On Thu, Oct 20, 2005 at 11:52:40AM -0400, Andrew Sullivan wrote:
> On Thu, Oct 20, 2005 at 06:10:40PM +0300, Andrus wrote:
> > >From this thread I got the regular expression
>
> [snipped]
>
> Note that that regular expression, which appears to be validating
> TLDs as well, is incredibly fragile. John Klensin has actually
> written an RFC about this very problem. Among other problems, what
> do you do when a country code ceases to be? (There's a similar
> problem that the naming bodies struggke with from time to time.)
>
> I suggest that if you want to validate TLDs, you pull them off when
> you write the data in your database, and use a lookup table to make
> sure they're valid (you can keep the table up to date regularly by
> checking the official IANA registry for them). At least that way you
> don't have to change a regex every time ICANN decides to add another
> TLD.
You need to maintain the data, certainly. To argue that it must
be in a table to be maintained is, well, wrong. My preference would
be to keep it in a table and regenerate the regex periodically, and
in the application layer I do exactly that, but to try and do that
in a check constraint would be painful. A cleaner approach would
be to have a regex that checks for general syntax and extracts the
TLD, which is then compared to a lookup table, perhaps, but that
adds a lot of complexity for no real benefit.
> (The regex is wrong anyway, I think: it doesn't have .mobi,
> which has been announced although isn't taking registrations yet, and
> it doesn't appear to have arpa, either.)
While there are valid deliverable email addresses in .arpa, you really
don't want to be accepting them from end users...
Cheers,
Steve