On 9/6/16 1:08 PM, Tom Lane wrote:
>> As just mentioned elsewhere, this accidentally introduces a failure if
>> > the PostgreSQL installation path contains LF/CR, because of the use of
>> > appendShellString().
> I think that's intentional, not accidental. What actual use case is
> there for allowing such paths?
There probably isn't one. But we ought to be introducing this change in
a more intentional and consistent way.
For example, pg_basebackup has no such restriction. So using
pg_basebackup, then promote, then pg_upgrade will (probably) fail now
for some paths.
More generally, I'm concerned that appendShellString() looks pretty
attractive for future use. It's not inconceivable that someone will
want to use it for say calling pg_dump from pg_dumpall or pg_upgrade at
some point, and then maybe we'll accidentally disallow LF/CR in
tablespace names, say.
Also, if we're concerned about the potential for confusion that these
characters can cause, maybe we should be disallowing more control
characters in similar places.
--
Peter Eisentraut http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services