Namespaces for disabled modules are registered during the first request after a module is disabled

This can be applied on top of #1784312-76: Stop doing so much pre-kernel bootstrapping.

About to commit the other one, let's get this in quick.

Let's try to test it then.

Ugh.

That's an unrelated test failure, HEAD is broken atm: #1842726: Transliteration component must not contain drupal_alter().

#4: 1846376.patch queued for re-testing.

#4: 1846376.patch queued for re-testing.

HEAD is working again, and #4 passed the retest, but isn't turning green for some reason, so here it is again.

#11 passed the BOT.

Thanks, looks good to me.

Committed/pushed to 8.x, thanks!

Does someone in here still remember this issue?

I am pretty sure this "fix" no longer works in current Drupal 8.
At least with the Composer ClassLoader, there is no way to "unregister" a namespace. Whatever is still happening there is bogus now...

http://drupalcode.org/project/drupal.git/blob/1f9a488b927e9548365bf4882e...

        // Revert the class loader to its prior state. However,
        // registerNamespaces() performs a merge rather than replace, so to
        // effectively remove erroneous registrations, we must replace them with
        // empty arrays.
        $namespaces_after = $this->classLoader->getPrefixes();
        $namespaces_before += array_fill_keys(array_diff(array_keys($namespaces_after), array_keys($namespaces_before)), array());
        $this->registerNamespaces($namespaces_before);

And here is what ClassLoader::add() actually does..
http://drupalcode.org/project/drupal.git/blob/1f9a488b927e9548365bf4882e...

This can only ever add to the registered namespaces, never remove any.

I am keeping this mechanic in the PSR-4 patch,
#2083547: PSR-4: Putting it all together
but I would prefer to remove it.

We can open a new issue for this, but I would like to get some feedback from the old days' participants first..
(and apologies if reopening is the wrong thing to do)

Since modules can only be uninstalled now, are the concerns in #16 still applicable?

I have reviewed the comment of donquixote, but cannot find any traces of the code he references.

Just as jhedstrom mentions, it also doesn't seem to be relevant anymore. Closing this issue, if it somehow is not ok maybe it's better to open a new one.