Re: [PATCH] daemon: return "access denied" if a service is not allowed

Jeff King wrote:

I'm tempted to suggest this on top:

-- >8 --
Subject: [PATCH] daemon: turn on informative errors by default

Very good idea, as long as we are cautious about making sure admins
know about the change before it comes (for example using release
notes).

[...]

Git is foremost an open system, and our defaults should
reflect that.

I think this is a lousy justification. :)

Sure, certain prominent users of git (like kernel.org) would probably
not want to set --no-informative-errors.  And you and I might prefer
that _nobody_ set --no-informative-errors.  But this does not mean the
design of Git has anything to do with that, and I am afraid of the
direction it would take us in if we start pretending it does.

The git daemon is primarily a functional, secure, admin-friendly and
client-friendly program whose defaults should make admins happy where
possible.  Luckily all that is consistent with --informative-errors.
In most use cases (i.e., not weird military security kinds of things),
although --informative-errors without --base-path could have negative
security impact as Andreas has explained before, --informative-errors
with --base-path is harmless as far as I can tell.

I am tempted to propose making --base-path mandatory when there is not
at least one path_ok argument, so in the unusual case, people would
have to explicitly say they want to serve repositories rooted at /.

Just my two cents,
Jonathan