[Dovecot] Allowing tilde at start of mailbox names

[pod]

>>>>> "TS" == Timo Sirainen <tss at iki.fi> writes:

    TS> It's mostly there just to make sure that ~/ or ~user/ can't be
    TS> used to open mailboxes where you weren't supposed to have access
    TS> to. I didn't think anyone would really want to have mailboxes
    TS> beginning with ~. Do you really want them? (I don't think they
    TS> were possible with UW-IMAP either?)

Unfortunately yes it would be useful in my case to have them because of
historical accident.  UW-IMAP does in fact allow it but maybe only if one
is using particular build-time options such as blackBoxDir; I forget the
precise details.  Suffice to say at some point users were erroneously told
to configure clients with '~/mail' as a prefix and this in combination
with the particular UW-IMAP build options means we have users with folder
hierarchies beginning "~/mail" ('~' uninterpreted).  I don't think I can
work around this with a hidden namespace since a client subsequently
configured not to use a prefix will be used to seeing folders under
"~/mail" ('~' uninterpreted).  I'm (increasingly forlornly it seems)
attempting to migrate with no user visible change.

    >> Why is it not possible to have mailbox names that begin with '~'
    >> but without '~' carrying any special meaning?

    TS> That's what the code does if you don't have
    TS> mail_full_filesystem_access=yes, but I thought I'd prevent this in
    TS> no case just in case of a bug somewhere, possibly even outside of
    TS> Dovecot itself.

I appreciate your reasoning regarding bugs elsewhere.  I was, however,
hoping to introduce a mode of behaviour that would allow me to accept the
risk.  I'm not using virtual users; every login runs under its own uid so
I still have filesystem permissions as a reasonable fall back.

    TS> There are too many settings
    TS> already. mail_full_filesystem_access=no should be enough for this.

I completely understand why you want to keep the number of setting down.
My motivation for introducing a new setting was a vain hope that allowing
uninterpreted (*name == '~') behaviour could go into the 1.0 codebase but
without also changing the meaning of mail_full_filesystem_access=no in
existing deployments.