更新

2026-07-22 10:18:59 +08:00
parent 2530ddada6
commit 0fb03d0bca
618 changed files with 19445 additions and 3 deletions
@@ -0,0 +1,175 @@
+<?php
+
+declare(strict_types=1);
+
+require dirname(__DIR__) . '/vendor/autoload.php';
+
+$rounds = max(1, (int) ($_SERVER['argv'][1] ?? 1000));
+$suite = strtolower(trim((string) ($_SERVER['argv'][2] ?? 'all')));
+
+$allowedSuites = ['all', 'agent', 'image', 'llm'];
+if (!in_array($suite, $allowedSuites, true)) {
+    fwrite(STDERR, 'Usage: php backend/tests/stability_1000_round.php [rounds] [all|agent|image|llm]' . PHP_EOL);
+    exit(2);
+}
+
+$scripts = [
+    'agent' => [
+        'rounds_per_run' => 100,
+        'command' => 'php tests/agent_100_round_regression.php',
+    ],
+    'llm' => [
+        'rounds_per_run' => 100,
+        'command' => 'php tests/llm_100_round_regression.php',
+    ],
+];
+
+$root = dirname(__DIR__);
+$exitCode = 0;
+
+$runCommand = static function (string $command, string $workingDir): array {
+    $descriptorSpec = [
+        0 => ['pipe', 'r'],
+        1 => ['pipe', 'w'],
+        2 => ['redirect', 1],
+    ];
+
+    $process = proc_open($command, $descriptorSpec, $pipes, $workingDir);
+    if (!is_resource($process)) {
+        throw new RuntimeException('Unable to start command: ' . $command);
+    }
+
+    fclose($pipes[0]);
+    $stdout = stream_get_contents($pipes[1]);
+    fclose($pipes[1]);
+    $status = proc_close($process);
+
+    return [
+        'stdout' => $stdout === false ? '' : $stdout,
+        'status' => $status,
+    ];
+};
+
+$runFixedSuite = static function (string $name, int $targetRounds, array $config) use ($root, $runCommand): int {
+    $completed = 0;
+    $runIndex = 0;
+
+    while ($completed < $targetRounds) {
+        $runIndex++;
+        $result = $runCommand($config['command'], $root);
+        $output = trim($result['stdout']);
+        if ($output !== '') {
+            echo $output . PHP_EOL;
+        }
+
+        if ((int) $result['status'] !== 0) {
+            throw new RuntimeException(sprintf(
+                '%s suite failed on batch %d after %d rounds',
+                strtoupper($name),
+                $runIndex,
+                $completed
+            ));
+        }
+
+        $completed += (int) $config['rounds_per_run'];
+        printf(
+            "%s PROGRESS %d/%d rounds\n",
+            strtoupper($name),
+            min($completed, $targetRounds),
+            $targetRounds
+        );
+    }
+
+    return $completed;
+};
+
+$runImageSuite = static function (int $targetRounds) use ($root, $runCommand): int {
+    $bestRegressionCount = null;
+    $bestIntegrationCount = null;
+
+    for ($regressionCount = intdiv($targetRounds, 79); $regressionCount >= 0; $regressionCount--) {
+        $remaining = $targetRounds - ($regressionCount * 79);
+        if ($remaining < 0 || $remaining % 2 !== 0) {
+            continue;
+        }
+        $bestRegressionCount = $regressionCount;
+        $bestIntegrationCount = intdiv($remaining, 2);
+        break;
+    }
+
+    if ($bestRegressionCount === null || $bestIntegrationCount === null) {
+        throw new RuntimeException('Unable to build an image-suite plan for ' . $targetRounds . ' rounds');
+    }
+
+    $completed = 0;
+    for ($index = 1; $index <= $bestRegressionCount; $index++) {
+        $result = $runCommand('php tests/comfy_image_edit_regression.php', $root);
+        $output = trim($result['stdout']);
+        if ($output !== '') {
+            echo $output . PHP_EOL;
+        }
+        if ((int) $result['status'] !== 0) {
+            throw new RuntimeException(sprintf(
+                'IMAGE regression batch %d failed after %d rounds',
+                $index,
+                $completed
+            ));
+        }
+        $completed += 79;
+        printf("IMAGE PROGRESS %d/%d rounds\n", min($completed, $targetRounds), $targetRounds);
+    }
+
+    for ($index = 1; $index <= $bestIntegrationCount; $index++) {
+        $result = $runCommand('php tests/comfy_image_edit_integration.php', $root);
+        $output = trim($result['stdout']);
+        if ($output !== '') {
+            echo $output . PHP_EOL;
+        }
+        if ((int) $result['status'] !== 0) {
+            throw new RuntimeException(sprintf(
+                'IMAGE integration batch %d failed after %d rounds',
+                $index,
+                $completed
+            ));
+        }
+        $completed += 2;
+        printf("IMAGE PROGRESS %d/%d rounds\n", min($completed, $targetRounds), $targetRounds);
+    }
+
+    return $completed;
+};
+
+$targets = $suite === 'all' ? ['agent', 'image', 'llm'] : [$suite];
+$summary = [];
+
+foreach ($targets as $target) {
+    try {
+        if ($target === 'image') {
+            $completed = $runImageSuite($rounds);
+        } else {
+            $completed = $runFixedSuite($target, $rounds, $scripts[$target]);
+        }
+        $summary[$target] = ['completed' => $completed, 'failed' => false];
+    } catch (Throwable $exception) {
+        $summary[$target] = [
+            'completed' => $summary[$target]['completed'] ?? 0,
+            'failed' => true,
+            'message' => $exception->getMessage(),
+        ];
+        fwrite(STDERR, strtoupper($target) . ' FAILURE: ' . $exception->getMessage() . PHP_EOL);
+        $exitCode = 1;
+        break;
+    }
+}
+
+foreach ($summary as $name => $item) {
+    printf(
+        "%s RESULT completed=%d failed=%s%s\n",
+        strtoupper($name),
+        (int) ($item['completed'] ?? 0),
+        ($item['failed'] ?? false) ? 'true' : 'false',
+        !empty($item['message']) ? ' message=' . $item['message'] : ''
+    );
+}
+
+exit($exitCode);