"Russ" == Russ Allbery <rra(_at_)stanford(_dot_)edu> writes:
Andrew Gierth <andrew(_at_)erlenstar(_dot_)demon(_dot_)co(_dot_)uk> writes:
but this proposal as it stands largely cripples the ability for
clients to do "LIST ACTIVE <wildmat>" unless _both_ the client and
the server are updated to support an extension for
punycode-wildmats. (i.e. without such an extension, the client
can't do pattern matches in ways that are meaningful in terms of
the human-readable name, only in terms of the encoded name)
Russ> I'd thought about this, but the degree of crippling didn't seem
Russ> severe enough to even mention. Per-component wildmat patterns
Russ> still work, and within-component matches of ASCII groups
Russ> continue to work, so this only affects cases where the client
Russ> is trying to use a wildmat to match a portion of a non-ASCII
Russ> name.
Russ> This is an extremely uncommon operation for a reader as I
Russ> understand it,
I have no data on whether or not this is common. I know that one of
the clients I use does it when I ask it to (even though it
theoretically has a local copy of the list available), usually when
I'm looking for some group the name of which I'm not certain of.
To gather data on this I would have to add some additional logging
into my reader servers, which isn't something I can do in the short
term.
Russ> How frequently do you see clients using wildmats other than *,
Russ> the exact name of a group, or component-wise? I thought that
Russ> most clients that supported a newsgroup search feature did so
Russ> by downloading the entire active file and then doing their
Russ> searches locally.
well, if I were writing a client I would not necessarily do it that
way, if for no other reason than that the entire active file is often
uncomfortably large.
--
Andrew.