Greetings,
* Bruce Momjian (bruce@momjian.us) wrote:
> On Tue, Jan 7, 2020 at 11:46:31AM +0100, Laurenz Albe wrote:
> > On Fri, 2019-12-27 at 12:16 -0500, Bruce Momjian wrote:
> > > On Fri, Dec 27, 2019 at 05:44:10AM +0000, PG Doc comments form wrote:
> > > > The following documentation comment has been logged on the website:
> > > >
> > > > Page: https://www.postgresql.org/docs/12/default-roles.html
> > > > Description:
> > > >
> > > > The title is wrong. The roles are not defaults; they are predefined and
> > > > privileged. The title suggests that a user should expect to be assigned
> > > > these roles. "21.5 Sub-Administrator Roles" would be accurate--improving
> > > > clarity over all and removing any need to explain why postgres is not in
> > > > this list of roles.
> > > >
> > >
> > > Good points. I have developed the attached documentation patch which
> > > includes your ideas.
> >
> > +1
> >
> > I think that "predefined role" is better than "default role".
>
> Thanks, patch applied through 9.6.
Erm, I didn't agree with this and pointed to reasons why it was based,
for starters, on a misunderstanding and further wasn't a particularly
good idea anyway. I'm not happy that it was committed, and to have been
back-patched strikes me as even worse. What about existing links to
things like: https://www.postgresql.org/docs/9.6/default-roles.html
which will now be broken, like from here?:
https://paquier.xyz/postgresql-2/postgres-11-new-system-roles/
Or that the documentation wasn't properly updated to reflect this change
as a simple "git grep 'default role'" would have shown? There's at
least 5 references still to 'default role' in the documentation after
this commit.
Not to mention that, with this patch, we now have confusion between
things like 'DEFAULT_ROLE_WRITE_SERVER_FILES' in the code vs. the
documentation.
In short, I don't agree with this change, which strikes me as looking
largely like it's trying to make PG look more like Oracle than anything
else, but if we're going to move in this direction we should only be
doing so in master and we should be much more careful making sure that
the documentation, at least, is updated and consistent and that
appropriate comments are made to the code to explain that DEFAULT_ROLE
in the code is referring to "predefined roles" (or we should change the
code, though I can understand if there's argument that doing so would
create unnecessary back-patching hazards.. though there isn't all
*that* much code, so I could go either way on that myself).
Thanks,
Stephen