (r.p) passed++; else {
        failed++;
        const e = r.e || '';
        let t = 'other';
        if (e === 'TIMEOUT') t = 'timeout';
        else if (e.includes('NOT IMPLEMENTED')) t = 'stub';
        else if (e.includes('callFn')) t = 'crash';
        else if (e.includes('Assertion')) t = 'assert-fail';
        else if (e.includes('Unhandled')) t = 'unhandled';
        else if (e.includes('Expected')) t = 'wrong-value';
        else if (e.includes('Cannot read')) t = 'null-ref';
        else if (e.includes('Undefined')) t = 'undef-sym';
        else if (e.includes('no thunk')) t = 'no-thunk';
        else if (e.includes('reboot')) t = 'reboot-fail';
        if (!errTypes[t]) errTypes[t] = 0;
        errTypes[t]++;
      }
      if (!cats[r.s]) cats[r.s] = { p: 0, f: 0 };
      if (r.p) cats[r.s].p++; else cats[r.s].f++;
    }
    console.log(`\n  Upstream conformance: ${passed}/${results.length} (${(100*passed/results.length).toFixed(0)}%)`);
    for (const [cat, s] of Object.entries(cats).sort((a,b) => b[1].p - a[1].p)) {
      const mark = s.f === 0 ? `✓ ${s.p}` : `${s.p}/${s.p+s.f}`;
      console.log(`    ${cat}: ${mark}`);
    }
    console.log(`  Failure types:`);
    for (const [t, n] of Object.entries(errTypes).sort((a,b) => b[1] - a[1])) {
      console.log(`    ${t}: ${n}`);
    }
    // Show ALL crash errors (deduplicated by error message)
    const uniqueErrors = {};
    for (const r of results.filter(r => !r.p)) {
      const e = (r.e || '').slice(0, 100);
      if (!uniqueErrors[e]) uniqueErrors[e] = { count: 0, example: r };
      uniqueErrors[e].count++;
    }
    console.log(`  Unique error messages (${Object.keys(uniqueErrors).length}):`);
    for (const [e, info] of Object.entries(uniqueErrors).sort((a,b) => b[1].count - a[1].count).slice(0, 25)) {
      console.log(`    [${info.count}x] ${e}`);
    }
    // Show samples of "bar" error specifically
    const barSamples = results.filter(r => !r.p && (r.e||'').includes('Expected , got')).slice(0, 15);
    if (barSamples.length > 0) {
      console.log(`  Expected-got failures (${barSamples.length}):`);
      for (const s of barSamples) console.log(`    ${s.s}/${s.n}`);
    }

    expect(results.length).toBeGreaterThanOrEqual(830);
    expect(passed).toBeGreaterThanOrEqual(300);
  });
});