[EvanK]

While #37 has fixed most url-related issues, it has some lingering side-effects:

1. Requests to production.$DOMAIN will have all links pointing to $DOMAIN (or www.$DOMAIN, whichever is the canonical)
2. Requests to $BRANCH.staging.$DOMAIN will have all links pointing to staging.$DOMAIN

There doesn't seem to be a good solution to this, without filtering all content through output buffering.

We could implement this so that output buffering is only turned on when WP_HOME and $_SERVER[SERVER_NAME] aren't equal, but that's still a hack I'd like to avoid..

[ericrasch]

I'm seeing this [staging] issue on a current project. Give me anything you want to test and I can put it on here to see if it works.

[EvanK]

This seems noteworthy...

On the staging environment, PHP's SERVER_NAME will match the domain on which it was requested, either staging.example.com or somebranchname.staging.example.com.

On the production environment, the SERVER_NAME will always match the canonical server name from apache's vhost (example.com or www.example.com depending on what was configured by the generator), meaning we'd need to check HTTP_HOST for this comparison.

The reason this is worth noting is that SERVER_NAME is significantly more secure as it's provided by the server, where HTTP_HOST is taken directly from the Host: header provided by the client.

[EvanK]

@ericrasch Implemented a workaround using output buffering (mostly copied from Genesis) that only kicks in when the constructed hostname doesn't match HTTP_HOST.

[ericrasch]

@EvanK you fixed this by modifying the following lines:

        $old_url = '://' . self::getHostname();
        $new_url = '://' . htmlspecialchars($_SERVER['HTTP_HOST']);

[EvanK]

I still feel dirty using output buffering + str_replace, but I genuinely don't see a better solution.

The upside, as previously stated, is that it only kicks in when the detected hostname (from Evolution::getHostname) doesn't match HTTP_HOST.

I've also confirmed that varnish uses the same http host during hashing, so we shouldn't see any overlapping caching

[EvanK]

Since I'm doing a minor version bump anyway (for #59), merging this one too