This is more a support question and should be on the support list but see below.
Granted this is could be a support issue, but I had intended to (mostly) gauge the validity of a patch to core. Is that not a development question?
My question back to you is why isn't HTTP_HOST properly setup?
AFAICT it is functioning exactly as designed. It returns the value of the Host: header as sent by the client. Numerous experiments with phpinfo() have confirmed that. According to the apache docs (and confirmed by the above mentioned experiments) if UseCanonicalName is set to 'off' the SERVER_NAME variable is set to equal HTTP_HOST. This works great in most all cases. If UseCanonicalName is set to 'on' then SERVER_NAME is set to whatever 'ServerName' is set to in the apache conf. Seeing as UseCanonicalName can be set in the virtual host entry there is no problem doing something like this: <VirtualHost *:80> UseCanonicalName On ServerName hostname.com ServerAlias hostname.dev </VirtualHost> If Drupal used the SERVER_NAME variable then accessing the site should load the hostname.com configuration regardless of what alias is accessed. Seeing as it uses HTTP_HOST it will fail to load a configuration.
Generally, for Drupal multisites to work under different Apache configurations, Drupal needs to rely more on the client's HTTP request rather than the server configuration. You can see a related issue in http://drupal.org/node/262920
I suppose it's swings and roundabouts really. Using HTTP_HOST provides a globally predictable outcome. But if the user has access to the apache configuration, either by support request or personally, then SERVER_NAME provides a more configurable outcome. I guess I'll just stick with symlinks (they can always be excluded from the version control anyway). Thanks Adam