Ouch. This goes against rule number 1 of software: don't lose your user's data. I'm sure they're smart enough that nothing's permanently lost, but even the perception that it could happen is bad enough to seriously tarnish their brand.
It didn't seem to hurt Reddit at all when Reddit lost my account data early in its existence. Similarly, MSN did that years ago when I was a charter member, but that snafu alone didn't seem seriously to hurt MSN's reputation. It's annoying to me to lose a username, and I don't give repeat business to services that trash my account data in a way that causes me to be unable to use my preferred username, but plenty of people sign up for online services based on the recommendation of a friend (or perhaps an advertisement) without doing exhaustive research on the company's record of technical competence.
The main thing is negative press. Google is already big enough that people don't need to look them up to hear about them. No news outlets really cared if you lost your very old Reddit username, but news outlets do care if it's Google losing your username because news about Google makes them money.
> rule number 1 of software: don't lose your user's data
That doesn't scale to the scope of Gmail. The appropriate precautions to preserve old medical records differ in kind from those appropriate for old shopping lists.
I'd like my Google account to have a "Backup URL" field, that I could set to <git:tux.example.org> or <venti:glenda.example.com>. Google would undertake to push all "my" data to that URL, to pull it back as needed to recover their systems, and to publish the format it was stored in. I'd be responsible for the rest.
You can have Gmail forward all your incoming mail (or even better, just the non-spam mail). Worst case you can then import that back into Gmail. I just have a copy kept on my computer and its various backups.
Now, I can already run "webcheckout $url" and get a git repository for any of the websites I run, as well as all of the wikis and blogs of customers at my startup, Branchable.com.
That is implemented via a proto-RFC called the "rel=vcs microformat", which simply adds a little data to web pages to indicate where the underlaying data can be cloned from. <http://kitenet.net/~joey/rfc/rel-vcs/>;
