On Tue, Dec 17, 2019 at 09:43:23AM -0800, Junio C Hamano wrote:

Emily Shaffer [off-list ref] writes:

quoted

quoted

But that is too simple-minded.  If we wanted to safelist foo.*.bar,
where '*' can be anything, walking on the list of safe variables
would not work.  We must have a hash table that knows "foo.*.bar" is
allowed, and while walking all the configuration keys, when we see
foo.a.bar, we consult "foo.*.bar" as well as "foo.a.bar" to see if
it is whitelisted, or something like that.

...unless we want to use wildcards like you suggest.

But I'm not sure it's a good idea. I envision someone writing another
Git add-on, which offers someone to specify "user.password" ...

Wildcarding the leaf level of two (or for that matter three) level
names like "user.*" in your example is of course a nonsense way to
use the safelist.  But think about three-level names where the
second level is designed to be used for user-supplied token to give
things of similar nature a name the user can use to distinguish
things.  If remote.origin.url is worth reporting, wouldn't
remote.upstream.url also be?  Shouldn't remote.*.url be the way to
say "the URL field for each and every remote is a safe thing to
report" in the safelist?

Bah, somehow this mail slipped through the cracks for me.

When you put it that way, I see how it can be useful. Although I admit
to fatigue on this patchset and now I'm hesitant about feature creep. :)

I'm going to send the reroll for v5 as I have it, not having seen this
mail, and think more about how I'd approach wildcarding in the middle
like this. I want to take a look at the other configs which use a middle
level as a user token and try to reason about whether we ought to be
collecting them. I'll hope to report back sooner (or later).

 - Emily