Check for existence of http(s):// resources is broken

[kasemir]

In #3246 we replaced this code for checking existence of a URL

final InputStream stream = openURL(resource_name);
stream.close();
return true;

in the ModelResourceUtil with

URL url = new URL(resource_name);
url.openConnection();
return true;

Turns out that breaks with our setup.
We still have some older *.opi files, and newer X.bob files.
An older action button might still try to open the older X.opi file, but the runtime first checks if there's already a newer X.bob version.

That functions fine with local files, but when using http(s) resources, a check for https://server/path/to/X.bob will now succeed even when the file doesn't actually exist on the web server?!
So then we try to actually open it, and fail.

With the previous openURL...close check, we would correctly detect if a file exists or not.

[klemenv]

It sounds like the url.openConnection() might not in fact verify the existence of the file when connecting through the web server: https://docs.oracle.com/javase/8/docs/api/java/net/URL.html#openConnection--

But maybe the connect() would, so adding in a third line something like below would fix the issue as the web server will reply with 4xx error which hopefully turns into a Java exception inside connect()?

URL url = new URL(resource_name);
url.openConnection();
url.connect();
return true;

[kasemir]

I tried adding 'connect':

URL resource = new URL(escaped);
// From javadoc:
// ".. URLConnection .. does not establish the actual network connection on creation.
//     This will happen only when calling URLConnection.connect()."
URLConnection connection = resource.openConnection();
connection.connect();

No progress. Still appears to succeed and then later fails when reading the non-existing document.
Also followed up with a check of the document size:

URL resource = new URL(escaped);
// From javadoc:
// ".. URLConnection .. does not establish the actual network connection on creation.
//     This will happen only when calling URLConnection.connect()."
URLConnection connection = resource.openConnection();
connection.connect();
if (connection.getContentLength() > 0)
    ... looks like we have a valid document...

Our problem might be site-specific and related to accessing the web resources via NGNX proxies. That content length is around 190 bytes, maybe some intermediate redirect information?

The only approach that works for us is the original one, opening a stream to the resource.
If the resource doesn't exist, that fails.
On success, it might look like we're then reading the resource twice, once to check, then again to really read it, but openURL uses our url_cache, so we only read once via the network.