>> * should we worry about case-insensitive matching?
>
> hgweb seems to accept only lowercase cs ids, so the regex can be limited
> to [0-9a-f].  OTOH cs ids are lowercase everywhere so someone would have
> to intentionally convert it to uppercase to break the link.

Yes, this makes sense. It might also reduce false-positives slightly.

>> * links to repository tags
>
> I would do this only if prefixed by 'hg:'.  hgweb accepts tags already,
> so you don't have to anything special to handle this (just have a regex
> that accepts the tags).

Agreed. I was thinking that all links would be prefixed by 'hg:' with 
the exception of the 12/40 character commit IDs (which would be optional)

>> * links to repository revision numbers
>
> Revision numbers (the sequential ones) shouldn't be used IMHO.

Because we would be only linking to a specific repository branch, 
linking of revision numbers _should_ be safe, but I am happy to leave 
them out. If someone specifically requests the feature later, then we 
can look it again in the future.

>> * links to specific files/lines
>
> I find this handy, so you could include it too.

I am thinking a scheme like:
   hg:<commit ref>:path/to/file

 From a little bit of experimenting it seems as though the <commit ref> 
can be a commit ID, tag, branch, or revision ID. It would also support 
adding '#l<line no>' to the end without requiring anything extra to be done.

Other suggestions welcome though.