Dimitri Fontaine wrote:
> Andrew Dunstan <andrew@dunslane.net> writes:
>> On 10/12/2010 05:02 PM, Dimitri Fontaine wrote:
>>> So, what you do is have a file per GUC, file name is the GUC name, first
>>> line contains *ONLY* current value, the rest of the file is comments.
>> You're joking, right?
>
> No. I just want both comments and SQL commands. If you refuse this
> simple file scheme, keep your postgresql.conf and don't remote edit it.
>
> That's my proposal, I'm happy that it comes with laughter :)
Maybe I missed something important, but why is it not possible to retain the
single existing postgres.conf file format (human writable) *and* have it
machine/SQL-editable *and* maintain the comments? I should think that it would
be possible to do all of these without too much trouble. All you would need is
for the file parser to retain the comments as metadata, include them in the
relations that the SQL commands see where the latter can also edit them as data,
and then write out the updated file with comments. The fact that Postgres
already explicitly supports comment metadata in its system catalog means it must
already know something about this. If something is missing, then expand the
catalog so it represents all the details you want to preserve. -- Darren Duncan