On Wed, Oct 05, 2022 at 03:32:20PM +0200, Drouvot, Bertrand wrote:
> On 10/5/22 9:24 AM, Michael Paquier wrote:
>> - test_role() -> test_conn() to be able to pass down a database name.
>> - reset_pg_hba() to control the host, db and user parts. The host
>> part does not really apply after moving the hosts checks to a more
>> secure location, so I guess that this had better be extended just for
>> the user and database, keeping host=local all the time.
>> I am planning to apply 0001 attached independently,
>
> 0001 looks good to me.
Thanks. I have applied this refactoring, leaving the host part out of
the equation as we should rely only on local connections for this
part of the test. The best fit I can think about for the checks on
the hostname patterns would be either the ssl, ldap or krb5 tests.
SSL is more widely tested than the two others.
>> reducing the
>> footprint of 0002, which is your previous patch left untouched
>> (mostly!).
>
> Thanks! I'll look at it and the comments you just made up-thread.
Cool, thanks. One thing that matters a lot IMO (that I forgot to
mention previously) is to preserve the order of the items parsed from
the configuration files.
Also, I am wondering whether we'd better have some regression tests
where a regex includes a comma and a role name itself has a comma,
actually, just to stress more the parsing of individual items in the
HBA file.
--
Michael