BatchAPI should be able to read fatal error and exception messages

Copying description from #363911: Batch API needs to return useful error messages, since this issue didn't come up in my search results.

This issue came up while testing #363262: Missing index on url_alias table, but works for really any Batch API-related thing. When Batch API encounters an error, instead of something actually helpful such as "I'm sorry, that index already exists," it gives you the completely unhelpful:

An unrecoverable error has occurred. You can find the error message below. It is advised to copy it to the clipboard for reference.
Please continue to the error page
An HTTP error 500 occurred.
http://localhost/core/update.php?id=7&op=do

An easy way to replicate this is to rename system_update_7018 it to system_update_7999 and run it again.

This is getting masked right now because we removed the ability to select past updates from update.php. It also happens in SimpleTest from time to time. Needs to be fixed before release or the support requests will be unbearable.

This works for me. Batch API wasn't handling a PDOException and was simply returning a "An HTTP error 500 occurred" error. Applying Damien's patch resulted in a useful error message (about the PDOException) being returned to the browser.

+      // When called from javascript, simply output the error message.

Please use "JavaScript" here.

+    // Force display of error messages in update.php or if the proper error
+    // reporting level is set.

Given the complexity of the following condition, "proper" sounds like something no one should understand. Can we change this, so developers understand?

Hm. Don't we have these already?

/**
 * Log message severity -- Emergency: system is unusable.
 *
 * @see watchdog()
 * @see watchdog_severity_levels()
 */
define('WATCHDOG_EMERG', 0);

/**
 * Log message severity -- Alert: action must be taken immediately.
 *
 * @see watchdog()
 * @see watchdog_severity_levels()
 */
define('WATCHDOG_ALERT', 1);

/**
 * Log message severity -- Critical: critical conditions.
 *
 * @see watchdog()
 * @see watchdog_severity_levels()
 */
define('WATCHDOG_CRITICAL', 2);

/**
 * Log message severity -- Error: error conditions.
 *
 * @see watchdog()
 * @see watchdog_severity_levels()
 */
define('WATCHDOG_ERROR', 3);

/**
 * Log message severity -- Warning: warning conditions.
 *
 * @see watchdog()
 * @see watchdog_severity_levels()
 */
define('WATCHDOG_WARNING', 4);

/**
 * Log message severity -- Notice: normal but significant condition.
 *
 * @see watchdog()
 * @see watchdog_severity_levels()
 */
define('WATCHDOG_NOTICE', 5);

/**
 * Log message severity -- Informational: informational messages.
 *
 * @see watchdog()
 * @see watchdog_severity_levels()
 */
define('WATCHDOG_INFO', 6);

/**
 * Log message severity -- Debug: debug-level messages.
 *
 * @see watchdog()
 * @see watchdog_severity_levels()
 */
define('WATCHDOG_DEBUG', 7);

IMO, it does not make sense to introduce yet another set of severity levels. Their intention and purpose are the same, just the target (watchdog/dsm) differs.

ok, agreed. Should be reworked in a separate issue.

-  return message.replace(/\n/g, '<br />');
-};
+  return message.replace(/\n/g, '<br />');;
+}

That, I don't understand. The old code was correct, so please revert. ;)

After that fix, mark as RTBC, please.

Committed to CVS HEAD. Thanks.